Cdn1.sph.harvard.edu



November 12, 2019DRAFT AVAILABLE FOR PUBLIC COMMENTPlease send your comments and suggestions to Dr. Eleanor Murray (ejmurray@bu.edu)Guidelines for estimating causal effects in pragmatic randomized trialsEleanor J. Murray1, Sonja A. Swanson2, Jessica Young3, Miguel A. Hernán1,4,51Department of Epidemiology, Harvard T.H. Chan School of Public Health, Boston, MA, USA2Department of Epidemiology, Erasmus Medical Center, Rotterdam, The Netherlands3Harvard Medical School & Harvard Pilgrim Health Care Institute, Boston, MA, USA4Department of Biostatistics, Harvard T.H. Chan School of Public Health, Boston, MA, USA5Harvard-MIT Division of Health Sciences and Technology, Cambridge, USA. Corresponding author: Eleanor MurrayDepartment of EpidemiologyBoston University School of Public Health322 Talbot EastBoston, MA, USA, 02118Email: ejmurray@bu.eduFunding: This study was supported through a Patient Centered Outcomes Research Institute (PCORI) award (ME-1503-8119). All statements in this report, including its findings and conclusions, are solely those of the authors and do not necessarily represent the views of the Patient-Centered Outcomes Research Institute (PCORI), its Board of Governors or Methodology Committee. The funder approved the protocol but had no role in the conduct, analysis, or reporting of study findings.Word count: 9882AbstractPragmatic randomized trials are designed to provide evidence for clinical decision-making rather than regulatory approval. Common features of these trials include the inclusion of heterogeneous or diverse patient populations in a wide range of care settings, the use of active treatment strategies as comparators, unblinded treatment assignment, and the study of long-term, clinically relevant outcomes. These features can greatly increase the usefulness of the trial results for patients, clinicians, and other stakeholders. However, these features also introduce an increased risk of non-adherence, which reduces the value of the intention-to-treat effect as a patient-centered measure of causal effect. In these settings, the per-protocol effect provides useful complementary information for decision making. Unfortunately, there is little guidance for valid estimation of the per-protocol effect. Here, we present our full guidelines for analyses of pragmatic trials that will result in more informative causal inferences for both the intention-to-treat effect and the per-protocol effect. 1. IntroductionPragmatic randomized trials are key tools for research on the comparative effectiveness of medical interventions. Like other randomized trials, pragmatic trials are designed to compare strategies for the prevention, diagnosis, and treatment of diseases. The CONSORT extension for pragmatic trials defines these trials as “designed to measure effectiveness; that is whether an intervention works when used in usual conditions of care” ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"zLJ92QBp","properties":{"formattedCitation":"(1,2)","plainCitation":"(1,2)","noteIndex":0},"citationItems":[{"id":2963,"uris":[""],"uri":[""],"itemData":{"id":2963,"type":"article-journal","title":"Improving the reporting of pragmatic trials: an extension of the CONSORT statement","container-title":"BMJ","page":"a2390","volume":"337","source":"","abstract":"Background The CONSORT statement is intended to improve reporting of randomised controlled trials and focuses on minimising the risk of bias (internal validity). The applicability of a trial’s results (generalisability or external validity) is also important, particularly for pragmatic trials. A pragmatic trial (a term first used in 1967 by Schwartz and Lellouch) can be broadly defined as a randomised controlled trial whose purpose is to inform decisions about practice. This extension of the CONSORT statement is intended to improve the reporting of such trials and focuses on applicability.\nMethods At two, two-day meetings held in Toronto in 2005 and 2008, we reviewed the CONSORT statement and its extensions, the literature on pragmatic trials and applicability, and our experiences in conducting pragmatic trials.\nRecommendations We recommend extending eight CONSORT checklist items for reporting of pragmatic trials: the background, participants, interventions, outcomes, sample size, blinding, participant flow, and generalisability of the findings. These extensions are presented, along with illustrative examples of reporting, and an explanation of each extension. Adherence to these reporting criteria will make it easier for decision makers to judge how applicable the results of randomised controlled trials are to their own conditions. Empirical studies are needed to ascertain the usefulness and comprehensiveness of these CONSORT checklist item extensions. In the meantime we recommend that those who support, conduct, and report pragmatic trials should use this extension of the CONSORT statement to facilitate the use of trial results in decisions about health care.","DOI":"10.1136/bmj.a2390","ISSN":"0959-8138, 1468-5833","note":"PMID: 19001484","title-short":"Improving the reporting of pragmatic trials","journalAbbreviation":"BMJ","language":"en","author":[{"family":"Zwarenstein","given":"Merrick"},{"family":"Treweek","given":"Shaun"},{"family":"Gagnier","given":"Joel J."},{"family":"Altman","given":"Douglas G."},{"family":"Tunis","given":"Sean"},{"family":"Haynes","given":"Brian"},{"family":"Oxman","given":"Andrew D."},{"family":"Moher","given":"David"}],"issued":{"date-parts":[["2008",11,11]]}}},{"id":2967,"uris":[""],"uri":[""],"itemData":{"id":2967,"type":"webpage","title":"Consort - Pragmatic Trials","URL":"","accessed":{"date-parts":[["2019",10,31]]}}}],"schema":""} (1,2) Unlike other randomized trials which are designed to seek regulatory approval, pragmatic trials are specifically designed to address real-world questions about options for care and therefore to guide decisions by patients, clinicians and other stakeholders. Therefore, characteristics of a pragmatic trial include heterogeneous or diverse patients and care settings, clinically relevant comparators (e.g., usual care rather than placebo), unconcealed assignment to treatment, and follow-up time long enough to study long-term clinical outcomes without having to rely on surrogates ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"iSgEZBf1","properties":{"formattedCitation":"(3,4)","plainCitation":"(3,4)","noteIndex":0},"citationItems":[{"id":942,"uris":[""],"uri":[""],"itemData":{"id":942,"type":"article-journal","title":"The PRECIS-2 tool: designing trials that are fit for purpose","container-title":"BMJ : British Medical Journal","volume":"350","URL":"","DOI":"10.1136/bmj.h2147","title-short":"The PRECIS-2 tool: designing trials that are fit for purpose","author":[{"family":"Loudon","given":"Kirsty"},{"family":"Treweek","given":"Shaun"},{"family":"Sullivan","given":"Frank"},{"family":"Donnan","given":"Peter"},{"family":"Thorpe","given":"Kevin E"},{"family":"Zwarenstein","given":"Merrick"}],"issued":{"date-parts":[["2015"]]}},"label":"page"},{"id":2682,"uris":[""],"uri":[""],"itemData":{"id":2682,"type":"article-journal","title":"Pragmatic randomized clinical trials: best practices and statistical guidance","container-title":"Health Services and Outcomes Research Methodology","page":"23-35","volume":"19","issue":"1","source":" (Crossref)","abstract":"Randomized clinical trials often serve the purpose of assessing the efficacy and safety of a compound. By combining real-world evidence and randomization, pragmatic randomized clinical trials (PrCTs) can be used to inform treatment effectiveness and healthcare decisions. PrCTs, referring to studies where several pragmatic elements are used (eligibility, endpoints, follow-up, etc.), pose unique challenges (Loudon et al. in BMJ 350:h2147, 2015). From a literature review, we propose a definition of PrCT and discuss strategies to overcome some PrCT challenges. Use of alternative data collection approaches may lead to uncertainties, and absence of blinding could potentially lead to non-random missing data at study endpoints such that randomization is no longer protected by an intent to treat. Therefore, more complex randomization strategies may be needed to minimize bias. Additional data sources could be used to synthesize information and create a more accurate endpoint definition, which may require tools such as natural language processing. The statistician must become familiar with the challenges and strengths of PrCTs, ranging from design to analysis to interpretation, in order to transform data into evidence (Califf in Clin Trials 13:471–477, 2016).","DOI":"10.1007/s10742-018-0192-5","ISSN":"1387-3741, 1572-9400","title-short":"Pragmatic randomized clinical trials","journalAbbreviation":"Health Serv Outcomes Res Method","language":"en","author":[{"family":"Gamerman","given":"Victoria"},{"family":"Cai","given":"Tianxi"},{"family":"Els??er","given":"Amelie"}],"issued":{"date-parts":[["2019",3]]}}}],"schema":""} (3,4).While pragmatic trials are useful to guide decision making, they are also especially vulnerable to post-randomization confounding from incomplete adherence and post-randomization selection bias from loss to follow-up ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"NaNoyUFS","properties":{"formattedCitation":"(5)","plainCitation":"(5)","noteIndex":0},"citationItems":[{"id":1153,"uris":[""],"uri":[""],"itemData":{"id":1153,"type":"article-journal","title":"Randomized trials analyzed as observational studies","container-title":"Annals of Internal Medicine","page":"560-2","volume":"159","issue":"8","source":"Nlm","archive_location":"24018844","DOI":"10.7326/0003-4819-159-8-201310150-00709","ISSN":"1539-3704 (Electronic) 0003-4819 (Linking)","title-short":"Randomized trials analyzed as observational studies","journalAbbreviation":"Ann Intern Med","language":"eng","author":[{"family":"Hernán","given":"M. A."},{"family":"Hernandez-Diaz","given":"S."},{"family":"Robins","given":"J. M."}],"issued":{"date-parts":[["2013",10,15]]}}}],"schema":""} (5). That is, pragmatic trials are especially subject to many of the biases that we have learned to associate exclusively with observational follow-up studies. Yet no standardized methodology for causal inference from pragmatic trials has been proposed, with most investigators exclusively relying on the intention-to-treat principle.The emphasis on the intention-to-treat principle is the historical consequence of many early randomized trials being short, small, double-blinded, tightly controlled experiments among highly selected patients who largely adhered to the assigned treatment and who were rarely lost to follow-up. Because these experiments are designed to minimize post-randomization confounding and selection bias, intention-to-treat analyses provide reasonable measures of the treatment effect. However, the almost exclusive reliance on the intention-to-treat principle in pragmatic trials is worrisome because, under real-world conditions, the effect estimates may be profoundly impacted by non-adherence and loss to follow-up. As a result, pragmatic trials exclusively analyzed under the intention-to-treat principle may not be quantifying the causal effect of primary interest for patients, clinicians, researchers, and other stakeholders ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"iErepyYd","properties":{"formattedCitation":"(6)","plainCitation":"(6)","noteIndex":0},"citationItems":[{"id":1156,"uris":[""],"uri":[""],"itemData":{"id":1156,"type":"article-journal","title":"Patients and investigators prefer measures of absolute risk in subgroups for pragmatic randomized trials","container-title":"Journal of Clinical Epidemiology","page":"10-21","volume":"103","source":"PubMed","abstract":"OBJECTIVES: Pragmatic randomized trials are important tools for shared decision-making, but no guidance exists on patients' preferences for types of causal information. We aimed to assess preferences of patients and investigators toward causal effects in pragmatic randomized trials.\nSTUDY DESIGN AND SETTING: We (a) held three focus groups with patients (n?=?23) in Boston, MA; (b) surveyed (n?=?12) and interviewed (n?=?5) investigators with experience conducting pragmatic trials; and (c) conducted a systematic literature review of pragmatic trials (n?=?63).\nRESULTS: Patients were distrustful of new-to-market medications unless substantially more effective than existing choices, preferred stratified absolute risks, and valued adherence-adjusted analyses when they expected to adhere. Investigators wanted both intention-to-treat and per-protocol effects but felt methods for estimating per-protocol effects were lacking. When estimating per-protocol effects, many pragmatic trials used inappropriate methods to adjust for adherence and loss to follow-up.\nCONCLUSION: We made four recommendations for pragmatic trials to improve patient centeredness: (1) focus on superiority in effectiveness or safety, rather than noninferiority; (2) involve patients in specifying a priori subgroups; (3) report absolute measures of risk; and (4) complement intention-to-treat effect estimates with valid per-protocol effect estimates.","DOI":"10.1016/j.jclinepi.2018.06.009","ISSN":"1878-5921","note":"PMID: 29966732\nPMCID: PMC6175611","journalAbbreviation":"J Clin Epidemiol","language":"eng","author":[{"family":"Murray","given":"Eleanor J."},{"family":"Caniglia","given":"Ellen C."},{"family":"Swanson","given":"Sonja A."},{"family":"Hernández-Díaz","given":"Sonia"},{"family":"Hernán","given":"Miguel A."}],"issued":{"date-parts":[["2018",11]]}}}],"schema":""} (6).Yet there is little methodological guidance for pragmatic trials beyond intention-to-treat. For example, the Biostatistics and Study Design Core of the National Institutes of Health Collaboratory has provided insightful guidance about many technical aspects of pragmatic trials ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"DZOpvdMy","properties":{"formattedCitation":"(7)","plainCitation":"(7)","noteIndex":0},"citationItems":[{"id":863,"uris":[""],"uri":[""],"itemData":{"id":863,"type":"webpage","title":"Experimental Designs and Randomization Schemes","URL":"","title-short":"Experimental Designs and Randomization Schemes","author":[{"literal":"NIH Collaboratory Health Care Biostatistics and Study Design Core"}],"issued":{"date-parts":[["2018"]]},"accessed":{"date-parts":[["2018",8,1]]}}}],"schema":""} (7), but it presupposes an intention-to-treat analysis. Similarly, the Clinical Trials Transformation Initiative—a public-private partnership comprising government agencies, industry, patient advocacy groups, professional societies, investigator groups, academic institutions, and other interested parties—promotes helpful proposals to improve clinical trials, but does not directly address causal inference standards ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"SXBhXxC9","properties":{"formattedCitation":"(8)","plainCitation":"(8)","noteIndex":0},"citationItems":[{"id":273,"uris":[""],"uri":[""],"itemData":{"id":273,"type":"webpage","title":"Clinical Trials Transformation Initiative","abstract":"CTTI is a public-private partnership to develop and drive adoption of practices that will increase the quality and efficiency of clinical trials","URL":"","title-short":"Clinical Trials Transformation Initiative","author":[{"family":"Initiative","given":"Clinical Trials Transformation"}],"issued":{"date-parts":[["2018"]]}}}],"schema":""} (8). Emphatic calls to overcome the barriers for the conduct of large, simple pragmatic trials also presuppose an intention-to-treat analysis ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"fKoZyV6U","properties":{"formattedCitation":"(9)","plainCitation":"(9)","noteIndex":0},"citationItems":[{"id":203,"uris":[""],"uri":[""],"itemData":{"id":203,"type":"article-journal","title":"The imperative of overcoming barriers to the conduct of large, simple trials","container-title":"JAMA","page":"1397-1398","volume":"311","issue":"14","abstract":"Randomized clinical trials remain the most reliable means of identifying the drugs, devices, and treatment strategies that will improve human health. There is increasing interest in the possibility that “personalized” medicine can be evaluated in much smaller trials because the average treatment effect is expected to be larger in highly selected cohorts. Smaller, biomarker-driven trials can provide major insights into whom to treat and may be sufficient for selected disease states in which considerable treatment effects may be observed. However, a precise biological understanding of most chronic illnesses and biomarkers that might predict response has eluded investigators. Moreover, treatment effect sizes in chronic conditions are expected to be modest in most cases. As a result, determining the long-term balance of risk and benefit, particularly in comparative effectiveness trials, often requires large numbers of clinical events in representative populations.","DOI":"10.1001/jama.2014.1030","ISSN":"0098-7484","title-short":"The imperative of overcoming barriers to the conduct of large, simple trials","author":[{"family":"Eapen","given":"Z. J."},{"family":"Lauer","given":"M. S."},{"family":"Temple","given":"R. J."}],"issued":{"date-parts":[["2014"]]}}}],"schema":""} (9). Here we propose causal inference guidelines for the analysis of pragmatic clinical trials. Because some of our recommended analyses require data on post-randomization treatment decisions and covariates, embracing these guidelines will require a revised framework for the design of pragmatic trials and other trials with substantial loss to follow-up or non-adherence. In fact, these guidelines are relevant for trials with individual-level randomization and parallel groups, even if they are not defined as pragmatic (conversely, some of the case studies described below are based on non-pragmatic trials but apply equally to pragmatic ones). Therefore, these guidelines complement recent efforts to go beyond the intention-to-treat principle in regulatory trials ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"8o1OrWeN","properties":{"formattedCitation":"(10)","plainCitation":"(10)","noteIndex":0},"citationItems":[{"id":674,"uris":[""],"uri":[""],"itemData":{"id":674,"type":"webpage","title":"ICH E9(R1) Addendum. Estimands and sensitivity analysis in clinical trials.","URL":".","title-short":"ICH E9(R1) Addendum. Estimands and sensitivity analysis in clinical trials.","author":[{"literal":"International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use"}],"issued":{"date-parts":[["2017"]]}}}],"schema":""} (10). Crossover trials and cluster randomized trials may also benefit from a consideration of these guidelines, but these trials have special features which are beyond the scope of this paper. Also, this document is primarily concerned with issues related to bias in the effect estimates of a pragmatic trial, rather than with generalizability and transportability of those effect estimates to other populations.This document is organized as follows. The next section discusses two options for causal effects that can be estimated from pragmatic trials: the intention-to-treat effect and the per-protocol effect. Section 3 emphasizes the need to measure effects on an absolute, rather than relative, scale. Sections 4, 5, and 6 provide guidelines for the estimation of the intention-to-treat effect, the per-protocol effect for point interventions, and the per-protocol effect for treatment strategies sustained over time. Throughout, we feature case studies to emphasize key points.2. Choice of causal effect: Intention-to-treat effect vs. Per-protocol effectThe first step towards a relevant causal analysis of pragmatic trials is a precise definition of the causal effect of interest (also known as the causal estimand). An explicit specification of the causal effect is important to engage (i) patients and stakeholders in a conversation about what they expect to learn from the trial, and (ii) investigators in a conversation about design and data analysis choices. A clear distinction should be made between the causal effect of interest and the method by which an estimate of that effect is obtained. Here, we will consider two causal effects:Intention-to-treat effect: the effect of being assigned to the treatment strategies, regardless of treatment actually received during the follow-up. In trials where initiation of the strategies co-occurs with randomization, the intention-to-treat effect is the effect of initiation of the treatment strategies, regardless of subsequent adherence to them.Per-protocol effect: the effect of receiving the assigned treatment strategies throughout the follow-up as specified in the study protocol.Some authors have referred to the intention-to-treat effect as the “treatment policy effect” ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"q8KkFp4z","properties":{"formattedCitation":"(10)","plainCitation":"(10)","noteIndex":0},"citationItems":[{"id":674,"uris":[""],"uri":[""],"itemData":{"id":674,"type":"webpage","title":"ICH E9(R1) Addendum. Estimands and sensitivity analysis in clinical trials.","URL":".","title-short":"ICH E9(R1) Addendum. Estimands and sensitivity analysis in clinical trials.","author":[{"literal":"International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use"}],"issued":{"date-parts":[["2017"]]}},"locator":"9"}],"schema":""} (10). This is an unfortunate term because, although the intention-to-treat effect is indeed a contrast of treatment policies or strategies (specifically the policies “assign to a treatment strategy, then do whatever you want”), so is the per-protocol effect. We now review the relative advantages and disadvantages of these two causal effects. The intention-to-treat effectThe intention-to-treat effect, which is the default target of many randomized trials is appealing for several reasons, included the perceived simplicity with which a valid estimate can be obtained. However, the strengths of the intention-to-treat effect come at the cost of limited utility for decision-making, interpretability, and external validity. The primary argument in favor of the intention-to-treat effect is, of course, that assignment is randomized. Therefore, we expect that a simple analysis that estimates the association between assignment and outcome—an intention-to-treat analysis—will yield an unconfounded estimate of the intention-to-treat effect. A second argument is that an intention-to-treat analysis provides a valid statistical test of the null hypothesis of no treatment effect in blinded trials. That is, if the treatment effect is null then the intention-to-treat effect is expected to be null regardless of the actual adherence pattern in the trial. This null preservation property is a desirable property. The intention-to-treat effect is said to be conservative (“biased towards the null”) in placebo-controlled trials because the magnitude of the intention-to-treat effect is somewhere in between the null value and the true effect of treatment. Conservativeness is often presented as a desirable property of the intention-to-treat effect. However, these arguments are less compelling for pragmatic trials. First, pragmatic trials are not double-blind randomized trials. Therefore, the effect of being assigned to treatment—the intention-to-treat effect—will typically be a combination of the effect of the treatment under study and of any other patient and physician’s behavioral changes triggered by the assignment itself. Quantifying this combination effect may be of practical interest in some settings, but it follows that intention-to-treat analyses may not preserve the null: the intention-to-treat effect estimate may not be null, even if the effect of treatment is null, because the intention-to-treat incorporates effects of assignment not mediated by treatment itself. Second, pragmatic trials are rarely placebo-controlled trials. Therefore, the intention-to-treat may not be conservative, even when the treatment effects are monotonic ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"wcxsDAZz","properties":{"formattedCitation":"(11)","plainCitation":"(11)","noteIndex":0},"citationItems":[{"id":511,"uris":[""],"uri":[""],"itemData":{"id":511,"type":"book","title":"Causal Inference: What If","publisher":"Chapman & Hill/CRC","publisher-place":"Boca Raton","event-place":"Boca Raton","title-short":"Causal Inference","author":[{"family":"Hernan","given":"M.A."},{"family":"Robins","given":"J."}],"issued":{"date-parts":[["2020"]]}}}],"schema":""} (11), because pragmatic trials usually compare treatment strategies with potentially differential adherence. Regardless, conservativeness is an undesirable property in trials studying harms because a conservative treatment effect can result in missed harms, and in non-inferiority trials because a conservative treatment effects can also lead to erroneous claims of non-inferiority ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"MYVCoE6m","properties":{"formattedCitation":"(12,13)","plainCitation":"(12,13)","noteIndex":0},"citationItems":[{"id":1172,"uris":[""],"uri":[""],"itemData":{"id":1172,"type":"article-journal","title":"Beyond the intention-to-treat in comparative effectiveness research","container-title":"Clinical Trials","page":"48-55","volume":"9","issue":"1","source":"Nlm","archive_location":"21948059","abstract":"BACKGROUND: The intention-to-treat comparison is the primary, if not the only, analytic approach of many randomized clinical trials. PURPOSE: To review the shortcomings of intention-to-treat analyses, and of 'as treated' and 'per protocol' analyses as commonly implemented, with an emphasis on problems that are especially relevant for comparative effectiveness research. METHODS AND RESULTS: In placebo-controlled randomized clinical trials, intention-to-treat analyses underestimate the treatment effect and are therefore nonconservative for both safety trials and noninferiority trials. In randomized clinical trials with an active comparator, intention-to-treat estimates can overestimate a treatment's effect in the presence of differential adherence. In either case, there is no guarantee that an intention-to-treat analysis estimates the clinical effectiveness of treatment. Inverse probability weighting, g-estimation, and instrumental variable estimation can reduce the bias introduced by nonadherence and loss to follow-up in 'as treated' and 'per protocol' analyses. LIMITATIONS: These analyse require untestable assumptions, a dose-response model, and time-varying data on confounders and adherence. CONCLUSIONS: We recommend that all randomized clinical trials with substantial lack of adherence or loss to follow-up are analyzed using different methods. These include an intention-to-treat analysis to estimate the effect of assigned treatment and 'as treated' and 'per protocol' analyses to estimate the effect of treatment after appropriate adjustment via inverse probability weighting or g-estimation.","DOI":"10.1177/1740774511420743","ISSN":"1740-7753 (Electronic) 1740-7745 (Linking)","title-short":"Beyond the intention-to-treat in comparative effectiveness research","language":"eng","author":[{"family":"Hernán","given":"M. A."},{"family":"Hernandez-Diaz","given":"S."}],"issued":{"date-parts":[["2012",2]]}},"label":"page"},{"id":520,"uris":[""],"uri":[""],"itemData":{"id":520,"type":"article-journal","title":"Cautions as regulators move to end exclusive reliance on intention to treat","container-title":"Annals of Internal Medicine","URL":"","DOI":"10.7326/m17-3354","ISSN":"0003-4819","title-short":"Cautions as regulators move to end exclusive reliance on intention to treat","author":[{"family":"Hernán","given":"M. A."},{"family":"Scharfstein","given":"D."}],"issued":{"date-parts":[["2018"]]}},"label":"page"}],"schema":""} (12,13). Intention-to-treat estimates are also difficult for patients, clinicians, and other decision-makers to interpret ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"pMeTvJJH","properties":{"formattedCitation":"(6)","plainCitation":"(6)","noteIndex":0},"citationItems":[{"id":1156,"uris":[""],"uri":[""],"itemData":{"id":1156,"type":"article-journal","title":"Patients and investigators prefer measures of absolute risk in subgroups for pragmatic randomized trials","container-title":"Journal of Clinical Epidemiology","page":"10-21","volume":"103","source":"PubMed","abstract":"OBJECTIVES: Pragmatic randomized trials are important tools for shared decision-making, but no guidance exists on patients' preferences for types of causal information. We aimed to assess preferences of patients and investigators toward causal effects in pragmatic randomized trials.\nSTUDY DESIGN AND SETTING: We (a) held three focus groups with patients (n?=?23) in Boston, MA; (b) surveyed (n?=?12) and interviewed (n?=?5) investigators with experience conducting pragmatic trials; and (c) conducted a systematic literature review of pragmatic trials (n?=?63).\nRESULTS: Patients were distrustful of new-to-market medications unless substantially more effective than existing choices, preferred stratified absolute risks, and valued adherence-adjusted analyses when they expected to adhere. Investigators wanted both intention-to-treat and per-protocol effects but felt methods for estimating per-protocol effects were lacking. When estimating per-protocol effects, many pragmatic trials used inappropriate methods to adjust for adherence and loss to follow-up.\nCONCLUSION: We made four recommendations for pragmatic trials to improve patient centeredness: (1) focus on superiority in effectiveness or safety, rather than noninferiority; (2) involve patients in specifying a priori subgroups; (3) report absolute measures of risk; and (4) complement intention-to-treat effect estimates with valid per-protocol effect estimates.","DOI":"10.1016/j.jclinepi.2018.06.009","ISSN":"1878-5921","note":"PMID: 29966732\nPMCID: PMC6175611","journalAbbreviation":"J Clin Epidemiol","language":"eng","author":[{"family":"Murray","given":"Eleanor J."},{"family":"Caniglia","given":"Ellen C."},{"family":"Swanson","given":"Sonja A."},{"family":"Hernández-Díaz","given":"Sonia"},{"family":"Hernán","given":"Miguel A."}],"issued":{"date-parts":[["2018",11]]}}}],"schema":""} (6) because the intention-to-treat effect is agnostic to any treatment decisions made after baseline, including discontinuation or initiation of the treatment strategies of interest, use of concomitant therapies, or any other deviations from protocol. As a result, the magnitude of the intention-to-treat from a given trial depends on the particular patterns of deviation from protocol that occur during the conduct of the trial. Two trials of the same treatment strategies, conducted in the same population, could have different intention-to-treat effects if adherence patterns differed, and both would be internally valid effects of assignment to treatment. The external validity of the intention-to-treat effect can be poor and difficult to assess because patterns of adherence in the trial may not reflect those outside the trial. In fact, the publication of the trial results may lead to alterations in adherence patterns outside the trial ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"46l9gYZq","properties":{"formattedCitation":"(14)","plainCitation":"(14)","noteIndex":0},"citationItems":[{"id":1154,"uris":[""],"uri":[""],"itemData":{"id":1154,"type":"article-journal","title":"Per-protocol analyses of pragmatic trials","container-title":"New England Journal of Medicine","page":"1391-1398","volume":"377","issue":"14","archive_location":"28976864","DOI":"10.1056/NEJMsm1605385","title-short":"Per-Protocol Analyses of Pragmatic Trials","journalAbbreviation":"N Eng J Med","author":[{"family":"Hernán","given":"Miguel A."},{"family":"Robins","given":"James M."}],"issued":{"date-parts":[["2017"]]}}}],"schema":""} (14).Further, the intention-to-treat effect is often considered to estimate treatment effectiveness (that is, the effect of treatment under realistic conditions or in everyday practice) rather than efficacy (that is, the effect of treatment under ideal conditions including perfect adherence). However, this distinction between efficacy and effectiveness is artificial (in fact, we don’t distinguish between “safety” and “safetiness” ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"Gn79FqSm","properties":{"formattedCitation":"(5)","plainCitation":"(5)","noteIndex":0},"citationItems":[{"id":1153,"uris":[""],"uri":[""],"itemData":{"id":1153,"type":"article-journal","title":"Randomized trials analyzed as observational studies","container-title":"Annals of Internal Medicine","page":"560-2","volume":"159","issue":"8","source":"Nlm","archive_location":"24018844","DOI":"10.7326/0003-4819-159-8-201310150-00709","ISSN":"1539-3704 (Electronic) 0003-4819 (Linking)","title-short":"Randomized trials analyzed as observational studies","journalAbbreviation":"Ann Intern Med","language":"eng","author":[{"family":"Hernán","given":"M. A."},{"family":"Hernandez-Diaz","given":"S."},{"family":"Robins","given":"J. M."}],"issued":{"date-parts":[["2013",10,15]]}}}],"schema":""} (5)) and of unclear relevance for decision makers. For example, a patient deciding whether to take cholesterol lowering medication is usually interested in its effectiveness, rather than the effectiveness estimated from a study in which many trial participants did not adhere to a correct use ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"MMkVHgaW","properties":{"formattedCitation":"(6,12)","plainCitation":"(6,12)","noteIndex":0},"citationItems":[{"id":1156,"uris":[""],"uri":[""],"itemData":{"id":1156,"type":"article-journal","title":"Patients and investigators prefer measures of absolute risk in subgroups for pragmatic randomized trials","container-title":"Journal of Clinical Epidemiology","page":"10-21","volume":"103","source":"PubMed","abstract":"OBJECTIVES: Pragmatic randomized trials are important tools for shared decision-making, but no guidance exists on patients' preferences for types of causal information. We aimed to assess preferences of patients and investigators toward causal effects in pragmatic randomized trials.\nSTUDY DESIGN AND SETTING: We (a) held three focus groups with patients (n?=?23) in Boston, MA; (b) surveyed (n?=?12) and interviewed (n?=?5) investigators with experience conducting pragmatic trials; and (c) conducted a systematic literature review of pragmatic trials (n?=?63).\nRESULTS: Patients were distrustful of new-to-market medications unless substantially more effective than existing choices, preferred stratified absolute risks, and valued adherence-adjusted analyses when they expected to adhere. Investigators wanted both intention-to-treat and per-protocol effects but felt methods for estimating per-protocol effects were lacking. When estimating per-protocol effects, many pragmatic trials used inappropriate methods to adjust for adherence and loss to follow-up.\nCONCLUSION: We made four recommendations for pragmatic trials to improve patient centeredness: (1) focus on superiority in effectiveness or safety, rather than noninferiority; (2) involve patients in specifying a priori subgroups; (3) report absolute measures of risk; and (4) complement intention-to-treat effect estimates with valid per-protocol effect estimates.","DOI":"10.1016/j.jclinepi.2018.06.009","ISSN":"1878-5921","note":"PMID: 29966732\nPMCID: PMC6175611","journalAbbreviation":"J Clin Epidemiol","language":"eng","author":[{"family":"Murray","given":"Eleanor J."},{"family":"Caniglia","given":"Ellen C."},{"family":"Swanson","given":"Sonja A."},{"family":"Hernández-Díaz","given":"Sonia"},{"family":"Hernán","given":"Miguel A."}],"issued":{"date-parts":[["2018",11]]}}},{"id":1172,"uris":[""],"uri":[""],"itemData":{"id":1172,"type":"article-journal","title":"Beyond the intention-to-treat in comparative effectiveness research","container-title":"Clinical Trials","page":"48-55","volume":"9","issue":"1","source":"Nlm","archive_location":"21948059","abstract":"BACKGROUND: The intention-to-treat comparison is the primary, if not the only, analytic approach of many randomized clinical trials. PURPOSE: To review the shortcomings of intention-to-treat analyses, and of 'as treated' and 'per protocol' analyses as commonly implemented, with an emphasis on problems that are especially relevant for comparative effectiveness research. METHODS AND RESULTS: In placebo-controlled randomized clinical trials, intention-to-treat analyses underestimate the treatment effect and are therefore nonconservative for both safety trials and noninferiority trials. In randomized clinical trials with an active comparator, intention-to-treat estimates can overestimate a treatment's effect in the presence of differential adherence. In either case, there is no guarantee that an intention-to-treat analysis estimates the clinical effectiveness of treatment. Inverse probability weighting, g-estimation, and instrumental variable estimation can reduce the bias introduced by nonadherence and loss to follow-up in 'as treated' and 'per protocol' analyses. LIMITATIONS: These analyse require untestable assumptions, a dose-response model, and time-varying data on confounders and adherence. CONCLUSIONS: We recommend that all randomized clinical trials with substantial lack of adherence or loss to follow-up are analyzed using different methods. These include an intention-to-treat analysis to estimate the effect of assigned treatment and 'as treated' and 'per protocol' analyses to estimate the effect of treatment after appropriate adjustment via inverse probability weighting or g-estimation.","DOI":"10.1177/1740774511420743","ISSN":"1740-7753 (Electronic) 1740-7745 (Linking)","title-short":"Beyond the intention-to-treat in comparative effectiveness research","language":"eng","author":[{"family":"Hernán","given":"M. A."},{"family":"Hernandez-Diaz","given":"S."}],"issued":{"date-parts":[["2012",2]]}}}],"schema":""} (6,12). In general, patients are mostly interested in the effect of treatment in everyday practice when taken as instructed, an effect that corresponds to the common definition of neither “efficacy” nor “effectiveness”. We therefore largely avoid the use of those terms here. Finally, it is not universally true that traditional intention-to-treat analyses provide a valid estimate of the intention-to-treat effect. In trials with incomplete follow-up of participants, losses to follow-up may introduce selection bias ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"xn1POp75","properties":{"formattedCitation":"(5,15)","plainCitation":"(5,15)","noteIndex":0},"citationItems":[{"id":1153,"uris":[""],"uri":[""],"itemData":{"id":1153,"type":"article-journal","title":"Randomized trials analyzed as observational studies","container-title":"Annals of Internal Medicine","page":"560-2","volume":"159","issue":"8","source":"Nlm","archive_location":"24018844","DOI":"10.7326/0003-4819-159-8-201310150-00709","ISSN":"1539-3704 (Electronic) 0003-4819 (Linking)","title-short":"Randomized trials analyzed as observational studies","journalAbbreviation":"Ann Intern Med","language":"eng","author":[{"family":"Hernán","given":"M. A."},{"family":"Hernandez-Diaz","given":"S."},{"family":"Robins","given":"J. M."}],"issued":{"date-parts":[["2013",10,15]]}},"label":"page"},{"id":1150,"uris":[""],"uri":[""],"itemData":{"id":1150,"type":"article-journal","title":"The prevention and treatment of missing data in clinical trials","container-title":"New England Journal of Medicine","page":"1355-60","volume":"367","issue":"14","source":"Nlm","archive_location":"23034025","DOI":"10.1056/NEJMsr1203730","ISSN":"1533-4406 (Electronic) 0028-4793 (Linking)","title-short":"The prevention and treatment of missing data in clinical trials","journalAbbreviation":"N Engl J Med","language":"eng","author":[{"family":"Little","given":"R. J."},{"family":"D'Agostino","given":"R."},{"family":"Cohen","given":"M. L."},{"family":"Dickersin","given":"K."},{"family":"Emerson","given":"S. S."},{"family":"Farrar","given":"J. T."},{"family":"Frangakis","given":"C."},{"family":"Hogan","given":"J. W."},{"family":"Molenberghs","given":"G."},{"family":"Murphy","given":"S. A."},{"family":"Neaton","given":"J. D."},{"family":"Rotnitzky","given":"A."},{"family":"Scharfstein","given":"D."},{"family":"Shih","given":"W. J."},{"family":"Siegel","given":"J. P."},{"family":"Stern","given":"H."}],"issued":{"date-parts":[["2012",10,4]]}},"label":"page"}],"schema":""} (5,15). Eliminating this bias requires adjusting for pre- and post-randomization predictors of loss to follow-up that are also prognostic factors. To summarize, the intention-to-treat effect from a given trial measures the effect of treatment assignment, not of treatment itself, under whatever level of adherence to protocol took place in that trial. Therefore, intention-to-treat estimates, even if appropriately adjusted for loss to follow-up, provide incomplete information for patients who are interested in the benefit-risk profile of the treatment when taken as instructed. The per-protocol effectThe per-protocol effect overcomes many limitations of the intention-to-treat effect. The per-protocol effect is closer than the intention-to-treat effect to what patients are mostly interested in learning from pragmatic trials (Case Study A), and is often the implicit target of inference for investigators too. When investigators say that the intention-to-treat is “biased towards the null”, the implication is that the intention-to-treat effect is a biased estimate of the per-protocol effect (a bias sometimes referred to as “performance bias” ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"Z5kqr7r7","properties":{"formattedCitation":"(16,17)","plainCitation":"(16,17)","noteIndex":0},"citationItems":[{"id":1073,"uris":[""],"uri":[""],"itemData":{"id":1073,"type":"article-journal","title":"The Cochrane Collaboration's tool for assessing risk of bias in randomised trials","container-title":"BMJ (Clinical research ed.)","page":"d5928","volume":"343","source":"PubMed","DOI":"10.1136/bmj.d5928","ISSN":"1756-1833","note":"PMID: 22008217\nPMCID: PMC3196245","journalAbbreviation":"BMJ","language":"eng","author":[{"family":"Higgins","given":"Julian P. T."},{"family":"Altman","given":"Douglas G."},{"family":"G?tzsche","given":"Peter C."},{"family":"Jüni","given":"Peter"},{"family":"Moher","given":"David"},{"family":"Oxman","given":"Andrew D."},{"family":"Savovic","given":"Jelena"},{"family":"Schulz","given":"Kenneth F."},{"family":"Weeks","given":"Laura"},{"family":"Sterne","given":"Jonathan A. C."},{"literal":"Cochrane Bias Methods Group"},{"literal":"Cochrane Statistical Methods Group"}],"issued":{"date-parts":[["2011",10,18]]}},"label":"page"},{"id":1171,"uris":[""],"uri":[""],"itemData":{"id":1171,"type":"article-journal","title":"Biases in Randomized Trials: A Conversation Between Trialists and Epidemiologists","container-title":"Epidemiology","page":"54-59","volume":"28","issue":"1","archive_location":"27748683","abstract":"Trialists and epidemiologists often employ different terminology to refer to biases in randomized trials and observational studies, even though many biases have a similar structure in both types of study. We use causal diagrams to represent the structure of biases, as described by Cochrane for randomized trials, and provide a translation to the usual epidemiologic terms of confounding, selection bias, and measurement bias. This structural approach clarifies that an explicit description of the inferential goal-the intention-to-treat effect or the per-protocol effect-is necessary to assess risk of bias in the estimates. Being aware of each other's terminologies will enhance communication between trialists and epidemiologists when considering key concepts and methods for causal inference.","DOI":"10.1097/EDE.0000000000000564","ISSN":"1531-5487","title-short":"Biases in Randomized Trials: A Conversation Between Trialists and Epidemiologists","language":"eng","author":[{"family":"Mansournia","given":"M. A."},{"family":"Higgins","given":"J. P."},{"family":"Sterne","given":"J. A."},{"family":"Hernán","given":"M. A."}],"issued":{"literal":"1"}},"label":"page"}],"schema":""} (16,17)). Thus, the investigators’ language indicates that the investigators are really interested in comparing the interventions implemented during the follow-up as specified in the protocol (i.e., the per-protocol effect) and not just in the effect of assignment to the interventions at baseline (i.e., the intention-to-treat effect). An added advantage of the per-protocol effect is that its interpretation does not depend on a trial-specific degree of adherence ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"ojQi8UFY","properties":{"formattedCitation":"(14)","plainCitation":"(14)","noteIndex":0},"citationItems":[{"id":1154,"uris":[""],"uri":[""],"itemData":{"id":1154,"type":"article-journal","title":"Per-protocol analyses of pragmatic trials","container-title":"New England Journal of Medicine","page":"1391-1398","volume":"377","issue":"14","archive_location":"28976864","DOI":"10.1056/NEJMsm1605385","title-short":"Per-Protocol Analyses of Pragmatic Trials","journalAbbreviation":"N Eng J Med","author":[{"family":"Hernán","given":"Miguel A."},{"family":"Robins","given":"James M."}],"issued":{"date-parts":[["2017"]]}}}],"schema":""} (14), which makes it a potentially more transportable effect. For example, in a cancer prevention trial, participants were randomized to either an invitation to receive a screening colonoscopy or to usual care (with no colonoscopy screening) ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"VFh6PwDo","properties":{"formattedCitation":"(18)","plainCitation":"(18)","noteIndex":0},"citationItems":[{"id":1026,"uris":[""],"uri":[""],"itemData":{"id":1026,"type":"article-journal","title":"Long-Term Effectiveness of Sigmoidoscopy Screening on Colorectal Cancer Incidence and Mortality in Women and Men: A Randomized Trial","container-title":"Ann Intern Med","page":"775-782","volume":"168","issue":"11","archive_location":"29710125","abstract":"The long-term effects of sigmoidoscopy screening on colorectal cancer (CRC) incidence and mortality in women and men are unclear.|To determine the effectiveness of flexible sigmoidoscopy screening after 15 years of follow-up in women and men.|Randomized controlled trial. (: NCT00119912).|Oslo and Telemark County, Norway.|Adults aged 50 to 64 years at baseline without prior CRC.|Screening (between 1999 and 2001) with flexible sigmoidoscopy with and without additional fecal blood testing versus no screening. Participants with positive screening results were offered colonoscopy.|Age-adjusted CRC incidence and mortality stratified by sex.|Of 98?678 persons, 20?552 were randomly assigned to screening and 78?126 to no screening. Adherence rates were 64.7% in women and 61.4% in men. Median follow-up was 14.8 years. The absolute risks for CRC in women were 1.86% in the screening group and 2.05% in the control group (risk difference, -0.19 percentage point [95% CI, -0.49 to 0.11 percentage point]; HR, 0.92 [CI, 0.79 to 1.07]). In men, the corresponding risks were 1.72% and 2.50%, respectively (risk difference, -0.78 percentage point [CI, -1.08 to -0.48 percentage points]; hazard ratio [HR], 0.66 [CI, 0.57 to 0.78]) (P for heterogeneity?= 0.004). The absolute risks for death from CRC in women were 0.60% in the screening group and 0.59% in the control group (risk difference, 0.01 percentage point [CI, -0.16 to 0.18 percentage point]; HR, 1.01 [CI, 0.77 to 1.33]). The corresponding risks for death from CRC in men were 0.49% and 0.81%, respectively (risk difference, -0.33 percentage point [CI, -0.49 to -0.16 percentage point]; HR, 0.63 [CI, 0.47 to 0.83]) (P for heterogeneity?= 0.014).|Follow-up through national registries.|Offering sigmoidoscopy screening in Norway reduced CRC incidence and mortality in men but had little or no effect in women.|Norwegian government and Norwegian Cancer Society.","DOI":"10.7326/M17-1441","ISSN":"1539-3704","title-short":"Long-Term Effectiveness of Sigmoidoscopy Screening on Colorectal Cancer Incidence and Mortality in Women and Men: A Randomized Trial","language":"eng","author":[{"family":"Holme","given":"?"},{"family":"L?berg","given":"M."},{"family":"Kalager","given":"M."},{"family":"Bretthauer","given":"M."},{"family":"Hernán","given":"M. A."},{"family":"Aas","given":"E."},{"family":"Eide","given":"T. J."},{"family":"Skovlund","given":"E."},{"family":"Lekven","given":"J."},{"family":"Schneede","given":"J."},{"family":"Tveit","given":"K. M."},{"family":"Vatn","given":"M."},{"family":"Ursin","given":"G."},{"family":"Hoff","given":"G."},{"family":"Group?","given":"NORCCAP Study"}],"issued":{"date-parts":[["2018",6]]}}}],"schema":""} (18). The intention-to-treat effect for this trial quantifies the effect of being invited to a screening colonoscopy in a population in which about 30% of individuals declined the invitation. Because the proportion and type of people who reject a colonoscopy varies across populations, the intention-to-treat effect from this trial is difficult to transport to other populations. In contrast, the per-protocol effect would quantify the effect of receiving the colonoscopy under perfect adherence in the studied population, which may be closer to the effect under perfect adherence in similar populations.Case Study A: Intention-to-treat vs. per-protocol effectThe Women’s Health Initiative (WHI) conducted a randomized trial to estimate the health effects of postmenopausal estrogen plus progestin hormone therapy versus placebo ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"gDpm8c2t","properties":{"formattedCitation":"(19,20)","plainCitation":"(19,20)","noteIndex":0},"citationItems":[{"id":752,"uris":[""],"uri":[""],"itemData":{"id":752,"type":"article-journal","title":"Risks and benefits of estrogen plus progestin in healthy postmenopausal women: principal results From the Women's Health Initiative randomized controlled trial","container-title":"JAMA","page":"321-33","volume":"288","issue":"3","archive_location":"12117397","abstract":"Despite decades of accumulated observational evidence, the balance of risks and benefits for hormone use in healthy postmenopausal women remains uncertain.|To assess the major health benefits and risks of the most commonly used combined hormone preparation in the United States.|Estrogen plus progestin component of the Women's Health Initiative, a randomized controlled primary prevention trial (planned duration, 8.5 years) in which 16608 postmenopausal women aged 50-79 years with an intact uterus at baseline were recruited by 40 US clinical centers in 1993-1998.|Participants received conjugated equine estrogens, 0.625 mg/d, plus medroxyprogesterone acetate, 2.5 mg/d, in 1 tablet (n = 8506) or placebo (n = 8102).|The primary outcome was coronary heart disease (CHD) (nonfatal myocardial infarction and CHD death), with invasive breast cancer as the primary adverse outcome. A global index summarizing the balance of risks and benefits included the 2 primary outcomes plus stroke, pulmonary embolism (PE), endometrial cancer, colorectal cancer, hip fracture, and death due to other causes.|On May 31, 2002, after a mean of 5.2 years of follow-up, the data and safety monitoring board recommended stopping the trial of estrogen plus progestin vs placebo because the test statistic for invasive breast cancer exceeded the stopping boundary for this adverse effect and the global index statistic supported risks exceeding benefits. This report includes data on the major clinical outcomes through April 30, 2002. Estimated hazard ratios (HRs) (nominal 95% confidence intervals [CIs]) were as follows: CHD, 1.29 (1.02-1.63) with 286 cases; breast cancer, 1.26 (1.00-1.59) with 290 cases; stroke, 1.41 (1.07-1.85) with 212 cases; PE, 2.13 (1.39-3.25) with 101 cases; colorectal cancer, 0.63 (0.43-0.92) with 112 cases; endometrial cancer, 0.83 (0.47-1.47) with 47 cases; hip fracture, 0.66 (0.45-0.98) with 106 cases; and death due to other causes, 0.92 (0.74-1.14) with 331 cases. Corresponding HRs (nominal 95% CIs) for composite outcomes were 1.22 (1.09-1.36) for total cardiovascular disease (arterial and venous disease), 1.03 (0.90-1.17) for total cancer, 0.76 (0.69-0.85) for combined fractures, 0.98 (0.82-1.18) for total mortality, and 1.15 (1.03-1.28) for the global index. Absolute excess risks per 10 000 person-years attributable to estrogen plus progestin were 7 more CHD events, 8 more strokes, 8 more PEs, and 8 more invasive breast cancers, while absolute risk reductions per 10 000 person-years were 6 fewer colorectal cancers and 5 fewer hip fractures. The absolute excess risk of events included in the global index was 19 per 10 000 person-years.|Overall health risks exceeded benefits from use of combined estrogen plus progestin for an average 5.2-year follow-up among healthy postmenopausal US women. All-cause mortality was not affected during the trial. The risk-benefit profile found in this trial is not consistent with the requirements for a viable intervention for primary prevention of chronic diseases, and the results indicate that this regimen should not be initiated or continued for primary prevention of CHD.","ISSN":"0098-7484","title-short":"Risks and benefits of estrogen plus progestin in healthy postmenopausal women: principal results From the Women's Health Initiative randomized controlled trial","language":"eng","author":[{"family":"Rossouw","given":"J. E."},{"family":"Anderson","given":"G. L."},{"family":"Prentice","given":"R. L."},{"family":"LaCroix","given":"A. Z."},{"family":"Kooperberg","given":"C."},{"family":"Stefanick","given":"M. L."},{"family":"Jackson","given":"R. D."},{"family":"Beresford","given":"S. A."},{"family":"Howard","given":"B. V."},{"family":"Johnson","given":"K. C."},{"family":"Kotchen","given":"J. M."},{"family":"Ockene","given":"J."},{"family":"Investigators","given":"Writing Group for the Women's Health Initiative"}],"issued":{"date-parts":[["2002",7]]}},"label":"page"},{"id":364,"uris":[""],"uri":[""],"itemData":{"id":364,"type":"article-journal","title":"Design of the Women's Health Initiative clinical trial and observational study. The Women's Health Initiative Study Group","container-title":"Control Clin Trials","page":"61-109","volume":"19","issue":"1","archive_location":"9492970","abstract":"The Women's Health Initiative (WHI) is a large and complex clinical investigation of strategies for the prevention and control of some of the most common causes of morbidity and mortality among postmenopausal women, including cancer, cardiovascular disease, and osteoporotic fractures. The WHI was initiated in 1992, with a planned completion date of 2007. Postmenopausal women ranging in age from 50 to 79 are enrolled at one of 40 WHI clinical centers nationwide into either a clinical trial (CT) that will include about 64,500 women or an observational study (OS) that will include about 100,000 women. The CT is designed to allow randomized controlled evaluation of three distinct interventions: a low-fat eating pattern, hypothesized to prevent breast cancer and colorectal cancer and, secondarily, coronary heart disease; hormone replacement therapy, hypothesized to reduce the risk of coronary heart disease and other cardiovascular diseases and, secondarily, to reduce the risk of hip and other fractures, with increased breast cancer risk as a possible adverse outcome; and calcium and vitamin D supplementation, hypothesized to prevent hip fractures and, secondarily, other fractures and colorectal cancer. Overall benefit-versus-risk assessment is a central focus in each of the three CT components. Women are screened for participation in one or both of the components--dietary modification (DM) or hormone replacement therapy (HRT)--of the CT, which will randomize 48,000 and 27,500 women, respectively. Women who prove to be ineligible for, or who are unwilling to enroll in, these CT components are invited to enroll in the OS. At their 1-year anniversary of randomization, CT women are invited to be further randomized into the calcium and vitamin D (CaD) trial component, which is projected to include 45,000 women. The average follow-up for women in either CT or OS is approximately 9 years. Concerted efforts are made to enroll women of racial and ethnic minority groups, with a target of 20% of overall enrollment in both the CT and OS. This article gives a brief description of the rationale for the interventions being studied in each of the CT components and for the inclusion of the OS component. Some detail is provided on specific study design choices, including eligibility criteria, recruitment strategy, and sample size, with attention to the partial factorial design of the CT. Some aspects of the CT monitoring approach are also outlined. The scientific and logistic complexity of the WHI implies particular leadership and management challenges. The WHI organization and committee structure employed to respond to these challenges is also briefly described.","ISSN":"0197-2456","title-short":"Design of the Women's Health Initiative clinical trial and observational study. The Women's Health Initiative Study Group","language":"eng","issued":{"date-parts":[["1998",2]]}},"label":"page"}],"schema":""} (19,20). Analyses to estimate both the intention-to-treat effect and the per-protocol effect were conducted (the latter ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"SiYsZhV4","properties":{"formattedCitation":"(21)","plainCitation":"(21)","noteIndex":0},"citationItems":[{"id":1155,"uris":[""],"uri":[""],"itemData":{"id":1155,"type":"article-journal","title":"Estimating absolute risks in the presence of nonadherence: an application to a follow-up study with baseline randomization","container-title":"Epidemiology","page":"528-39","volume":"21","issue":"4","source":"Nlm","archive_location":"20526200","abstract":"The intention-to-treat (ITT) analysis provides a valid test of the null hypothesis and naturally results in both absolute and relative measures of risk. However, this analytic approach may miss the occurrence of serious adverse effects that would have been detected under full adherence to the assigned treatment. Inverse probability weighting of marginal structural models has been used to adjust for nonadherence, but most studies have provided only relative measures of risk. In this study, we used inverse probability weighting to estimate both absolute and relative measures of risk of invasive breast cancer under full adherence to the assigned treatment in the Women's Health Initiative estrogen-plus-progestin trial. In contrast to an ITT hazard ratio (HR) of 1.25 (95% confidence interval [CI] = 1.01 to 1.54), the HR for 8-year continuous estrogen-plus-progestin use versus no use was 1.68 (1.24 to 2.28). The estimated risk difference (cases/100 women) at year 8 was 0.83 (-0.03 to 1.69) in the ITT analysis, compared with 1.44 (0.52 to 2.37) in the adherence-adjusted analysis. Results were robust across various dose-response models. We also compared the dynamic treatment regimen \"take hormone therapy until certain adverse events become apparent, then stop taking hormone therapy\" with no use (HR = 1.64; 95% CI = 1.24 to 2.18). The methods described here are also applicable to observational studies with time-varying treatments.","DOI":"10.1097/EDE.0b013e3181df1b69","ISSN":"1531-5487 (Electronic) 1044-3983 (Linking)","title-short":"Estimating absolute risks in the presence of nonadherence: an application to a follow-up study with baseline randomization","language":"eng","author":[{"family":"Toh","given":"S."},{"family":"Hernandez-Diaz","given":"S."},{"family":"Logan","given":"R."},{"family":"Robins","given":"J. M."},{"family":"Hernán","given":"M. A."}],"issued":{"date-parts":[["2010",7]]}}}],"schema":""} (21) made the simplifying, but plausible, assumption that the number of women with contraindications for hormone therapy during the follow-up was negligible.) The estimated intention-to-treat hazard ratio of breast cancer was 1.25 (95% CI: 1.01, 1.54). This 25% increased risk reflects both the actual effect of hormone therapy and the incomplete adherence to the assigned treatment: only 58% of women in the hormone therapy arm, and 62% in the placebo arm, were still taking their assigned treatment at 6 years. In contrast, the estimated per-protocol hazard ratio of breast cancer was 1.68 (95% CI: 1.24, 2.28). This effect estimate suggests that continued use of postmenopausal estrogen plus progestin hormone use increases the risk of breast cancer by 68% rather than 25%. Women presented with the intention-to-treat effect estimate only might reasonably argue that they did not receive full information about their increased risk of breast cancer if they took treatment as instructed.Yet, despite its relevance for decision making by patients and clinicians, the per-protocol effect has been historically eschewed in randomized trials. A fundamental problem is that valid estimation of the per-protocol effect cannot be guaranteed by randomization because unbiased estimation generally requires adjustment for prognostic factors that predict adherence. Faced with the choice between a less relevant effect that is expected to be unbiased in the absence of substantial loss to follow-up (the intention-to-treat effect) and a more relevant effect that may be biased (the per-protocol effect), researchers chose the former. A consequence of the general skepticism about the per-protocol effect is that little emphasis has been given to studying the conditions under which it can be correctly estimated. This in turn has led to the design and conduct of randomized trials that are intrinsically ill-equipped to quantify the per-protocol effect, which often results in the intention-to-treat effect being the only viable option. To break this vicious circle, one needs to realize that unbiased estimation of the per-protocol effect in a randomized trial generally requires 3 elements:Unambiguous specification of the treatment strategies in the trial protocol. A common misconception is that trials compare treatments; rather, they compare treatment strategies ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"jrUcEJfT","properties":{"formattedCitation":"(13,14)","plainCitation":"(13,14)","noteIndex":0},"citationItems":[{"id":520,"uris":[""],"uri":[""],"itemData":{"id":520,"type":"article-journal","title":"Cautions as regulators move to end exclusive reliance on intention to treat","container-title":"Annals of Internal Medicine","URL":"","DOI":"10.7326/m17-3354","ISSN":"0003-4819","title-short":"Cautions as regulators move to end exclusive reliance on intention to treat","author":[{"family":"Hernán","given":"M. A."},{"family":"Scharfstein","given":"D."}],"issued":{"date-parts":[["2018"]]}},"label":"page"},{"id":1154,"uris":[""],"uri":[""],"itemData":{"id":1154,"type":"article-journal","title":"Per-protocol analyses of pragmatic trials","container-title":"New England Journal of Medicine","page":"1391-1398","volume":"377","issue":"14","archive_location":"28976864","DOI":"10.1056/NEJMsm1605385","title-short":"Per-Protocol Analyses of Pragmatic Trials","journalAbbreviation":"N Eng J Med","author":[{"family":"Hernán","given":"Miguel A."},{"family":"Robins","given":"James M."}],"issued":{"date-parts":[["2017"]]}},"label":"page"}],"schema":""} (13,14). Individuals assigned to a particular treatment must be allowed to discontinue, modify, or switch the treatment when clinically indicated (e.g., due to toxicity or lack of effectiveness) and still remain on protocol, both for ethical reasons and in order to ensure the protocol is of interest given current knowledge and medical practice ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"ENWF5vh3","properties":{"formattedCitation":"(14)","plainCitation":"(14)","noteIndex":0},"citationItems":[{"id":1154,"uris":[""],"uri":[""],"itemData":{"id":1154,"type":"article-journal","title":"Per-protocol analyses of pragmatic trials","container-title":"New England Journal of Medicine","page":"1391-1398","volume":"377","issue":"14","archive_location":"28976864","DOI":"10.1056/NEJMsm1605385","title-short":"Per-Protocol Analyses of Pragmatic Trials","journalAbbreviation":"N Eng J Med","author":[{"family":"Hernán","given":"Miguel A."},{"family":"Robins","given":"James M."}],"issued":{"date-parts":[["2017"]]}}}],"schema":""} (14). Therefore, adherence to the protocol is not defined as “taking treatment continuously no matter what” but as, for example, “taking treatment until serious side effects arise” (Case Study B). Collection of data on prognostic factors that predict adherence to the protocol. Most pragmatic trials compare treatment strategies that are sustained over the follow-up. As a result, adherence to the strategies may be affected by both pre- and post-randomization factors. When the treatment strategy is a point intervention, prognostic factors that affect adherence may not be required for effect estimation but are useful for characterizing the compliers (see Section 5.2).Adjustment for prognostic factors that predict adherence to the protocol. When the value of the post-randomization factors is affected by prior adherence, the adjustment needs to be carried out via g-methods ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"3VvQmGuH","properties":{"formattedCitation":"(11)","plainCitation":"(11)","noteIndex":0},"citationItems":[{"id":511,"uris":[""],"uri":[""],"itemData":{"id":511,"type":"book","title":"Causal Inference: What If","publisher":"Chapman & Hill/CRC","publisher-place":"Boca Raton","event-place":"Boca Raton","title-short":"Causal Inference","author":[{"family":"Hernan","given":"M.A."},{"family":"Robins","given":"J."}],"issued":{"date-parts":[["2020"]]}},"label":"page"}],"schema":""} (11), as discussed below.To summarize, valid estimation of the per-protocol effect generally requires a specification of the treatment strategies, adequate data collection, and appropriate adjustment. In addition, at the design stage, researchers may want to ensure a sufficient sample size to estimate the per-protocol effect with sufficient precision. Given the complexities involved in per-protocol effect estimation (see below), simulations of the data generation process will often be necessary for approximate sample size calculations. Case study B: Specification of the treatment strategies in the protocolThe Candesartan in Heart Failure: Morbidity and Mortality (CHARM) trial was a secondary prevention trial aimed at reducing mortality and cardiac-related hospitalization among individuals with symptomatic congestive heart failure ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"vEPqnTA1","properties":{"formattedCitation":"(22)","plainCitation":"(22)","noteIndex":0},"citationItems":[{"id":816,"uris":[""],"uri":[""],"itemData":{"id":816,"type":"article-journal","title":"Effects of candesartan on mortality and morbidity in patients with chronic heart failure: the CHARM-Overall programme","container-title":"Lancet","page":"759-66","volume":"362","issue":"9386","source":"Nlm","archive_location":"13678868","abstract":"BACKGROUND: Patients with chronic heart failure (CHF) are at high risk of cardiovascular death and recurrent hospital admissions. We aimed to find out whether the use of an angiotensin-receptor blocker could reduce mortality and morbidity. METHODS: In parallel, randomised, double-blind, controlled, clinical trials we compared candesartan with placebo in three distinct populations. We studied patients with left-ventricular ejection fraction (LVEF) 40% or less who were not receiving angiotensin-converting-enzyme inhibitors because of previous intolerance or who were currently receiving angiotensin-converting-enzyme inhibitors, and patients with LVEF higher than 40%. Overall, 7601 patients (7599 with data) were randomly assigned candesartan (n=3803, titrated to 32 mg once daily) or matching placebo (n=3796), and followed up for at least 2 years. The primary outcome of the overall programme was all-cause mortality, and for all the component trials was cardiovascular death or hospital admission for CHF. Analysis was by intention to treat. FINDINGS: Median follow-up was 37.7 months. 886 (23%) patients in the candesartan and 945 (25%) in the placebo group died (unadjusted hazard ratio 0.91 [95% CI 0.83-1.00], p=0.055; covariate adjusted 0.90 [0.82-0.99], p=0.032), with fewer cardiovascular deaths (691 [18%] vs 769 [20%], unadjusted 0.88 [0.79-0.97], p=0.012; covariate adjusted 0.87 [0.78-0.96], p=0.006) and hospital admissions for CHF (757 [20%] vs 918 [24%], p<0.0001) in the candesartan group. There was no significant heterogeneity for candesartan results across the component trials. More patients discontinued candesartan than placebo because of concerns about renal function, hypotension, and hyperkalaemia. INTERPRETATION: Candesartan was generally well tolerated and significantly reduced cardiovascular deaths and hospital admissions for heart failure. Ejection fraction or treatment at baseline did not alter these effects.","ISSN":"1474-547X (Electronic) 0140-6736 (Linking)","title-short":"Effects of candesartan on mortality and morbidity in patients with chronic heart failure: the CHARM-Overall programme","language":"eng","author":[{"family":"Pfeffer","given":"M. A."},{"family":"Swedberg","given":"K."},{"family":"Granger","given":"C. B."},{"family":"Held","given":"P."},{"family":"McMurray","given":"J. J."},{"family":"Michelson","given":"E. L."},{"family":"Olofsson","given":"B."},{"family":"Ostergren","given":"J."},{"family":"Yusuf","given":"S."},{"family":"Pocock","given":"S."}],"issued":{"date-parts":[["2003",9,6]]}}}],"schema":""} (22). The estimated intention-to-treat mortality hazard ratio for candesartan vs. placebo was 0.89 (95% CI: 0.82, 0.97) ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"QKOoio0I","properties":{"formattedCitation":"(22)","plainCitation":"(22)","noteIndex":0},"citationItems":[{"id":816,"uris":[""],"uri":[""],"itemData":{"id":816,"type":"article-journal","title":"Effects of candesartan on mortality and morbidity in patients with chronic heart failure: the CHARM-Overall programme","container-title":"Lancet","page":"759-66","volume":"362","issue":"9386","source":"Nlm","archive_location":"13678868","abstract":"BACKGROUND: Patients with chronic heart failure (CHF) are at high risk of cardiovascular death and recurrent hospital admissions. We aimed to find out whether the use of an angiotensin-receptor blocker could reduce mortality and morbidity. METHODS: In parallel, randomised, double-blind, controlled, clinical trials we compared candesartan with placebo in three distinct populations. We studied patients with left-ventricular ejection fraction (LVEF) 40% or less who were not receiving angiotensin-converting-enzyme inhibitors because of previous intolerance or who were currently receiving angiotensin-converting-enzyme inhibitors, and patients with LVEF higher than 40%. Overall, 7601 patients (7599 with data) were randomly assigned candesartan (n=3803, titrated to 32 mg once daily) or matching placebo (n=3796), and followed up for at least 2 years. The primary outcome of the overall programme was all-cause mortality, and for all the component trials was cardiovascular death or hospital admission for CHF. Analysis was by intention to treat. FINDINGS: Median follow-up was 37.7 months. 886 (23%) patients in the candesartan and 945 (25%) in the placebo group died (unadjusted hazard ratio 0.91 [95% CI 0.83-1.00], p=0.055; covariate adjusted 0.90 [0.82-0.99], p=0.032), with fewer cardiovascular deaths (691 [18%] vs 769 [20%], unadjusted 0.88 [0.79-0.97], p=0.012; covariate adjusted 0.87 [0.78-0.96], p=0.006) and hospital admissions for CHF (757 [20%] vs 918 [24%], p<0.0001) in the candesartan group. There was no significant heterogeneity for candesartan results across the component trials. More patients discontinued candesartan than placebo because of concerns about renal function, hypotension, and hyperkalaemia. INTERPRETATION: Candesartan was generally well tolerated and significantly reduced cardiovascular deaths and hospital admissions for heart failure. Ejection fraction or treatment at baseline did not alter these effects.","ISSN":"1474-547X (Electronic) 0140-6736 (Linking)","title-short":"Effects of candesartan on mortality and morbidity in patients with chronic heart failure: the CHARM-Overall programme","language":"eng","author":[{"family":"Pfeffer","given":"M. A."},{"family":"Swedberg","given":"K."},{"family":"Granger","given":"C. B."},{"family":"Held","given":"P."},{"family":"McMurray","given":"J. J."},{"family":"Michelson","given":"E. L."},{"family":"Olofsson","given":"B."},{"family":"Ostergren","given":"J."},{"family":"Yusuf","given":"S."},{"family":"Pocock","given":"S."}],"issued":{"date-parts":[["2003",9,6]]}}}],"schema":""} (22). To quantify the extent to which this intention-to-treat effect may underestimate the effect of treatment, we could estimate the per-protocol effect. But the per-protocol effect cannot be estimated unless it is explicitly defined. Importantly, the per-protocol effect is not the effect of taking candesartan continuously but rather the effect of adhering continuously to the treatment strategy specified in the protocol. In CHARM, the per-protocol effect is the effect of taking candesartan continuously until the occurrence of abnormal renal function or hypotension. Individuals who discontinued their assigned treatment after experiencing either of these conditions need to be considered adherent to the protocol for the remainder of the trial, despite no longer taking their assigned medication ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"epEYVp4Y","properties":{"formattedCitation":"(23)","plainCitation":"(23)","noteIndex":0},"citationItems":[{"id":1072,"uris":[""],"uri":[""],"itemData":{"id":1072,"type":"article-journal","title":"Adherence-adjustment in placebo-controlled randomized trials: an application to the Candesartan in Heart Failure randomized trial","author":[{"family":"Murray","given":"E. J."},{"family":"Claggett","given":"B."},{"family":"Solomon","given":"S. D."},{"family":"Pfeffer","given":"M. A."},{"family":"Hernan","given":"M. A."}],"issued":{"date-parts":[["2019"]],"season":"submitted"}}}],"schema":""} (23).The trial protocol also specified that treating clinicians could approve the discontinuation or dosage reduction of an individual’s assigned treatment at their discretion based on a perception of medication intolerance ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"kiWpqlun","properties":{"formattedCitation":"(22)","plainCitation":"(22)","noteIndex":0},"citationItems":[{"id":816,"uris":[""],"uri":[""],"itemData":{"id":816,"type":"article-journal","title":"Effects of candesartan on mortality and morbidity in patients with chronic heart failure: the CHARM-Overall programme","container-title":"Lancet","page":"759-66","volume":"362","issue":"9386","source":"Nlm","archive_location":"13678868","abstract":"BACKGROUND: Patients with chronic heart failure (CHF) are at high risk of cardiovascular death and recurrent hospital admissions. We aimed to find out whether the use of an angiotensin-receptor blocker could reduce mortality and morbidity. METHODS: In parallel, randomised, double-blind, controlled, clinical trials we compared candesartan with placebo in three distinct populations. We studied patients with left-ventricular ejection fraction (LVEF) 40% or less who were not receiving angiotensin-converting-enzyme inhibitors because of previous intolerance or who were currently receiving angiotensin-converting-enzyme inhibitors, and patients with LVEF higher than 40%. Overall, 7601 patients (7599 with data) were randomly assigned candesartan (n=3803, titrated to 32 mg once daily) or matching placebo (n=3796), and followed up for at least 2 years. The primary outcome of the overall programme was all-cause mortality, and for all the component trials was cardiovascular death or hospital admission for CHF. Analysis was by intention to treat. FINDINGS: Median follow-up was 37.7 months. 886 (23%) patients in the candesartan and 945 (25%) in the placebo group died (unadjusted hazard ratio 0.91 [95% CI 0.83-1.00], p=0.055; covariate adjusted 0.90 [0.82-0.99], p=0.032), with fewer cardiovascular deaths (691 [18%] vs 769 [20%], unadjusted 0.88 [0.79-0.97], p=0.012; covariate adjusted 0.87 [0.78-0.96], p=0.006) and hospital admissions for CHF (757 [20%] vs 918 [24%], p<0.0001) in the candesartan group. There was no significant heterogeneity for candesartan results across the component trials. More patients discontinued candesartan than placebo because of concerns about renal function, hypotension, and hyperkalaemia. INTERPRETATION: Candesartan was generally well tolerated and significantly reduced cardiovascular deaths and hospital admissions for heart failure. Ejection fraction or treatment at baseline did not alter these effects.","ISSN":"1474-547X (Electronic) 0140-6736 (Linking)","title-short":"Effects of candesartan on mortality and morbidity in patients with chronic heart failure: the CHARM-Overall programme","language":"eng","author":[{"family":"Pfeffer","given":"M. A."},{"family":"Swedberg","given":"K."},{"family":"Granger","given":"C. B."},{"family":"Held","given":"P."},{"family":"McMurray","given":"J. J."},{"family":"Michelson","given":"E. L."},{"family":"Olofsson","given":"B."},{"family":"Ostergren","given":"J."},{"family":"Yusuf","given":"S."},{"family":"Pocock","given":"S."}],"issued":{"date-parts":[["2003",9,6]]}}}],"schema":""} (22). Therefore, one can define an alternate per-protocol effect in which patients are considered to be adherent upon clinician-approved treatment discontinuation ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"OD6jIXXC","properties":{"formattedCitation":"(23)","plainCitation":"(23)","noteIndex":0},"citationItems":[{"id":1072,"uris":[""],"uri":[""],"itemData":{"id":1072,"type":"article-journal","title":"Adherence-adjustment in placebo-controlled randomized trials: an application to the Candesartan in Heart Failure randomized trial","author":[{"family":"Murray","given":"E. J."},{"family":"Claggett","given":"B."},{"family":"Solomon","given":"S. D."},{"family":"Pfeffer","given":"M. A."},{"family":"Hernan","given":"M. A."}],"issued":{"date-parts":[["2019"]],"season":"submitted"}}}],"schema":""} (23).Guideline: To adequately guide decision making by all stakeholders, report estimates of both the intention-to-treat effect and the per-protocol effect, as well as methods and key conditions underlying the estimation procedures.3. Assessment of effect magnitude and heterogeneityMany pragmatic trials compare the risk of developing a health outcome, as opposed to non-binary outcomes like blood pressure or quality of life, under different treatment strategies. In these trials, the effects are often reported as relative risks, or related measures such as hazard ratios. In addition, most trials also report absolute risks, or related measures such as cumulative incidence curves or survival curves, but reporting absolute difference measures is less common ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"GHE1nWJo","properties":{"formattedCitation":"(6)","plainCitation":"(6)","noteIndex":0},"citationItems":[{"id":1156,"uris":[""],"uri":[""],"itemData":{"id":1156,"type":"article-journal","title":"Patients and investigators prefer measures of absolute risk in subgroups for pragmatic randomized trials","container-title":"Journal of Clinical Epidemiology","page":"10-21","volume":"103","source":"PubMed","abstract":"OBJECTIVES: Pragmatic randomized trials are important tools for shared decision-making, but no guidance exists on patients' preferences for types of causal information. We aimed to assess preferences of patients and investigators toward causal effects in pragmatic randomized trials.\nSTUDY DESIGN AND SETTING: We (a) held three focus groups with patients (n?=?23) in Boston, MA; (b) surveyed (n?=?12) and interviewed (n?=?5) investigators with experience conducting pragmatic trials; and (c) conducted a systematic literature review of pragmatic trials (n?=?63).\nRESULTS: Patients were distrustful of new-to-market medications unless substantially more effective than existing choices, preferred stratified absolute risks, and valued adherence-adjusted analyses when they expected to adhere. Investigators wanted both intention-to-treat and per-protocol effects but felt methods for estimating per-protocol effects were lacking. When estimating per-protocol effects, many pragmatic trials used inappropriate methods to adjust for adherence and loss to follow-up.\nCONCLUSION: We made four recommendations for pragmatic trials to improve patient centeredness: (1) focus on superiority in effectiveness or safety, rather than noninferiority; (2) involve patients in specifying a priori subgroups; (3) report absolute measures of risk; and (4) complement intention-to-treat effect estimates with valid per-protocol effect estimates.","DOI":"10.1016/j.jclinepi.2018.06.009","ISSN":"1878-5921","note":"PMID: 29966732\nPMCID: PMC6175611","journalAbbreviation":"J Clin Epidemiol","language":"eng","author":[{"family":"Murray","given":"Eleanor J."},{"family":"Caniglia","given":"Ellen C."},{"family":"Swanson","given":"Sonja A."},{"family":"Hernández-Díaz","given":"Sonia"},{"family":"Hernán","given":"Miguel A."}],"issued":{"date-parts":[["2018",11]]}}}],"schema":""} (6). Relative risks provide information on the direction of the causal effect, but not on the absolute benefit (or harm) of the treatment. For example, a relative risk of 3 for treatment versus no treatment indicates a 3-fold increased risk under treatment compared with no treatment, but carries no information on the absolute increase in risk. If the absolute risk under no treatment were 10%, then a relative risk of 3 would translate into an increased risk of 20 percentage points. In contrast, if the absolute risk under no treatment were 1%, then a relative risk of 3 would translate into an increased risk of only 2 percentage points. Since the primary purpose of pragmatic trials is to guide decision-making for patients and clinicians, absolute risks and their differences on the additive scale should always be presented (Case Study C). Patients prefer absolute measures of occurrence when participating in shared decision-making ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"88HhDuBJ","properties":{"formattedCitation":"(6,24)","plainCitation":"(6,24)","noteIndex":0},"citationItems":[{"id":1156,"uris":[""],"uri":[""],"itemData":{"id":1156,"type":"article-journal","title":"Patients and investigators prefer measures of absolute risk in subgroups for pragmatic randomized trials","container-title":"Journal of Clinical Epidemiology","page":"10-21","volume":"103","source":"PubMed","abstract":"OBJECTIVES: Pragmatic randomized trials are important tools for shared decision-making, but no guidance exists on patients' preferences for types of causal information. We aimed to assess preferences of patients and investigators toward causal effects in pragmatic randomized trials.\nSTUDY DESIGN AND SETTING: We (a) held three focus groups with patients (n?=?23) in Boston, MA; (b) surveyed (n?=?12) and interviewed (n?=?5) investigators with experience conducting pragmatic trials; and (c) conducted a systematic literature review of pragmatic trials (n?=?63).\nRESULTS: Patients were distrustful of new-to-market medications unless substantially more effective than existing choices, preferred stratified absolute risks, and valued adherence-adjusted analyses when they expected to adhere. Investigators wanted both intention-to-treat and per-protocol effects but felt methods for estimating per-protocol effects were lacking. When estimating per-protocol effects, many pragmatic trials used inappropriate methods to adjust for adherence and loss to follow-up.\nCONCLUSION: We made four recommendations for pragmatic trials to improve patient centeredness: (1) focus on superiority in effectiveness or safety, rather than noninferiority; (2) involve patients in specifying a priori subgroups; (3) report absolute measures of risk; and (4) complement intention-to-treat effect estimates with valid per-protocol effect estimates.","DOI":"10.1016/j.jclinepi.2018.06.009","ISSN":"1878-5921","note":"PMID: 29966732\nPMCID: PMC6175611","journalAbbreviation":"J Clin Epidemiol","language":"eng","author":[{"family":"Murray","given":"Eleanor J."},{"family":"Caniglia","given":"Ellen C."},{"family":"Swanson","given":"Sonja A."},{"family":"Hernández-Díaz","given":"Sonia"},{"family":"Hernán","given":"Miguel A."}],"issued":{"date-parts":[["2018",11]]}},"label":"page"},{"id":76,"uris":[""],"uri":[""],"itemData":{"id":76,"type":"article-journal","title":"Medical decision making and the importance of baseline risk","container-title":"Br J Gen Pract","page":"e795-7","volume":"63","issue":"616","archive_location":"24267863","DOI":"10.3399/bjgp13X674585","ISSN":"1478-5242","title-short":"Medical decision making and the importance of baseline risk","language":"eng","author":[{"family":"Stovitz","given":"S. D."},{"family":"Shrier","given":"I."}],"issued":{"date-parts":[["2013",11]]}},"label":"page"}],"schema":""} (6,24).Case Study C: Absolute vs. relative risksCase Study #1 described the intention-to-treat and per-protocol estimates of the effect of postmenopausal hormone therapy on breast cancer. We concluded that women considering the initiation of therapy would be misinformed if provided with the intention-to-treat effect (25% increased risk) but not the per-protocol effect (65% increased risk). However, our reasoning was based on relative risk estimates only. Had we used absolute risks, our conclusion would not have been as strong. The estimated intention-to-treat difference in 8-year risk of breast cancer incidence if all women had been assigned to hormone therapy versus if all had been assigned to placebo was only 0.83 percentage points (95% CI: -0.03, 1.69), while the estimated per-protocol difference in 8-year risk if all women had continuously used hormone therapy versus all women had used no hormone therapy was 1.44 percentage points (95% CI: 0.52, 2.37) ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"exgOEle7","properties":{"formattedCitation":"(21)","plainCitation":"(21)","noteIndex":0},"citationItems":[{"id":1155,"uris":[""],"uri":[""],"itemData":{"id":1155,"type":"article-journal","title":"Estimating absolute risks in the presence of nonadherence: an application to a follow-up study with baseline randomization","container-title":"Epidemiology","page":"528-39","volume":"21","issue":"4","source":"Nlm","archive_location":"20526200","abstract":"The intention-to-treat (ITT) analysis provides a valid test of the null hypothesis and naturally results in both absolute and relative measures of risk. However, this analytic approach may miss the occurrence of serious adverse effects that would have been detected under full adherence to the assigned treatment. Inverse probability weighting of marginal structural models has been used to adjust for nonadherence, but most studies have provided only relative measures of risk. In this study, we used inverse probability weighting to estimate both absolute and relative measures of risk of invasive breast cancer under full adherence to the assigned treatment in the Women's Health Initiative estrogen-plus-progestin trial. In contrast to an ITT hazard ratio (HR) of 1.25 (95% confidence interval [CI] = 1.01 to 1.54), the HR for 8-year continuous estrogen-plus-progestin use versus no use was 1.68 (1.24 to 2.28). The estimated risk difference (cases/100 women) at year 8 was 0.83 (-0.03 to 1.69) in the ITT analysis, compared with 1.44 (0.52 to 2.37) in the adherence-adjusted analysis. Results were robust across various dose-response models. We also compared the dynamic treatment regimen \"take hormone therapy until certain adverse events become apparent, then stop taking hormone therapy\" with no use (HR = 1.64; 95% CI = 1.24 to 2.18). The methods described here are also applicable to observational studies with time-varying treatments.","DOI":"10.1097/EDE.0b013e3181df1b69","ISSN":"1531-5487 (Electronic) 1044-3983 (Linking)","title-short":"Estimating absolute risks in the presence of nonadherence: an application to a follow-up study with baseline randomization","language":"eng","author":[{"family":"Toh","given":"S."},{"family":"Hernandez-Diaz","given":"S."},{"family":"Logan","given":"R."},{"family":"Robins","given":"J. M."},{"family":"Hernán","given":"M. A."}],"issued":{"date-parts":[["2010",7]]}}}],"schema":""} (21). When providing effect estimates on the additive scale, both the intention-to-treat and per-protocol estimates suggest that the 8-year risk increases by less than 1.5 percentage points.The above discussion is especially relevant when a goal of the trial is to quantify whether there is treatment effect heterogeneity, that is, whether the effect varies across subgroups of the study population defined by their baseline characteristics, e.g., men and women, younger and older, diabetics and nondiabetics. Estimating treatment effects in different strata of trial participants is a natural goal of pragmatic trials, since this can improve decision-making. In fact, patients report a high degree of interest in stratum-specific treatment effects that provide them with some guidance on how a medical intervention may work for individuals similar to themselves ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"n3w7Jqyo","properties":{"formattedCitation":"(6)","plainCitation":"(6)","noteIndex":0},"citationItems":[{"id":1156,"uris":[""],"uri":[""],"itemData":{"id":1156,"type":"article-journal","title":"Patients and investigators prefer measures of absolute risk in subgroups for pragmatic randomized trials","container-title":"Journal of Clinical Epidemiology","page":"10-21","volume":"103","source":"PubMed","abstract":"OBJECTIVES: Pragmatic randomized trials are important tools for shared decision-making, but no guidance exists on patients' preferences for types of causal information. We aimed to assess preferences of patients and investigators toward causal effects in pragmatic randomized trials.\nSTUDY DESIGN AND SETTING: We (a) held three focus groups with patients (n?=?23) in Boston, MA; (b) surveyed (n?=?12) and interviewed (n?=?5) investigators with experience conducting pragmatic trials; and (c) conducted a systematic literature review of pragmatic trials (n?=?63).\nRESULTS: Patients were distrustful of new-to-market medications unless substantially more effective than existing choices, preferred stratified absolute risks, and valued adherence-adjusted analyses when they expected to adhere. Investigators wanted both intention-to-treat and per-protocol effects but felt methods for estimating per-protocol effects were lacking. When estimating per-protocol effects, many pragmatic trials used inappropriate methods to adjust for adherence and loss to follow-up.\nCONCLUSION: We made four recommendations for pragmatic trials to improve patient centeredness: (1) focus on superiority in effectiveness or safety, rather than noninferiority; (2) involve patients in specifying a priori subgroups; (3) report absolute measures of risk; and (4) complement intention-to-treat effect estimates with valid per-protocol effect estimates.","DOI":"10.1016/j.jclinepi.2018.06.009","ISSN":"1878-5921","note":"PMID: 29966732\nPMCID: PMC6175611","journalAbbreviation":"J Clin Epidemiol","language":"eng","author":[{"family":"Murray","given":"Eleanor J."},{"family":"Caniglia","given":"Ellen C."},{"family":"Swanson","given":"Sonja A."},{"family":"Hernández-Díaz","given":"Sonia"},{"family":"Hernán","given":"Miguel A."}],"issued":{"date-parts":[["2018",11]]}}}],"schema":""} (6).When a goal is the identification of effect heterogeneity, the trial protocol will typically specify the participants’ strata within which the effects will be estimated. Ideally, clinicians, patients, and other stakeholders will be consulted to pre-specify the subgroups of interest. Then the effect estimates will be obtained separately in each of the subgroups of interest (Case Study D). New methods that allow estimation of subgroup-specific causal effects without requiring pre-specification of the strata of interest are also being developed ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"nXeCQO1V","properties":{"formattedCitation":"(25,26)","plainCitation":"(25,26)","noteIndex":0},"citationItems":[{"id":1054,"uris":[""],"uri":[""],"itemData":{"id":1054,"type":"article-journal","title":"Recursive Partitioning for Heterogeneous Causal Effects","container-title":"arXiv:1504.01132 [econ, stat]","source":"","abstract":"In this paper we study the problems of estimating heterogeneity in causal effects in experimental or observational studies and conducting inference about the magnitude of the differences in treatment effects across subsets of the population. In applications, our method provides a data-driven approach to determine which subpopulations have large or small treatment effects and to test hypotheses about the differences in these effects. For experiments, our method allows researchers to identify heterogeneity in treatment effects that was not specified in a pre-analysis plan, without concern about invalidating inference due to multiple testing. In most of the literature on supervised machine learning (e.g. regression trees, random forests, LASSO, etc.), the goal is to build a model of the relationship between a unit's attributes and an observed outcome. A prominent role in these methods is played by cross-validation which compares predictions to actual outcomes in test samples, in order to select the level of complexity of the model that provides the best predictive power. Our method is closely related, but it differs in that it is tailored for predicting causal effects of a treatment rather than a unit's outcome. The challenge is that the \"ground truth\" for a causal effect is not observed for any individual unit: we observe the unit with the treatment, or without the treatment, but not both at the same time. Thus, it is not obvious how to use cross-validation to determine whether a causal effect has been accurately predicted. We propose several novel cross-validation criteria for this problem and demonstrate through simulations the conditions under which they perform better than standard methods for the problem of causal effects. We then apply the method to a large-scale field experiment re-ranking results on a search engine.","URL":"","note":"arXiv: 1504.01132","author":[{"family":"Athey","given":"Susan"},{"family":"Imbens","given":"Guido"}],"issued":{"date-parts":[["2015",4,5]]},"accessed":{"date-parts":[["2019",3,5]]}},"label":"page"},{"id":1053,"uris":[""],"uri":[""],"itemData":{"id":1053,"type":"article-journal","title":"Estimation and Inference of Heterogeneous Treatment Effects using Random Forests","container-title":"arXiv:1510.04342 [math, stat]","source":"","abstract":"Many scientific and engineering challenges -- ranging from personalized medicine to customized marketing recommendations -- require an understanding of treatment effect heterogeneity. In this paper, we develop a non-parametric causal forest for estimating heterogeneous treatment effects that extends Breiman's widely used random forest algorithm. In the potential outcomes framework with unconfoundedness, we show that causal forests are pointwise consistent for the true treatment effect, and have an asymptotically Gaussian and centered sampling distribution. We also discuss a practical method for constructing asymptotic confidence intervals for the true treatment effect that are centered at the causal forest estimates. Our theoretical results rely on a generic Gaussian theory for a large family of random forest algorithms. To our knowledge, this is the first set of results that allows any type of random forest, including classification and regression forests, to be used for provably valid statistical inference. In experiments, we find causal forests to be substantially more powerful than classical methods based on nearest-neighbor matching, especially in the presence of irrelevant covariates.","URL":"","note":"arXiv: 1510.04342","author":[{"family":"Wager","given":"Stefan"},{"family":"Athey","given":"Susan"}],"issued":{"date-parts":[["2015",10,14]]},"accessed":{"date-parts":[["2019",3,5]]}},"label":"page"}],"schema":""} (25,26). Importantly, when the goal is to identify the patient subgroups that are more likely to benefit (or be harmed) by treatment, the effect estimates must be measured on the additive, not the multiplicative, scale ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"NVjG4yZY","properties":{"formattedCitation":"(27)","plainCitation":"(27)","noteIndex":0},"citationItems":[{"id":1062,"uris":[""],"uri":[""],"itemData":{"id":1062,"type":"article-journal","title":"Concepts of interaction","container-title":"American Journal of Epidemiology","page":"467-470","volume":"112","issue":"4","source":"PubMed","ISSN":"0002-9262","note":"PMID: 7424895","journalAbbreviation":"Am. J. Epidemiol.","language":"eng","author":[{"family":"Rothman","given":"K. J."},{"family":"Greenland","given":"S."},{"family":"Walker","given":"A. M."}],"issued":{"date-parts":[["1980",10]]}}}],"schema":""} (27). For example, the comparison of risk differences, not risk ratios, across subgroups is the correct procedure in this setting. A final comment. though most trials already present comparison of absolute risks to quantify the intention-to-treat effect, it is important to bear in mind these considerations also when estimating per-protocol effects. The valid estimation of per-protocol effects typically requires adjustment for prognostic factors, and some researchers seem to be under the impression that adjusted analyses can only provide relative risk measures. On the contrary, as discussed below, adjusted analyses can yield adequately standardized absolute risks and risk differences. The choice of the scale (additive or multiplicative) on which the effect is measured does not dictate the analytic approach ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"CXWw0jrG","properties":{"formattedCitation":"(28)","plainCitation":"(28)","noteIndex":0},"citationItems":[{"id":561,"uris":[""],"uri":[""],"itemData":{"id":561,"type":"article-journal","title":"The hazards of hazard ratios","container-title":"Epidemiology","page":"13-5","volume":"21","issue":"1","archive_location":"20010207","DOI":"10.1097/EDE.0b013e3181c1ea43","ISSN":"1531-5487","title-short":"The hazards of hazard ratios","language":"eng","author":[{"family":"Hernán","given":"M. A."}],"issued":{"date-parts":[["2010",1]]}}}],"schema":""} (28). Case Study D: Heterogeneity of treatment effectsIn the Women’s Health Initiative (WHI) estrogen plus progestin trial (Case Study #1), the intention-to-treat analysis was conducted in 23 pre-specified subgroups ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"93GbM3AE","properties":{"formattedCitation":"(29)","plainCitation":"(29)","noteIndex":0},"citationItems":[{"id":928,"uris":[""],"uri":[""],"itemData":{"id":928,"type":"article-journal","title":"Estrogen plus progestin and the risk of coronary heart disease","container-title":"N Engl J Med","page":"523-34","volume":"349","issue":"6","source":"NLM","archive_location":"12904517","abstract":"BACKGROUND: Recent randomized clinical trials have suggested that estrogen plus progestin does not confer cardiac protection and may increase the risk of coronary heart disease (CHD). In this report, we provide the final results with regard to estrogen plus progestin and CHD from the Women's Health Initiative (WHI). METHODS: The WHI included a randomized primary-prevention trial of estrogen plus progestin in 16,608 postmenopausal women who were 50 to 79 years of age at base line. Participants were randomly assigned to receive conjugated equine estrogens (0.625 mg per day) plus medroxyprogesterone acetate (2.5 mg per day) or placebo. The primary efficacy outcome of the trial was CHD (nonfatal myocardial infarction or death due to CHD). RESULTS: After a mean follow-up of 5.2 years (planned duration, 8.5 years), the data and safety monitoring board recommended terminating the estrogen-plus-progestin trial because the overall risks exceeded the benefits. Combined hormone therapy was associated with a hazard ratio for CHD of 1.24 (nominal 95 percent confidence interval, 1.00 to 1.54; 95 percent confidence interval after adjustment for sequential monitoring, 0.97 to 1.60). The elevation in risk was most apparent at one year (hazard ratio, 1.81 [95 percent confidence interval, 1.09 to 3.01]). Although higher base-line levels of low-density lipoprotein cholesterol were associated with an excess risk of CHD among women who received hormone therapy, higher base-line levels of C-reactive protein, other biomarkers, and other clinical characteristics did not significantly modify the treatment-related risk of CHD. CONCLUSIONS: Estrogen plus progestin does not confer cardiac protection and may increase the risk of CHD among generally healthy postmenopausal women, especially during the first year after the initiation of hormone use. This treatment should not be prescribed for the prevention of cardiovascular disease.","DOI":"10.1056/NEJMoa030808","ISSN":"0028-4793","title-short":"Estrogen plus progestin and the risk of coronary heart disease","journalAbbreviation":"The New England journal of medicine","language":"eng","author":[{"family":"Manson","given":"J. E."},{"family":"Hsia","given":"J."},{"family":"Johnson","given":"K. C."},{"family":"Rossouw","given":"J. E."},{"family":"Assaf","given":"A. R."},{"family":"Lasser","given":"N. L."},{"family":"Trevisan","given":"M."},{"family":"Black","given":"H. R."},{"family":"Heckbert","given":"S. R."},{"family":"Detrano","given":"R."},{"family":"Strickland","given":"O. L."},{"family":"Wong","given":"N. D."},{"family":"Crouse","given":"J. R."},{"family":"Stein","given":"E."},{"family":"Cushman","given":"M."}],"issued":{"date-parts":[["2003",8,7]]}}}],"schema":""} (29) defined by demographic and clinical characteristics for which heterogeneity of treatment effects was expected.For example, the intention-to-treat odds ratio of coronary heart disease for hormone therapy versus placebo ranged from 0.76 among women with low cholesterol at baseline to 2.03 among women with high cholesterol at baseline. Unfortunately, absolute risks or rates were not presented for these sub-groups so heterogeneity could not be evaluated on the additive scale. Guideline: Report absolute risks and their differences, as well as their ratios, for discrete outcomes. Heterogeneity of treatment effects can be reported using subgroup analyses that use the additive scale to measure the effect of interest. Patients and advocates should be included in a priori specification of subgroups.4. Estimation of the intention-to-treat effectWe refer to the analyses aimed at estimating the intention-to-treat effect as intention-to-treat analyses. In a large pragmatic trial with complete follow-up and no competing events for the outcome, the intention-to-treat analysis is straightforward: compare the observed outcome distribution between trial arms. For example, in a trial of aspirin and mortality, the intention-to-treat risk ratio at 5 years is the ratio of the observed 5-year mortality risk in those assigned to daily aspirin divided by the 5-year mortality risk in those assigned to no aspirin. That is, in this setting the intention-to-treat effect can be validly estimated without adjustment for covariates. We now review three settings in which analyses to estimate the intention-to-treat effect require additional measures—imbalanced prognostic factors at baseline, competing events, and losses to follow-up—and provide guidelines. These guidelines also apply to the estimation of per-protocol effects (which is discussed in Sections 5 and 6).4.1 Imbalanced prognostic factors at baselineRandomization cannot ensure that the distribution of all prognostic factors is identical between the trial arms. For example, in an aspirin trial the proportion of smokers may be, just by chance, greater in the group assigned to aspirin than in the group assigned to no aspirin. As a result of these chance imbalances, the effect estimate may be far from the true effect. Some authors refer to the difference between a study-specific effect estimate and the true effect as “random confounding” because, in a particular study, the problem is indistinguishable from the systematic confounding that results from systematic imbalances of prognostic factors ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"jZhcrY2z","properties":{"formattedCitation":"(30)","plainCitation":"(30)","noteIndex":0},"citationItems":[{"id":1157,"uris":[""],"uri":[""],"itemData":{"id":1157,"type":"article-journal","title":"Limitations of individual causal models, causal graphs, and ignorability assumptions, as illustrated by random confounding and design unfaithfulness","container-title":"European Journal of Epidemiology","page":"1101-1110","volume":"30","issue":"10","source":"PubMed","abstract":"We describe how ordinary interpretations of causal models and causal graphs fail to capture important distinctions among ignorable allocation mechanisms for subject selection or allocation. We illustrate these limitations in the case of random confounding and designs that prevent such confounding. In many experimental designs individual treatment allocations are dependent, and explicit population models are needed to show this dependency. In particular, certain designs impose unfaithful covariate-treatment distributions to prevent random confounding, yet ordinary causal graphs cannot discriminate between these unconfounded designs and confounded studies. Causal models for populations are better suited for displaying these phenomena than are individual-level models, because they allow representation of allocation dependencies as well as outcome dependencies across individuals. Nonetheless, even with this extension, ordinary graphical models still fail to capture distinctions between hypothetical superpopulations (sampling distributions) and observed populations (actual distributions), although potential-outcome models can be adapted to show these distinctions and their consequences.","DOI":"10.1007/s10654-015-9995-7","ISSN":"1573-7284","note":"PMID: 25687168","journalAbbreviation":"Eur. J. Epidemiol.","language":"eng","author":[{"family":"Greenland","given":"Sander"},{"family":"Mansournia","given":"Mohammad Ali"}],"issued":{"date-parts":[["2015",10]]}}}],"schema":""} (30). Practically speaking, it is advisable to deal with any imbalances in intention-to-treat analyses, regardless of whether they are random or systematic, via adjustment for the imbalanced prognostic factors ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"3zOhjs7s","properties":{"formattedCitation":"(31)","plainCitation":"(31)","noteIndex":0},"citationItems":[{"id":509,"uris":[""],"uri":[""],"itemData":{"id":509,"type":"chapter","title":"Chapter 10: Random variability","container-title":"Causal Inference","publisher":"Chapman & Hall/CRC, forthcoming.","publisher-place":"Boca Raton","event-place":"Boca Raton","title-short":"Chapter 10: Random variability","author":[{"family":"Hernan","given":"MA"},{"family":"Robins","given":"JM"}],"issued":{"date-parts":[["2018"]]}}}],"schema":""} (31). Of course, in well conducted randomized trials, only random and not systematic imbalances are expected.How to best choose the prognostic factors that need to be adjusted for in intention-to-treat analyses remains debated. While there is consensus that the use of significance testing is strongly discouraged ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"Ez0JvKIb","properties":{"formattedCitation":"(32)","plainCitation":"(32)","noteIndex":0},"citationItems":[{"id":327,"uris":[""],"uri":[""],"itemData":{"id":327,"type":"article-journal","title":"Comparability of Randomised Groups","container-title":"Journal of the Royal Statististical Society. Series D","page":"125--36","volume":"34","issue":"1","title-short":"Comparability of Randomised Groups","author":[{"family":"Altman","given":"DG"}],"issued":{"date-parts":[["1985"]]}}}],"schema":""} (32), several other options exist. One option is to pre-specify important prognostic factors for adjustment (Case Study E1). Another possibility is to pre-specify a maximum acceptable imbalance which, if exceeded, will trigger adjustment for that factor. Pre-specification of a maximum acceptable imbalance could be based on discussion with subject matter experts about the degree of confounding likely in an observational study, or estimated using an e-value calculation based on assumptions about likely unadjusted intention-to-treat estimates ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"oSw71LK3","properties":{"formattedCitation":"(33)","plainCitation":"(33)","noteIndex":0},"citationItems":[{"id":1271,"uris":[""],"uri":[""],"itemData":{"id":1271,"type":"article-journal","title":"Sensitivity analysis in observational research: Introducing the e-value","container-title":"Annals of Internal Medicine","page":"268-274","volume":"167","issue":"4","abstract":"Sensitivity analysis is useful in assessing how robust an association is to potential unmeasured or uncontrolled confounding. This article introduces a new measure called the “E-value,” which is related to the evidence for causality in observational studies that are potentially subject to confounding. The E-value is defined as the minimum strength of association, on the risk ratio scale, that an unmeasured confounder would need to have with both the treatment and the outcome to fully explain away a specific treatment–outcome association, conditional on the measured covariates. A large E-value implies that considerable unmeasured confounding would be needed to explain away an effect estimate. A small E-value implies little unmeasured confounding would be needed to explain away an effect estimate. The authors propose that in all observational studies intended to produce evidence for causality, the E-value be reported or some other sensitivity analysis be used. They suggest calculating the E-value for both the observed association estimate (after adjustments for measured confounders) and the limit of the confidence interval closest to the null. If this were to become standard practice, the ability of the scientific community to assess evidence from observational studies would improve considerably, and ultimately, science would be strengthened.","DOI":"10.7326/m16-2607","ISSN":"0003-4819","title-short":"Sensitivity analysis in observational research: Introducing the e-value","author":[{"family":"VanderWeele","given":"T. J."},{"family":"Ding","given":"P."}],"issued":{"date-parts":[["2017"]]}}}],"schema":""} (33). Yet neither option provides guidance about what to do when a large imbalance is found for a strong prognostic factor that was not pre-specified. In this case, the recommended course of action is to adjust for that factor ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"ifaXiR9w","properties":{"formattedCitation":"(31,34)","plainCitation":"(31,34)","noteIndex":0},"citationItems":[{"id":509,"uris":[""],"uri":[""],"itemData":{"id":509,"type":"chapter","title":"Chapter 10: Random variability","container-title":"Causal Inference","publisher":"Chapman & Hall/CRC, forthcoming.","publisher-place":"Boca Raton","event-place":"Boca Raton","title-short":"Chapter 10: Random variability","author":[{"family":"Hernan","given":"MA"},{"family":"Robins","given":"JM"}],"issued":{"date-parts":[["2018"]]}},"label":"page"},{"id":1146,"uris":[""],"uri":[""],"itemData":{"id":1146,"type":"article-journal","title":"Confidence intervals for causal parameters","container-title":"Statistics in Medicine","page":"773-785","volume":"7","issue":"7","source":"Wiley Online Library","abstract":"Consider an unbiased follow-up study designed to investigate the causal effect of a dichotomous exposure on a dichotomous disease outcome. Under a deterministic outcome model, a standard ‘95 per cent binomial confidence interval’ may fail to cover the causal parameter of interest at the nominal rate when we take the causal parameter to be a parameter associated with the observed study population (regardless of whether the observed study population was sampled from a larger superpopulation). I propose new interval estimators that, in this setting, improve upon the performance of the standard ‘binomial confidence interval.’","DOI":"10.1002/sim.4780070707","ISSN":"1097-0258","language":"en","author":[{"family":"Robins","given":"James M."}],"issued":{"date-parts":[["1988",7,1]]}},"label":"page"}],"schema":""} (31,34), even if only as a sensitivity analysis (Case Studies E2, E3). However, such adjustment raises concerns about the validity of the statistical inferences. A principled approach to adjustment that preserves the inferential properties of the statistical analysis would need to be based on doubly robust procedures ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"2Jbjx2LO","properties":{"formattedCitation":"(35\\uc0\\u8211{}37)","plainCitation":"(35–37)","noteIndex":0},"citationItems":[{"id":2683,"uris":[""],"uri":[""],"itemData":{"id":2683,"type":"article-journal","title":"Covariate adjustment for two-sample treatment comparisons in randomized clinical trials: a principled yet flexible approach","container-title":"Statistics in Medicine","page":"4658-4677","volume":"27","issue":"23","source":"PubMed","abstract":"There is considerable debate regarding whether and how covariate-adjusted analyses should be used in the comparison of treatments in randomized clinical trials. Substantial baseline covariate information is routinely collected in such trials, and one goal of adjustment is to exploit covariates associated with outcome to increase precision of estimation of the treatment effect. However, concerns are routinely raised over the potential for bias when the covariates used are selected post hoc and the potential for adjustment based on a model of the relationship between outcome, covariates, and treatment to invite a 'fishing expedition' for that leading to the most dramatic effect estimate. By appealing to the theory of semiparametrics, we are led naturally to a characterization of all treatment effect estimators and to principled, practically feasible methods for covariate adjustment that yield the desired gains in efficiency and that allow covariate relationships to be identified and exploited while circumventing the usual concerns. The methods and strategies for their implementation in practice are presented. Simulation studies and an application to data from an HIV clinical trial demonstrate the performance of the techniques relative to the existing methods.","DOI":"10.1002/sim.3113","ISSN":"0277-6715","note":"PMID: 17960577\nPMCID: PMC2562926","title-short":"Covariate adjustment for two-sample treatment comparisons in randomized clinical trials","journalAbbreviation":"Stat Med","language":"eng","author":[{"family":"Tsiatis","given":"Anastasios A."},{"family":"Davidian","given":"Marie"},{"family":"Zhang","given":"Min"},{"family":"Lu","given":"Xiaomin"}],"issued":{"date-parts":[["2008",10,15]]}}},{"id":2686,"uris":[""],"uri":[""],"itemData":{"id":2686,"type":"article-journal","title":"Improving efficiency of inferences in randomized clinical trials using auxiliary covariates","container-title":"Biometrics","page":"707-715","volume":"64","issue":"3","source":"PubMed","abstract":"The primary goal of a randomized clinical trial is to make comparisons among two or more treatments. For example, in a two-arm trial with continuous response, the focus may be on the difference in treatment means; with more than two treatments, the comparison may be based on pairwise differences. With binary outcomes, pairwise odds ratios or log odds ratios may be used. In general, comparisons may be based on meaningful parameters in a relevant statistical model. Standard analyses for estimation and testing in this context typically are based on the data collected on response and treatment assignment only. In many trials, auxiliary baseline covariate information may also be available, and it is of interest to exploit these data to improve the efficiency of inferences. Taking a semiparametric theory perspective, we propose a broadly applicable approach to adjustment for auxiliary covariates to achieve more efficient estimators and tests for treatment parameters in the analysis of randomized clinical trials. Simulations and applications demonstrate the performance of the methods.","DOI":"10.1111/j.1541-0420.2007.00976.x","ISSN":"1541-0420","note":"PMID: 18190618\nPMCID: PMC2574960","journalAbbreviation":"Biometrics","language":"eng","author":[{"family":"Zhang","given":"Min"},{"family":"Tsiatis","given":"Anastasios A."},{"family":"Davidian","given":"Marie"}],"issued":{"date-parts":[["2008",9]]}}},{"id":2689,"uris":[""],"uri":[""],"itemData":{"id":2689,"type":"article-journal","title":"Covariate adjustment in randomized trials with binary outcomes: targeted maximum likelihood estimation","container-title":"Statistics in Medicine","page":"39-64","volume":"28","issue":"1","source":"PubMed","abstract":"Covariate adjustment using linear models for continuous outcomes in randomized trials has been shown to increase efficiency and power over the unadjusted method in estimating the marginal effect of treatment. However, for binary outcomes, investigators generally rely on the unadjusted estimate as the literature indicates that covariate-adjusted estimates based on the logistic regression models are less efficient. The crucial step that has been missing when adjusting for covariates is that one must integrate/average the adjusted estimate over those covariates in order to obtain the marginal effect. We apply the method of targeted maximum likelihood estimation (tMLE) to obtain estimators for the marginal effect using covariate adjustment for binary outcomes. We show that the covariate adjustment in randomized trials using the logistic regression models can be mapped, by averaging over the covariate(s), to obtain a fully robust and efficient estimator of the marginal effect, which equals a targeted maximum likelihood estimator. This tMLE is obtained by simply adding a clever covariate to a fixed initial regression. We present simulation studies that demonstrate that this tMLE increases efficiency and power over the unadjusted method, particularly for smaller sample sizes, even when the regression model is mis-specified.","DOI":"10.1002/sim.3445","ISSN":"0277-6715","note":"PMID: 18985634\nPMCID: PMC2857590","title-short":"Covariate adjustment in randomized trials with binary outcomes","journalAbbreviation":"Stat Med","language":"eng","author":[{"family":"Moore","given":"K. L."},{"family":"Laan","given":"M. J.","non-dropping-particle":"van der"}],"issued":{"date-parts":[["2009",1,15]]}}}],"schema":""} (35–37). However, software for doubly robust methods is not yet readily applicable to all the settings (e.g., survival analysis with time-varying treatments and covariates) that investigators may encounter when designing and analyzing pragmatic trials.To preserve the marginal (unconditional) interpretation of intention-to-treat effect estimates in the study population, adjustment for baseline covariates is better carried out via standardization, inverse probability weighting (Case Study E1) or, preferably, doubly-robust methods ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"npUUil1D","properties":{"formattedCitation":"(11,38,39)","plainCitation":"(11,38,39)","noteIndex":0},"citationItems":[{"id":511,"uris":[""],"uri":[""],"itemData":{"id":511,"type":"book","title":"Causal Inference: What If","publisher":"Chapman & Hill/CRC","publisher-place":"Boca Raton","event-place":"Boca Raton","title-short":"Causal Inference","author":[{"family":"Hernan","given":"M.A."},{"family":"Robins","given":"J."}],"issued":{"date-parts":[["2020"]]}},"label":"page"},{"id":1161,"uris":[""],"uri":[""],"itemData":{"id":1161,"type":"article-journal","title":"Estimating causal effects from epidemiological data","container-title":"Journal of Epidemiology and Community Health","page":"578-586","volume":"60","issue":"7","archive":"Pmc","archive_location":"PMC2652882","abstract":"In ideal randomised experiments, association is causation: association measures can be interpreted as effect measures because randomisation ensures that the exposed and the unexposed are exchangeable. On the other hand, in observational studies, association is not generally causation: association measures cannot be interpreted as effect measures because the exposed and the unexposed are not generally exchangeable. However, observational research is often the only alternative for causal inference. This article reviews a condition that permits the estimation of causal effects from observational data, and two methods—standardisation and inverse probability weighting—to estimate population causal effects under that condition. For simplicity, the main description is restricted to dichotomous variables and assumes that no random error attributable to sampling variability exists. The appendix provides a generalisation of inverse probability weighting.","DOI":"10.1136/jech.2004.029496","ISSN":"0143-005X 1470-2738","title-short":"Estimating causal effects from epidemiological data","journalAbbreviation":"J Epidemiol Community Health","author":[{"family":"Hernán","given":"Miguel A."},{"family":"Robins","given":"James M."}],"issued":{"date-parts":[["2006"]],"season":"accepted"}},"label":"page"},{"id":769,"uris":[""],"uri":[""],"itemData":{"id":769,"type":"article-journal","title":"Marginal structural models and causal inference in epidemiology","container-title":"Epidemiology","page":"550-60","volume":"11","issue":"5","archive_location":"10955408","abstract":"In observational studies with exposures or treatments that vary over time, standard approaches for adjustment of confounding are biased when there exist time-dependent confounders that are also affected by previous treatment. This paper introduces marginal structural models, a new class of causal models that allow for improved adjustment of confounding in those situations. The parameters of a marginal structural model can be consistently estimated using a new class of estimators, the inverse-probability-of-treatment weighted estimators.","ISSN":"1044-3983","title-short":"Marginal structural models and causal inference in epidemiology","language":"eng","author":[{"family":"Robins","given":"J. M."},{"family":"Hernán","given":"M. A."},{"family":"Brumback","given":"B."}],"issued":{"date-parts":[["2000",9]]}},"label":"page"}],"schema":""} (11,38,39). Adjustment methods like outcome regression or propensity score analysis may result in more precise intention-to-treat effect estimates ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"lo7QjboL","properties":{"formattedCitation":"(40\\uc0\\u8211{}42)","plainCitation":"(40–42)","noteIndex":0},"citationItems":[{"id":706,"uris":[""],"uri":[""],"itemData":{"id":706,"type":"article-journal","title":"Covariate imbalance and random allocation in clinical trials","container-title":"Statistics in Medicine","page":"467-75","volume":"8","DOI":"10.1002/sim.4780080410","title-short":"Covariate imbalance and random allocation in clinical trials","author":[{"family":"Senn","given":"Stephen"}],"issued":{"date-parts":[["1989"]]}},"label":"page"},{"id":705,"uris":[""],"uri":[""],"itemData":{"id":705,"type":"article-journal","title":"Testing for baseline balance in clinical trials","container-title":"Statistics in Medicine","page":"1715-26","volume":"13","DOI":"10.1002/sim.4780131703","title-short":"Testing for baseline balance in clinical trials","author":[{"family":"Senn","given":"Stephen"}],"issued":{"date-parts":[["1994"]]}},"label":"page"},{"id":703,"uris":[""],"uri":[""],"itemData":{"id":703,"type":"article-journal","title":"Seven myths of randomisation in clinical trials","container-title":"Statistics in Medicine","page":"1439-50","volume":"32","DOI":"10.1002/sim.5713","title-short":"Seven myths of randomisation in clinical trials","author":[{"family":"Senn","given":"Stephen"}],"issued":{"date-parts":[["2013"]]}}}],"schema":""} (40–42) but the validity of these estimates rely on the assumption of effect homogeneity across levels of the covariates. Case study E1: Adjusting for pre-specified baseline prognostic factorsIn the Candesartan in Heart Failure: Morbidity and Mortality (CHARM) trial, a list of important baseline prognostic factors for all-cause mortality among individuals with heart failure was pre-specified, including heart disease risk factors, medical history, and use of medical treatments. Table 1 described the distribution of these prognostic factors separately in the candesartan and placebo arms. The unadjusted intention-to-treat estimate of the effect of randomization to candesartan versus placebo on all-cause mortality was a hazard ratio of 0.91 (95% CI: 0.83, 1.00) ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"TdnCUA7g","properties":{"formattedCitation":"(22)","plainCitation":"(22)","noteIndex":0},"citationItems":[{"id":816,"uris":[""],"uri":[""],"itemData":{"id":816,"type":"article-journal","title":"Effects of candesartan on mortality and morbidity in patients with chronic heart failure: the CHARM-Overall programme","container-title":"Lancet","page":"759-66","volume":"362","issue":"9386","source":"Nlm","archive_location":"13678868","abstract":"BACKGROUND: Patients with chronic heart failure (CHF) are at high risk of cardiovascular death and recurrent hospital admissions. We aimed to find out whether the use of an angiotensin-receptor blocker could reduce mortality and morbidity. METHODS: In parallel, randomised, double-blind, controlled, clinical trials we compared candesartan with placebo in three distinct populations. We studied patients with left-ventricular ejection fraction (LVEF) 40% or less who were not receiving angiotensin-converting-enzyme inhibitors because of previous intolerance or who were currently receiving angiotensin-converting-enzyme inhibitors, and patients with LVEF higher than 40%. Overall, 7601 patients (7599 with data) were randomly assigned candesartan (n=3803, titrated to 32 mg once daily) or matching placebo (n=3796), and followed up for at least 2 years. The primary outcome of the overall programme was all-cause mortality, and for all the component trials was cardiovascular death or hospital admission for CHF. Analysis was by intention to treat. FINDINGS: Median follow-up was 37.7 months. 886 (23%) patients in the candesartan and 945 (25%) in the placebo group died (unadjusted hazard ratio 0.91 [95% CI 0.83-1.00], p=0.055; covariate adjusted 0.90 [0.82-0.99], p=0.032), with fewer cardiovascular deaths (691 [18%] vs 769 [20%], unadjusted 0.88 [0.79-0.97], p=0.012; covariate adjusted 0.87 [0.78-0.96], p=0.006) and hospital admissions for CHF (757 [20%] vs 918 [24%], p<0.0001) in the candesartan group. There was no significant heterogeneity for candesartan results across the component trials. More patients discontinued candesartan than placebo because of concerns about renal function, hypotension, and hyperkalaemia. INTERPRETATION: Candesartan was generally well tolerated and significantly reduced cardiovascular deaths and hospital admissions for heart failure. Ejection fraction or treatment at baseline did not alter these effects.","ISSN":"1474-547X (Electronic) 0140-6736 (Linking)","title-short":"Effects of candesartan on mortality and morbidity in patients with chronic heart failure: the CHARM-Overall programme","language":"eng","author":[{"family":"Pfeffer","given":"M. A."},{"family":"Swedberg","given":"K."},{"family":"Granger","given":"C. B."},{"family":"Held","given":"P."},{"family":"McMurray","given":"J. J."},{"family":"Michelson","given":"E. L."},{"family":"Olofsson","given":"B."},{"family":"Ostergren","given":"J."},{"family":"Yusuf","given":"S."},{"family":"Pocock","given":"S."}],"issued":{"date-parts":[["2003",9,6]]}}}],"schema":""} (22). After adjustment for all pre-specified baseline prognostic factors, the estimated effect of randomization to candesartan versus placebo on all-cause mortality was a hazard ratio of 0.90 (95% CI: 0.82, 0.99) ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"ZeQw3hca","properties":{"formattedCitation":"(22)","plainCitation":"(22)","noteIndex":0},"citationItems":[{"id":816,"uris":[""],"uri":[""],"itemData":{"id":816,"type":"article-journal","title":"Effects of candesartan on mortality and morbidity in patients with chronic heart failure: the CHARM-Overall programme","container-title":"Lancet","page":"759-66","volume":"362","issue":"9386","source":"Nlm","archive_location":"13678868","abstract":"BACKGROUND: Patients with chronic heart failure (CHF) are at high risk of cardiovascular death and recurrent hospital admissions. We aimed to find out whether the use of an angiotensin-receptor blocker could reduce mortality and morbidity. METHODS: In parallel, randomised, double-blind, controlled, clinical trials we compared candesartan with placebo in three distinct populations. We studied patients with left-ventricular ejection fraction (LVEF) 40% or less who were not receiving angiotensin-converting-enzyme inhibitors because of previous intolerance or who were currently receiving angiotensin-converting-enzyme inhibitors, and patients with LVEF higher than 40%. Overall, 7601 patients (7599 with data) were randomly assigned candesartan (n=3803, titrated to 32 mg once daily) or matching placebo (n=3796), and followed up for at least 2 years. The primary outcome of the overall programme was all-cause mortality, and for all the component trials was cardiovascular death or hospital admission for CHF. Analysis was by intention to treat. FINDINGS: Median follow-up was 37.7 months. 886 (23%) patients in the candesartan and 945 (25%) in the placebo group died (unadjusted hazard ratio 0.91 [95% CI 0.83-1.00], p=0.055; covariate adjusted 0.90 [0.82-0.99], p=0.032), with fewer cardiovascular deaths (691 [18%] vs 769 [20%], unadjusted 0.88 [0.79-0.97], p=0.012; covariate adjusted 0.87 [0.78-0.96], p=0.006) and hospital admissions for CHF (757 [20%] vs 918 [24%], p<0.0001) in the candesartan group. There was no significant heterogeneity for candesartan results across the component trials. More patients discontinued candesartan than placebo because of concerns about renal function, hypotension, and hyperkalaemia. INTERPRETATION: Candesartan was generally well tolerated and significantly reduced cardiovascular deaths and hospital admissions for heart failure. Ejection fraction or treatment at baseline did not alter these effects.","ISSN":"1474-547X (Electronic) 0140-6736 (Linking)","title-short":"Effects of candesartan on mortality and morbidity in patients with chronic heart failure: the CHARM-Overall programme","language":"eng","author":[{"family":"Pfeffer","given":"M. A."},{"family":"Swedberg","given":"K."},{"family":"Granger","given":"C. B."},{"family":"Held","given":"P."},{"family":"McMurray","given":"J. J."},{"family":"Michelson","given":"E. L."},{"family":"Olofsson","given":"B."},{"family":"Ostergren","given":"J."},{"family":"Yusuf","given":"S."},{"family":"Pocock","given":"S."}],"issued":{"date-parts":[["2003",9,6]]}}}],"schema":""} (22). In a secondary analysis of the CHARM data, we also estimated the intention-to-treat effect standardized across these baseline prognostic factors to improve interpretation – standardization allows the interpretation as an average effect among the trial population, rather than as a conditional effect “holding covariates constant”. The average intention-to-treat effect of randomization to candesartan versus placebo was 0.89 (95% CI: 0.82, 0.97) ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"cVa2mxNl","properties":{"formattedCitation":"(23)","plainCitation":"(23)","noteIndex":0},"citationItems":[{"id":1072,"uris":[""],"uri":[""],"itemData":{"id":1072,"type":"article-journal","title":"Adherence-adjustment in placebo-controlled randomized trials: an application to the Candesartan in Heart Failure randomized trial","author":[{"family":"Murray","given":"E. J."},{"family":"Claggett","given":"B."},{"family":"Solomon","given":"S. D."},{"family":"Pfeffer","given":"M. A."},{"family":"Hernan","given":"M. A."}],"issued":{"date-parts":[["2019"]],"season":"submitted"}}}],"schema":""} (23). The absence of material differences across of all these estimates is expected in randomized trials which, like CHARM, have small imbalances in baseline covariates. Case study E2: Adjusting for imbalanced baseline prognostic factors The Norwegian Colorectal Cancer Prevention (NORCCAP) trial was a primary prevention trial for colorectal cancer incidence and mortality comparing a one-time screening sigmoidoscopy with no screening ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"Juu7SHdZ","properties":{"formattedCitation":"(18)","plainCitation":"(18)","noteIndex":0},"citationItems":[{"id":1026,"uris":[""],"uri":[""],"itemData":{"id":1026,"type":"article-journal","title":"Long-Term Effectiveness of Sigmoidoscopy Screening on Colorectal Cancer Incidence and Mortality in Women and Men: A Randomized Trial","container-title":"Ann Intern Med","page":"775-782","volume":"168","issue":"11","archive_location":"29710125","abstract":"The long-term effects of sigmoidoscopy screening on colorectal cancer (CRC) incidence and mortality in women and men are unclear.|To determine the effectiveness of flexible sigmoidoscopy screening after 15 years of follow-up in women and men.|Randomized controlled trial. (: NCT00119912).|Oslo and Telemark County, Norway.|Adults aged 50 to 64 years at baseline without prior CRC.|Screening (between 1999 and 2001) with flexible sigmoidoscopy with and without additional fecal blood testing versus no screening. Participants with positive screening results were offered colonoscopy.|Age-adjusted CRC incidence and mortality stratified by sex.|Of 98?678 persons, 20?552 were randomly assigned to screening and 78?126 to no screening. Adherence rates were 64.7% in women and 61.4% in men. Median follow-up was 14.8 years. The absolute risks for CRC in women were 1.86% in the screening group and 2.05% in the control group (risk difference, -0.19 percentage point [95% CI, -0.49 to 0.11 percentage point]; HR, 0.92 [CI, 0.79 to 1.07]). In men, the corresponding risks were 1.72% and 2.50%, respectively (risk difference, -0.78 percentage point [CI, -1.08 to -0.48 percentage points]; hazard ratio [HR], 0.66 [CI, 0.57 to 0.78]) (P for heterogeneity?= 0.004). The absolute risks for death from CRC in women were 0.60% in the screening group and 0.59% in the control group (risk difference, 0.01 percentage point [CI, -0.16 to 0.18 percentage point]; HR, 1.01 [CI, 0.77 to 1.33]). The corresponding risks for death from CRC in men were 0.49% and 0.81%, respectively (risk difference, -0.33 percentage point [CI, -0.49 to -0.16 percentage point]; HR, 0.63 [CI, 0.47 to 0.83]) (P for heterogeneity?= 0.014).|Follow-up through national registries.|Offering sigmoidoscopy screening in Norway reduced CRC incidence and mortality in men but had little or no effect in women.|Norwegian government and Norwegian Cancer Society.","DOI":"10.7326/M17-1441","ISSN":"1539-3704","title-short":"Long-Term Effectiveness of Sigmoidoscopy Screening on Colorectal Cancer Incidence and Mortality in Women and Men: A Randomized Trial","language":"eng","author":[{"family":"Holme","given":"?"},{"family":"L?berg","given":"M."},{"family":"Kalager","given":"M."},{"family":"Bretthauer","given":"M."},{"family":"Hernán","given":"M. A."},{"family":"Aas","given":"E."},{"family":"Eide","given":"T. J."},{"family":"Skovlund","given":"E."},{"family":"Lekven","given":"J."},{"family":"Schneede","given":"J."},{"family":"Tveit","given":"K. M."},{"family":"Vatn","given":"M."},{"family":"Ursin","given":"G."},{"family":"Hoff","given":"G."},{"family":"Group?","given":"NORCCAP Study"}],"issued":{"date-parts":[["2018",6]]}}}],"schema":""} (18). In the original trial protocol, all individuals aged 55 to 64 years old living in two regions of Norway in 1998 were eligible for participation in the NORCCAP trial. A random sample of these individuals were sent an invitation for a screening sigmoidoscopy. However, in 2000, the funding agencies decided to increase the eligibility to include individuals aged 50 to 54 years old. The same random selection process was used for this new age group, but because of the larger size of this post-war birth cohort, the ratio of invitation group to control group participants was lower among younger individuals. Since age is a strong prognostic factor for colorectal cancer, this design change artificially induced confounding by age—which had not been pre-specified for adjustment in the original trial design. However, since there was a substantial imbalance in age between trial arms, the primary analysis reported age-standardized effect measures. For example, the age-standardized intention-to-treat difference for the effect of randomization to screening invitation versus randomization to no invitation was a 28.4 percentage point reduction in the rate of colorectal cancer diagnoses over the 10 years of follow-up.Case study E3: Adjusting for pre-specified and imbalanced baseline prognostic factorsIn the Women’s Health Initiative (WHI) estrogen plus progestin trial (Case Study #1), intention-to-treat analyses were adjusted for a set of pre-specified baseline covariates: age, evidence of coronary heart disease at enrollment, and randomization status in a low-fat diet sub-trial ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"k5t6Zw4N","properties":{"formattedCitation":"(43)","plainCitation":"(43)","noteIndex":0},"citationItems":[{"id":1059,"uris":[""],"uri":[""],"itemData":{"id":1059,"type":"article-journal","title":"Risks and Benefits of Estrogen Plus Progestin in Healthy Postmenopausal WomenPrincipal Results From the Women's Health Initiative Randomized Controlled Trial","container-title":"JAMA","page":"321-333","volume":"288","issue":"3","abstract":"ContextDespite decades of accumulated observational evidence, the balance of\nrisks and benefits for hormone use in healthy postmenopausal women remains\nuncertain.ObjectiveTo assess the major health benefits and risks of the most commonly used\ncombined hormone preparation in the United States.DesignEstrogen plus progestin component of the Women's Health Initiative,\na randomized controlled primary prevention trial (planned duration, 8.5 years)\nin which 16608 postmenopausal women aged 50-79 years with an intact uterus\nat baseline were recruited by 40 US clinical centers in 1993-1998.InterventionsParticipants received conjugated equine estrogens, 0.625 mg/d, plus\nmedroxyprogesterone acetate, 2.5 mg/d, in 1 tablet (n = 8506) or placebo (n\n= 8102).Main Outcomes MeasuresThe primary outcome was coronary heart disease (CHD) (nonfatal myocardial\ninfarction and CHD death), with invasive breast cancer as the primary adverse\noutcome. A global index summarizing the balance of risks and benefits included\nthe 2 primary outcomes plus stroke, pulmonary embolism (PE), endometrial cancer,\ncolorectal cancer, hip fracture, and death due to other causes.ResultsOn May 31, 2002, after a mean of 5.2 years of follow-up, the data and\nsafety monitoring board recommended stopping the trial of estrogen plus progestin\nvs placebo because the test statistic for invasive breast cancer exceeded\nthe stopping boundary for this adverse effect and the global index statistic\nsupported risks exceeding benefits. This report includes data on the major\nclinical outcomes through April 30, 2002. Estimated hazard ratios (HRs) (nominal\n95% confidence intervals [CIs]) were as follows: CHD, 1.29 (1.02-1.63) with\n286 cases; breast cancer, 1.26 (1.00-1.59) with 290 cases; stroke, 1.41 (1.07-1.85)\nwith 212 cases; PE, 2.13 (1.39-3.25) with 101 cases; colorectal cancer, 0.63\n(0.43-0.92) with 112 cases; endometrial cancer, 0.83 (0.47-1.47) with 47 cases;\nhip fracture, 0.66 (0.45-0.98) with 106 cases; and death due to other causes,\n0.92 (0.74-1.14) with 331 cases. Corresponding HRs (nominal 95% CIs) for composite\noutcomes were 1.22 (1.09-1.36) for total cardiovascular disease (arterial\nand venous disease), 1.03 (0.90-1.17) for total cancer, 0.76 (0.69-0.85) for\ncombined fractures, 0.98 (0.82-1.18) for total mortality, and 1.15 (1.03-1.28)\nfor the global index. Absolute excess risks per 10?000 person-years attributable\nto estrogen plus progestin were 7 more CHD events, 8 more strokes, 8 more\nPEs, and 8 more invasive breast cancers, while absolute risk reductions per\n10?000 person-years were 6 fewer colorectal cancers and 5 fewer hip fractures.\nThe absolute excess risk of events included in the global index was 19 per\n10?000 person-years.ConclusionsOverall health risks exceeded benefits from use of combined estrogen\nplus progestin for an average 5.2-year follow-up among healthy postmenopausal\nUS women. All-cause mortality was not affected during the trial. The risk-benefit\nprofile found in this trial is not consistent with the requirements for a\nviable intervention for primary prevention of chronic diseases, and the results\nindicate that this regimen should not be initiated or continued for primary\nprevention of CHD.","DOI":"10.1001/jama.288.3.321","ISSN":"0098-7484","journalAbbreviation":"JAMA","author":[{"literal":"Writing Group for the Women's Health Initiative Investigators"}],"issued":{"date-parts":[["2002",7,17]]}}}],"schema":""} (43). The intention-to-treat analysis for coronary heart disease was further adjusted for history of coronary artery bypass graft (CABG) or percutaneous transluminal coronary angioplasty (PTCA) at enrollment ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"oEdJRDHe","properties":{"formattedCitation":"(29)","plainCitation":"(29)","noteIndex":0},"citationItems":[{"id":928,"uris":[""],"uri":[""],"itemData":{"id":928,"type":"article-journal","title":"Estrogen plus progestin and the risk of coronary heart disease","container-title":"N Engl J Med","page":"523-34","volume":"349","issue":"6","source":"NLM","archive_location":"12904517","abstract":"BACKGROUND: Recent randomized clinical trials have suggested that estrogen plus progestin does not confer cardiac protection and may increase the risk of coronary heart disease (CHD). In this report, we provide the final results with regard to estrogen plus progestin and CHD from the Women's Health Initiative (WHI). METHODS: The WHI included a randomized primary-prevention trial of estrogen plus progestin in 16,608 postmenopausal women who were 50 to 79 years of age at base line. Participants were randomly assigned to receive conjugated equine estrogens (0.625 mg per day) plus medroxyprogesterone acetate (2.5 mg per day) or placebo. The primary efficacy outcome of the trial was CHD (nonfatal myocardial infarction or death due to CHD). RESULTS: After a mean follow-up of 5.2 years (planned duration, 8.5 years), the data and safety monitoring board recommended terminating the estrogen-plus-progestin trial because the overall risks exceeded the benefits. Combined hormone therapy was associated with a hazard ratio for CHD of 1.24 (nominal 95 percent confidence interval, 1.00 to 1.54; 95 percent confidence interval after adjustment for sequential monitoring, 0.97 to 1.60). The elevation in risk was most apparent at one year (hazard ratio, 1.81 [95 percent confidence interval, 1.09 to 3.01]). Although higher base-line levels of low-density lipoprotein cholesterol were associated with an excess risk of CHD among women who received hormone therapy, higher base-line levels of C-reactive protein, other biomarkers, and other clinical characteristics did not significantly modify the treatment-related risk of CHD. CONCLUSIONS: Estrogen plus progestin does not confer cardiac protection and may increase the risk of CHD among generally healthy postmenopausal women, especially during the first year after the initiation of hormone use. This treatment should not be prescribed for the prevention of cardiovascular disease.","DOI":"10.1056/NEJMoa030808","ISSN":"0028-4793","title-short":"Estrogen plus progestin and the risk of coronary heart disease","journalAbbreviation":"The New England journal of medicine","language":"eng","author":[{"family":"Manson","given":"J. E."},{"family":"Hsia","given":"J."},{"family":"Johnson","given":"K. C."},{"family":"Rossouw","given":"J. E."},{"family":"Assaf","given":"A. R."},{"family":"Lasser","given":"N. L."},{"family":"Trevisan","given":"M."},{"family":"Black","given":"H. R."},{"family":"Heckbert","given":"S. R."},{"family":"Detrano","given":"R."},{"family":"Strickland","given":"O. L."},{"family":"Wong","given":"N. D."},{"family":"Crouse","given":"J. R."},{"family":"Stein","given":"E."},{"family":"Cushman","given":"M."}],"issued":{"date-parts":[["2003",8,7]]}}}],"schema":""} (29). These covariates, which were not pre-specified, were slightly imbalanced between randomization arms: the prevalence of CABG/PTCA was 1.1% in the estrogen plus progesterone arm was 1.1% compared with 1.5% in the placebo arm. This variable was chosen because the 0.4% difference was “statistically significant” (p=0.04) whereas other variables with greater differences were not adjusted for ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"SgYdSfTm","properties":{"formattedCitation":"(43)","plainCitation":"(43)","noteIndex":0},"citationItems":[{"id":1059,"uris":[""],"uri":[""],"itemData":{"id":1059,"type":"article-journal","title":"Risks and Benefits of Estrogen Plus Progestin in Healthy Postmenopausal WomenPrincipal Results From the Women's Health Initiative Randomized Controlled Trial","container-title":"JAMA","page":"321-333","volume":"288","issue":"3","abstract":"ContextDespite decades of accumulated observational evidence, the balance of\nrisks and benefits for hormone use in healthy postmenopausal women remains\nuncertain.ObjectiveTo assess the major health benefits and risks of the most commonly used\ncombined hormone preparation in the United States.DesignEstrogen plus progestin component of the Women's Health Initiative,\na randomized controlled primary prevention trial (planned duration, 8.5 years)\nin which 16608 postmenopausal women aged 50-79 years with an intact uterus\nat baseline were recruited by 40 US clinical centers in 1993-1998.InterventionsParticipants received conjugated equine estrogens, 0.625 mg/d, plus\nmedroxyprogesterone acetate, 2.5 mg/d, in 1 tablet (n = 8506) or placebo (n\n= 8102).Main Outcomes MeasuresThe primary outcome was coronary heart disease (CHD) (nonfatal myocardial\ninfarction and CHD death), with invasive breast cancer as the primary adverse\noutcome. A global index summarizing the balance of risks and benefits included\nthe 2 primary outcomes plus stroke, pulmonary embolism (PE), endometrial cancer,\ncolorectal cancer, hip fracture, and death due to other causes.ResultsOn May 31, 2002, after a mean of 5.2 years of follow-up, the data and\nsafety monitoring board recommended stopping the trial of estrogen plus progestin\nvs placebo because the test statistic for invasive breast cancer exceeded\nthe stopping boundary for this adverse effect and the global index statistic\nsupported risks exceeding benefits. This report includes data on the major\nclinical outcomes through April 30, 2002. Estimated hazard ratios (HRs) (nominal\n95% confidence intervals [CIs]) were as follows: CHD, 1.29 (1.02-1.63) with\n286 cases; breast cancer, 1.26 (1.00-1.59) with 290 cases; stroke, 1.41 (1.07-1.85)\nwith 212 cases; PE, 2.13 (1.39-3.25) with 101 cases; colorectal cancer, 0.63\n(0.43-0.92) with 112 cases; endometrial cancer, 0.83 (0.47-1.47) with 47 cases;\nhip fracture, 0.66 (0.45-0.98) with 106 cases; and death due to other causes,\n0.92 (0.74-1.14) with 331 cases. Corresponding HRs (nominal 95% CIs) for composite\noutcomes were 1.22 (1.09-1.36) for total cardiovascular disease (arterial\nand venous disease), 1.03 (0.90-1.17) for total cancer, 0.76 (0.69-0.85) for\ncombined fractures, 0.98 (0.82-1.18) for total mortality, and 1.15 (1.03-1.28)\nfor the global index. Absolute excess risks per 10?000 person-years attributable\nto estrogen plus progestin were 7 more CHD events, 8 more strokes, 8 more\nPEs, and 8 more invasive breast cancers, while absolute risk reductions per\n10?000 person-years were 6 fewer colorectal cancers and 5 fewer hip fractures.\nThe absolute excess risk of events included in the global index was 19 per\n10?000 person-years.ConclusionsOverall health risks exceeded benefits from use of combined estrogen\nplus progestin for an average 5.2-year follow-up among healthy postmenopausal\nUS women. All-cause mortality was not affected during the trial. The risk-benefit\nprofile found in this trial is not consistent with the requirements for a\nviable intervention for primary prevention of chronic diseases, and the results\nindicate that this regimen should not be initiated or continued for primary\nprevention of CHD.","DOI":"10.1001/jama.288.3.321","ISSN":"0098-7484","journalAbbreviation":"JAMA","author":[{"literal":"Writing Group for the Women's Health Initiative Investigators"}],"issued":{"date-parts":[["2002",7,17]]}}}],"schema":""} (43). Adjustment did not materially change the intention-to-treat estimates.Guideline: Pre-specify important prognostic factors for the outcome and the maximum acceptable difference in the distribution of these factors between groups. When one or more prognostic factor meets the threshold for imbalance, adjust via standardization, inverse probability weighting or, preferably, doubly-robust methods.In sensitivity analyses, adjust for large imbalances in any important prognostic factors, regardless of whether they have been pre-specified. 4.2 Competing events A competing event is an event that precludes the outcome from happening ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"uMw7Vc2l","properties":{"formattedCitation":"(44)","plainCitation":"(44)","noteIndex":0},"citationItems":[{"id":1197,"uris":[""],"uri":[""],"itemData":{"id":1197,"type":"article-journal","title":"The choice to define competing risk events as censoring events and implications for causal inference","URL":"","title-short":"The choice to define competing risk events as censoring events and implications for causal inference","author":[{"family":"Young","given":"Jessica G."},{"family":"Tchetgen","given":"Eric J. Tchetgen"},{"family":"Hernán","given":"Miguel A."}],"issued":{"date-parts":[["2018",6,15]]}}}],"schema":""} (44). For example, if we are interested in studying whether daily aspirin use decreases the risk of stroke, then death from causes other than stroke is a competing event because individuals who die are no longer at risk for stroke. In these settings, informed decision analysis requires information about both the risk of the competing event by treatment group and the risk of the event interest among those who survived the competing events.The presence of competing events implies that the definition of intention-to-treat effect provided in Section 2 is incomplete. In order to decide how to account for competing events in intention-to-treat analyses, we first need to choose among several possible definitions of intention-to-treat effect (i.e., causal estimands) ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"0tuFYd4g","properties":{"formattedCitation":"(11,44)","plainCitation":"(11,44)","noteIndex":0},"citationItems":[{"id":511,"uris":[""],"uri":[""],"itemData":{"id":511,"type":"book","title":"Causal Inference: What If","publisher":"Chapman & Hill/CRC","publisher-place":"Boca Raton","event-place":"Boca Raton","title-short":"Causal Inference","author":[{"family":"Hernan","given":"M.A."},{"family":"Robins","given":"J."}],"issued":{"date-parts":[["2020"]]}},"label":"page"},{"id":1197,"uris":[""],"uri":[""],"itemData":{"id":1197,"type":"article-journal","title":"The choice to define competing risk events as censoring events and implications for causal inference","URL":"","title-short":"The choice to define competing risk events as censoring events and implications for causal inference","author":[{"family":"Young","given":"Jessica G."},{"family":"Tchetgen","given":"Eric J. Tchetgen"},{"family":"Hernán","given":"Miguel A."}],"issued":{"date-parts":[["2018",6,15]]}},"label":"page"}],"schema":""} (11,44). For example, in our aspirin example, we consider any of the following definitions of intention-to-treat effect:The total effect of assignment to aspirin versus no aspirin on stroke in a population with the same death rate as the trial populationThe direct effect of assignment to aspirin versus no aspirin on stroke if no one had died before having a stroke;The total effect of assignment to aspirin versus no aspirin on stroke among those individuals who would never have died before having a stroke during the trial regardless of which assignment they had been givenThe total effect of assignment to aspirin versus no aspirin on death or stroke, whichever happens firstThe choice among these (or other ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"0W2BDEtB","properties":{"formattedCitation":"(45)","plainCitation":"(45)","noteIndex":0},"citationItems":[{"id":1070,"uris":[""],"uri":[""],"itemData":{"id":1070,"type":"article-journal","title":"Separable Effects for Causal Inference in the Presence of Competing Risks","container-title":"arXiv:1901.09472 [stat]","source":"","abstract":"In time-to-event settings, the presence of competing events complicates the definition of causal effects. Here we propose the new separable effects to study the causal effect of a treatment on an event of interest. The separable direct effect is the treatment effect on the event of interest not mediated by its effect on the competing event. The separable indirect effect is the treatment effect on the event of interest only through its effect on the competing event. Similar to Robins and Richardson's extended graphical approach for mediation analysis, the separable effects can only be identified under the assumption that the treatment can be decomposed into two distinct components that exert effects through distinct causal pathways. Unlike existing definitions of causal effects in the presence of competing risks, our estimands do not require cross-world contrasts or hypothetical interventions to prevent death. As an illustration, we implement our approach in a randomized clinical trial on estrogen therapy in individuals with prostate cancer.","URL":"","note":"arXiv: 1901.09472","author":[{"family":"Stensrud","given":"Mats J."},{"family":"Young","given":"Jessica G."},{"family":"Didelez","given":"Vanessa"},{"family":"Robins","given":"James M."},{"family":"Hernán","given":"Miguel A."}],"issued":{"date-parts":[["2019",1,27]]},"accessed":{"date-parts":[["2019",2,25]]}}}],"schema":""} (45)) effects implies trade-offs between ease of estimation, tenability of the assumptions, and interpretability of the effect estimate (Case study F). The first effect, which is estimated by a simple contrast of the cumulative incidence in each group, does not allow us to distinguish whether the effect on the outcome of interest is mediated through the effect on death (e.g., the more people die in a treatment group, the fewer people can develop stroke). The second effect, which is estimated by censoring individuals if/when they develop the competing event and adjusting for potential selection bias, measures the direct effect on the outcome that is not mediated by the competing event. This effect, however, is hard to interpret as no well-defined intervention exists to prevent all deaths from causes other than the event of interest. The third effect, which requires strong assumptions, is also hard to of limited relevance to clinical decision-making because it is restricted to an unidentifiable subset of the population (known as a principal stratum). Finally, the fourth effect eliminates the competing events problem by estimating the risk of the combined outcome, but can fundamentally change the research question of interest ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"ZfMkbpOG","properties":{"formattedCitation":"(11,44)","plainCitation":"(11,44)","noteIndex":0},"citationItems":[{"id":511,"uris":[""],"uri":[""],"itemData":{"id":511,"type":"book","title":"Causal Inference: What If","publisher":"Chapman & Hill/CRC","publisher-place":"Boca Raton","event-place":"Boca Raton","title-short":"Causal Inference","author":[{"family":"Hernan","given":"M.A."},{"family":"Robins","given":"J."}],"issued":{"date-parts":[["2020"]]}},"label":"page"},{"id":1197,"uris":[""],"uri":[""],"itemData":{"id":1197,"type":"article-journal","title":"The choice to define competing risk events as censoring events and implications for causal inference","URL":"","title-short":"The choice to define competing risk events as censoring events and implications for causal inference","author":[{"family":"Young","given":"Jessica G."},{"family":"Tchetgen","given":"Eric J. Tchetgen"},{"family":"Hernán","given":"Miguel A."}],"issued":{"date-parts":[["2018",6,15]]}},"label":"page"}],"schema":""} (11,44). Given that none of the available approaches to deal with competing events is clearly superior to the others, it may be advisable to choose the simplest estimand (the total effect on the event of interest in the entire study population; the first effect listed above) as the target of the primary analysis and to complement it with sensitivity analyses that estimate the direct effect after confirming that treatment assignment has an effect on the competing event. Importantly, the choice among several causal estimands is generally only possible in survival (e.g., failure time) analyses. When the outcome is a variable measured at the end of follow-up (e.g., blood pressure), then the only options are the direct effect estimated after censoring and the effect in those who would always have survived ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"S28xhHRZ","properties":{"formattedCitation":"(46)","plainCitation":"(46)","noteIndex":0},"citationItems":[{"id":1047,"uris":[""],"uri":[""],"itemData":{"id":1047,"type":"article-journal","title":"A Simple Regression-based Approach to Account for Survival Bias in Birth Outcomes Research","container-title":"Epidemiology (Cambridge, Mass.)","page":"473-480","volume":"26","issue":"4","source":"PubMed","abstract":"In perinatal epidemiology, birth outcomes such as small for gestational age (SGA) may not be observed for a pregnancy ending with a stillbirth. It is then said that SGA is truncated by stillbirth, which may give rise to survival bias when evaluating the effects on SGA of an exposure known also to influence the risk of a stillbirth. In this article, we consider the causal effects of maternal infection with human immunodeficiency virus (HIV) on the risk of SGA, in a sample of pregnant women in Botswana. We hypothesize that previously estimated effects of HIV on SGA may be understated because they fail to appropriately account for the over-representation of live births among HIV negative mothers, relative to HIV positive mothers. A simple yet novel regression-based approach is proposed to adjust effect estimates for survival bias for an outcome that is either continuous or binary. Under certain straightforward assumptions, the approach produces an estimate that may be interpreted as the survivor average causal effect of maternal HIV, which is, the average effect of maternal HIV on SGA among births that would be live irrespective of maternal HIV status. The approach is particularly appealing, because it recovers an exposure effect which is robust to survival bias, even if the association between the risk of SGA and that of a stillbirth cannot be completely explained by adjusting for observed shared risk factors. The approach also gives a formal statistical test of the null hypothesis of no survival bias in the regression framework.","DOI":"10.1097/EDE.0000000000000317","ISSN":"1531-5487","note":"PMID: 26011373","journalAbbreviation":"Epidemiology","language":"eng","author":[{"family":"Tchetgen Tchetgen","given":"Eric J."},{"family":"Phiri","given":"Kelesitse"},{"family":"Shapiro","given":"Roger"}],"issued":{"date-parts":[["2015",7]]}}}],"schema":""} (46). Case study F: The intention-to-treat effect in the presence of competing eventsThe Assessment of Multiple Intrauterine Gestations from Ovarian Stimulation (AMIGOS) trial compared maternal use of gonadotropins versus clomiphene during fertility treatments for the prevention of neonatal complications. In this trial, the outcome is only defined among pregnancies which ended in a live birth, that is, “no live birth” is a competing event for the event of interest ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"nG6EZSKT","properties":{"formattedCitation":"(47)","plainCitation":"(47)","noteIndex":0},"citationItems":[{"id":1105,"uris":[""],"uri":[""],"itemData":{"id":1105,"type":"article-journal","title":"Letrozole, Gonadotropin, or Clomiphene for Unexplained Infertility","container-title":"New England Journal of Medicine","page":"1230-1240","volume":"373","issue":"13","DOI":"10.1056/NEJMoa1414827","note":"PMID: 26398071","author":[{"family":"Diamond","given":"Michael P."},{"family":"Legro","given":"Richard S."},{"family":"Coutifaris","given":"Christos"},{"family":"Alvero","given":"Ruben"},{"family":"Robinson","given":"Randal D."},{"family":"Casson","given":"Peter"},{"family":"Christman","given":"Gregory M."},{"family":"Ager","given":"Joel"},{"family":"Huang","given":"Hao"},{"family":"Hansen","given":"Karl R."},{"family":"Baker","given":"Valerie"},{"family":"Usadi","given":"Rebecca"},{"family":"Seungdamrong","given":"Aimee"},{"family":"Bates","given":"G. Wright"},{"family":"Rosen","given":"R. Mitchell"},{"family":"Haisenleder","given":"Daniel"},{"family":"Krawetz","given":"Stephen A."},{"family":"Barnhart","given":"Kurt"},{"family":"Trussell","given":"J.C."},{"family":"Ohl","given":"Dana"},{"family":"Jin","given":"Yufeng"},{"family":"Santoro","given":"Nanette"},{"family":"Eisenberg","given":"Esther"},{"family":"Zhang","given":"Heping"}],"issued":{"date-parts":[["2015"]]}}}],"schema":""} (47).We can consider several causal estimands for the intention-to-treat effect in this trial ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"N5Ts4mcG","properties":{"formattedCitation":"(48)","plainCitation":"(48)","noteIndex":0},"citationItems":[{"id":1104,"uris":[""],"uri":[""],"itemData":{"id":1104,"type":"article-journal","title":"Dealing with competing events in the estimation of the health effects of fertility treatment on the offspring","author":[{"family":"Chiu","given":"Yu-Han"},{"family":"Hsu","given":"John"},{"family":"Stensrud","given":"Mats Julius"},{"family":"Rinaudo","given":"Paolo"},{"family":"Hernandez-Diaz","given":"Sonia"},{"family":"Hernan","given":"Miguel A."}]}}],"schema":""} (48), including: the total effect of gonadotropins versus clomiphene on neonatal complications assuming that no complications can occur without live birth the total effect of gonadotropins versus clomiphene on the composite outcome of either neonatal complications or no live birth the direct effect of gonadotropins versus clomiphene on neonatal complications under the assumption that censoring due to failure to produce a live birth can be abolished and that all shared causes of the competing events and the event of interest can be adjusted for the total effect of gonadotropins versus clomiphene on neonatal complications in pregnancies which would always result in a live births regardless of whether they were assigned to gonadotropins or clomiphene (an unidentifiable subset of the study population) under several untestable assumptionsEach causal estimand will result in a different estimate of the intention-to-treat effect. Note that the last two estimands require strong assumptions that are not guaranteed to hold in any randomized trial.Guidelines:In survival analyses with competing events, report both the risk of the competing event by treatment group and the risk of the event of interest among those who survived the competing event by treatment group.In survival analyses with competing events, specify the intention-to-treat effect as the total effect of treatment assignment on the outcome of interest (the simplest analysis), and justify interest in any additional effects that are estimated.Loss to follow-upIntention-to-treat analyses require that the outcomes of all trial participants are ascertained. If some participants are lost to follow-up (e.g., because they drop out of the study) and therefore their outcomes are unknown, a true intention-to-treat analysis is not possible. In the presence of losses to follow-up, a common strategy is to conduct a complete case analysis in which participants are censored if/when they are lost to follow-up (or, for non-failure time outcomes, in which a participant’s last available outcome measurement is carried forward). These pseudo-intention-to-treat analyses may introduce selection bias in the effect estimates ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"BWMBykXC","properties":{"formattedCitation":"(49)","plainCitation":"(49)","noteIndex":0},"citationItems":[{"id":1176,"uris":[""],"uri":[""],"itemData":{"id":1176,"type":"article-journal","title":"A structural approach to selection bias","container-title":"Epidemiology","page":"615-25","volume":"15","issue":"5","archive_location":"15308962","abstract":"The term \"selection bias\" encompasses various biases in epidemiology. We describe examples of selection bias in case-control studies (eg, inappropriate selection of controls) and cohort studies (eg, informative censoring). We argue that the causal structure underlying the bias in each example is essentially the same: conditioning on a common effect of 2 variables, one of which is either exposure or a cause of exposure and the other is either the outcome or a cause of the outcome. This structure is shared by other biases (eg, adjustment for variables affected by prior exposure). A structural classification of bias distinguishes between biases resulting from conditioning on common effects (\"selection bias\") and those resulting from the existence of common causes of exposure and outcome (\"confounding\"). This classification also leads to a unified approach to adjust for selection bias.","ISSN":"1044-3983","title-short":"A structural approach to selection bias","language":"eng","author":[{"family":"Hernán","given":"M. A."},{"family":"Hernández-Díaz","given":"S."},{"family":"Robins","given":"J. M."}],"issued":{"date-parts":[["2004",9]]}},"label":"page"}],"schema":""} (49). For example, if individuals are more likely to be lost to follow-up in the aspirin group and those who are lost to follow-up tend to have a worse health status, then a pseudo-intention-to-treat analysis will find better outcomes in the aspirin group compared with the no aspirin group, even if aspirin had no effect on the outcome.Pseudo-intention to treat analyses provide valid estimates of the intention-to-treat effect if censoring due to loss to follow-up is completely at random with respect to the outcome ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"UMKFk3dv","properties":{"formattedCitation":"(50)","plainCitation":"(50)","noteIndex":0},"citationItems":[{"id":1162,"uris":[""],"uri":[""],"itemData":{"id":1162,"type":"article-journal","title":"Causal inference from longitudinal studies with baseline randomization","container-title":"The International Journal of Biostatistics","page":"22","volume":"4","issue":"1","archive":"Pmc","archive_location":"PMC2835458","abstract":"We describe analytic approaches for study designs that, like large simple trials, can be better characterized as longitudinal studies with baseline randomization than as either a pure randomized experiment or a purely observational study. We (i) discuss the intention-to-treat effect as an effect measure for randomized studies, (ii) provide a formal definition of causal effect for longitudinal studies, (iii) describe several methods -- based on inverse probability weighting and g-estimation -- to estimate such effect, (iv) present an application of these methods to a naturalistic trial of antipsychotics on symptom severity of schizophrenia, and (v) discuss the relative advantages and disadvantages of each method.","DOI":"10.2202/1557-4679.1117","ISSN":"1557-4679","title-short":"Causal Inference from Longitudinal Studies with Baseline Randomization","author":[{"family":"Toh","given":"Sengwee"},{"family":"Hernán","given":"Miguel A."}],"issued":{"literal":"10"}}}],"schema":""} (50). However, this assumption of non-informative censoring is untenable when, as it is common, loss to follow-up depends on pre- and post-randomization prognostic factors. Rather, one can make the weaker assumption that censoring is non-informative conditional on pre- and post-randomization covariates, and make sure that the pseudo-intention-to-treat analysis adjusts for those pre- and post-randomization prognostic factors ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"3sFS5eTZ","properties":{"formattedCitation":"(15,51)","plainCitation":"(15,51)","noteIndex":0},"citationItems":[{"id":870,"uris":[""],"uri":[""],"itemData":{"id":870,"type":"book","title":"The Prevention and Treatment of Missing Data in Clinical Trials","publisher":"National Academies Press","publisher-place":"Washington, DC","event-place":"Washington, DC","abstract":"Randomized clinical trials are the primary tool for evaluating new medical interventions. Randomization provides for a fair comparison between treatment and control groups, balancing out, on average, distributions of known and unknown factors among the participants. Unfortunately, these studies often lack a substantial percentage of data. This missing data reduces the benefit provided by the randomization and introduces potential biases in the comparison of the treatment groups. Missing data can arise for a variety of reasons, including the inability or unwillingness of participants to meet appointments for evaluation. And in some studies, some or all of data collection ceases when participants discontinue study treatment. Existing guidelines for the design and conduct of clinical trials, and the analysis of the resulting data, provide only limited advice on how to handle missing data. Thus, approaches to the analysis of data with an appreciable amount of missing values tend to be ad hoc and variable. The Prevention and Treatment of Missing Data in Clinical Trials concludes that a more principled approach to design and analysis in the presence of missing data is both needed and possible. Such an approach needs to focus on two critical elements: (1) careful design and conduct to limit the amount and impact of missing data and (2) analysis that makes full use of information on all randomized participants and is based on careful attention to the assumptions about the nature of the missing data underlying estimates of treatment effects. In addition to the highest priority recommendations, the book offers more detailed recommendations on the conduct of clinical trials and techniques for analysis of trial data.","URL":"","title-short":"The Prevention and Treatment of Missing Data in Clinical Trials","author":[{"literal":"National Research Council"}],"issued":{"date-parts":[["2010"]]}},"label":"page"},{"id":1150,"uris":[""],"uri":[""],"itemData":{"id":1150,"type":"article-journal","title":"The prevention and treatment of missing data in clinical trials","container-title":"New England Journal of Medicine","page":"1355-60","volume":"367","issue":"14","source":"Nlm","archive_location":"23034025","DOI":"10.1056/NEJMsr1203730","ISSN":"1533-4406 (Electronic) 0028-4793 (Linking)","title-short":"The prevention and treatment of missing data in clinical trials","journalAbbreviation":"N Engl J Med","language":"eng","author":[{"family":"Little","given":"R. J."},{"family":"D'Agostino","given":"R."},{"family":"Cohen","given":"M. L."},{"family":"Dickersin","given":"K."},{"family":"Emerson","given":"S. S."},{"family":"Farrar","given":"J. T."},{"family":"Frangakis","given":"C."},{"family":"Hogan","given":"J. W."},{"family":"Molenberghs","given":"G."},{"family":"Murphy","given":"S. A."},{"family":"Neaton","given":"J. D."},{"family":"Rotnitzky","given":"A."},{"family":"Scharfstein","given":"D."},{"family":"Shih","given":"W. J."},{"family":"Siegel","given":"J. P."},{"family":"Stern","given":"H."}],"issued":{"date-parts":[["2012",10,4]]}},"label":"page"}],"schema":""} (15,51). Adjusted pseudo-intention-to-treat analyses provide valid estimates of the intention-to-treat effect if censoring due to loss to follow-up is not informative within levels of the measured prognostic factors.Because post-randomization prognostic factors may be affected by treatment assignment, adjustment methods that appropriately account for time-varying covariates, such as inverse probability weighting or the parametric g-formula, are usually preferable ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"jd0XgekF","properties":{"formattedCitation":"(5)","plainCitation":"(5)","noteIndex":0},"citationItems":[{"id":1153,"uris":[""],"uri":[""],"itemData":{"id":1153,"type":"article-journal","title":"Randomized trials analyzed as observational studies","container-title":"Annals of Internal Medicine","page":"560-2","volume":"159","issue":"8","source":"Nlm","archive_location":"24018844","DOI":"10.7326/0003-4819-159-8-201310150-00709","ISSN":"1539-3704 (Electronic) 0003-4819 (Linking)","title-short":"Randomized trials analyzed as observational studies","journalAbbreviation":"Ann Intern Med","language":"eng","author":[{"family":"Hernán","given":"M. A."},{"family":"Hernandez-Diaz","given":"S."},{"family":"Robins","given":"J. M."}],"issued":{"date-parts":[["2013",10,15]]}}}],"schema":""} (5). Inverse probability weighting is typically easier to implement (Case Study G). Note that the implementation of these methods requires that post-randomization factors are measured regularly over follow-up to ensure that data will be available close to the time of censoring and that the model for censoring is correctly specified. Case study G: Adjusting for loss to follow-up when estimating the intention-to-treat effectIn an open-label, multicenter randomized trial of antipsychotic medications to reduce self-reported symptoms among individuals with schizophrenia ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"VqSP1vLr","properties":{"formattedCitation":"(52)","plainCitation":"(52)","noteIndex":0},"citationItems":[{"id":7,"uris":[""],"uri":[""],"itemData":{"id":7,"type":"article-journal","title":"Cost-effectiveness of olanzapine as first-line treatment for schizophrenia: results from a randomized, open-label, 1-year trial","container-title":"Value Health","page":"77-89","volume":"9","issue":"2","archive_location":"16626411","abstract":"This randomized, open-label trial was designed to help inform antipsychotic treatment policies. It compared the 1-year cost-effectiveness of initial treatment with olanzapine (OLZ) (n = 229) versus a \"fail-first\" algorithm on conventional antipsychotics (then olanzapine if indicated) (CON) (n = 214); and versus initial treatment with risperidone (RIS) (n = 221).|Individuals with schizophrenia or schizoaffective disorder were recruited from May 1998 to September 2001. Clinical, functioning, and resource utilization data were collected at baseline and five postbaseline visits. Brief Psychiatric Rating Scale scores defined \"clinical effectiveness;\" Lehman Quality of Life Scale social relations scores defined \"social effectiveness.\"|Requiring failure on less expensive antipsychotics before use of olanzapine did not result in total cost savings, despite significantly higher antipsychotic costs with OLZ. Total 1-year mean costs were 21,283 dollars for CON; 20,891 dollars for OLZ; and 21,347 dollars for RIS (pair-wise comparisons nonsignificant). Intent-to-treat effectiveness comparisons (nonsignificant) were augmented by analyses that adjusted for duration on initial antipsychotic treatment, and by comparisons of patients remaining on initial antipsychotic treatment versus those who required switching. When accounting for differential switching rates (OLZ 0.14 vs. CON 0.53, P < 0.0001; vs. RIS 0.31, P < 0.0001), OLZ was significantly more effective than CON on clinical (P = 0.025) and social (P = 0.043) measures, and significantly more effective than RIS on the social (P = 0.002) measure. Further, patients initiated on an antipsychotic from which they needed to switch required additional resources for hospitalization (P = 0.036) and crisis services (P = 0.029).|Approaches that integrate costs, effectiveness, and treatment patterns are important for providing optimal information regarding the value of first-line antipsychotic options for schizophrenia.","DOI":"10.1111/j.1524-4733.2006.00083.x","ISSN":"1098-3015","title-short":"Cost-effectiveness of olanzapine as first-line treatment for schizophrenia: results from a randomized, open-label, 1-year trial","language":"eng","author":[{"family":"Tunis","given":"S. L."},{"family":"Faries","given":"D. E."},{"family":"Nyhuis","given":"A. W."},{"family":"Kinon","given":"B. J."},{"family":"Ascher-Svanum","given":"H."},{"family":"Aquila","given":"R."}],"issued":{"date-parts":[["2006",4]]}}}],"schema":""} (52), the primary outcome of interest was Brief Psychiatric Rating Scale (BPRS) score at the end of follow-up (12 months). 450 individuals were assigned to one of two atypical antipsychotic treatments (combined here for simplicity), and 214 to conventional antipsychotics. However, only 430 individuals assigned to atypical treatments and 204 assigned to conventional completed at least one follow-up visit, and only 235 atypical arm and 130 conventional arm participants completed all follow-up visits ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"U9yuksgK","properties":{"formattedCitation":"(50)","plainCitation":"(50)","noteIndex":0},"citationItems":[{"id":1162,"uris":[""],"uri":[""],"itemData":{"id":1162,"type":"article-journal","title":"Causal inference from longitudinal studies with baseline randomization","container-title":"The International Journal of Biostatistics","page":"22","volume":"4","issue":"1","archive":"Pmc","archive_location":"PMC2835458","abstract":"We describe analytic approaches for study designs that, like large simple trials, can be better characterized as longitudinal studies with baseline randomization than as either a pure randomized experiment or a purely observational study. We (i) discuss the intention-to-treat effect as an effect measure for randomized studies, (ii) provide a formal definition of causal effect for longitudinal studies, (iii) describe several methods -- based on inverse probability weighting and g-estimation -- to estimate such effect, (iv) present an application of these methods to a naturalistic trial of antipsychotics on symptom severity of schizophrenia, and (v) discuss the relative advantages and disadvantages of each method.","DOI":"10.2202/1557-4679.1117","ISSN":"1557-4679","title-short":"Causal Inference from Longitudinal Studies with Baseline Randomization","author":[{"family":"Toh","given":"Sengwee"},{"family":"Hernán","given":"Miguel A."}],"issued":{"literal":"10"}}}],"schema":""} (50). This high level of loss to follow-up could result in a biased intention-to-treat effect estimate. A pseudo-intention-to-treat analysis yielded a difference in BPRS score of 0.42 units (95% CI: -2.36, 3.19) ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"R5srpNdw","properties":{"formattedCitation":"(50)","plainCitation":"(50)","noteIndex":0},"citationItems":[{"id":1162,"uris":[""],"uri":[""],"itemData":{"id":1162,"type":"article-journal","title":"Causal inference from longitudinal studies with baseline randomization","container-title":"The International Journal of Biostatistics","page":"22","volume":"4","issue":"1","archive":"Pmc","archive_location":"PMC2835458","abstract":"We describe analytic approaches for study designs that, like large simple trials, can be better characterized as longitudinal studies with baseline randomization than as either a pure randomized experiment or a purely observational study. We (i) discuss the intention-to-treat effect as an effect measure for randomized studies, (ii) provide a formal definition of causal effect for longitudinal studies, (iii) describe several methods -- based on inverse probability weighting and g-estimation -- to estimate such effect, (iv) present an application of these methods to a naturalistic trial of antipsychotics on symptom severity of schizophrenia, and (v) discuss the relative advantages and disadvantages of each method.","DOI":"10.2202/1557-4679.1117","ISSN":"1557-4679","title-short":"Causal Inference from Longitudinal Studies with Baseline Randomization","author":[{"family":"Toh","given":"Sengwee"},{"family":"Hernán","given":"Miguel A."}],"issued":{"literal":"10"}}}],"schema":""} (50). However, loss to follow-up may have been related to pre- and post-randomization covariates. Adjusting for these via inverse probability weighting yielded an estimated intention-to-treat difference in BPRS score of -0.86 units (95% CI: -3.88, 2.15). Note that, as expected, the adjusted estimate is more imprecise than the unadjusted one.Guidelines: Ensure that the trial protocol specifies the collection of post-randomization time-varying prognostic factors that predict loss to follow-up, and appropriately adjust for these factors to reduce selection bias.4.4 External validityAn important goal of pragmatic randomized trials is to provide relevant and generalizable information to guide clinical decision-making. As such, the inclusion and exclusion criteria of pragmatic trials is often more broad than those of traditional randomized controlled trials. In order to ensure that pragmatic trial results are generalizable to the target population of interest, this population should be clearly and explicitly defined before the inclusion and exclusion criteria are defined. In some cases, it will also be of interest to estimate what the results of a pragmatic randomized trial would have been had the trial been conducted in a different population, that is to transport the results from the current target population under study to a new target population. Methods for transporting effect estimates (standardization, inverse weighting, or doubly robust methods) require adjustment for all causes of the outcome that are differently distributed between the trial sample and the new target population ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"NKNj5beE","properties":{"formattedCitation":"(53\\uc0\\u8211{}57)","plainCitation":"(53–57)","noteIndex":0},"citationItems":[{"id":2700,"uris":[""],"uri":[""],"itemData":{"id":2700,"type":"article-journal","title":"External Validity: From Do-Calculus to Transportability Across Populations","container-title":"Statistical Science","page":"579-595","volume":"29","issue":"4","source":"Project Euclid","abstract":"The generalizability of empirical findings to new environments, settings or populations, often called “external validity,” is essential in most scientific explorations. This paper treats a particular problem of generalizability, called “transportability,” defined as a license to transfer causal effects learned in experimental studies to a new population, in which only observational studies can be conducted. We introduce a formal representation called “selection diagrams” for expressing knowledge about differences and commonalities between populations of interest and, using this representation, we reduce questions of transportability to symbolic derivations in the do-calculus. This reduction yields graph-based procedures for deciding, prior to observing any data, whether causal effects in the target population can be inferred from experimental findings in the study population. When the answer is affirmative, the procedures identify what experimental and observational findings need be obtained from the two populations, and how they can be combined to ensure bias-free transport.","DOI":"10.1214/14-STS486","ISSN":"0883-4237, 2168-8745","note":"MR: MR3300360\nZbl: 1331.62326","title-short":"External Validity","journalAbbreviation":"Statist. Sci.","language":"EN","author":[{"family":"Pearl","given":"Judea"},{"family":"Bareinboim","given":"Elias"}],"issued":{"date-parts":[["2014",11]]}}},{"id":1083,"uris":[""],"uri":[""],"itemData":{"id":1083,"type":"article-journal","title":"Transporting inferences from a randomized trial to a new target population","container-title":"arXiv:1805.00550 [stat]","source":"","abstract":"When variables that are treatment effect modifiers also influence the decision to participate in a clinical trial, the average effect among trial participants will differ from the effect in other populations of trial-eligible individuals. In this tutorial, we consider methods for transporting inferences about a time-fixed treatment from trial participants to a new target population of trial-eligible individuals, using data from a completed randomized trial along with baseline covariate data from a sample of non-participants. We examine methods based on modeling the expectation of the outcome, the probability of participation, or both (doubly robust). We compare the finite-sample performance of different methods in a simulation study and provide example code to implement the methods in software. We illustrate the application of the methods to the Coronary Artery Surgery Study, a randomized trial nested within a cohort of trial-eligible patients to compare coronary artery surgery plus medical therapy versus medical therapy alone for patients with chronic coronary artery disease. Lastly, we discuss issues that arise when using the methods in applied transportability analyses.","URL":"","note":"arXiv: 1805.00550","author":[{"family":"Dahabreh","given":"Issa J."},{"family":"Robertson","given":"Sarah E."},{"family":"Stuart","given":"Elizabeth A."},{"family":"Hernan","given":"Miguel A."}],"issued":{"date-parts":[["2018",5,1]]},"accessed":{"date-parts":[["2019",2,21]]}}},{"id":2693,"uris":[""],"uri":[""],"itemData":{"id":2693,"type":"article-journal","title":"Generalizing causal inferences from individuals in randomized trials to all trial-eligible individuals","container-title":"Biometrics","source":"PubMed","abstract":"We consider methods for causal inference in randomized trials nested within cohorts of trial-eligible individuals, including those who are not randomized. We show how baseline covariate data from the entire cohort, and treatment and outcome data only from randomized individuals, can be used to identify potential (counterfactual) outcome means and average treatment effects in the target population of all eligible individuals. We review identifiability conditions, propose estimators, and assess the estimators' finite-sample performance in simulation studies. As an illustration, we apply the estimators in a trial nested within a cohort of trial-eligible individuals to compare coronary artery bypass grafting surgery plus medical therapy vs. medical therapy alone for chronic coronary artery disease.","DOI":"10.1111/biom.13009","ISSN":"1541-0420","note":"PMID: 30488513","journalAbbreviation":"Biometrics","language":"eng","author":[{"family":"Dahabreh","given":"Issa J."},{"family":"Robertson","given":"Sarah E."},{"family":"Tchetgen","given":"Eric J."},{"family":"Stuart","given":"Elizabeth A."},{"family":"Hernán","given":"Miguel A."}],"issued":{"date-parts":[["2018",11,29]]}}},{"id":2709,"uris":[""],"uri":[""],"itemData":{"id":2709,"type":"article-journal","title":"Generalizing Evidence From Randomized Clinical Trials to Target Populations","container-title":"American Journal of Epidemiology","page":"107-115","volume":"172","issue":"1","source":"PubMed Central","abstract":"Properly planned and conducted randomized clinical trials remain susceptible to a lack of external validity. The authors illustrate a model-based method to standardize observed trial results to a specified target population using a seminal human immunodeficiency virus (HIV) treatment trial, and they provide Monte Carlo simulation evidence supporting the method. The example trial enrolled 1,156 HIV-infected adult men and women in the United States in 1996, randomly assigned 577 to a highly active antiretroviral therapy and 579 to a largely ineffective combination therapy, and followed participants for 52 weeks. The target population was US people infected with HIV in 2006, as estimated by the Centers for Disease Control and Prevention. Results from the trial apply, albeit muted by 12%, to the target population, under the assumption that the authors have measured and correctly modeled the determinants of selection that reflect heterogeneity in the treatment effect. In simulations with a heterogeneous treatment effect, a conventional intent-to-treat estimate was biased with poor confidence limit coverage, but the proposed estimate was largely unbiased with appropriate confidence limit coverage. The proposed method standardizes observed trial results to a specified target population and thereby provides information regarding the generalizability of trial results.","DOI":"10.1093/aje/kwq084","ISSN":"0002-9262","note":"PMID: 20547574\nPMCID: PMC2915476","journalAbbreviation":"Am J Epidemiol","author":[{"family":"Cole","given":"Stephen R."},{"family":"Stuart","given":"Elizabeth A."}],"issued":{"date-parts":[["2010",7,1]]}}},{"id":2703,"uris":[""],"uri":[""],"itemData":{"id":2703,"type":"article-journal","title":"Generalizing causal inferences from randomized trials: counterfactual and graphical identification","container-title":"arXiv:1906.10792 [stat]","source":"","abstract":"When engagement with a randomized trial is driven by factors that affect the outcome or when trial engagement directly affects the outcome independent of treatment, the average treatment effect among trial participants is unlikely to generalize to a target population. In this paper, we use counterfactual and graphical causal models to examine under what conditions we can generalize causal inferences from a randomized trial to the target population of trial-eligible individuals. We offer an interpretation of generalizability analyses using the notion of a hypothetical intervention to \"scale-up\" trial engagement to the target population. We consider the interpretation of generalizability analyses when trial engagement does or does not directly affect the outcome, highlight connections with censoring in longitudinal studies, and discuss identification of the distribution of counterfactual outcomes via g-formula computation and inverse probability weighting. Last, we show how the methods can be extended to address time-varying treatments, non-adherence, and censoring.","URL":"","note":"arXiv: 1906.10792","title-short":"Generalizing causal inferences from randomized trials","author":[{"family":"Dahabreh","given":"Issa J."},{"family":"Robins","given":"James M."},{"family":"Haneuse","given":"Sebastien J.-P. A."},{"family":"Hernán","given":"Miguel A."}],"issued":{"date-parts":[["2019",6,25]]},"accessed":{"date-parts":[["2019",7,1]]}}}],"schema":""} (53–57). We do not make specific recommendations about external validity since this is an evolving area of research. However, it is important that the target population of interest be considered during the design phase of a pragmatic randomized trial and efforts made to ensure that the results of the trial will be applicable to that population. 5. Estimation of the per-protocol effect for point interventionsA point intervention is an intervention that takes place only once at the start of follow-up. A sustained intervention or treatment strategy, on the other hand, is sustained over time after baseline. For example, an intervention that consists of a screening test at baseline and lets individuals do whatever they wish after baseline is a point intervention, whereas an intervention that consists of a screening test at baseline and no additional tests during the follow-up is a sustained strategy. Note that treatment assignment is a point intervention (it only occurs once at the time of randomization) and thus the intention-to-treat effect is a contrast of point interventions. This section focuses on the estimation for per-protocol effects for point interventions. The next section focuses on sustained strategies. Because a point intervention is delivered at or close to the time of randomization, only covariates at or before the time of randomization can influence adherence to a point intervention. To validly estimate the per-protocol effect, baseline variables which predict adherence and are prognostic for the outcome need to be accounted for, either through direct adjustment or via an instrumental variable analysis. Yet two commonly used analytic approaches do not incorporate any such adjustment: Na?ve per-protocol analysis, that is, restricting the analytic subset to adherent individualsAs-treated analysis, that is, comparing individuals based on the treatment they choose Therefore, neither approach can generally provide unbiased estimates of the per-protocol effect when adherence does not occur at random (see Appendix) ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"UAH0S7DM","properties":{"formattedCitation":"(12)","plainCitation":"(12)","noteIndex":0},"citationItems":[{"id":1172,"uris":[""],"uri":[""],"itemData":{"id":1172,"type":"article-journal","title":"Beyond the intention-to-treat in comparative effectiveness research","container-title":"Clinical Trials","page":"48-55","volume":"9","issue":"1","source":"Nlm","archive_location":"21948059","abstract":"BACKGROUND: The intention-to-treat comparison is the primary, if not the only, analytic approach of many randomized clinical trials. PURPOSE: To review the shortcomings of intention-to-treat analyses, and of 'as treated' and 'per protocol' analyses as commonly implemented, with an emphasis on problems that are especially relevant for comparative effectiveness research. METHODS AND RESULTS: In placebo-controlled randomized clinical trials, intention-to-treat analyses underestimate the treatment effect and are therefore nonconservative for both safety trials and noninferiority trials. In randomized clinical trials with an active comparator, intention-to-treat estimates can overestimate a treatment's effect in the presence of differential adherence. In either case, there is no guarantee that an intention-to-treat analysis estimates the clinical effectiveness of treatment. Inverse probability weighting, g-estimation, and instrumental variable estimation can reduce the bias introduced by nonadherence and loss to follow-up in 'as treated' and 'per protocol' analyses. LIMITATIONS: These analyse require untestable assumptions, a dose-response model, and time-varying data on confounders and adherence. CONCLUSIONS: We recommend that all randomized clinical trials with substantial lack of adherence or loss to follow-up are analyzed using different methods. These include an intention-to-treat analysis to estimate the effect of assigned treatment and 'as treated' and 'per protocol' analyses to estimate the effect of treatment after appropriate adjustment via inverse probability weighting or g-estimation.","DOI":"10.1177/1740774511420743","ISSN":"1740-7753 (Electronic) 1740-7745 (Linking)","title-short":"Beyond the intention-to-treat in comparative effectiveness research","language":"eng","author":[{"family":"Hernán","given":"M. A."},{"family":"Hernandez-Diaz","given":"S."}],"issued":{"date-parts":[["2012",2]]}}}],"schema":""} (12). For example, in a trial of usual care vs. a new treatment, suppose that clinicians often put individuals with more severe disease on the new treatment (e.g., due to a perceived greater need) than in usual care, regardless of their randomized assignment. Then individuals on the new treatment will be, on average, sicker than the ones using usual care, and any analytic approach that directly compares users of new treatment vs. usual care will result in invalid per-protocol effect estimates.Here we consider two approaches to the estimation of per-protocol effects: direct adjustment for measured confounders and indirect adjustment based on instrumental variable estimation. In what follows we use the term “per-protocol analysis” to refer to analyses whose goal is the estimation of the per-protocol effect. Also, note that valid estimation of the per-protocol effect, like that of the intention-to-treat effect, requires appropriate adjustment for selection bias due to losses to follow-up and a choice of causal estimand when competing events exist. Therefore, the guidelines described above for intention-to-treat effects also apply to per-protocol effects and we will not repeat them here. Direct adjustment for confoundersOne option for validly estimating the per-protocol effect in a pragmatic trial with a point intervention is to directly adjust for baseline prognostic factors that are also predictors of adherence, i.e. baseline confounders. Many statistical approaches are valid to adjust for confounders in per-protocol analyses. Two of them, inverse probability and standardization ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"qzrKVz6v","properties":{"formattedCitation":"(58\\uc0\\u8211{}61)","plainCitation":"(58–61)","noteIndex":0},"citationItems":[{"id":1135,"uris":[""],"uri":[""],"itemData":{"id":1135,"type":"chapter","title":"Chapter 12: IP weighting and marginal structural models","container-title":"Causal Inference","publisher":"Chapman & Hall/CRC, forthcoming.","publisher-place":"Boca Raton","event-place":"Boca Raton","title-short":"Chapter 10: Random variability","author":[{"family":"Hernan","given":"MA"},{"family":"Robins","given":"JM"}],"issued":{"date-parts":[["2018"]]}},"label":"page"},{"id":1136,"uris":[""],"uri":[""],"itemData":{"id":1136,"type":"chapter","title":"Chapter 2: Randomized experiments","container-title":"Causal Inference","publisher":"Chapman & Hall/CRC, forthcoming.","publisher-place":"Boca Raton","event-place":"Boca Raton","title-short":"Chapter 10: Random variability","author":[{"family":"Hernan","given":"MA"},{"family":"Robins","given":"JM"}],"issued":{"date-parts":[["2018"]]}},"label":"page"},{"id":1134,"uris":[""],"uri":[""],"itemData":{"id":1134,"type":"chapter","title":"Chapter 13: Standardization and the parametric g-formula","container-title":"Causal Inference","publisher":"Chapman & Hall/CRC, forthcoming.","publisher-place":"Boca Raton","event-place":"Boca Raton","title-short":"Chapter 10: Random variability","author":[{"family":"Hernan","given":"MA"},{"family":"Robins","given":"JM"}],"issued":{"date-parts":[["2018"]]}},"label":"page"},{"id":1131,"uris":[""],"uri":[""],"itemData":{"id":1131,"type":"chapter","title":"Chapter 17: Causal survival analysis","container-title":"Causal Inference","publisher":"Chapman & Hall/CRC, forthcoming.","publisher-place":"Boca Raton","event-place":"Boca Raton","title-short":"Chapter 10: Random variability","author":[{"family":"Hernan","given":"MA"},{"family":"Robins","given":"JM"}],"issued":{"date-parts":[["2018"]]}},"label":"page"}],"schema":""} (58–61), allow calculation of absolute risks in the study population and are therefore preferable. Other commonly used adjustment methods, like outcome regression and propensity score adjustment or matching, typically make strong assumptions about no effect heterogeneity, and do not easily yield unconditional absolute risks.A good practice for any per-protocol analysis is to include an outcome that can serve as a negative control ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"lhJyctfZ","properties":{"formattedCitation":"(62,63)","plainCitation":"(62,63)","noteIndex":0},"citationItems":[{"id":1101,"uris":[""],"uri":[""],"itemData":{"id":1101,"type":"article-journal","title":"A New Method for Partial Correction of Residual Confounding in Time-Series and Other Observational Studies","container-title":"American Journal of Epidemiology","page":"941-949","volume":"185","issue":"10","source":"PubMed","abstract":"Methods exist to detect residual confounding in epidemiologic studies. One requires a negative control exposure with 2 key properties: 1) conditional independence of the negative control and the outcome (given modeled variables) absent confounding and other model misspecification, and 2) associations of the negative control with uncontrolled confounders and the outcome. We present a new method to partially correct for residual confounding: When confounding is present and our assumptions hold, we argue that estimators from models that include a negative control exposure with these 2 properties tend to be less biased than those from models without it. Using regression theory, we provide theoretical arguments that support our claims. In simulations, we empirically evaluated the approach using a time-series study of ozone effects on asthma emergency department visits. In simulations, effect estimators from models that included the negative control exposure (ozone concentrations 1 day after the emergency department visit) had slightly or modestly less residual confounding than those from models without it. Theory and simulations show that including the negative control can reduce residual confounding, if our assumptions hold. Our method differs from available methods because it uses a regression approach involving an exposure-based indicator rather than a negative control outcome to partially correct for confounding.","DOI":"10.1093/aje/kwx013","ISSN":"1476-6256","note":"PMID: 28430842","journalAbbreviation":"Am. J. Epidemiol.","language":"eng","author":[{"family":"Flanders","given":"W. Dana"},{"family":"Strickland","given":"Matthew J."},{"family":"Klein","given":"Mitchel"}],"issued":{"date-parts":[["2017"]],"season":"15"}},"label":"page"},{"id":1103,"uris":[""],"uri":[""],"itemData":{"id":1103,"type":"article-journal","title":"Negative Controls: A Tool for Detecting Confounding and Bias in Observational Studies","container-title":"Epidemiology (Cambridge, Mass.)","page":"383-388","volume":"21","issue":"3","source":"PubMed Central","abstract":"Non-causal associations between exposures and outcomes are a threat to validity of causal inference in observational studies. Many techniques have been developed for study design and analysis to identify and eliminate such errors. Such problems are not expected to compromise experimental studies, where careful standardization of conditions (for laboratory work) and randomization (for population studies) should, if applied properly, eliminate most such non-causal associations. We argue, however, that a routine precaution taken in the design of biological laboratory experiments—the use of “negative controls”—is designed to detect both suspected and unsuspected sources of spurious causal inference. In epidemiology, analogous negative controls help to identify and resolve confounding as well as other sources of error, including recall bias or analytic flaws. We distinguish two types of negative controls (exposure controls and outcome controls), describe examples of each type from the epidemiologic literature, and identify the conditions for the use of such negative controls to detect confounding. We conclude that negative controls should be more commonly employed in observational studies, and that additional work is needed to specify the conditions under which negative controls will be sensitive detectors of other sources of error in observational studies.","DOI":"10.1097/EDE.0b013e3181d61eeb","ISSN":"1044-3983","note":"PMID: 20335814\nPMCID: PMC3053408","title-short":"Negative Controls","journalAbbreviation":"Epidemiology","author":[{"family":"Lipsitch","given":"Marc"},{"family":"Tchetgen","given":"Eric Tchetgen"},{"family":"Cohen","given":"Ted"}],"issued":{"date-parts":[["2010",5]]}},"label":"page"}],"schema":""} (62,63). That is, an outcome known to be unaffected by treatment and whose association with treatment would suggest the presence of residual confounding for the primary outcome (see Appendix). Guideline: When sufficient data on baseline confounders exist, estimate the per-protocol effect of point interventions via adjustment by inverse probability weighting, standardization, doubly-robust estimation, or other methods. Using the randomized assignment as an instrumental variableWhen information on baseline confounders is not available, or we do not know why people adhere, sometimes a per-protocol effect for point interventions can be quantified via an instrumental variable analysis with randomization assignment as the instrument ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"ELZw8z01","properties":{"formattedCitation":"(64)","plainCitation":"(64)","noteIndex":0},"citationItems":[{"id":1040,"uris":[""],"uri":[""],"itemData":{"id":1040,"type":"article-journal","title":"Identification of causal effects using instrumental variables","container-title":"Journal of the American Statistical Association","page":"444-455","volume":"91","issue":"434","title-short":"Identification of causal effects using instrumental variables","author":[{"family":"Angrist","given":"J."},{"family":"Imbens","given":"G W."},{"family":"Rubin","given":"D B."}],"issued":{"date-parts":[["1996",6]]}}}],"schema":""} (64). The validity of instrumental variable methods requires several conditions, not all of which are guaranteed to hold in randomized trials. Specifically, the randomized assignment needs to meet three conditions to qualify as an instrumental variable (or instrument). An informal description of these instrumental conditions follows. First, randomization assignment must be predictive of treatment received. This is expected to hold in all trials, and can be empirically confirmed. Second, the effect of randomization assignment must be unconfounded. This cannot be confirmed empirically but is expected to hold in randomized trials. Note, however, that loss to follow-up and competing events can affect the validity of this assumption for instrumental variable analysis and may require covariate adjustment similar to that described in Sections 4.2 and 4.3. Third, the effect of treatment assignment on the outcome must be entirely mediated through the received treatment. This third condition, often referred to as the exclusion restriction, cannot be empirically verified and is not guaranteed to hold by design alone. In particular, this condition is less likely to hold in pragmatic trials that are unblinded (and therefore participants can alter their outcome risk through changes to behaviors or treatments) or which compare active treatments (where some participants do not adhere by taking neither treatment) ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"RG4sbEcQ","properties":{"formattedCitation":"(65,66)","plainCitation":"(65,66)","noteIndex":0},"citationItems":[{"id":64,"uris":[""],"uri":[""],"itemData":{"id":64,"type":"article-journal","title":"Selecting on treatment: a pervasive form of bias in instrumental variable analyses","container-title":"Am J Epidemiol","page":"191-7","volume":"181","issue":"3","archive_location":"25609096","abstract":"Instrumental variable (IV) methods are increasingly being used in comparative effectiveness research. Studies using these methods often compare 2 particular treatments, and the researchers perform their IV analyses conditional on patients' receiving this subset of treatments (while ignoring the third option of \"neither treatment\"). The ensuing selection bias that occurs due to this restriction has gone relatively unnoticed in interpretations and discussions of these studies' results. In this paper we describe the structure of this selection bias with examples drawn from commonly proposed instruments such as calendar time and preference, illustrate the bias with causal diagrams, and estimate the magnitude and direction of possible bias using simulations. A noncausal association between the proposed instrument and the outcome can occur in analyses restricted to patients receiving a subset of the possible treatments. This results in bias in the numerator for the standard IV estimator; the bias is amplified in the treatment effect estimate. The direction and magnitude of the bias in the treatment effect estimate are functions of the distribution of and relationships between the proposed instrument, treatment values, unmeasured confounders, and outcome. IV methods used to compare a subset of treatment options are prone to substantial biases, even when the proposed instrument appears relatively strong.","DOI":"10.1093/aje/kwu284","ISSN":"1476-6256","title-short":"Selecting on treatment: a pervasive form of bias in instrumental variable analyses","language":"eng","author":[{"family":"Swanson","given":"S. A."},{"family":"Robins","given":"J. M."},{"family":"Miller","given":"M."},{"family":"Hernán","given":"M. A."}],"issued":{"date-parts":[["2015",2]]}}},{"id":1158,"uris":[""],"uri":[""],"itemData":{"id":1158,"type":"article-journal","title":"Instruments for causal inference: an epidemiologist's dream?","container-title":"Epidemiology","page":"360-72","volume":"17","issue":"4","source":"Nlm","archive_location":"16755261","abstract":"The use of instrumental variable (IV) methods is attractive because, even in the presence of unmeasured confounding, such methods may consistently estimate the average causal effect of an exposure on an outcome. However, for this consistent estimation to be achieved, several strong conditions must hold. We review the definition of an instrumental variable, describe the conditions required to obtain consistent estimates of causal effects, and explore their implications in the context of a recent application of the instrumental variables approach. We also present (1) a description of the connection between 4 causal models-counterfactuals, causal directed acyclic graphs, nonparametric structural equation models, and linear structural equation models-that have been used to describe instrumental variables methods; (2) a unified presentation of IV methods for the average causal effect in the study population through structural mean models; and (3) a discussion and new extensions of instrumental variables methods based on assumptions of monotonicity.","DOI":"10.1097/01.ede.0000222409.00878.37","ISSN":"1044-3983 (Print) 1044-3983","title-short":"Instruments for causal inference: an epidemiologist's dream?","journalAbbreviation":"Epidemiology (Cambridge, Mass.)","language":"eng","author":[{"family":"Hernan","given":"M. A."},{"family":"Robins","given":"J. M."}],"issued":{"date-parts":[["2006",7]]}}}],"schema":""} (65,66). When the three instrumental conditions are expected to hold, randomized assignment can be used as an instrument to estimate upper and lower bounds for the per-protocol effect (Case Study H). That is, an instrumental variable does not contain enough information to obtain a point estimate of the causal effect, but only estimates of the upper and lower bounds of the causal effect (each bound has its own 95% confidence interval). These bounds are usually too wide and therefore relatively unhelpful to guide decision making, but the bounds can be narrowed down by making additional assumptions ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"aQicYx0s","properties":{"formattedCitation":"(67,68)","plainCitation":"(67,68)","noteIndex":0},"citationItems":[{"id":1068,"uris":[""],"uri":[""],"itemData":{"id":1068,"type":"article-journal","title":"Partial Identification of the Average Treatment Effect Using Instrumental Variables: Review of Methods for Binary Instruments, Treatments, and Outcomes","container-title":"Journal of the American Statistical Association","page":"933-947","volume":"113","issue":"522","source":"amstat. (Atypon)","abstract":"Several methods have been proposed for partially or point identifying the average treatment effect (ATE) using instrumental variable (IV) type assumptions. The descriptions of these methods are widespread across the statistical, economic, epidemiologic, and computer science literature, and the connections between the methods have not been readily apparent. In the setting of a binary instrument, treatment, and outcome, we review proposed methods for partial and point identification of the ATE under IV assumptions, express the identification results in a common notation and terminology, and propose a taxonomy that is based on sets of identifying assumptions. We further demonstrate and provide software for the application of these methods to estimate bounds. Supplementary materials for this article are available online.","DOI":"10.1080/01621459.2018.1434530","ISSN":"0162-1459","title-short":"Partial Identification of the Average Treatment Effect Using Instrumental Variables","journalAbbreviation":"Journal of the American Statistical Association","author":[{"family":"Swanson","given":"Sonja A."},{"family":"Hernán","given":"Miguel A."},{"family":"Miller","given":"Matthew"},{"family":"Robins","given":"James M."},{"family":"Richardson","given":"Thomas S."}],"issued":{"date-parts":[["2018",4,3]]}},"label":"page"},{"id":1169,"uris":[""],"uri":[""],"itemData":{"id":1169,"type":"article-journal","title":"Bounding the per-protocol effect in randomized trials: an application to colorectal cancer screening","container-title":"Trials","page":"541","volume":"16","archive_location":"26620120","abstract":"The per-protocol effect is the effect that would have been observed in a randomized trial had everybody followed the protocol. Though obtaining a valid point estimate for the per-protocol effect requires assumptions that are unverifiable and often implausible, lower and upper bounds for the per-protocol effect may be estimated under more plausible assumptions. Strategies for obtaining bounds, known as \"partial identification\" methods, are especially promising in randomized trials.|We estimated bounds for the per-protocol effect of colorectal cancer screening in the Norwegian Colorectal Cancer Prevention trial, a randomized trial of one-time sigmoidoscopy screening in 98,792 men and women aged 50-64 years. The screening was not available to the control arm, while approximately two thirds of individuals in the treatment arm attended the screening. Study outcomes included colorectal cancer incidence and mortality over 10 years of follow-up. Without any assumptions, the data alone provide little information about the size of the effect. Under the assumption that randomization had no effect on the outcome except through screening, a point estimate for the risk under no screening and bounds for the risk under screening are achievable. Thus, the 10-year risk difference for colorectal cancer was estimated to be at least -0.6 % but less than 37.0 %. Bounds for the risk difference for colorectal cancer mortality (-0.2 to 37.4 %) and all-cause mortality (-5.1 to 32.6 %) had similar widths. These bounds appear helpful in quantifying the maximum possible effectiveness, but cannot rule out harm. By making further assumptions about the effect in the subpopulation who would not attend screening regardless of their randomization arm, narrower bounds can be achieved.|Bounding the per-protocol effect under several sets of assumptions illuminates our reliance on unverifiable assumptions, highlights the range of effect sizes we are most confident in, and can sometimes demonstrate whether to expect certain subpopulations to receive more benefit or harm than others.| identifier NCT00119912 (registered 6 July 2005).","DOI":"10.1186/s13063-015-1056-8","ISSN":"1745-6215","title-short":"Bounding the per-protocol effect in randomized trials: an application to colorectal cancer screening","language":"eng","author":[{"family":"Swanson","given":"S. A."},{"family":"Holme","given":"?"},{"family":"L?berg","given":"M."},{"family":"Kalager","given":"M."},{"family":"Bretthauer","given":"M."},{"family":"Hoff","given":"G."},{"family":"Aas","given":"E."},{"family":"Hernán","given":"M. A."}],"issued":{"date-parts":[["2015",11]]}},"label":"page"}],"schema":""} (67,68). Case study H: Estimating bounds on the per-protocol effectThe Norwegian Colorectal Cancer Prevention (NORCCAP) trial was a primary prevention trial for colorectal cancer incidence and mortality comparing a one-time screening sigmoidoscopy with no screening ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"pmDV8Aq2","properties":{"formattedCitation":"(18)","plainCitation":"(18)","noteIndex":0},"citationItems":[{"id":1026,"uris":[""],"uri":[""],"itemData":{"id":1026,"type":"article-journal","title":"Long-Term Effectiveness of Sigmoidoscopy Screening on Colorectal Cancer Incidence and Mortality in Women and Men: A Randomized Trial","container-title":"Ann Intern Med","page":"775-782","volume":"168","issue":"11","archive_location":"29710125","abstract":"The long-term effects of sigmoidoscopy screening on colorectal cancer (CRC) incidence and mortality in women and men are unclear.|To determine the effectiveness of flexible sigmoidoscopy screening after 15 years of follow-up in women and men.|Randomized controlled trial. (: NCT00119912).|Oslo and Telemark County, Norway.|Adults aged 50 to 64 years at baseline without prior CRC.|Screening (between 1999 and 2001) with flexible sigmoidoscopy with and without additional fecal blood testing versus no screening. Participants with positive screening results were offered colonoscopy.|Age-adjusted CRC incidence and mortality stratified by sex.|Of 98?678 persons, 20?552 were randomly assigned to screening and 78?126 to no screening. Adherence rates were 64.7% in women and 61.4% in men. Median follow-up was 14.8 years. The absolute risks for CRC in women were 1.86% in the screening group and 2.05% in the control group (risk difference, -0.19 percentage point [95% CI, -0.49 to 0.11 percentage point]; HR, 0.92 [CI, 0.79 to 1.07]). In men, the corresponding risks were 1.72% and 2.50%, respectively (risk difference, -0.78 percentage point [CI, -1.08 to -0.48 percentage points]; hazard ratio [HR], 0.66 [CI, 0.57 to 0.78]) (P for heterogeneity?= 0.004). The absolute risks for death from CRC in women were 0.60% in the screening group and 0.59% in the control group (risk difference, 0.01 percentage point [CI, -0.16 to 0.18 percentage point]; HR, 1.01 [CI, 0.77 to 1.33]). The corresponding risks for death from CRC in men were 0.49% and 0.81%, respectively (risk difference, -0.33 percentage point [CI, -0.49 to -0.16 percentage point]; HR, 0.63 [CI, 0.47 to 0.83]) (P for heterogeneity?= 0.014).|Follow-up through national registries.|Offering sigmoidoscopy screening in Norway reduced CRC incidence and mortality in men but had little or no effect in women.|Norwegian government and Norwegian Cancer Society.","DOI":"10.7326/M17-1441","ISSN":"1539-3704","title-short":"Long-Term Effectiveness of Sigmoidoscopy Screening on Colorectal Cancer Incidence and Mortality in Women and Men: A Randomized Trial","language":"eng","author":[{"family":"Holme","given":"?"},{"family":"L?berg","given":"M."},{"family":"Kalager","given":"M."},{"family":"Bretthauer","given":"M."},{"family":"Hernán","given":"M. A."},{"family":"Aas","given":"E."},{"family":"Eide","given":"T. J."},{"family":"Skovlund","given":"E."},{"family":"Lekven","given":"J."},{"family":"Schneede","given":"J."},{"family":"Tveit","given":"K. M."},{"family":"Vatn","given":"M."},{"family":"Ursin","given":"G."},{"family":"Hoff","given":"G."},{"family":"Group?","given":"NORCCAP Study"}],"issued":{"date-parts":[["2018",6]]}}}],"schema":""} (18). Since this trial assessed a point intervention, bounds for the per-protocol effect could be obtained if the randomized assignment meets the three instrumental conditions.The first condition holds because individuals randomized to usual care were unable to obtain a screening sigmoidoscopy (not available in Norway) and because 65% of individuals randomized to the screening arm received a screening sigmoidoscopy. The second condition is expected to hold because randomization arm will be unconfounded by design. Finally, we assume that the exclusion restriction holds because it is unlikely that individuals could or would alter their risk of death or colorectal cancer after receiving the invitation letter for a sigmoidoscopy. In the NORCCAP trial, the age-standardized intention-to-treat difference in 10-year risk of all-cause mortality was -0.2 percentage points (95% CI: -0.6, 0.2). Under the three instrumental variable assumptions, the age-standardized per-protocol difference in the 10-year risk of all-cause mortality was bounded between -5.9 percentage points and 29.3 percentage points ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"tutdO5cb","properties":{"formattedCitation":"(68)","plainCitation":"(68)","noteIndex":0},"citationItems":[{"id":1169,"uris":[""],"uri":[""],"itemData":{"id":1169,"type":"article-journal","title":"Bounding the per-protocol effect in randomized trials: an application to colorectal cancer screening","container-title":"Trials","page":"541","volume":"16","archive_location":"26620120","abstract":"The per-protocol effect is the effect that would have been observed in a randomized trial had everybody followed the protocol. Though obtaining a valid point estimate for the per-protocol effect requires assumptions that are unverifiable and often implausible, lower and upper bounds for the per-protocol effect may be estimated under more plausible assumptions. Strategies for obtaining bounds, known as \"partial identification\" methods, are especially promising in randomized trials.|We estimated bounds for the per-protocol effect of colorectal cancer screening in the Norwegian Colorectal Cancer Prevention trial, a randomized trial of one-time sigmoidoscopy screening in 98,792 men and women aged 50-64 years. The screening was not available to the control arm, while approximately two thirds of individuals in the treatment arm attended the screening. Study outcomes included colorectal cancer incidence and mortality over 10 years of follow-up. Without any assumptions, the data alone provide little information about the size of the effect. Under the assumption that randomization had no effect on the outcome except through screening, a point estimate for the risk under no screening and bounds for the risk under screening are achievable. Thus, the 10-year risk difference for colorectal cancer was estimated to be at least -0.6 % but less than 37.0 %. Bounds for the risk difference for colorectal cancer mortality (-0.2 to 37.4 %) and all-cause mortality (-5.1 to 32.6 %) had similar widths. These bounds appear helpful in quantifying the maximum possible effectiveness, but cannot rule out harm. By making further assumptions about the effect in the subpopulation who would not attend screening regardless of their randomization arm, narrower bounds can be achieved.|Bounding the per-protocol effect under several sets of assumptions illuminates our reliance on unverifiable assumptions, highlights the range of effect sizes we are most confident in, and can sometimes demonstrate whether to expect certain subpopulations to receive more benefit or harm than others.| identifier NCT00119912 (registered 6 July 2005).","DOI":"10.1186/s13063-015-1056-8","ISSN":"1745-6215","title-short":"Bounding the per-protocol effect in randomized trials: an application to colorectal cancer screening","language":"eng","author":[{"family":"Swanson","given":"S. A."},{"family":"Holme","given":"?"},{"family":"L?berg","given":"M."},{"family":"Kalager","given":"M."},{"family":"Bretthauer","given":"M."},{"family":"Hoff","given":"G."},{"family":"Aas","given":"E."},{"family":"Hernán","given":"M. A."}],"issued":{"date-parts":[["2015",11]]}}}],"schema":""} (68). Guideline: When the three instrumental conditions are expected to hold for treatment assignment, estimate bounds for the per-protocol effect of point interventions. Provide a justification for why you believe the exclusion restriction holds, including performing appropriate falsification tests.When the three instrumental conditions and another unverifiable fourth condition is met, randomized assignment can be used as an instrument to obtain a point estimate of a per-protocol effect. A variety of fourth conditions exist, including different versions of effect homogeneity and monotonicity. Importantly, the choice of fourth condition determines the type of per-protocol effect that is targeted. Estimation under different forms of effect homogeneity leads to the per-protocol effect that we have considered throughout this section: the average effect in all individuals in the study population. However, effect homogeneity is often perceived as unrealistic because it requires that there is either a constant treatment effect among everyone or, informally, that there are no unmeasured modifiers of the effect of treatment on the outcome ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"m2FrQW6M","properties":{"formattedCitation":"(66)","plainCitation":"(66)","noteIndex":0},"citationItems":[{"id":1158,"uris":[""],"uri":[""],"itemData":{"id":1158,"type":"article-journal","title":"Instruments for causal inference: an epidemiologist's dream?","container-title":"Epidemiology","page":"360-72","volume":"17","issue":"4","source":"Nlm","archive_location":"16755261","abstract":"The use of instrumental variable (IV) methods is attractive because, even in the presence of unmeasured confounding, such methods may consistently estimate the average causal effect of an exposure on an outcome. However, for this consistent estimation to be achieved, several strong conditions must hold. We review the definition of an instrumental variable, describe the conditions required to obtain consistent estimates of causal effects, and explore their implications in the context of a recent application of the instrumental variables approach. We also present (1) a description of the connection between 4 causal models-counterfactuals, causal directed acyclic graphs, nonparametric structural equation models, and linear structural equation models-that have been used to describe instrumental variables methods; (2) a unified presentation of IV methods for the average causal effect in the study population through structural mean models; and (3) a discussion and new extensions of instrumental variables methods based on assumptions of monotonicity.","DOI":"10.1097/01.ede.0000222409.00878.37","ISSN":"1044-3983 (Print) 1044-3983","title-short":"Instruments for causal inference: an epidemiologist's dream?","journalAbbreviation":"Epidemiology (Cambridge, Mass.)","language":"eng","author":[{"family":"Hernan","given":"M. A."},{"family":"Robins","given":"J. M."}],"issued":{"date-parts":[["2006",7]]}}}],"schema":""} (66). Therefore, monotonicity is (implicitly or explicitly) used in most instrumental variable analyses. Informally, monotonicity means that no individuals in the trial would take exactly the treatment they were not assigned to, regardless of which treatment they were assigned to ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"4Z58h7IV","properties":{"formattedCitation":"(64,69)","plainCitation":"(64,69)","noteIndex":0},"citationItems":[{"id":1040,"uris":[""],"uri":[""],"itemData":{"id":1040,"type":"article-journal","title":"Identification of causal effects using instrumental variables","container-title":"Journal of the American Statistical Association","page":"444-455","volume":"91","issue":"434","title-short":"Identification of causal effects using instrumental variables","author":[{"family":"Angrist","given":"J."},{"family":"Imbens","given":"G W."},{"family":"Rubin","given":"D B."}],"issued":{"date-parts":[["1996",6]]}},"label":"page"},{"id":1066,"uris":[""],"uri":[""],"itemData":{"id":1066,"type":"article-journal","title":"The paired availability design: A proposal for evaluating epidural analgesia during labor","container-title":"Statistics in Medicine","page":"2269-2278","volume":"13","issue":"21","source":"Wiley Online Library","abstract":"The paired availability design (PAD) can reduce selection bias when it is not possible to randomize subjects. PAD consists of independent pairs of experimental and control groups. Within each pair, the intervention is the availability of treatment not its receipt. In the experimental group, the new treatment is made available to all subjects although some may not receive it. In the control group, the experimental treatment is generally not available to subjects although some may receive it in special circumstances. We present a statistic to test a null hypothesis that the receipt of intervention will increase response by a specified non-zero amount δ. We propose this design for use in a study of the effect of epidural analgesia on the rate of Caesarean section.","DOI":"10.1002/sim.4780132108","ISSN":"1097-0258","title-short":"The paired availability design","language":"en","author":[{"family":"Baker","given":"Stuart G."},{"family":"Lindeman","given":"Karen S."}],"issued":{"date-parts":[["1994"]]}},"label":"page"}],"schema":""} (64,69). The absence of these “defiers” seems to be a reasonable assumption in many trials, but estimation under monotonicity leads to a different per-protocol effect: the average effect in the subset of individuals who have the unobserved characteristic that they adhered to their assigned intervention and would have also adhered if they had adhered to the other intervention. Individuals in this unidentifiable subset of the population are referred to as “compliers”, in the sense that they would have complied with any treatment they would have been assigned to, and the effect in the compliers as the local average treatment effect, or LATE (Case Study I). Unfortunately, the subset of “compliers” is not usually identifiable. We can never know if a given adherent individual is someone who would have always adhered regardless of which treatment they had been assigned, or whether they only adhered in the actual trial because of the treatment they were in fact assigned. For example, some individuals who adhere to the comparator treatment may not have adhered to the investigational treatment because they would have experienced intolerable side effects on that treatment. Further, identifying those individuals outside of the trial participants for whom the LATE is relevant can be even more challenging.Therefore, in instrumental variable analyses under monotonicity, a discussion is needed as to why a per-protocol effect in the “compliers” is of interest. Also, the analysis needs to quantify the proportion of “compliers” in the study population and their characteristics compared with the rest of the population (interestingly, both these things can be achieved even though identifying individual “compliers” is not possible) ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"tssNblcn","properties":{"formattedCitation":"(70)","plainCitation":"(70)","noteIndex":0},"citationItems":[{"id":1151,"uris":[""],"uri":[""],"itemData":{"id":1151,"type":"article-journal","title":"The challenging interpretation of instrumental variable estimates under monotonicity","container-title":"International Journal of Epidemiology","page":"dyx038-dyx038","abstract":"Background: Instrumental variable (IV) methods are often used to identify ‘local’ causal effects in a subgroup of the population of interest. Such ‘local’ effects may not be ideal for informing clinical or policy decision making. When the instrument is non-causal, additional difficulties arise for interpreting ‘local’ effects. Little attention has been paid to these difficulties, even though commonly proposed instruments in epidemiology are non-causal (e.g. proxies for physician’s preference; genetic variants in some Mendelian randomization studies).Methods: For IV estimates obtained from both causal and non-causal instruments under monotonicity, we present results to help investigators pose four questions about the local effect estimates obtained in their studies. (1) To what subgroup of the population does the effect pertain? Can we (2) estimate the size of or (3) describe the characteristics of this subgroup relative to the study population? (4) Can the sensitivity of the effect estimate to deviations from monotonicity be quantified?Results: We show that the common interpretations and approaches for answering these four questions are generally only appropriate in the case of causal instruments.Conclusions: Appropriate interpretation of an IV estimate under monotonicity as a ‘local’ effect critically depends on whether the proposed instrument is causal or non-causal. The results and formal proofs presented here can help in the transparent reporting of IV results and in enhancing the use of IV estimates in informing decision-making efforts.","DOI":"10.1093/ije/dyx038","ISSN":"0300-5771","title-short":"The challenging interpretation of instrumental variable estimates under monotonicity","journalAbbreviation":"Int J Epidemiol","author":[{"family":"Swanson","given":"Sonja A."},{"family":"Hernán","given":"Miguel A."}],"issued":{"date-parts":[["2017"]]}}}],"schema":""} (70).Case study I: Estimating the per-protocol effect in a subset using randomization as an instrumentIn the NORCCAP trial, it is also reasonable to assume monotonicity of the instrument (randomization). In fact, the one-sided non-compliance in the NORCCAP trial guarantees that monotonicity holds – that is, that there are no individuals who would have refused a screening sigmoidoscopy if randomized to that arm, but have somehow obtained one if randomized to usual care. Under this assumption and the three instrumental conditions, we can use randomization as an instrument to estimate the per-protocol effect among the compliers Among women, the estimated per-protocol difference in 15-year risk of colorectal cancer incidence in the treated was -0.27 percentage points (95% CI: -0.72, 0.18) ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"GDq67nqe","properties":{"formattedCitation":"(18)","plainCitation":"(18)","noteIndex":0},"citationItems":[{"id":1026,"uris":[""],"uri":[""],"itemData":{"id":1026,"type":"article-journal","title":"Long-Term Effectiveness of Sigmoidoscopy Screening on Colorectal Cancer Incidence and Mortality in Women and Men: A Randomized Trial","container-title":"Ann Intern Med","page":"775-782","volume":"168","issue":"11","archive_location":"29710125","abstract":"The long-term effects of sigmoidoscopy screening on colorectal cancer (CRC) incidence and mortality in women and men are unclear.|To determine the effectiveness of flexible sigmoidoscopy screening after 15 years of follow-up in women and men.|Randomized controlled trial. (: NCT00119912).|Oslo and Telemark County, Norway.|Adults aged 50 to 64 years at baseline without prior CRC.|Screening (between 1999 and 2001) with flexible sigmoidoscopy with and without additional fecal blood testing versus no screening. Participants with positive screening results were offered colonoscopy.|Age-adjusted CRC incidence and mortality stratified by sex.|Of 98?678 persons, 20?552 were randomly assigned to screening and 78?126 to no screening. Adherence rates were 64.7% in women and 61.4% in men. Median follow-up was 14.8 years. The absolute risks for CRC in women were 1.86% in the screening group and 2.05% in the control group (risk difference, -0.19 percentage point [95% CI, -0.49 to 0.11 percentage point]; HR, 0.92 [CI, 0.79 to 1.07]). In men, the corresponding risks were 1.72% and 2.50%, respectively (risk difference, -0.78 percentage point [CI, -1.08 to -0.48 percentage points]; hazard ratio [HR], 0.66 [CI, 0.57 to 0.78]) (P for heterogeneity?= 0.004). The absolute risks for death from CRC in women were 0.60% in the screening group and 0.59% in the control group (risk difference, 0.01 percentage point [CI, -0.16 to 0.18 percentage point]; HR, 1.01 [CI, 0.77 to 1.33]). The corresponding risks for death from CRC in men were 0.49% and 0.81%, respectively (risk difference, -0.33 percentage point [CI, -0.49 to -0.16 percentage point]; HR, 0.63 [CI, 0.47 to 0.83]) (P for heterogeneity?= 0.014).|Follow-up through national registries.|Offering sigmoidoscopy screening in Norway reduced CRC incidence and mortality in men but had little or no effect in women.|Norwegian government and Norwegian Cancer Society.","DOI":"10.7326/M17-1441","ISSN":"1539-3704","title-short":"Long-Term Effectiveness of Sigmoidoscopy Screening on Colorectal Cancer Incidence and Mortality in Women and Men: A Randomized Trial","language":"eng","author":[{"family":"Holme","given":"?"},{"family":"L?berg","given":"M."},{"family":"Kalager","given":"M."},{"family":"Bretthauer","given":"M."},{"family":"Hernán","given":"M. A."},{"family":"Aas","given":"E."},{"family":"Eide","given":"T. J."},{"family":"Skovlund","given":"E."},{"family":"Lekven","given":"J."},{"family":"Schneede","given":"J."},{"family":"Tveit","given":"K. M."},{"family":"Vatn","given":"M."},{"family":"Ursin","given":"G."},{"family":"Hoff","given":"G."},{"family":"Group?","given":"NORCCAP Study"}],"issued":{"date-parts":[["2018",6]]}}}],"schema":""} (18). For comparison, the estimated intention-to-treat difference in 15-year risk of colorectal cancer incidence was -0.19 percentage points (95% CI: -0.49, 0.11) ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"bCLF1AYA","properties":{"formattedCitation":"(18)","plainCitation":"(18)","noteIndex":0},"citationItems":[{"id":1026,"uris":[""],"uri":[""],"itemData":{"id":1026,"type":"article-journal","title":"Long-Term Effectiveness of Sigmoidoscopy Screening on Colorectal Cancer Incidence and Mortality in Women and Men: A Randomized Trial","container-title":"Ann Intern Med","page":"775-782","volume":"168","issue":"11","archive_location":"29710125","abstract":"The long-term effects of sigmoidoscopy screening on colorectal cancer (CRC) incidence and mortality in women and men are unclear.|To determine the effectiveness of flexible sigmoidoscopy screening after 15 years of follow-up in women and men.|Randomized controlled trial. (: NCT00119912).|Oslo and Telemark County, Norway.|Adults aged 50 to 64 years at baseline without prior CRC.|Screening (between 1999 and 2001) with flexible sigmoidoscopy with and without additional fecal blood testing versus no screening. Participants with positive screening results were offered colonoscopy.|Age-adjusted CRC incidence and mortality stratified by sex.|Of 98?678 persons, 20?552 were randomly assigned to screening and 78?126 to no screening. Adherence rates were 64.7% in women and 61.4% in men. Median follow-up was 14.8 years. The absolute risks for CRC in women were 1.86% in the screening group and 2.05% in the control group (risk difference, -0.19 percentage point [95% CI, -0.49 to 0.11 percentage point]; HR, 0.92 [CI, 0.79 to 1.07]). In men, the corresponding risks were 1.72% and 2.50%, respectively (risk difference, -0.78 percentage point [CI, -1.08 to -0.48 percentage points]; hazard ratio [HR], 0.66 [CI, 0.57 to 0.78]) (P for heterogeneity?= 0.004). The absolute risks for death from CRC in women were 0.60% in the screening group and 0.59% in the control group (risk difference, 0.01 percentage point [CI, -0.16 to 0.18 percentage point]; HR, 1.01 [CI, 0.77 to 1.33]). The corresponding risks for death from CRC in men were 0.49% and 0.81%, respectively (risk difference, -0.33 percentage point [CI, -0.49 to -0.16 percentage point]; HR, 0.63 [CI, 0.47 to 0.83]) (P for heterogeneity?= 0.014).|Follow-up through national registries.|Offering sigmoidoscopy screening in Norway reduced CRC incidence and mortality in men but had little or no effect in women.|Norwegian government and Norwegian Cancer Society.","DOI":"10.7326/M17-1441","ISSN":"1539-3704","title-short":"Long-Term Effectiveness of Sigmoidoscopy Screening on Colorectal Cancer Incidence and Mortality in Women and Men: A Randomized Trial","language":"eng","author":[{"family":"Holme","given":"?"},{"family":"L?berg","given":"M."},{"family":"Kalager","given":"M."},{"family":"Bretthauer","given":"M."},{"family":"Hernán","given":"M. A."},{"family":"Aas","given":"E."},{"family":"Eide","given":"T. J."},{"family":"Skovlund","given":"E."},{"family":"Lekven","given":"J."},{"family":"Schneede","given":"J."},{"family":"Tveit","given":"K. M."},{"family":"Vatn","given":"M."},{"family":"Ursin","given":"G."},{"family":"Hoff","given":"G."},{"family":"Group?","given":"NORCCAP Study"}],"issued":{"date-parts":[["2018",6]]}}}],"schema":""} (18).Guideline: When the three instrumental conditions and monotonicity are expected to hold, discuss whether the effect in the “compliers” is of interest. If so, estimate it and provide information on the relative size and characteristics of the “compliers” subset.Estimation of the per-protocol effect for sustained treatment strategiesSustained treatment strategies are those that start at baseline and continue during the follow-up. For example, “take daily medication” or “never receive a screening colonoscopy during follow-up”. Sustained treatment strategies can be further classified into static strategies, under which all individuals take the same treatment during the follow-up (e.g., “take aspirin every day no matter what happens”), or dynamic strategies, under which the treatment received by each individual depends on his/her evolving characteristics (e.g., “take aspirin every day until a diagnosis of stroke”). Dynamic treatment strategies are usually the most relevant for clinical decision-making. When estimating the per-protocol effect of sustained treatment strategies the first step is to specify the strategies. Ideally, a precise definition of the strategies would be found in the trial protocol. However, it is not uncommon to encounter situations in which the protocol is ambiguous regarding the strategies being studied. For example, many protocols give study clinicians discretion to advise individuals who experience mild side effects to discontinue treatment. Then the treatment strategy may be a dynamic strategy that allows discontinuation after side effects than a static strategy of continuous use of trial medications. In fact, the dynamic strategy is arguably the most clinically relevant and therefore the one that should be considered when estimating the per-protocol effect (Case study J).In addition, realistic treatment strategies often need to incorporate a grace period during which treatment decisions, such as initiation, dosage adjustment, discontinuation, or switching, are allowed to occur. For example, consider a protocol that specifies that treatment initiation should occur after reaching a particular threshold (e.g., exceeding a systolic blood pressure of 120 mm Hg). Because it would be unrealistic to expect that treatment initiation will occur exactly on the same day as the threshold is reached, the protocol should also specify a grace period (say, 2 months after the threshold) during which a treatment initiation will be considered adherent. The existence of a grace period means that the treatment strategy is sustained. Case study J: Specifying the protocol when estimating per-protocol effects for safety outcomesThe Saxagliptin Assessment of Vascular Outcomes Recorded in Patients with Diabetes Mellitus—Thrombolysis in Myocardial Infarction 53 (SAVOR-TIMI 53) trial was a double-blind, placebo-controlled randomized clinical trial designed to assess the safety and efficacy of saxagliptin for cardiovascular outcomes among patients with diabetes mellitus ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"qTVlM5vV","properties":{"formattedCitation":"(71)","plainCitation":"(71)","noteIndex":0},"citationItems":[{"id":711,"uris":[""],"uri":[""],"itemData":{"id":711,"type":"article-journal","title":"Saxagliptin and cardiovascular outcomes in patients with type 2 diabetes mellitus","container-title":"N Engl J Med","page":"1317-26","volume":"369","issue":"14","source":"NLM","archive_location":"23992601","DOI":"10.1056/NEJMoa1307684","ISSN":"0028-4793","title-short":"Saxagliptin and cardiovascular outcomes in patients with type 2 diabetes mellitus","journalAbbreviation":"The New England journal of medicine","language":"eng","author":[{"family":"Scirica","given":"B. M."},{"family":"Bhatt","given":"D. L."},{"family":"Braunwald","given":"E."},{"family":"Steg","given":"P. G."},{"family":"Davidson","given":"J."},{"family":"Hirshberg","given":"B."},{"family":"Ohman","given":"P."},{"family":"Frederich","given":"R."},{"family":"Wiviott","given":"S. D."},{"family":"Hoffman","given":"E. B."},{"family":"Cavender","given":"M. A."},{"family":"Udell","given":"J. A."},{"family":"Desai","given":"N. R."},{"family":"Mosenzon","given":"O."},{"family":"McGuire","given":"D. K."},{"family":"Ray","given":"K. K."},{"family":"Leiter","given":"L. A."},{"family":"Raz","given":"I."}],"issued":{"date-parts":[["2013",10,3]]}}}],"schema":""} (71). Participants were randomized to initiate either saxagliptin or placebo and continue receiving treatment until symptoms of renal impairment developed, at which point they would have a single dosage adjustment, and no patients were allowed to take other DPP-4 inhibitors or glucagon-like peptide 1 agonists ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"ROxFP5cq","properties":{"formattedCitation":"(71)","plainCitation":"(71)","noteIndex":0},"citationItems":[{"id":711,"uris":[""],"uri":[""],"itemData":{"id":711,"type":"article-journal","title":"Saxagliptin and cardiovascular outcomes in patients with type 2 diabetes mellitus","container-title":"N Engl J Med","page":"1317-26","volume":"369","issue":"14","source":"NLM","archive_location":"23992601","DOI":"10.1056/NEJMoa1307684","ISSN":"0028-4793","title-short":"Saxagliptin and cardiovascular outcomes in patients with type 2 diabetes mellitus","journalAbbreviation":"The New England journal of medicine","language":"eng","author":[{"family":"Scirica","given":"B. M."},{"family":"Bhatt","given":"D. L."},{"family":"Braunwald","given":"E."},{"family":"Steg","given":"P. G."},{"family":"Davidson","given":"J."},{"family":"Hirshberg","given":"B."},{"family":"Ohman","given":"P."},{"family":"Frederich","given":"R."},{"family":"Wiviott","given":"S. D."},{"family":"Hoffman","given":"E. B."},{"family":"Cavender","given":"M. A."},{"family":"Udell","given":"J. A."},{"family":"Desai","given":"N. R."},{"family":"Mosenzon","given":"O."},{"family":"McGuire","given":"D. K."},{"family":"Ray","given":"K. K."},{"family":"Leiter","given":"L. A."},{"family":"Raz","given":"I."}],"issued":{"date-parts":[["2013",10,3]]}}}],"schema":""} (71). One possible interpretation of the trial protocol is that non-adherence includes initiation of any other DPP-4 inhibitor or glucagon-like peptide 1 agonist, or discontinuation of the drug at any time. Another possible interpretation of the trial protocol is that non-adherence also includes failure to decrease study drug dose upon incident renal impairment, or alterations in dosage of study drug other than at the time of incident renal impairment. Each of these definitions would have different implications for data analysis.Guideline: To estimate the per-protocol effect of sustained treatment strategies, specify a priori a treatment protocol that incorporates real world clinical decision-making, including discontinuation, switching, or dose-reduction rules. When there is sufficient ambiguity about the appropriate treatment strategies, more than one protocol strategy can be specified. Because a sustained strategy is delivered during the follow-up, covariates at any time before or after randomization can influence adherence. That is, the estimation of per-protocol effects for sustained interventions require adjustment for pre- and post-randomization prognostic factors that predict adherence during the follow-up.As with point interventions, there are two basic approaches to estimating per-protocol effects of sustained treatment strategies—direct and indirect adjustment for post-randomization confounding. First, we can directly adjust for pre- and post-randomization confounders using g-methods—inverse probability weighting, the g-formula, or g-estimation of structural nested models—which require the same untestable assumptions that are needed for causal analyses of observational studies. Second, we can use extensions of g-estimation which generalizes instrumental variable estimation to the setting of sustained treatment strategies. This second approach, which requires the instrumental assumptions plus detailed modeling assumptions about the effect of treatment on the outcome of interest ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"gudosYKK","properties":{"formattedCitation":"(72\\uc0\\u8211{}74)","plainCitation":"(72–74)","noteIndex":0},"citationItems":[{"id":1130,"uris":[""],"uri":[""],"itemData":{"id":1130,"type":"chapter","title":"Chapter 21: G-methods for time-varying treatments","container-title":"Causal Inference","publisher":"Chapman & Hall/CRC, forthcoming.","publisher-place":"Boca Raton","event-place":"Boca Raton","title-short":"Chapter 10: Random variability","author":[{"family":"Hernan","given":"MA"},{"family":"Robins","given":"JM"}],"issued":{"date-parts":[["2018"]]}},"label":"page"},{"id":1147,"uris":[""],"uri":[""],"itemData":{"id":1147,"type":"article-journal","title":"Correcting for non-compliance in randomized trials using structural nested mean models","container-title":"Communications in Statistics","page":"2379-2412","volume":"23","author":[{"family":"Robins","given":"James M."}],"issued":{"date-parts":[["1994"]]}},"label":"page"},{"id":1133,"uris":[""],"uri":[""],"itemData":{"id":1133,"type":"chapter","title":"Chapter 14: G-estimation of structural nested models","container-title":"Causal Inference","publisher":"Chapman & Hall/CRC, forthcoming.","publisher-place":"Boca Raton","event-place":"Boca Raton","title-short":"Chapter 10: Random variability","author":[{"family":"Hernan","given":"MA"},{"family":"Robins","given":"JM"}],"issued":{"date-parts":[["2018"]]}},"label":"page"}],"schema":""} (72–74), has been rarely used and will not be discussed here ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"vcXSxeMR","properties":{"formattedCitation":"(75\\uc0\\u8211{}78)","plainCitation":"(75–78)","noteIndex":0},"citationItems":[{"id":919,"uris":[""],"uri":[""],"itemData":{"id":919,"type":"article-journal","title":"A method for the analysis of randomized trials with compliance information: an application to the Multiple Risk Factor Intervention Trial","container-title":"Controlled Clinical Trials","page":"79-97","volume":"14","issue":"2","source":"Nlm","archive_location":"8500308","abstract":"The standard approach to analyzing randomized trials ignores information on postrandomization compliance. Application of these methods results in estimates that may lack the desired causal interpretation. We employ a new method of estimation and analyze data from the Multiple Risk Factor Intervention Trial (MRFIT) to estimate the causal effect of quitting cigarette smoking. Our procedure utilizes a method proposed by Robins and Tsiatis and allows us to take advantage of postrandomization smoking history without requiring untenable assumptions about the comparability of compliers and noncompliers. We contrast the performance of our method and the standard intent-to-treat analysis in the MRFIT data and in simulated data in which compliance rates are varied.","ISSN":"0197-2456 (Print) 0197-2456","title-short":"A method for the analysis of randomized trials with compliance information: an application to the Multiple Risk Factor Intervention Trial","journalAbbreviation":"Controlled clinical trials","language":"eng","author":[{"family":"Mark","given":"S. D."},{"family":"Robins","given":"J. M."}],"issued":{"date-parts":[["1993",4]]}},"label":"page"},{"id":1095,"uris":[""],"uri":[""],"itemData":{"id":1095,"type":"article-journal","title":"Correcting for non-compliance in randomized trials: an application to the ATBC study","container-title":"Statistics in Medicine","page":"2879-2897","volume":"18","issue":"21","source":"Wiley Online Library","abstract":"Different methods for estimating the effect of treatment actually received in a longitudinal placebo-controlled trial with non-compliance are discussed. Total mortality from the ATBC Study is used as an illustrative example. In the ATBC Study some 25 per cent of the participants dropped out from active follow-up prior to the scheduled end of the study. The ‘intention-to-treat’ analysis showed an increased death risk in the beta-carotene arm when compared with the no beta-carotene arm. Owing to considerable non-compliance it is also of interest to estimate the effect of beta-carotene actually received. We use a simple model for the treatment action and discuss three methods for estimation of the treatment effect under the model – the ‘intention-to-treat’ approach, the ‘as-treated’ approach and the g-estimation approach. These approaches are compared in a simulation study under different settings for non-compliance. Finally, the data from the ATBC Study are analysed using the proposed methods. Copyright ? 1999 John Wiley & Sons, Ltd.","DOI":"10.1002/(SICI)1097-0258(19991115)18:21<2879::AID-SIM190>3.0.CO;2-K","ISSN":"1097-0258","title-short":"Correcting for non-compliance in randomized trials","language":"en","author":[{"family":"Korhonen","given":"Pasi A."},{"family":"Laird","given":"Nan M."},{"family":"Palmgren","given":"Juni"}],"issued":{"date-parts":[["1999"]]}},"label":"page"},{"id":1097,"uris":[""],"uri":[""],"itemData":{"id":1097,"type":"article-journal","title":"Estimating effects from randomized trials with discontinuations: the need for intent-to-treat design and G-estimation","container-title":"Clinical Trials","page":"5-13","volume":"5","issue":"1","source":"SAGE Journals","abstract":"Background Randomized trials provide pivotal evidence for evaluation and approval of therapies. Nonetheless, such trials are often plagued by noncompliance, especially in the form of premature discontinuation of treatment. While intent-to-treat (ITT) analysis can provide valid tests of no-effect hypotheses, some trials may make ITT analysis impossible by ceasing follow-up when patients go off assigned treatment. Furthermore, estimates based on ITT, on-treatment, or per-protocol comparisons can seriously understate harm or benefit., Purpose To show how g-estimation based on randomization status is a natural generalization of ITT null testing to estimating efficacy from trials with important discontinuation or noncompliance., Methods We contrast with an analysis of the effect of a tiotropium inhaler on the occurrence of chronic obstructive pulmonary disease (COPD) events in a six-month double-blind placebo-controlled trial of 1829 patients with good but imperfect compliance., Results The covariate-adjusted point estimates, 95% confidence limits (CL), and null P-values comparing expected COPD event times in placebo versus tiotropium patients were: ITT, 1.21, CL = 1.02, 1.43, P = 0.027; on-treatment, 1.27, CL = 1.06, 1.52, P = 0.009; per-protocol, 1.36, CL = 1.13, 1.63, P = 0.001; and g-estimation, 1.31, CL = 1.03,1.72, P = 0.027. Thus g-estimation preserved the ITT test of the null, but exhibited more uncertainty about the size of the tiotropium effect than the other methods. In particular, it allowed for a much larger potential effect than did ITT analysis, but produced a much larger null P than exhibited by per-protocol analysis. Limitations Like ITT analysis, g-estimation requires all patients be followed to the end of the trial protocol, regardless of whether they comply with the protocol. Like on-treatment and per-protocol analyses, it also requires accurate compliance information be recorded., Conclusion G-estimation should become a standard procedure for the analysis of trials with noncompliance. Software to do so is available in major packages, and the procedure is easily coded for other packages. Clinical Trials 2008; 5: 5—13. ","DOI":"10.1177/1740774507087703","ISSN":"1740-7745","title-short":"Estimating effects from randomized trials with discontinuations","journalAbbreviation":"Clinical Trials","language":"en","author":[{"family":"Greenland","given":"Sander"},{"family":"Lanes","given":"Stephan"},{"family":"Jara","given":"Michele"}],"issued":{"date-parts":[["2008",2,1]]}},"label":"page"},{"id":1098,"uris":[""],"uri":[""],"itemData":{"id":1098,"type":"article-journal","title":"Effect of acyclovir on herpetic ocular recurrence using a structural nested model","container-title":"Contemporary Clinical Trials","page":"300-310","volume":"26","issue":"3","source":"ScienceDirect","abstract":"Noncompliance with assigned therapies is ubiquitous in randomized clinical trials. Treatment effects may be corrected for noncompliance using Robins' structural nested models, but few examples have been published. The Herpetic Eye Disease Study randomized 703 ocular herpes patients to 365 days of acyclovir or placebo between 1992 and 1996, and achieved over 90% compliance in both arms. The hazard of recurrence in the acyclovir arm was 0.55 times the hazard in the placebo arm using an intent-to-treat approach (95% confidence interval [CI]: 0.41, 0.75). Assuming a structural nested model with a Weibull distribution, the hazard of recurrence under constant exposure to acyclovir was 0.41 times that of the non-exposed (test-based 95% CI: 0.28, 0.72), or 34% larger than the intent-to-treat estimate. Notwithstanding excellent compliance, intent-to-treat estimates may notably undervalue the causal effect of a treatment.","DOI":"10.1016/t.2005.01.009","ISSN":"1551-7144","journalAbbreviation":"Contemporary Clinical Trials","author":[{"family":"Cole","given":"Stephen R."},{"family":"Chu","given":"Haitao"}],"issued":{"date-parts":[["2005",6,1]]}},"label":"page"}],"schema":""} (75–78). 6.1 Data requirementsIn addition to the information required to estimate the per-protocol effects of point interventions, valid estimation of per-protocol effects of sustained treatment strategies requires accurate information on a variety of post-randomization variables. These include variables that are needed to determine adherence to the treatment strategy (e.g., treatment use and dose, contraindications, adverse effects) and post-randomization confounders.Trials with sparse data or non-systematic collection after randomization are less likely to be useful for the estimation of per-protocol effects. Simulation studies confirm that the bias for the per-protocol effect increases as the inter-visit interval gets larger ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"Fmo1wFms","properties":{"formattedCitation":"(79)","plainCitation":"(79)","noteIndex":0},"citationItems":[{"id":2881,"uris":[""],"uri":[""],"itemData":{"id":2881,"type":"article-journal","title":"Interval-cohort designs and bias in the estimation of per-protocol effects: a simulation study","container-title":"Trials","page":"552","volume":"20","issue":"1","abstract":"Randomized trials are considered the gold standard for making inferences about the causal effects of treatments. However, when protocol deviations occur, the baseline randomization of the trial is no longer sufficient to ensure unbiased estimation of the per-protocol effect: post-randomization, time-varying confounders must be sufficiently measured and adjusted for in the analysis. Given the historical emphasis on intention-to-treat effects in randomized trials, measurement of post-randomization confounders is typically infrequent. This may induce bias in estimates of the per-protocol effect, even using methods such as inverse probability weighting, which appropriately account for time-varying confounders affected by past treatment.","DOI":"10.1186/s13063-019-3577-z","ISSN":"1745-6215","journalAbbreviation":"Trials","author":[{"family":"Young","given":"Jessica G."},{"family":"Vatsa","given":"Rajet"},{"family":"Murray","given":"Eleanor J."},{"family":"Hernán","given":"Miguel A."}],"issued":{"date-parts":[["2019",9,5]]}}}],"schema":""} (79), whereas the intention-to-treat effect estimate is not affected by the frequency of study visits, since it does not depend on adjustment for post-randomization confounders.When frequent visits are not possible, linking to electronic health records may be an alternative method of collecting data on post-randomization variables. However, careful thought should be given to the completeness, accuracy, and usefulness of this information source for the purposes of the trial (see Appendix). Guideline: Ensure that sufficient data are collected to determine whether participants adhered to their assigned strategies throughout the follow-up, and to adjust for time-varying prognostic factors that predict adherence to the assigned treatment strategies. Methods to adjust for post-randomization confoundersWhen comparing sustained treatment strategies, the post-baseline treatment is time-varying and therefore the post-baseline confounders are also time-varying. In many clinical settings, the time-varying confounders are themselves affected by prior treatment, and thus we say that there is treatment-confounder feedback. For example, when comparing the effect of epoetin dosing strategies on the mortality of individuals with end-stage renal disease, the hemoglobin value is a time-varying confounder because hemoglobin is a prognostic factor that is used to decide the epoetin dose and, in addition, there is treatment-confounder feedback because prior epoetin doses affect subsequent hemoglobin values.Conventional adjustment methods cannot handle time-varying confounders when there is treatment-confounder feedback. In fact, conventional methods such as multivariate outcome regression, stratified analyses, propensity score regression and matching, and others will be biased when time-varying confounders are omitted and may introduce bias when time-varying confounders are included in the models ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"96a9wIkT","properties":{"unsorted":false,"formattedCitation":"(49,80\\uc0\\u8211{}82)","plainCitation":"(49,80–82)","noteIndex":0},"citationItems":[{"id":1176,"uris":[""],"uri":[""],"itemData":{"id":1176,"type":"article-journal","title":"A structural approach to selection bias","container-title":"Epidemiology","page":"615-25","volume":"15","issue":"5","archive_location":"15308962","abstract":"The term \"selection bias\" encompasses various biases in epidemiology. We describe examples of selection bias in case-control studies (eg, inappropriate selection of controls) and cohort studies (eg, informative censoring). We argue that the causal structure underlying the bias in each example is essentially the same: conditioning on a common effect of 2 variables, one of which is either exposure or a cause of exposure and the other is either the outcome or a cause of the outcome. This structure is shared by other biases (eg, adjustment for variables affected by prior exposure). A structural classification of bias distinguishes between biases resulting from conditioning on common effects (\"selection bias\") and those resulting from the existence of common causes of exposure and outcome (\"confounding\"). This classification also leads to a unified approach to adjust for selection bias.","ISSN":"1044-3983","title-short":"A structural approach to selection bias","language":"eng","author":[{"family":"Hernán","given":"M. A."},{"family":"Hernández-Díaz","given":"S."},{"family":"Robins","given":"J. M."}],"issued":{"date-parts":[["2004",9]]}},"label":"page"},{"id":1087,"uris":[""],"uri":[""],"itemData":{"id":1087,"type":"article-journal","title":"A new approach to causal inference in mortality studies with a sustained exposure period — Application to the healthy worker survivor effect","container-title":"Mathematical Modelling","page":"1393-1512","volume":"7","author":[{"family":"Robins","given":"JM"}],"issued":{"date-parts":[["1986"]]}},"label":"page"},{"id":1088,"uris":[""],"uri":[""],"itemData":{"id":1088,"type":"article-journal","title":"Addendum to \"A new approach to causal inference in mortality studies with a sustained exposure period — Application to the healthy worker survivor effect\" [published errata appear in Computers Math Applic 1989:18;477]","container-title":"Computers and Mathematics with Applications","page":"923-45","volume":"14","title-short":"Addendum to \"A new approach to causal inference in mortality studies with a sustained exposure period — Application to the healthy worker survivor effect\" [published errata appear in Computers Math Applic 1989:18;477]","author":[{"family":"Robins","given":"James M"}],"issued":{"date-parts":[["1987"]]}},"label":"page"},{"id":1148,"uris":[""],"uri":[""],"itemData":{"id":1148,"type":"article-journal","title":"A graphical approach to the identification and estimation of causal parameters in mortality studies with sustained exposure periods.","container-title":"Journal of chronic diseases","page":"139S-161S","volume":"40 Suppl 2","abstract":"In observational cohort mortality studies with prolonged periods of exposure to the agent under study, independent risk factors for death commonly determine subsequent exposure to the study agent. For example, in occupational mortality studies, date of termination of employment is both a determinant of subsequent exposure to the chemical agent under study (since terminated individuals receive no further exposure) and an independent risk factor for death (since disabled individuals tend to leave employment). When a risk factor determines subsequent exposure and is determined by previous exposure, standard analyses that estimate age-specific mortality rates as a function of cumulative exposure can underestimate the true effect of exposure on mortality, whether or not one adjusts for the risk factor in the analysis. This observation raises the question, \"Which, if any, empirical population parameter can be causally interpreted as the true effect of exposure in observational mortality studies?\" In answer, we offer a graphical approach to the identification and estimation of causal parameters in mortality studies with sustained exposure periods. We reanalyze the mortality experience of a cohort of arsenic-exposed copper smelter workers using our approach and compare our results with those obtained using standard methods. We find an adverse effect of arsenic exposure on all cause and lung cancer mortality, which standard methods failed to detect. The analytic approach introduced in this paper may be necessary to control bias in any epidemiologic study in which there exists a risk factor which both determines subsequent exposure and is determined by previous exposure to the agent under study.","ISSN":"0021-9681 0021-9681","note":"PMID: 3667861","journalAbbreviation":"J Chronic Dis","language":"eng","author":[{"family":"Robins","given":"J."}],"issued":{"date-parts":[["1987"]]}},"label":"page"}],"schema":""} (49,80–82).In contrast, g-methods, developed by Robins and collaborators since 1986, can appropriately adjust for measured time-varying confounders in the presence of treatment-confounder feedback ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"AYonCzbk","properties":{"formattedCitation":"(11,80)","plainCitation":"(11,80)","noteIndex":0},"citationItems":[{"id":511,"uris":[""],"uri":[""],"itemData":{"id":511,"type":"book","title":"Causal Inference: What If","publisher":"Chapman & Hill/CRC","publisher-place":"Boca Raton","event-place":"Boca Raton","title-short":"Causal Inference","author":[{"family":"Hernan","given":"M.A."},{"family":"Robins","given":"J."}],"issued":{"date-parts":[["2020"]]}}},{"id":1087,"uris":[""],"uri":[""],"itemData":{"id":1087,"type":"article-journal","title":"A new approach to causal inference in mortality studies with a sustained exposure period — Application to the healthy worker survivor effect","container-title":"Mathematical Modelling","page":"1393-1512","volume":"7","author":[{"family":"Robins","given":"JM"}],"issued":{"date-parts":[["1986"]]}}}],"schema":""} (11,80). The three classes of g-methods are inverse probability weighting (Case Study K), the plug-in g-formula (Case Study L), and g-estimation of structural nested models (Case Study M), as well as their doubly robust versions. The Table in reference (3) provides a comparison of the requirement of each of the g-methods. Of course, g-methods will result in biased estimates of the per-protocol effect if not all important pre- and post-baseline confounders are available for adjustment. Case study K: Estimating the per-protocol effect of sustained strategies using inverse probability weightingParticipants in the Prevención con Dieta Mediterránea (PREDIMED) trial were randomized to their usual diet with either supplemental olive oil or nuts, or to a low-fat diet ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"cVJ4dFgN","properties":{"formattedCitation":"(83)","plainCitation":"(83)","noteIndex":0},"citationItems":[{"id":1027,"uris":[""],"uri":[""],"itemData":{"id":1027,"type":"article-journal","title":"Primary Prevention of Cardiovascular Disease with a Mediterranean Diet Supplemented with Extra-Virgin Olive Oil or Nuts","container-title":"New England Journal of Medicine","page":"e34","volume":"378","issue":"25","source":"Crossref","abstract":"BACKGROUND Observational cohort studies and a secondary prevention trial have shown inverse associations between adherence to the Mediterranean diet and cardiovascular risk.\nMETHODS In a multicenter trial in Spain, we assigned 7447 participants (55 to 80 years of age, 57% women) who were at high cardiovascular risk, but with no cardiovascular disease at enrollment, to one of three diets: a Mediterranean diet supplemented with extra-virgin olive oil, a Mediterranean diet supplemented with mixed nuts, or a control diet (advice to reduce dietary fat). Participants received quarterly educational sessions and, depending on group assignment, free provision of extra-virgin olive oil, mixed nuts, or small nonfood gifts. The primary end point was a major cardiovascular event (myocardial infarction, stroke, or death from cardiovascular causes). After a median follow-up of 4.8 years, the trial was stopped on the basis of a prespecified interim analysis. In 2013, we reported the results for the primary end point in the Journal. We subsequently identified protocol deviations, including enrollment of household members without randomization, assignment to a study group without randomization of some participants at 1 of 11 study sites, and apparent inconsistent use of randomization tables at another site. We have withdrawn our previously published report and now report revised effect estimates based on analyses that do not rely exclusively on the assumption that all the participants were randomly assigned. The authors’ full names, academic degrees, and affiliations are listed in the Appendix. Address reprint requests to Dr. Martínez-González at the Department of Preventive Medicine and Public Health, Facultad de Medicina–Clínica Universidad de Navarra, Irunlarrea 1, 31008 Pamplona, Spain, or at m? amartinez@u? nav.?es. *The PREDIMED study investigators are listed in the Supplementary Appendix, available at . Drs. Estruch and Martínez-González contributed equally to this article. This article was published on June 13, 2018, at . N Engl J Med 2018;378:e34. DOI: 10.1056/NEJMoa1800389 Copyright ? 2018 Massachusetts Medical Society.\nRESULTS A primary end-point event occurred in 288 participants; there were 96 events in the group assigned to a Mediterranean diet with extra-virgin olive oil (3.8%), 83 in the group assigned to a Mediterranean diet with nuts (3.4%), and 109 in the control group (4.4%). In the intention-to-treat analysis including all the participants and adjusting for baseline characteristics and propensity scores, the hazard ratio was 0.69 (95% confidence interval [CI], 0.53 to 0.91) for a Mediterranean diet with extra-virgin olive oil and 0.72 (95% CI, 0.54 to 0.95) for a Mediterranean diet with nuts, as compared with the control diet. Results were similar after the omission of 1588 participants whose study-group assignments were known or suspected to have departed from the protocol.\nCONCLUSIONS In this study involving persons at high cardiovascular risk, the incidence of major cardiovascular events was lower among those assigned to a Mediterranean diet supplemented with extra-virgin olive oil or nuts than among those assigned to a reduced-fat diet. (Funded by Instituto de Salud Carlos III, Spanish Ministry of Health, and others; Current Controlled Trials number, ISRCTN35739639.)","DOI":"10.1056/NEJMoa1800389","ISSN":"0028-4793, 1533-4406","language":"en","author":[{"family":"Estruch","given":"Ramón"},{"family":"Ros","given":"Emilio"},{"family":"Salas-Salvadó","given":"Jordi"},{"family":"Covas","given":"Maria-Isabel"},{"family":"Corella","given":"Dolores"},{"family":"Arós","given":"Fernando"},{"family":"Gómez-Gracia","given":"Enrique"},{"family":"Ruiz-Gutiérrez","given":"Valentina"},{"family":"Fiol","given":"Miquel"},{"family":"Lapetra","given":"José"},{"family":"Lamuela-Raventos","given":"Rosa M."},{"family":"Serra-Majem","given":"Lluís"},{"family":"Pintó","given":"Xavier"},{"family":"Basora","given":"Josep"},{"family":"Mu?oz","given":"Miguel A."},{"family":"Sorlí","given":"José V."},{"family":"Martínez","given":"J. Alfredo"},{"family":"Fitó","given":"Montserrat"},{"family":"Gea","given":"Alfredo"},{"family":"Hernán","given":"Miguel A."},{"family":"Martínez-González","given":"Miguel A."}],"issued":{"date-parts":[["2018",6,21]]}}}],"schema":""} (83). The primary outcome was a composite of coronary heart disease events and death from cardiovascular causes over the 6-year follow-up period. The per-protocol analysis used inverse probability weighting to adjust for confounding by baseline and post-randomization variables associated with adherence and prognostic for the outcome. In addition, some participants did not attend all study visits. Therefore, the per-protocol analysis also used inverse probability weights to adjust for selection bias due to loss to follow-up. The intention-to-treat rate difference was 12.9 fewer cases (95% CI: 5.4, 21.1) of the combined endpoint per 1000 persons after 3 years for Mediterranean vs, low-fat diet. The corresponding per-protocol estimate was 21.3 fewer cases (95% CI: 3.8 to 44.8) per 1000 persons.The g-formula and inverse probability weighting can be used to obtain estimates of absolute risks and risks differences. Inverse probability weighting is a simpler method to implement but it may result in imprecise estimates when it cannot be combined with a dose-response marginal structural model, which is often the case when comparing dynamic strategies ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"gkEkx71A","properties":{"formattedCitation":"(21)","plainCitation":"(21)","noteIndex":0},"citationItems":[{"id":1155,"uris":[""],"uri":[""],"itemData":{"id":1155,"type":"article-journal","title":"Estimating absolute risks in the presence of nonadherence: an application to a follow-up study with baseline randomization","container-title":"Epidemiology","page":"528-39","volume":"21","issue":"4","source":"Nlm","archive_location":"20526200","abstract":"The intention-to-treat (ITT) analysis provides a valid test of the null hypothesis and naturally results in both absolute and relative measures of risk. However, this analytic approach may miss the occurrence of serious adverse effects that would have been detected under full adherence to the assigned treatment. Inverse probability weighting of marginal structural models has been used to adjust for nonadherence, but most studies have provided only relative measures of risk. In this study, we used inverse probability weighting to estimate both absolute and relative measures of risk of invasive breast cancer under full adherence to the assigned treatment in the Women's Health Initiative estrogen-plus-progestin trial. In contrast to an ITT hazard ratio (HR) of 1.25 (95% confidence interval [CI] = 1.01 to 1.54), the HR for 8-year continuous estrogen-plus-progestin use versus no use was 1.68 (1.24 to 2.28). The estimated risk difference (cases/100 women) at year 8 was 0.83 (-0.03 to 1.69) in the ITT analysis, compared with 1.44 (0.52 to 2.37) in the adherence-adjusted analysis. Results were robust across various dose-response models. We also compared the dynamic treatment regimen \"take hormone therapy until certain adverse events become apparent, then stop taking hormone therapy\" with no use (HR = 1.64; 95% CI = 1.24 to 2.18). The methods described here are also applicable to observational studies with time-varying treatments.","DOI":"10.1097/EDE.0b013e3181df1b69","ISSN":"1531-5487 (Electronic) 1044-3983 (Linking)","title-short":"Estimating absolute risks in the presence of nonadherence: an application to a follow-up study with baseline randomization","language":"eng","author":[{"family":"Toh","given":"S."},{"family":"Hernandez-Diaz","given":"S."},{"family":"Logan","given":"R."},{"family":"Robins","given":"J. M."},{"family":"Hernán","given":"M. A."}],"issued":{"date-parts":[["2010",7]]}}}],"schema":""} (21). The g-formula is a more flexible approach that results in more precise estimates, but its validity depends on the correct specification of models for all time-varying confounders and the outcome. Theoretically, the g-formula effect estimates will not be null even if the null hypothesis is true. However, in practice, random variability appears to overwhelm any bias due to the g-null paradox.Case Study L: Estimating the per-protocol effect of sustained strategies using the parametric g-formulaThe Strategic Timing of AntiRetroviral Treatment (START) trial was designed to assess the impact of antiretroviral treatment initiation strategies on a combined primary outcome of AIDS events or all-cause mortality, among individuals with HIV who were treatment na?ve and had good immune function at baseline (CD4 count above 500 cells/ml). Individuals were randomly assigned to either immediate treatment initiation at enrollment or delayed initiation when immune function decreased or upon diagnosis of AIDS. After specifying the protocol, including a grace period for treatment initiation, the per-protocol effect was estimated using the parametric g-formula. The estimated intention-to-treat difference in 5-year risk of AIDS or death was -3.1 percentage points (95% CI: -5.2, -0.8). The corresponding per-protocol difference was -3.8 percentage points (95% CI: -6.7, -1.5) ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"S2nefuF3","properties":{"formattedCitation":"(84)","plainCitation":"(84)","noteIndex":0},"citationItems":[{"id":949,"uris":[""],"uri":[""],"itemData":{"id":949,"type":"article-journal","title":"The per-protocol effect of immediate versus deferred antiretroviral therapy initiation","container-title":"AIDS","page":"2659-2663","volume":"30","issue":"17","archive_location":"27782964","abstract":"The Strategic Timing of AntiRetroviral Treatment (START) trial found a lower risk of a composite clinical outcome in HIV-positive individuals assigned to immediate initiation of antiretroviral therapy (ART) compared with those assigned to deferred initiation. However, 30% of those assigned to deferred initiation started ART earlier than the protocol specified. To supplement the published intention-to-treat (ITT) effect estimates, here we estimate the per-protocol effect of immediate versus deferred ART initiation in START.|The START trial randomized 4685 HIV-positive participants with CD4 cell counts more than 500 cells/μl to start ART immediately after randomization (immediate initiation group) or to wait until the CD4 cell count dropped below 350 cells/μl or an AIDS diagnosis (deferred initiation group).|We used the parametric g-formula to estimate and compare the cumulative 5-year risk of the composite clinical outcome in the immediate initiation group, and deferred initiation groups had all the trial participants adhered to the protocol.|We estimated that the 5-year risk of the composite outcome would have been 3.2% under immediate ART initiation and 7.0% under deferred initiation. The difference of 3.8% (95% confidence interval 1.5, 6.5) was larger than the ITT effect estimate of 3.1%, corresponding to a difference in effect estimates of 0.72% (-0.35, 2.35).|The ITT effect estimate may underestimate the benefit of immediate ART initiation by 23%. This estimate can be used by patients and policy-makers who need to understand the full extent of the benefit of changes in ART initiation policies.","DOI":"10.1097/QAD.0000000000001243","ISSN":"1473-5571","title-short":"The per-protocol effect of immediate versus deferred antiretroviral therapy initiation","language":"eng","author":[{"family":"Lodi","given":"S."},{"family":"Sharma","given":"S."},{"family":"Lundgren","given":"J. D."},{"family":"Phillips","given":"A. N."},{"family":"Cole","given":"S. R."},{"family":"Logan","given":"R."},{"family":"Agan","given":"B. K."},{"family":"Babiker","given":"A."},{"family":"Klinker","given":"H."},{"family":"Chu","given":"H."},{"family":"Law","given":"M."},{"family":"Neaton","given":"J. D."},{"family":"Hernán","given":"M. A."},{"family":"group","given":"INSIGHT Strategic Timing of AntiRetroviral Treatment (START)","dropping-particle":"study"}],"issued":{"literal":"11"}}}],"schema":""} (84). G-estimation of structural nested models is a more limited approach for the estimation of per-protocol effects because it does not readily yield estimates of absolute risk and it cannot easily accommodate complex dynamic strategies. It also needs to be combine with inverse probability weighting when adjustment for selection bias due to loss to follow-up is necessary ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"CRM7UM3L","properties":{"formattedCitation":"(50,85)","plainCitation":"(50,85)","noteIndex":0},"citationItems":[{"id":1162,"uris":[""],"uri":[""],"itemData":{"id":1162,"type":"article-journal","title":"Causal inference from longitudinal studies with baseline randomization","container-title":"The International Journal of Biostatistics","page":"22","volume":"4","issue":"1","archive":"Pmc","archive_location":"PMC2835458","abstract":"We describe analytic approaches for study designs that, like large simple trials, can be better characterized as longitudinal studies with baseline randomization than as either a pure randomized experiment or a purely observational study. We (i) discuss the intention-to-treat effect as an effect measure for randomized studies, (ii) provide a formal definition of causal effect for longitudinal studies, (iii) describe several methods -- based on inverse probability weighting and g-estimation -- to estimate such effect, (iv) present an application of these methods to a naturalistic trial of antipsychotics on symptom severity of schizophrenia, and (v) discuss the relative advantages and disadvantages of each method.","DOI":"10.2202/1557-4679.1117","ISSN":"1557-4679","title-short":"Causal Inference from Longitudinal Studies with Baseline Randomization","author":[{"family":"Toh","given":"Sengwee"},{"family":"Hernán","given":"Miguel A."}],"issued":{"literal":"10"}},"label":"page"},{"id":1152,"uris":[""],"uri":[""],"itemData":{"id":1152,"type":"article-journal","title":"Structural accelerated failure time models for survival analysis in studies with time-varying treatments","container-title":"Pharmacoepidemiology and Drug Safety","page":"477-91","volume":"14","issue":"7","source":"Nlm","archive_location":"15660442","abstract":"BACKGROUND: In the absence of unmeasured confounding factors and model misspecification, standard methods for estimating the causal effect of time-varying treatments on survival are biased when (i) there exists a time-dependent risk factor for survival that also predicts subsequent treatment and (ii) past treatment history predicts subsequent risk factor level. In contrast, structural models provide consistent estimates of causal effects when unmeasured confounding and model misspecification are absent. The parameters of nested structural models are estimated by g-estimation and those of marginal structural models by inverse probability weighting. METHODS: We describe a nested structural accelerated failure time model and use it to estimate the total causal effect of highly active antiretroviral therapy (HAART) on the time to AIDS or death among human immunodeficiency virus (HIV)-infected participants of the Multicenter AIDS Cohort and Women's Interagency HIV Studies. The Appendix describes g-estimation and methods to deal with censoring. RESULTS: Comparing the regime 'always treated' to 'never treated,' the AIDS-free survival time ratio was 2.5 (95% confidence interval [CI]: 1.7, 3.3). CONCLUSIONS: Our finding of a strongly beneficial effect is consistent with results from randomized trials and from a previous analysis of the same data using a marginal structural Cox model. In contrast, a previous analysis using a standard (non-structural) model did not find an effect of treatment on survival.","DOI":"10.1002/pds.1064","ISSN":"1053-8569 (Print) 1053-8569","title-short":"Structural accelerated failure time models for survival analysis in studies with time-varying treatments","journalAbbreviation":"Pharmacoepidemiol Drug Saf","language":"eng","author":[{"family":"Hernan","given":"M. A."},{"family":"Cole","given":"S. R."},{"family":"Margolick","given":"J."},{"family":"Cohen","given":"M."},{"family":"Robins","given":"J. M."}],"issued":{"date-parts":[["2005",7]]}},"label":"page"}],"schema":""} (50,85).Case Study M: Estimating the per-protocol effect of sustained strategies using g-estimation of structural nested modelsIn Case Study #6, we described a randomized trial of antipsychotic use in which individuals with schizophrenia were randomized to either atypical or conventional antipsychotics. After adjustment or loss to follow-up via inverse probability weighting, the estimated intention-to-treat difference in BPRS score was -0.86-units (95% CI: -3.88, 2.15) for atypical vs. conventional antipsychotics. The estimated per-protocol difference in BPRS score was -1.50-units (95% CI: -6.84, 3.84) after adding confounding adjustment via g-estimation of a structural nested model ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"0MOLY4B3","properties":{"formattedCitation":"(50)","plainCitation":"(50)","noteIndex":0},"citationItems":[{"id":1162,"uris":[""],"uri":[""],"itemData":{"id":1162,"type":"article-journal","title":"Causal inference from longitudinal studies with baseline randomization","container-title":"The International Journal of Biostatistics","page":"22","volume":"4","issue":"1","archive":"Pmc","archive_location":"PMC2835458","abstract":"We describe analytic approaches for study designs that, like large simple trials, can be better characterized as longitudinal studies with baseline randomization than as either a pure randomized experiment or a purely observational study. We (i) discuss the intention-to-treat effect as an effect measure for randomized studies, (ii) provide a formal definition of causal effect for longitudinal studies, (iii) describe several methods -- based on inverse probability weighting and g-estimation -- to estimate such effect, (iv) present an application of these methods to a naturalistic trial of antipsychotics on symptom severity of schizophrenia, and (v) discuss the relative advantages and disadvantages of each method.","DOI":"10.2202/1557-4679.1117","ISSN":"1557-4679","title-short":"Causal Inference from Longitudinal Studies with Baseline Randomization","author":[{"family":"Toh","given":"Sengwee"},{"family":"Hernán","given":"Miguel A."}],"issued":{"literal":"10"}}}],"schema":""} (50). Guideline: Use g-methods to adjust for time-varying confounders when there is treatment-confounder feedback. Choose inverse probability weighting or the g-formula to obtain adjusted estimates of absolute risks and risk differences. DiscussionPragmatic randomized trials are a useful tool for estimating the comparative effectiveness of treatment options. However, the features of these trials which are most useful for clinical decision-making make estimation more challenging by introducing the possibility of confounding, selection bias, and competing risks. Until now, there have not been any clear guidelines for the valid estimation of causal effects from pragmatic trials. We propose fourteen guidelines for the analysis of pragmatic randomized trials to address this gap (Table 1). These guidelines fall under four broad categories: Choice of causal effect Estimating the intention-to-treat effect (these guidelines are also relevant to the per-protocol effect)Estimating the per-protocol effect of point interventionsEstimating the per-protocol effect of sustained interventionsSince the intention-to-treat effect will often provide limited information for clinical decision-making, we strongly urge that all pragmatic randomized trials should be designed to allow for valid estimation of both the intention-to-treat and the per-protocol effect.A number of additional concerns may arise in the design and analysis of pragmatic trials for which insufficient theoretical basis is available to provide clear guidelines. In particular, there are currently no simple methods for estimating the required sample size under an expected degree of non-adherence or loss to follow-up when using g-methods. Similarly, more work is needed to compare the available methods for g-estimation when sustained treatment strategies are of interest and minimal covariate information is available. Finally, a priori specification of the statistical analysis plan is an important tool for ensuring replicability and preventing p-hacking or ‘hypothesizing after results are known’ (HARK-ing) ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"NTsg6vfy","properties":{"formattedCitation":"(86)","plainCitation":"(86)","noteIndex":0},"citationItems":[{"id":1044,"uris":[""],"uri":[""],"itemData":{"id":1044,"type":"article-journal","title":"Why Most Published Research Findings Are False","container-title":"PLOS Medicine","page":"e124","volume":"2","issue":"8","abstract":"Published research findings are sometimes refuted by subsequent evidence, says Ioannidis, with ensuing confusion and disappointment.","DOI":"10.1371/journal.pmed.0020124","journalAbbreviation":"PLOS Medicine","author":[{"family":"Ioannidis","given":"John P. A."}],"issued":{"date-parts":[["2005",8,30]]}}}],"schema":""} (86). However, for per-protocol effect estimation, full a priori specification of the statistical analysis plan is often difficult since the degree of non-adherence or loss to follow-up, and reasons for these post-randomization events, that will be observed in the trial is unknown before the trial begins. The statistical analysis plan may therefore need to include adaptive features, such as rules for modeling inverse probability weights, sensitivity analyses for the assumptions required in per-protocol effect estimation, or the use of alternative methods which make use of different assumptions ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"WipKYwpi","properties":{"formattedCitation":"(14)","plainCitation":"(14)","noteIndex":0},"citationItems":[{"id":1154,"uris":[""],"uri":[""],"itemData":{"id":1154,"type":"article-journal","title":"Per-protocol analyses of pragmatic trials","container-title":"New England Journal of Medicine","page":"1391-1398","volume":"377","issue":"14","archive_location":"28976864","DOI":"10.1056/NEJMsm1605385","title-short":"Per-Protocol Analyses of Pragmatic Trials","journalAbbreviation":"N Eng J Med","author":[{"family":"Hernán","given":"Miguel A."},{"family":"Robins","given":"James M."}],"issued":{"date-parts":[["2017"]]}}}],"schema":""} (14). Further work is needed to clarify best practices for these adaptive statistical analysis plans in pragmatic randomized trials.Table 1. Guidelines for the estimation of causal effects from pragmatic trialsCategoryGuidelineChoice of causal effect To adequately guide decision making by all stakeholders, report estimates of both the intention-to-treat effect and the per-protocol effect, as well as methods and key conditions underlying the estimation procedures. Report absolute risks and their differences, as well as their ratios, for discrete outcomes. Heterogeneity of treatment effects can be reported using subgroup analyses that use the additive scale to measure the effect of interest. Patients and advocates should be included in a priori specification of subgroups.Estimating the intention-to-treat effect*Pre-specify important prognostic factors for the outcome and the maximum acceptable difference in the distribution of these factors between groups. When one or more prognostic factor meets the threshold for imbalance, adjust via standardization, inverse probability weighting or, preferably, doubly-robust methods.In sensitivity analyses, adjust for large imbalances in any important prognostic factors, regardless of whether they have been pre-specified. In survival analyses with competing events, report both the risk of the competing event and the risk of the event of interest among those who survived the competing event by treatment group.In survival analyses with competing events, specify the intention-to-treat effect as the total effect of treatment assignment on the outcome of interest (the simplest analysis), and justify interest in any additional effects that are estimated.Ensure that the trial protocol specifies the collection of post-randomization time-varying prognostic factors that predict loss to follow-up, and appropriately adjust for these factors to reduce selection bias.Estimating the per-protocol effect of point interventionsWhen sufficient data on baseline confounders exist, estimate the per-protocol effect of point interventions via adjustment by inverse probability weighting, standardization, doubly-robust estimation, or other methods.When the three instrumental conditions are expected to hold for treatment assignment, estimate bounds for the per-protocol effect of point interventions. Provide a justification for why you believe the exclusion restriction holds, including performing appropriate falsification tests.When the three instrumental conditions and monotonicity are expected to hold, discuss whether the effect in the “compliers” is of interest. If so, estimate it and provide information on the relative size and characteristics of the “compliers” subset.Estimating the per-protocol effect of sustained interventionsTo estimate the per-protocol effect of sustained treatment strategies, specify a priori a treatment protocol that incorporates real world clinical decision-making, including discontinuation, switching, or dose-reduction rules. When there is sufficient ambiguity about the appropriate treatment strategies, more than one protocol strategy can be specified.Ensure that sufficient data are collected to determine whether participants adhered to their assigned strategies throughout the follow-up, and to adjust for time-varying prognostic factors that predict adherence to the assigned treatment strategies.Use g-methods to adjust for time-varying confounders when there is treatment-confounder feedback. Choose inverse probability weighting or the g-formula to obtain adjusted estimates of absolute risks and risk differences.* These guidelines are relevant to both the intention-to-treat and per-protocol effectsReferences ADDIN ZOTERO_BIBL {"uncited":[],"omitted":[],"custom":[]} CSL_BIBLIOGRAPHY 1. Zwarenstein M, Treweek S, Gagnier JJ, Altman DG, Tunis S, Haynes B, et al. Improving the reporting of pragmatic trials: an extension of the CONSORT statement. BMJ. 2008 Nov 11;337:a2390. 2. Consort - Pragmatic Trials [Internet]. [cited 2019 Oct 31]. Available from: . Loudon K, Treweek S, Sullivan F, Donnan P, Thorpe KE, Zwarenstein M. The PRECIS-2 tool: designing trials that are fit for purpose. BMJ [Internet]. 2015;350. Available from: . Gamerman V, Cai T, Els??er A. Pragmatic randomized clinical trials: best practices and statistical guidance. Health Serv Outcomes Res Methodol. 2019 Mar;19(1):23–35. 5. Hernán MA, Hernandez-Diaz S, Robins JM. Randomized trials analyzed as observational studies. Ann Intern Med. 2013 Oct 15;159(8):560–2. 6. Murray EJ, Caniglia EC, Swanson SA, Hernández-Díaz S, Hernán MA. Patients and investigators prefer measures of absolute risk in subgroups for pragmatic randomized trials. J Clin Epidemiol. 2018 Nov;103:10–21. 7. NIH Collaboratory Health Care Biostatistics and Study Design Core. Experimental Designs and Randomization Schemes [Internet]. 2018 [cited 2018 Aug 1]. Available from: . Initiative CTT. Clinical Trials Transformation Initiative [Internet]. 2018. Available from: . Eapen ZJ, Lauer MS, Temple RJ. The imperative of overcoming barriers to the conduct of large, simple trials. JAMA. 2014;311(14):1397–8. 10. International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use. ICH E9(R1) Addendum. Estimands and sensitivity analysis in clinical trials. [Internet]. 2017. Available from: . Hernan MA, Robins J. Causal Inference: What If. Boca Raton: Chapman & Hill/CRC; 2020. 12. Hernán MA, Hernandez-Diaz S. Beyond the intention-to-treat in comparative effectiveness research. Clin Trials. 2012 Feb;9(1):48–55. 13. Hernán MA, Scharfstein D. Cautions as regulators move to end exclusive reliance on intention to treat. Ann Intern Med [Internet]. 2018; Available from: . Hernán MA, Robins JM. Per-protocol analyses of pragmatic trials. N Engl J Med. 2017;377(14):1391–8. 15. Little RJ, D’Agostino R, Cohen ML, Dickersin K, Emerson SS, Farrar JT, et al. The prevention and treatment of missing data in clinical trials. N Engl J Med. 2012 Oct 4;367(14):1355–60. 16. Higgins JPT, Altman DG, G?tzsche PC, Jüni P, Moher D, Oxman AD, et al. The Cochrane Collaboration’s tool for assessing risk of bias in randomised trials. BMJ. 2011 Oct 18;343:d5928. 17. Mansournia MA, Higgins JP, Sterne JA, Hernán MA. Biases in Randomized Trials: A Conversation Between Trialists and Epidemiologists. Epidemiology. 1;28(1):54–9. 18. Holme ?, L?berg M, Kalager M, Bretthauer M, Hernán MA, Aas E, et al. Long-Term Effectiveness of Sigmoidoscopy Screening on Colorectal Cancer Incidence and Mortality in Women and Men: A Randomized Trial. Ann Intern Med. 2018 Jun;168(11):775–82. 19. Rossouw JE, Anderson GL, Prentice RL, LaCroix AZ, Kooperberg C, Stefanick ML, et al. Risks and benefits of estrogen plus progestin in healthy postmenopausal women: principal results From the Women’s Health Initiative randomized controlled trial. JAMA. 2002 Jul;288(3):321–33. 20. Design of the Women’s Health Initiative clinical trial and observational study. The Women’s Health Initiative Study Group. Control Clin Trials. 1998 Feb;19(1):61–109. 21. Toh S, Hernandez-Diaz S, Logan R, Robins JM, Hernán MA. Estimating absolute risks in the presence of nonadherence: an application to a follow-up study with baseline randomization. Epidemiology. 2010 Jul;21(4):528–39. 22. Pfeffer MA, Swedberg K, Granger CB, Held P, McMurray JJ, Michelson EL, et al. Effects of candesartan on mortality and morbidity in patients with chronic heart failure: the CHARM-Overall programme. Lancet. 2003 Sep 6;362(9386):759–66. 23. Murray EJ, Claggett B, Solomon SD, Pfeffer MA, Hernan MA. Adherence-adjustment in placebo-controlled randomized trials: an application to the Candesartan in Heart Failure randomized trial. 2019 submitted; 24. Stovitz SD, Shrier I. Medical decision making and the importance of baseline risk. Br J Gen Pr. 2013 Nov;63(616):e795-7. 25. Athey S, Imbens G. Recursive Partitioning for Heterogeneous Causal Effects. ArXiv150401132 Econ Stat [Internet]. 2015 Apr 5 [cited 2019 Mar 5]; Available from: . Wager S, Athey S. Estimation and Inference of Heterogeneous Treatment Effects using Random Forests. ArXiv151004342 Math Stat [Internet]. 2015 Oct 14 [cited 2019 Mar 5]; Available from: . Rothman KJ, Greenland S, Walker AM. Concepts of interaction. Am J Epidemiol. 1980 Oct;112(4):467–70. 28. Hernán MA. The hazards of hazard ratios. Epidemiology. 2010 Jan;21(1):13–5. 29. Manson JE, Hsia J, Johnson KC, Rossouw JE, Assaf AR, Lasser NL, et al. Estrogen plus progestin and the risk of coronary heart disease. N Engl J Med. 2003 Aug 7;349(6):523–34. 30. Greenland S, Mansournia MA. Limitations of individual causal models, causal graphs, and ignorability assumptions, as illustrated by random confounding and design unfaithfulness. Eur J Epidemiol. 2015 Oct;30(10):1101–10. 31. Hernan M, Robins J. Chapter 10: Random variability. In: Causal Inference. Boca Raton: Chapman & Hall/CRC, forthcoming.; 2018. 32. Altman D. Comparability of Randomised Groups. J R Stat Soc Ser D. 1985;34(1):125--36. 33. VanderWeele TJ, Ding P. Sensitivity analysis in observational research: Introducing the e-value. Ann Intern Med. 2017;167(4):268–74. 34. Robins JM. Confidence intervals for causal parameters. Stat Med. 1988 Jul 1;7(7):773–85. 35. Tsiatis AA, Davidian M, Zhang M, Lu X. Covariate adjustment for two-sample treatment comparisons in randomized clinical trials: a principled yet flexible approach. Stat Med. 2008 Oct 15;27(23):4658–77. 36. Zhang M, Tsiatis AA, Davidian M. Improving efficiency of inferences in randomized clinical trials using auxiliary covariates. Biometrics. 2008 Sep;64(3):707–15. 37. Moore KL, van der Laan MJ. Covariate adjustment in randomized trials with binary outcomes: targeted maximum likelihood estimation. Stat Med. 2009 Jan 15;28(1):39–64. 38. Hernán MA, Robins JM. Estimating causal effects from epidemiological data. J Epidemiol Community Health. 2006 accepted;60(7):578–86. 39. Robins JM, Hernán MA, Brumback B. Marginal structural models and causal inference in epidemiology. Epidemiology. 2000 Sep;11(5):550–60. 40. Senn S. Covariate imbalance and random allocation in clinical trials. Stat Med. 1989;8:467–75. 41. Senn S. Testing for baseline balance in clinical trials. Stat Med. 1994;13:1715–26. 42. Senn S. Seven myths of randomisation in clinical trials. Stat Med. 2013;32:1439–50. 43. Writing Group for the Women’s Health Initiative Investigators. Risks and Benefits of Estrogen Plus Progestin in Healthy Postmenopausal WomenPrincipal Results From the Women’s Health Initiative Randomized Controlled Trial. JAMA. 2002 Jul 17;288(3):321–33. 44. Young JG, Tchetgen EJT, Hernán MA. The choice to define competing risk events as censoring events and implications for causal inference. 2018 Jun 15; Available from: . Stensrud MJ, Young JG, Didelez V, Robins JM, Hernán MA. Separable Effects for Causal Inference in the Presence of Competing Risks. ArXiv190109472 Stat [Internet]. 2019 Jan 27 [cited 2019 Feb 25]; Available from: . Tchetgen Tchetgen EJ, Phiri K, Shapiro R. A Simple Regression-based Approach to Account for Survival Bias in Birth Outcomes Research. Epidemiol Camb Mass. 2015 Jul;26(4):473–80. 47. Diamond MP, Legro RS, Coutifaris C, Alvero R, Robinson RD, Casson P, et al. Letrozole, Gonadotropin, or Clomiphene for Unexplained Infertility. N Engl J Med. 2015;373(13):1230–40. 48. Chiu Y-H, Hsu J, Stensrud MJ, Rinaudo P, Hernandez-Diaz S, Hernan MA. Dealing with competing events in the estimation of the health effects of fertility treatment on the offspring. 49. Hernán MA, Hernández-Díaz S, Robins JM. A structural approach to selection bias. Epidemiology. 2004 Sep;15(5):615–25. 50. Toh S, Hernán MA. Causal inference from longitudinal studies with baseline randomization. Int J Biostat. 10;4(1):22. 51. National Research Council. The Prevention and Treatment of Missing Data in Clinical Trials [Internet]. Washington, DC: National Academies Press; 2010. Available from: . Tunis SL, Faries DE, Nyhuis AW, Kinon BJ, Ascher-Svanum H, Aquila R. Cost-effectiveness of olanzapine as first-line treatment for schizophrenia: results from a randomized, open-label, 1-year trial. Value Health. 2006 Apr;9(2):77–89. 53. Pearl J, Bareinboim E. External Validity: From Do-Calculus to Transportability Across Populations. Stat Sci. 2014 Nov;29(4):579–95. 54. Dahabreh IJ, Robertson SE, Stuart EA, Hernan MA. Transporting inferences from a randomized trial to a new target population. ArXiv180500550 Stat [Internet]. 2018 May 1 [cited 2019 Feb 21]; Available from: . Dahabreh IJ, Robertson SE, Tchetgen EJ, Stuart EA, Hernán MA. Generalizing causal inferences from individuals in randomized trials to all trial-eligible individuals. Biometrics. 2018 Nov 29; 56. Cole SR, Stuart EA. Generalizing Evidence From Randomized Clinical Trials to Target Populations. Am J Epidemiol. 2010 Jul 1;172(1):107–15. 57. Dahabreh IJ, Robins JM, Haneuse SJ-PA, Hernán MA. Generalizing causal inferences from randomized trials: counterfactual and graphical identification. ArXiv190610792 Stat [Internet]. 2019 Jun 25 [cited 2019 Jul 1]; Available from: . Hernan M, Robins J. Chapter 12: IP weighting and marginal structural models. In: Causal Inference. Boca Raton: Chapman & Hall/CRC, forthcoming.; 2018. 59. Hernan M, Robins J. Chapter 2: Randomized experiments. In: Causal Inference. Boca Raton: Chapman & Hall/CRC, forthcoming.; 2018. 60. Hernan M, Robins J. Chapter 13: Standardization and the parametric g-formula. In: Causal Inference. Boca Raton: Chapman & Hall/CRC, forthcoming.; 2018. 61. Hernan M, Robins J. Chapter 17: Causal survival analysis. In: Causal Inference. Boca Raton: Chapman & Hall/CRC, forthcoming.; 2018. 62. Flanders WD, Strickland MJ, Klein M. A New Method for Partial Correction of Residual Confounding in Time-Series and Other Observational Studies. Am J Epidemiol. 2017 15;185(10):941–9. 63. Lipsitch M, Tchetgen ET, Cohen T. Negative Controls: A Tool for Detecting Confounding and Bias in Observational Studies. Epidemiol Camb Mass. 2010 May;21(3):383–8. 64. Angrist J, Imbens GW, Rubin DB. Identification of causal effects using instrumental variables. J Am Stat Assoc. 1996 Jun;91(434):444–55. 65. Swanson SA, Robins JM, Miller M, Hernán MA. Selecting on treatment: a pervasive form of bias in instrumental variable analyses. Am J Epidemiol. 2015 Feb;181(3):191–7. 66. Hernan MA, Robins JM. Instruments for causal inference: an epidemiologist’s dream? Epidemiology. 2006 Jul;17(4):360–72. 67. Swanson SA, Hernán MA, Miller M, Robins JM, Richardson TS. Partial Identification of the Average Treatment Effect Using Instrumental Variables: Review of Methods for Binary Instruments, Treatments, and Outcomes. J Am Stat Assoc. 2018 Apr 3;113(522):933–47. 68. Swanson SA, Holme ?, L?berg M, Kalager M, Bretthauer M, Hoff G, et al. Bounding the per-protocol effect in randomized trials: an application to colorectal cancer screening. Trials. 2015 Nov;16:541. 69. Baker SG, Lindeman KS. The paired availability design: A proposal for evaluating epidural analgesia during labor. Stat Med. 1994;13(21):2269–78. 70. Swanson SA, Hernán MA. The challenging interpretation of instrumental variable estimates under monotonicity. Int J Epidemiol. 2017;dyx038–dyx038. 71. Scirica BM, Bhatt DL, Braunwald E, Steg PG, Davidson J, Hirshberg B, et al. Saxagliptin and cardiovascular outcomes in patients with type 2 diabetes mellitus. N Engl J Med. 2013 Oct 3;369(14):1317–26. 72. Hernan M, Robins J. Chapter 21: G-methods for time-varying treatments. In: Causal Inference. Boca Raton: Chapman & Hall/CRC, forthcoming.; 2018. 73. Robins JM. Correcting for non-compliance in randomized trials using structural nested mean models. Commun Stat. 1994;23:2379–412. 74. Hernan M, Robins J. Chapter 14: G-estimation of structural nested models. In: Causal Inference. Boca Raton: Chapman & Hall/CRC, forthcoming.; 2018. 75. Mark SD, Robins JM. A method for the analysis of randomized trials with compliance information: an application to the Multiple Risk Factor Intervention Trial. Control Clin Trials. 1993 Apr;14(2):79–97. 76. Korhonen PA, Laird NM, Palmgren J. Correcting for non-compliance in randomized trials: an application to the ATBC study. Stat Med. 1999;18(21):2879–97. 77. Greenland S, Lanes S, Jara M. Estimating effects from randomized trials with discontinuations: the need for intent-to-treat design and G-estimation. Clin Trials. 2008 Feb 1;5(1):5–13. 78. Cole SR, Chu H. Effect of acyclovir on herpetic ocular recurrence using a structural nested model. Contemp Clin Trials. 2005 Jun 1;26(3):300–10. 79. Young JG, Vatsa R, Murray EJ, Hernán MA. Interval-cohort designs and bias in the estimation of per-protocol effects: a simulation study. Trials. 2019 Sep 5;20(1):552. 80. Robins J. A new approach to causal inference in mortality studies with a sustained exposure period — Application to the healthy worker survivor effect. Math Model. 1986;7:1393–512. 81. Robins JM. Addendum to “A new approach to causal inference in mortality studies with a sustained exposure period — Application to the healthy worker survivor effect” [published errata appear in Computers Math Applic 1989:18;477]. Comput Math Appl. 1987;14:923–45. 82. Robins J. A graphical approach to the identification and estimation of causal parameters in mortality studies with sustained exposure periods. J Chronic Dis. 1987;40 Suppl 2:139S-161S. 83. Estruch R, Ros E, Salas-Salvadó J, Covas M-I, Corella D, Arós F, et al. Primary Prevention of Cardiovascular Disease with a Mediterranean Diet Supplemented with Extra-Virgin Olive Oil or Nuts. N Engl J Med. 2018 Jun 21;378(25):e34. 84. Lodi S, Sharma S, Lundgren JD, Phillips AN, Cole SR, Logan R, et al. The per-protocol effect of immediate versus deferred antiretroviral therapy initiation. AIDS. 11;30(17):2659–63. 85. Hernan MA, Cole SR, Margolick J, Cohen M, Robins JM. Structural accelerated failure time models for survival analysis in studies with time-varying treatments. Pharmacoepidemiol Drug Saf. 2005 Jul;14(7):477–91. 86. Ioannidis JPA. Why Most Published Research Findings Are False. PLOS Med. 2005 Aug 30;2(8):e124. 87. Morden JP, Lambert PC, Latimer N, Abrams KR, Wailoo AJ. Assessing methods for dealing with treatment switching in randomised controlled trials: a simulation study. BMC Med Res Methodol. 2011 Jan 11;11(1):4. 88. Huitfeldt A. Emulation of Target Trials to Study the Effectiveness and Safety of Medical Interventions [Internet] [text]. Harvard University; 2015. Available from: . Robins JM, Hernán MA. Estimation of the causal effects of time-varying exposures. In: Fitzmaurice G, Davidian M, Verbeke G, Molenberghs G, editors. Longitudinal Data Analysis. New York: Chapman and Hall/CRC Press; 2008. p. 553–99. 90. Coronary Drug Project Research Group. Influence of adherence to treatment and response of cholesterol on mortality in the coronary drug project. N Engl J Med. 1980 Oct 30;303(18):1038–41. 91. Murray EJ, Hernán MA. Adherence adjustment in the Coronary Drug Project: A call for better per-protocol effect estimates in randomized trials. Clin Trials Lond Engl. 2016;13(4):372–8. 92. Murray EJ, Hernán MA. Improved adherence adjustment in the Coronary Drug Project. Trials. 2018 Mar 5;19(1):158. 93. Swanson SA, Hernán MA. Commentary: how to report instrumental variable analyses (suggestions welcome). Epidemiology. 2013 May;24(3):370–4. 94. Glymour MM, Tchetgen Tchetgen EJ, Robins JM. Credible Mendelian randomization studies: approaches for evaluating the instrumental variable assumptions. Am J Epidemiol. 2012 Feb 15;175(4):332–9. AppendixA.1. Methods that are generally not valid for estimating the per-protocol effectA variety of other per-protocol analysis methods exist but generally do not provide valid estimates of the per-protocol effect. In addition, the terminology used to describe these methods can introduce unnecessary complexity and confusion because the target causal effect (or estimand) is generally the same regardless of analytic approach. We briefly summarize these approaches and describe why they do not provide valid estimates of the target estimand. A na?ve per-protocol analysis, also called an on-treatment analysis, is one in which non-adherent person-time is censored from the analytic dataset and no adjustment is made ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"9Wq72X1r","properties":{"formattedCitation":"(6)","plainCitation":"(6)","noteIndex":0},"citationItems":[{"id":1156,"uris":[""],"uri":[""],"itemData":{"id":1156,"type":"article-journal","title":"Patients and investigators prefer measures of absolute risk in subgroups for pragmatic randomized trials","container-title":"Journal of Clinical Epidemiology","page":"10-21","volume":"103","source":"PubMed","abstract":"OBJECTIVES: Pragmatic randomized trials are important tools for shared decision-making, but no guidance exists on patients' preferences for types of causal information. We aimed to assess preferences of patients and investigators toward causal effects in pragmatic randomized trials.\nSTUDY DESIGN AND SETTING: We (a) held three focus groups with patients (n?=?23) in Boston, MA; (b) surveyed (n?=?12) and interviewed (n?=?5) investigators with experience conducting pragmatic trials; and (c) conducted a systematic literature review of pragmatic trials (n?=?63).\nRESULTS: Patients were distrustful of new-to-market medications unless substantially more effective than existing choices, preferred stratified absolute risks, and valued adherence-adjusted analyses when they expected to adhere. Investigators wanted both intention-to-treat and per-protocol effects but felt methods for estimating per-protocol effects were lacking. When estimating per-protocol effects, many pragmatic trials used inappropriate methods to adjust for adherence and loss to follow-up.\nCONCLUSION: We made four recommendations for pragmatic trials to improve patient centeredness: (1) focus on superiority in effectiveness or safety, rather than noninferiority; (2) involve patients in specifying a priori subgroups; (3) report absolute measures of risk; and (4) complement intention-to-treat effect estimates with valid per-protocol effect estimates.","DOI":"10.1016/j.jclinepi.2018.06.009","ISSN":"1878-5921","note":"PMID: 29966732\nPMCID: PMC6175611","journalAbbreviation":"J Clin Epidemiol","language":"eng","author":[{"family":"Murray","given":"Eleanor J."},{"family":"Caniglia","given":"Ellen C."},{"family":"Swanson","given":"Sonja A."},{"family":"Hernández-Díaz","given":"Sonia"},{"family":"Hernán","given":"Miguel A."}],"issued":{"date-parts":[["2018",11]]}}}],"schema":""} (6). An as-treated analysis is but allows individuals to switch (or cross-over) between treatment arms without censoring, and individuals or person-time are analyzed based on treatment received, rather than by randomization arm. These analyses are both attempts to estimate the effect of continuous receipt of the investigational treatment versus the comparator treatment on the outcome, but failure to adjust for predictors of treatment received leads to bias whenever these variables are also prognostic for the outcome ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"oCsvP6Pr","properties":{"formattedCitation":"(12)","plainCitation":"(12)","noteIndex":0},"citationItems":[{"id":1172,"uris":[""],"uri":[""],"itemData":{"id":1172,"type":"article-journal","title":"Beyond the intention-to-treat in comparative effectiveness research","container-title":"Clinical Trials","page":"48-55","volume":"9","issue":"1","source":"Nlm","archive_location":"21948059","abstract":"BACKGROUND: The intention-to-treat comparison is the primary, if not the only, analytic approach of many randomized clinical trials. PURPOSE: To review the shortcomings of intention-to-treat analyses, and of 'as treated' and 'per protocol' analyses as commonly implemented, with an emphasis on problems that are especially relevant for comparative effectiveness research. METHODS AND RESULTS: In placebo-controlled randomized clinical trials, intention-to-treat analyses underestimate the treatment effect and are therefore nonconservative for both safety trials and noninferiority trials. In randomized clinical trials with an active comparator, intention-to-treat estimates can overestimate a treatment's effect in the presence of differential adherence. In either case, there is no guarantee that an intention-to-treat analysis estimates the clinical effectiveness of treatment. Inverse probability weighting, g-estimation, and instrumental variable estimation can reduce the bias introduced by nonadherence and loss to follow-up in 'as treated' and 'per protocol' analyses. LIMITATIONS: These analyse require untestable assumptions, a dose-response model, and time-varying data on confounders and adherence. CONCLUSIONS: We recommend that all randomized clinical trials with substantial lack of adherence or loss to follow-up are analyzed using different methods. These include an intention-to-treat analysis to estimate the effect of assigned treatment and 'as treated' and 'per protocol' analyses to estimate the effect of treatment after appropriate adjustment via inverse probability weighting or g-estimation.","DOI":"10.1177/1740774511420743","ISSN":"1740-7753 (Electronic) 1740-7745 (Linking)","title-short":"Beyond the intention-to-treat in comparative effectiveness research","language":"eng","author":[{"family":"Hernán","given":"M. A."},{"family":"Hernandez-Diaz","given":"S."}],"issued":{"date-parts":[["2012",2]]}}}],"schema":""} (12). Further, the continuous treatment effect is not necessarily the per-protocol effect of interest.Finally, a modified intention-to-treat analysis is generally a form of per-protocol effect estimation rather than intention-to-treat effect estimation. In this analysis, any individual who does not initiate at least one dose of assigned treatment is removed from the analytic dataset. The target estimand for a point intervention is the effect of adherence to the investigational treatment versus adherence to the comparator treatment on the outcome, and for a sustained intervention is the effect of adherence to initiation of the investigational treatment versus adherence to initiation of the comparator treatment. In general, no adjustment is made for confounders of treatment initiation leading to a biased estimate of this causal estimand.These conventional methods for per-protocol effect estimation generally do not provide valid estimates of their target estimands because they are commonly implemented without adjustment for predictors of adherence, switching, or initiation that are also prognostic for the outcome. Unless these confounders are appropriately measured and adjusted for, the estimates obtained via these methods will therefore be biased. Only when treatment decisions are made completely at random are these four approaches valid; these methods then become special cases of the per-protocol effect estimation methods described in the main text. A.2. Effects under other protocolsMany trials may also be interested in estimating causal effects which deviate from the randomized interventions. We briefly discuss three examples of these effects. In many oncology trials, there are ethical concerns about withholding effective treatments from patients randomized to the comparator arm. These trials may be designed with a protocol that allows participants randomized to the comparator arm to switch to the investigational treatment after some pre-specified time period, such as upon disease progression. In these trials, the per-protocol effect would consider individuals who switch as adherent. In some cases, investigators may still be interested in knowing the effect had no one switched, which we could call the switching-free effect ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"6ON8m7hN","properties":{"formattedCitation":"(87)","plainCitation":"(87)","noteIndex":0},"citationItems":[{"id":666,"uris":[""],"uri":[""],"itemData":{"id":666,"type":"article-journal","title":"Assessing methods for dealing with treatment switching in randomised controlled trials: a simulation study","container-title":"BMC Medical Research Methodology","page":"4","volume":"11","issue":"1","abstract":"We investigate methods used to analyse the results of clinical trials with survival outcomes in which some patients switch from their allocated treatment to another trial treatment. These included simple methods which are commonly used in medical literature and may be subject to selection bias if patients switching are not typical of the population as a whole. Methods which attempt to adjust the estimated treatment effect, either through adjustment to the hazard ratio or via accelerated failure time models, were also considered. A simulation study was conducted to assess the performance of each method in a number of different scenarios. 16 different scenarios were identified which differed by the proportion of patients switching, underlying prognosis of switchers and the size of true treatment effect. 1000 datasets were simulated for each of these and all methods applied. Selection bias was observed in simple methods when the difference in survival between switchers and non-switchers were large. A number of methods, particularly the AFT method of Branson and Whitehead were found to give less biased estimates of the true treatment effect in these situations. Simple methods are often not appropriate to deal with treatment switching. Alternative approaches such as the Branson & Whitehead method to adjust for switching should be considered.","DOI":"10.1186/1471-2288-11-4","ISSN":"1471-2288","title-short":"Assessing methods for dealing with treatment switching in randomised controlled trials: a simulation study","language":"En","author":[{"family":"Morden","given":"James P"},{"family":"Lambert","given":"Paul C"},{"family":"Latimer","given":"Nicholas"},{"family":"Abrams","given":"Keith R"},{"family":"Wailoo","given":"Allan J"}],"issued":{"date-parts":[["2011",1,11]]}}}],"schema":""} (87). This switching-free effect requires many of the same assumptions as the per-protocol effect, but also requires information on time of switching, and on confounders for switching and the outcome. However, it should be noted that the switching-free effect may not be clinically relevant since it may be unlikely that oncologists would ever retain patients on their primary treatment after disease progression.The (controlled) direct effect of treatment is the effect to receiving the intervention that is not mediated through a specific third variable (a mediator). For example, the initial results of Women’s Health Initiative (WHI) trial were surprising to many. The estimated intention-to-treat effect on coronary heart disease incidence if all women had been assigned to hormone therapy versus if they had been assigned to placebo was a hazard ratio of 1.24 (95% CI: 1.00, 1.54) ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"q5HTSvVa","properties":{"formattedCitation":"(29)","plainCitation":"(29)","noteIndex":0},"citationItems":[{"id":928,"uris":[""],"uri":[""],"itemData":{"id":928,"type":"article-journal","title":"Estrogen plus progestin and the risk of coronary heart disease","container-title":"N Engl J Med","page":"523-34","volume":"349","issue":"6","source":"NLM","archive_location":"12904517","abstract":"BACKGROUND: Recent randomized clinical trials have suggested that estrogen plus progestin does not confer cardiac protection and may increase the risk of coronary heart disease (CHD). In this report, we provide the final results with regard to estrogen plus progestin and CHD from the Women's Health Initiative (WHI). METHODS: The WHI included a randomized primary-prevention trial of estrogen plus progestin in 16,608 postmenopausal women who were 50 to 79 years of age at base line. Participants were randomly assigned to receive conjugated equine estrogens (0.625 mg per day) plus medroxyprogesterone acetate (2.5 mg per day) or placebo. The primary efficacy outcome of the trial was CHD (nonfatal myocardial infarction or death due to CHD). RESULTS: After a mean follow-up of 5.2 years (planned duration, 8.5 years), the data and safety monitoring board recommended terminating the estrogen-plus-progestin trial because the overall risks exceeded the benefits. Combined hormone therapy was associated with a hazard ratio for CHD of 1.24 (nominal 95 percent confidence interval, 1.00 to 1.54; 95 percent confidence interval after adjustment for sequential monitoring, 0.97 to 1.60). The elevation in risk was most apparent at one year (hazard ratio, 1.81 [95 percent confidence interval, 1.09 to 3.01]). Although higher base-line levels of low-density lipoprotein cholesterol were associated with an excess risk of CHD among women who received hormone therapy, higher base-line levels of C-reactive protein, other biomarkers, and other clinical characteristics did not significantly modify the treatment-related risk of CHD. CONCLUSIONS: Estrogen plus progestin does not confer cardiac protection and may increase the risk of CHD among generally healthy postmenopausal women, especially during the first year after the initiation of hormone use. This treatment should not be prescribed for the prevention of cardiovascular disease.","DOI":"10.1056/NEJMoa030808","ISSN":"0028-4793","title-short":"Estrogen plus progestin and the risk of coronary heart disease","journalAbbreviation":"The New England journal of medicine","language":"eng","author":[{"family":"Manson","given":"J. E."},{"family":"Hsia","given":"J."},{"family":"Johnson","given":"K. C."},{"family":"Rossouw","given":"J. E."},{"family":"Assaf","given":"A. R."},{"family":"Lasser","given":"N. L."},{"family":"Trevisan","given":"M."},{"family":"Black","given":"H. R."},{"family":"Heckbert","given":"S. R."},{"family":"Detrano","given":"R."},{"family":"Strickland","given":"O. L."},{"family":"Wong","given":"N. D."},{"family":"Crouse","given":"J. R."},{"family":"Stein","given":"E."},{"family":"Cushman","given":"M."}],"issued":{"date-parts":[["2003",8,7]]}}}],"schema":""} (29). However, women in the hormone therapy arm were less likely than women in the placebo arm to initiate statins during follow-up. Therefore, a secondary analysis of the trial data attempted to assess whether the apparent effect of hormone therapy on coronary heart disease could be explained via statin initiation, using a structural nested model approach to estimate the controlled direct effect of hormone therapy if everyone had initiated statins ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"PXjb9pyG","properties":{"formattedCitation":"(88)","plainCitation":"(88)","noteIndex":0},"citationItems":[{"id":681,"uris":[""],"uri":[""],"itemData":{"id":681,"type":"thesis","title":"Emulation of Target Trials to Study the Effectiveness and Safety of Medical Interventions","publisher":"Harvard University","genre":"text","abstract":"Ideally, clinical guidelines would be informed by well-designed randomized experiments. However, it is generally not possible to conduct a randomized trial for every clinically relevant decision. Decision makers therefore often have to rely on observational data. Guidelines that rely on observational data due to the absence of randomized trials benefit when the analysis mimics the analysis of a hypothetical target trial. This can be achieved by explicitly formulating the protocol of the target trial, and thoroughly discussing the feasibility of the conditions that must be met in order to validly emulate the target trial using observational data. \n\n\n\n\n\n\n\nIn chapter one, we discuss the emulation of trials that compare the effects of different timing strategies, that is, strategies that vary the frequency of delivery of a medical intervention or procedures, and provide an application to surveillance for colorectal cancer. In chapter two, we discuss a study design that attempts to avoid bias by comparing initiators of the treatment of interest with initiators of an “active comparator” that is believed to be inactive for the outcome, in order to emulate a randomized trial that compares the treatment of interest with an inactive comparator. In chapter three, we describe a new method that combines randomized trial data and external information to emulate a different target trial. We apply this method to a randomized trial of postmenopausal hormone therapy in order to emulate a trial of a joint intervention on hormone therapy and statin therapy.Causal Inference; Comparative Effectiveness; Trial emulation; Inverse probability weighting","URL":"","title-short":"Emulation of Target Trials to Study the Effectiveness and Safety of Medical Interventions","language":"en","author":[{"family":"Huitfeldt","given":"Anders"}],"issued":{"date-parts":[["2015"]]}}}],"schema":""} (88). Interestingly, the controlled direct effect did not differ meaningfully from the main trial results, and so it appeared that trial findings were unlikely to be explained through differences in statin use during follow-up ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"7x1oGxMJ","properties":{"formattedCitation":"(88)","plainCitation":"(88)","noteIndex":0},"citationItems":[{"id":681,"uris":[""],"uri":[""],"itemData":{"id":681,"type":"thesis","title":"Emulation of Target Trials to Study the Effectiveness and Safety of Medical Interventions","publisher":"Harvard University","genre":"text","abstract":"Ideally, clinical guidelines would be informed by well-designed randomized experiments. However, it is generally not possible to conduct a randomized trial for every clinically relevant decision. Decision makers therefore often have to rely on observational data. Guidelines that rely on observational data due to the absence of randomized trials benefit when the analysis mimics the analysis of a hypothetical target trial. This can be achieved by explicitly formulating the protocol of the target trial, and thoroughly discussing the feasibility of the conditions that must be met in order to validly emulate the target trial using observational data. \n\n\n\n\n\n\n\nIn chapter one, we discuss the emulation of trials that compare the effects of different timing strategies, that is, strategies that vary the frequency of delivery of a medical intervention or procedures, and provide an application to surveillance for colorectal cancer. In chapter two, we discuss a study design that attempts to avoid bias by comparing initiators of the treatment of interest with initiators of an “active comparator” that is believed to be inactive for the outcome, in order to emulate a randomized trial that compares the treatment of interest with an inactive comparator. In chapter three, we describe a new method that combines randomized trial data and external information to emulate a different target trial. We apply this method to a randomized trial of postmenopausal hormone therapy in order to emulate a trial of a joint intervention on hormone therapy and statin therapy.Causal Inference; Comparative Effectiveness; Trial emulation; Inverse probability weighting","URL":"","title-short":"Emulation of Target Trials to Study the Effectiveness and Safety of Medical Interventions","language":"en","author":[{"family":"Huitfeldt","given":"Anders"}],"issued":{"date-parts":[["2015"]]}}}],"schema":""} (88). Another scenario where the controlled direct effect of treatment is of interest is when changes to the standard of care occurs during the trial follow-up period ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"xAnZDIcw","properties":{"formattedCitation":"(5)","plainCitation":"(5)","noteIndex":0},"citationItems":[{"id":1153,"uris":[""],"uri":[""],"itemData":{"id":1153,"type":"article-journal","title":"Randomized trials analyzed as observational studies","container-title":"Annals of Internal Medicine","page":"560-2","volume":"159","issue":"8","source":"Nlm","archive_location":"24018844","DOI":"10.7326/0003-4819-159-8-201310150-00709","ISSN":"1539-3704 (Electronic) 0003-4819 (Linking)","title-short":"Randomized trials analyzed as observational studies","journalAbbreviation":"Ann Intern Med","language":"eng","author":[{"family":"Hernán","given":"M. A."},{"family":"Hernandez-Diaz","given":"S."},{"family":"Robins","given":"J. M."}],"issued":{"date-parts":[["2013",10,15]]}}}],"schema":""} (5). For example, in the ACTG 002 (AIDS Clinical Trials Group 002) trial, participants with HIV were randomized to the use of high- versus low-dose zidovudine, and no guidance was made about the use of prophylaxis for opportunistic infections. During the trial, the use of prophylaxis for Pneumocystis pneumonia differed between trial arms, but by the end of the trial the use of prophylaxis had become standard practice. Therefore, a relevant clinical question was whether the low-dose group (which had better survival in the intention-to-treat effect) would have still had better survival than the high-dose group had everyone in the trial received prophylaxis. Analysis of this direct effect requires the use of g-methods ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"LXysdYBS","properties":{"formattedCitation":"(5)","plainCitation":"(5)","noteIndex":0},"citationItems":[{"id":1153,"uris":[""],"uri":[""],"itemData":{"id":1153,"type":"article-journal","title":"Randomized trials analyzed as observational studies","container-title":"Annals of Internal Medicine","page":"560-2","volume":"159","issue":"8","source":"Nlm","archive_location":"24018844","DOI":"10.7326/0003-4819-159-8-201310150-00709","ISSN":"1539-3704 (Electronic) 0003-4819 (Linking)","title-short":"Randomized trials analyzed as observational studies","journalAbbreviation":"Ann Intern Med","language":"eng","author":[{"family":"Hernán","given":"M. A."},{"family":"Hernandez-Diaz","given":"S."},{"family":"Robins","given":"J. M."}],"issued":{"date-parts":[["2013",10,15]]}}}],"schema":""} (5).Finally, investigators may be interested in estimating the effects of other, non-randomized treatments within the trial population. In this case, the trial population can be considered an observational cohort and adjustment for baseline and post-baseline confounders using g-methods is required ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"8dALcICc","properties":{"formattedCitation":"(89)","plainCitation":"(89)","noteIndex":0},"citationItems":[{"id":766,"uris":[""],"uri":[""],"itemData":{"id":766,"type":"chapter","title":"Estimation of the causal effects of time-varying exposures","container-title":"Longitudinal Data Analysis","publisher":"Chapman and Hall/CRC Press","publisher-place":"New York","page":"553-599","event-place":"New York","title-short":"Estimation of the causal effects of time-varying exposures","editor":[{"family":"Fitzmaurice","given":"G."},{"family":"Davidian","given":"M."},{"family":"Verbeke","given":"G."},{"family":"Molenberghs","given":"G."}],"author":[{"family":"Robins","given":"J.M."},{"family":"Hernán","given":"M.A."}],"issued":{"date-parts":[["2008"]]}}}],"schema":""} (89). A.3. Missing adherence dataUnlike the intention-to-treat effect where treatment assignment is almost always known for all individuals enrolled in the trial, per-protocol effects rely on information about exposure status collected over time after randomization. It is not only possible but common that exposure information will be missing for at least some visits for at least some individuals. In order to estimate the per-protocol effect including these individuals some decision must be made about the missing exposure data (Case Study N). A common approach is to either censor these individuals without adjustment or to assign them an exposure value often based on single-value imputation (for example, to assume that all missing adherence implies non-adherence). Both of these approaches can lead to bias whenever adherence information is missing conditionally at random or missing not at random. However, if sufficient baseline and post-randomization data on predictors for adherence measurement exist, inverse probability of measurement weights can be used to up-weight those individuals or time points with complete adherence data. Case study N: adjusting for missing adherence data when estimating the per-protocol effectThe Coronary Drug Project (CDP) trial compared the mortality of adherers and non-adherers among those assigned to the placebo arm. In order to maximize available data, they categorized any person-time with missing adherence information as non-adherent. This made the strong assumptions that missingness was highly informative of adherence value, but not associated with any measured or unmeasured covariates. These assumptions were likely not correct and introduced potential bias into their analysis, because missingness was also highly predictive of mortality ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"Clo4dlQd","properties":{"formattedCitation":"(90)","plainCitation":"(90)","noteIndex":0},"citationItems":[{"id":1159,"uris":[""],"uri":[""],"itemData":{"id":1159,"type":"article-journal","title":"Influence of adherence to treatment and response of cholesterol on mortality in the coronary drug project","container-title":"New England Journal of Medicine","page":"1038-41","volume":"303","issue":"18","source":"Nlm","archive_location":"6999345","abstract":"The Coronary Drug Project was carried out to evaluate the efficacy and safety of several lipid-influencing drugs in the long-term treatment of coronary heart disease. The five-year mortality in 1103 men treated with clofibrate was 20.0 per cent, as compared with 20.9 per cent in 2789 men given placebo (P = 0.55). Good adherers to clofibrate, i.e., patients who took 80 per cent of more of the protocol prescription during the five-year follow-up period, had a substantially lower five-year mortality than did poor adherers to clofibrate (15.0 vs. 24.6 per cent; P = 0.00011). However, similar findings were noted in the placebo group, i.e., 15.1 per cent mortality for good adherers and 28.3 per cent for poor adherers (P = 4.7x10-16). These findings and various other analyses of mortality in the clofibrate and placebo groups of the project show the serious difficulty, if not impossibility, of evaluating treatment efficacy in subgroups determined by patient responses (e.g., adherence or cholesterol change) to the treatment protocol after randomization.","DOI":"10.1056/nejm198010303031804","ISSN":"0028-4793 (Print) 0028-4793 (Linking)","title-short":"Influence of adherence to treatment and response of cholesterol on mortality in the coronary drug project","journalAbbreviation":"N Engl J Med","language":"eng","author":[{"family":"Coronary Drug Project Research Group","given":""}],"issued":{"date-parts":[["1980",10,30]]}}}],"schema":""} (90). In a re-analysis of this trial, we instead censored individuals when their adherence status was unknown for more than a year and used inverse probability of censoring weights to adjust for loss to follow-up by individuals who stopped reporting adherence information. This removed the bias caused by inappropriate imputation of missing adherence ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"Bk71z7Zw","properties":{"formattedCitation":"(91,92)","plainCitation":"(91,92)","noteIndex":0},"citationItems":[{"id":1175,"uris":[""],"uri":[""],"itemData":{"id":1175,"type":"article-journal","title":"Adherence adjustment in the Coronary Drug Project: A call for better per-protocol effect estimates in randomized trials","container-title":"Clinical Trials (London, England)","page":"372-378","volume":"13","issue":"4","source":"PubMed","abstract":"BACKGROUND: In many randomized controlled trials, patients and doctors are more interested in the per-protocol effect than in the intention-to-treat effect. However, valid estimation of the per-protocol effect generally requires adjustment for prognostic factors associated with adherence. These adherence adjustments have been strongly questioned in the clinical trials community, especially after 1980 when the Coronary Drug Project team found that adherers to placebo had lower 5-year mortality than non-adherers to placebo.\nMETHODS: We replicated the original Coronary Drug Project findings from 1980 and re-analyzed the Coronary Drug Project data using technical and conceptual developments that have become established since 1980. Specifically, we used logistic models for binary outcomes, decoupled the definition of adherence from loss to follow-up, and adjusted for pre-randomization covariates via standardization and for post-randomization covariates via inverse probability weighting.\nRESULTS: The original Coronary Drug Project analysis reported a difference in 5-year mortality between adherers and non-adherers in the placebo arm of 9.4?percentage points. Using modern approaches, we found that this difference was reduced to 2.5 (95% confidence interval: -2.1 to 7.0).\nCONCLUSION: Valid estimation of per-protocol effects may be possible in randomized clinical trials when analysts use appropriate methods to adjust for post-randomization variables.","DOI":"10.1177/1740774516634335","ISSN":"1740-7753","note":"PMID: 26951361\nPMCID: PMC4942353","title-short":"Adherence adjustment in the Coronary Drug Project","journalAbbreviation":"Clin Trials","language":"eng","author":[{"family":"Murray","given":"Eleanor J."},{"family":"Hernán","given":"Miguel A."}],"issued":{"date-parts":[["2016"]]}},"label":"page"},{"id":1160,"uris":[""],"uri":[""],"itemData":{"id":1160,"type":"article-journal","title":"Improved adherence adjustment in the Coronary Drug Project","container-title":"Trials","page":"158","volume":"19","issue":"1","source":"PubMed","abstract":"BACKGROUND: The survival difference between adherers and non-adherers to placebo in the Coronary Drug Project has been used to support the thesis that adherence adjustment in randomized trials is not generally possible and, therefore, that only intention-to-treat analyses should be trusted. We previously demonstrated that adherence adjustment can be validly conducted in the Coronary Drug Project using a simplistic approach. Here, we re-analyze the data using an approach that takes full advantage of recent methodological developments.\nMETHODS: We used inverse-probability weighted hazards models to estimate the 5-year survival and mortality risk when individuals in the placebo arm of the Coronary Drug Project adhere to at least 80% of the drug continuously or never during the 5-year follow-up period.\nRESULTS: Adjustment for post-randomization covariates resulted in 5-year mortality risk difference estimates ranging from -?0.7 (95% confidence intervals (CI), -?12.2, 10.7) to 4.5 (95% CI, -?6.3, 15.3) percentage points.\nCONCLUSIONS: Our analysis confirms that appropriate adjustment for post-randomization predictors of adherence largely removes the association between adherence to placebo and mortality originally described in this trial.\nTRIAL REGISTRATION: , Identifier: NCT00000482 . Registered retrospectively on 27 October 1999.","DOI":"10.1186/s13063-018-2519-5","ISSN":"1745-6215","note":"PMID: 29506561\nPMCID: PMC5836455","journalAbbreviation":"Trials","language":"eng","author":[{"family":"Murray","given":"Eleanor J."},{"family":"Hernán","given":"Miguel A."}],"issued":{"date-parts":[["2018",3,5]]}},"label":"page"}],"schema":""} (91,92). The unadjusted 5-year mortality risk difference comparing placebo adherence to non-adherence when missing data was assumed to be non-adherence was 14.3 percentage points (95% CI: 10.8, 17.8), whereas the unadjusted risk difference was 11.0 percentage points (95% CI: 6.5, 15.6) when missing data was carried forward for intermittent missingness and censored after 3 consecutive missed visits. (Note that improving the assumptions about missing adherence data was not sufficient to fully address the bias in the unadjusted analysis – adjustment for baseline and post-randomization confounders was still required.)A.4. Sensitivity analyses and falsification testsThe recommended analytic methods for estimating intention-to-treat and per-protocol effects under loss to follow-up, competing events, and non-adherence discussed in these guidelines require strong assumptions more commonly applied to observational studies. Few of these studies can be empirically verified, and good statistical practice therefore requires the use and reporting of sensitivity analyses to determine the impact of potential violations of these assumptions on the estimated effects. For point interventions, sensitivity analyses for unmeasured confounding, such as the e-value, can be useful tools for assessing the potential for residual bias ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"jtCsC51G","properties":{"formattedCitation":"(33)","plainCitation":"(33)","noteIndex":0},"citationItems":[{"id":1271,"uris":[""],"uri":[""],"itemData":{"id":1271,"type":"article-journal","title":"Sensitivity analysis in observational research: Introducing the e-value","container-title":"Annals of Internal Medicine","page":"268-274","volume":"167","issue":"4","abstract":"Sensitivity analysis is useful in assessing how robust an association is to potential unmeasured or uncontrolled confounding. This article introduces a new measure called the “E-value,” which is related to the evidence for causality in observational studies that are potentially subject to confounding. The E-value is defined as the minimum strength of association, on the risk ratio scale, that an unmeasured confounder would need to have with both the treatment and the outcome to fully explain away a specific treatment–outcome association, conditional on the measured covariates. A large E-value implies that considerable unmeasured confounding would be needed to explain away an effect estimate. A small E-value implies little unmeasured confounding would be needed to explain away an effect estimate. The authors propose that in all observational studies intended to produce evidence for causality, the E-value be reported or some other sensitivity analysis be used. They suggest calculating the E-value for both the observed association estimate (after adjustments for measured confounders) and the limit of the confidence interval closest to the null. If this were to become standard practice, the ability of the scientific community to assess evidence from observational studies would improve considerably, and ultimately, science would be strengthened.","DOI":"10.7326/m16-2607","ISSN":"0003-4819","title-short":"Sensitivity analysis in observational research: Introducing the e-value","author":[{"family":"VanderWeele","given":"T. J."},{"family":"Ding","given":"P."}],"issued":{"date-parts":[["2017"]]}}}],"schema":""} (33). When using inverse probability weighting (for loss to follow-up or non-adherence) or propensity score adjustment, appropriate sensitivity analyses will include a description of the distribution of the weights or propensity scores as well as a range of models testing alternative modeling assumptions. Graphical assessment of balance in weights or propensity scores can be useful but does not guarantee unconfoundedness. When using the parametric g-formula, sensitivity analyses should include comparing the observed covariate distribution to the distribution simulated under the treatment assignment probability, as well as assessment of the sensitivity of results to model specification assumptions. For all confounding-adjustment based methods, negative control outcomes can be a useful tool for assessing the potential for residual confounding. In trials with a placebo arm, adherence to placebo versus non-adherence to placebo can be used as a negative control exposure to partially assess the assumptions required for estimating the per-protocol effect ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"N2Iu4345","properties":{"formattedCitation":"(23,91)","plainCitation":"(23,91)","noteIndex":0},"citationItems":[{"id":1072,"uris":[""],"uri":[""],"itemData":{"id":1072,"type":"article-journal","title":"Adherence-adjustment in placebo-controlled randomized trials: an application to the Candesartan in Heart Failure randomized trial","author":[{"family":"Murray","given":"E. J."},{"family":"Claggett","given":"B."},{"family":"Solomon","given":"S. D."},{"family":"Pfeffer","given":"M. A."},{"family":"Hernan","given":"M. A."}],"issued":{"date-parts":[["2019"]],"season":"submitted"}}},{"id":1175,"uris":[""],"uri":[""],"itemData":{"id":1175,"type":"article-journal","title":"Adherence adjustment in the Coronary Drug Project: A call for better per-protocol effect estimates in randomized trials","container-title":"Clinical Trials (London, England)","page":"372-378","volume":"13","issue":"4","source":"PubMed","abstract":"BACKGROUND: In many randomized controlled trials, patients and doctors are more interested in the per-protocol effect than in the intention-to-treat effect. However, valid estimation of the per-protocol effect generally requires adjustment for prognostic factors associated with adherence. These adherence adjustments have been strongly questioned in the clinical trials community, especially after 1980 when the Coronary Drug Project team found that adherers to placebo had lower 5-year mortality than non-adherers to placebo.\nMETHODS: We replicated the original Coronary Drug Project findings from 1980 and re-analyzed the Coronary Drug Project data using technical and conceptual developments that have become established since 1980. Specifically, we used logistic models for binary outcomes, decoupled the definition of adherence from loss to follow-up, and adjusted for pre-randomization covariates via standardization and for post-randomization covariates via inverse probability weighting.\nRESULTS: The original Coronary Drug Project analysis reported a difference in 5-year mortality between adherers and non-adherers in the placebo arm of 9.4?percentage points. Using modern approaches, we found that this difference was reduced to 2.5 (95% confidence interval: -2.1 to 7.0).\nCONCLUSION: Valid estimation of per-protocol effects may be possible in randomized clinical trials when analysts use appropriate methods to adjust for post-randomization variables.","DOI":"10.1177/1740774516634335","ISSN":"1740-7753","note":"PMID: 26951361\nPMCID: PMC4942353","title-short":"Adherence adjustment in the Coronary Drug Project","journalAbbreviation":"Clin Trials","language":"eng","author":[{"family":"Murray","given":"Eleanor J."},{"family":"Hernán","given":"Miguel A."}],"issued":{"date-parts":[["2016"]]}}}],"schema":""} (23,91). Finally, for instrumental variable methods, falsification tests should be performed to assess the potential for residual bias in the estimate ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"I38S5b9s","properties":{"formattedCitation":"(93,94)","plainCitation":"(93,94)","noteIndex":0},"citationItems":[{"id":68,"uris":[""],"uri":[""],"itemData":{"id":68,"type":"article-journal","title":"Commentary: how to report instrumental variable analyses (suggestions welcome)","container-title":"Epidemiology","page":"370-4","volume":"24","issue":"3","archive_location":"23549180","DOI":"10.1097/EDE.0b013e31828d0590","ISSN":"1531-5487","title-short":"Commentary: how to report instrumental variable analyses (suggestions welcome)","language":"eng","author":[{"family":"Swanson","given":"S. A."},{"family":"Hernán","given":"M. A."}],"issued":{"date-parts":[["2013",5]]}}},{"id":423,"uris":[""],"uri":[""],"itemData":{"id":423,"type":"article-journal","title":"Credible Mendelian randomization studies: approaches for evaluating the instrumental variable assumptions","container-title":"American Journal of Epidemiology","page":"332-9","volume":"175","issue":"4","source":"Nlm","archive_location":"22247045","abstract":"As with other instrumental variable (IV) analyses, Mendelian randomization (MR) studies rest on strong assumptions. These assumptions are not routinely systematically evaluated in MR applications, although such evaluation could add to the credibility of MR analyses. In this article, the authors present several methods that are useful for evaluating the validity of an MR study. They apply these methods to a recent MR study that used fat mass and obesity-associated (FTO) genotype as an IV to estimate the effect of obesity on mental disorder. These approaches to evaluating assumptions for valid IV analyses are not fail-safe, in that there are situations where the approaches might either fail to identify a biased IV or inappropriately suggest that a valid IV is biased. Therefore, the authors describe the assumptions upon which the IV assessments rely. The methods they describe are relevant to any IV analysis, regardless of whether it is based on a genetic IV or other possible sources of exogenous variation. Methods that assess the IV assumptions are generally not conclusive, but routinely applying such methods is nonetheless likely to improve the scientific contributions of MR studies.","DOI":"10.1093/aje/kwr323","ISSN":"1476-6256 (Electronic) 0002-9262 (Linking)","title-short":"Credible Mendelian randomization studies: approaches for evaluating the instrumental variable assumptions","language":"eng","author":[{"family":"Glymour","given":"M. M."},{"family":"Tchetgen Tchetgen","given":"E. J."},{"family":"Robins","given":"J. M."}],"issued":{"date-parts":[["2012",2,15]]}}}],"schema":""} (93,94). ................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download