Population Attributable Risk of Postmenopausal Breast Cancer



November 3, 2009 at Brigham and Women’s Hospital Offices, 14.15-17.00

THEMATIC WORKING GROUP SESSIONS (pre-registration required)

Evidence for an Emerging Breast Cancer Epidemic: Why the Big Numbers? Why the Younger Ages?

(previously called “Explaining Differences in the Age Distribution of Breast Cancer Across Countries ")

Co-leaders: Donna Spiegelman, Professor of Epidemiological Methods, Department of Epidemiology, Department of Biostatistics, Harvard School of Public Health

Clement Adebamowo, Associate Professor, Department of Epidemiology and Preventive Medicine, School of Medicine and Institute of Human Virology, University of Maryland, Baltimore and Director, Office of Strategic Information and Research, Institute of Human Virology in Nigeria

Rapporteur: Elysia Álvarez, Harvard School of Public Health

Reading Materials

Spiegelman D, et al., Population Attributable Risk of Postmenopausal Breast Cancer

Due to Six Breast Cancer Risk Factors. Rough draft, October 2009.

Population Attributable Risk of Postmenopausal Breast Cancer

Due to Six Breast Cancer Risk Factors

Stephanie A. Smith-Warner, Ph.D.

Donna Spiegelman, Sc.D.

Shiaw-Shyuan Yaun, M.P.H.

Hans-Olov Adami, M.D.

Lawrence Beeson, M.S.P.H.

Piet A. van den Brandt, Ph.D.

Graham A. Colditz, M.D.

Aaron R. Folsom, M.D.

Gary E. Fraser, M.D.

R. Alexandra Goldbohm, Ph.D.

Anthony B. Miller, M.B., B.Ch.

John D. Potter, M.B., B.S., Ph.D.

Thomas E. Rohan, M.B., Ph.D.

Walter C. Willett, M.D.

Alicja Wolk, Dr.Med.Sci.

David J. Hunter, M.B., B.S.

Author affiliations:

( Harvard School of Public Health, Departments of Nutrition (S.A.S.-W., W.C.W.), Epidemiology (D.S., G.A.C., W.C.W., D.J.H.), Biostatistics (D.S.), Boston, MA

( Channing Laboratory, Department of Medicine, Brigham and Women’s Hospital and Harvard Medical School, Boston, MA (W.C.W., G.A.C., S.-S.Y., D.J.H.)

( Harvard Center for Cancer Prevention, Boston, MA (W.C.W., G.A.C., D.J.H.)

( Department of Medical Epidemiology, Karolinska Institutet, Stockholm, Sweden (H.-O.A., A.W.)

( The Center for Health Research, Loma Linda University School of Medicine, Loma Linda, CA (L.B., G.E.F.)

( Department of Epidemiology, University of Maastricht, Maastricht, The Netherlands (P.A.B.)

( Division of Epidemiology, School of Public Health, University of Minnesota, Minneapolis, MN (A.R.F.)

( Department of Epidemiology, TNO Nutrition and Food Research Institute, Zeist, The Netherlands (R.A.G.)

( NCIC Epidemiology Unit, Department of Preventive Medicine and Biostatistics, Faculty of Medicine, University of Toronto, Toronto, ON, Canada (A.B.M., T.E.R.)

( Cancer Prevention Research Program, Fred Hutchinson Cancer Research Center, Seattle, WA (J.D.P.)

Correspondence to: Stephanie A. Smith-Warner, Ph.D., Harvard School of Public Health, Department of Nutrition, 665 Huntington Avenue, Boston, MA 02115 (e-mail: stephanie.smith-warner@channing.harvard.edu; phone: 617-432-4655; fax: 617-432-2435).

Supported by research grants NIH CA55075 and CA50597, by the Wallace Genetic Foundation, Inc. and by the Cancer Research Foundation of America/American Society of Preventive Oncology research fellowship to Dr. Smith-Warner, and by a Faculty Research Award (FRA-455) to Dr. Hunter from the American Cancer Society.

Abstract

Background: The population attributable risk (PAR) is the proportion of cases of a disease that would be avoided if a population experienced a shift in the distribution of causal risk factors to an overall lower risk level.

Methods: We calculated PARs for postmenopausal breast cancer for age at menarche, parity, age at first birth, body mass index and, height. Relative risks were estimated from the Pooling Project of Prospective Studies of Diet and Cancer; risk factor prevalences were estimated from the 1987 U.S. National Health Interview Survey, a 1989 survey in rural China, and from the studies comprising the Pooling Project.

Results: For these five risk factors, the PAR for a simultaneous change in the distributions observed in U.S. women to below the median values observed in women in rural China was 57% (95% CI 44-71%). Differences in the distribution of these five factors accounted for 35% (95% CI 25-45%) of the difference in breast cancer incidence rates between the U.S. and rural China. The estimated reduction in breast cancer risk for a simultaneous change in the median value of each factor to the median in rural China ranged from 37-44% across the countries represented by the Pooling Project studies.

Conclusions: We found that a moderate proportion of the difference in breast cancer incidence between China and the U.S. is accounted for by the five variables as we measured them. However, because these variables may be measured with considerable error, or are only indirect surrogates of the biologically relevant variables, our findings probably represent a serious underestimate of the contribution of these factors to international differences in breast cancer rates.

INTRODUCTION

Breast cancer is a leading cause of cancer in women worldwide 1. However, breast cancer incidence and mortality rates vary considerably across countries. Rates tend to be higher in North America and Europe and lower in Asia and Africa 1. Breast cancer incidence rates also vary within countries. For example, the 1988-1992 breast cancer incidence rates were 90.7/100,000/year in the United States (U.S.), 26.5/100,000/year in Shanghai, one of the largest cities in China, and 11.2/100,000/year in Qidong, a more rural city near Shanghai 2. Migrant studies demonstrate that differences in lifestyle and reproductive factors contribute to the differences in incidence rates 3-5. However, the proportion of the international variation in breast cancer incidence that is explained by differences in breast cancer risk factors is uncertain.

The potential impact of risk factors on breast cancer occurrence can be evaluated by estimating the population attributable risk (PAR) due to individual risk factors and combinations of risk factors. The PAR is the proportion of cases that would be avoided if the risk factor distribution of a high risk population switched to that of a low risk population 6, 7. PARs may be used to compare disease incidence rates among countries with different risk factor distributions, to evaluate the impact of established risk factors on a disease within a population, and to examine the potential impact of interventions designed to decrease the risk of a disease 6. Calculation of PARs assumes that the variables examined are causally associated with the disease of interest. The most common risk factors that have been used to estimate PARs for breast cancer have been age at first birth, number of breast biopsies/history of benign breast disease, family history, and alcohol consumption 6, 8-15.

We evaluated the proportion of postmenopausal breast cancer in the United States that theoretically could be avoided if the distributions for age at menarche parity, age at first birth, body mass index, and height for women living in the U.S. changed to the distributions seen among women living in rural China. In addition, we estimated the change in breast cancer risk associated with a shift in the median values of these five risk factors for each of the four high-risk countries represented in the Pooling Project of Prospective Studies of Diet and Cancer (the U.S., Canada, the Netherlands, and Sweden) to the median values of a lower-risk country (China). We focused on postmenopausal breast cancer only because most breast cancer occurs among postmenopausal women 16, some risk factors differ for premenopausal and postmenopausal breast cancer 17-20 and because only four cohorts in the Pooling Project included premenopausal participants, resulting in limited power for analyses of premenopausal breast cancer. We chose China as an example of a lower-risk country because information on these risk factors at an individual level was available from a national survey (Dr. Banoo Parpia, personal communication). The five risk factors were selected because they have been consistently associated with breast cancer risk in epidemiologic studies 17, 21-23 and there is biological evidence that these risk factors are causal 24, 25. Although alcohol consumption has been shown to increase the risk of breast cancer 12, 22, we did not include alcohol consumption in these analyses because the median alcohol consumption was similar between women living in rural China and the U.S. However, including alcohol consumption in our analyses probably would not have materially changed the results because the PAR for switching from any alcohol consumption to none has been estimated as 2% in the United States 26.

METHODS

In these analyses, we used the PAR as an estimate of the excess fraction, the proportion of cases that would not have occurred if the exposure of interest had not been present 27. We estimated the PAR for postmenopausal breast cancer associated with five breast cancer risk factors: age at menarche, parity, age at first birth, body mass index, and height. We used the term “summary PAR” to refer to a PAR calculated for the combination of one or more risk factors.

The full PAR (PARF), estimated the proportion of breast cancer cases that would be avoided if the distribution of each of the risk factors in the model was switched to its corresponding low-risk category. The PARF was computed as

[pic]

where RRs is the multivariate-adjusted relative risk determined from the Pooling Project 28 and ps is the U.S. population prevalence obtained from the 1987 U.S. National Health Interview Survey (NHIS) 29 for the sth combination of levels of the six risk factors, s=1,...,S 30. The variance for the estimate of PARF using external population prevalences is derived in Appendix 1.

The partial PAR (PARP) evaluated the percent reduction expected in the crude breast cancer incidence rate of the target population if some, but not all, of the risk factors in the model were eliminated from the target population 6. Appendix 2 shows the derivation of the formula for estimating the PARP when the relative risks and prevalences are estimated in different populations. The corresponding variance for the PARp is derived in Appendix 3.

Because PAR estimates obtained using multiple exposure categories are roughly equivalent to estimates obtained using a single dichotomous exposure category grouped over these multiple categories 31, age at menarche, body mass index, and height, were each defined as categorical variables with two levels. Parity and age at first birth were modelled jointly as an interaction term with five levels. The reference group for each variable was defined as the group with the lowest risk. The cutpoint for each risk factor was chosen using the median level observed in a 1989 communication, 1989 Chinese ecologic survey of 69 rural countries in China 32.

To estimate the proportion of breast cancer cases in the U.S. that would be avoided if the distributions of age at menarche, parity, age at first birth, body mass index, and height in women living in the U.S. were changed to the distributions of women living in rural China (rather than to the low-risk category which results in a larger change in the risk factor distribution), we calculated the 2-country PAR (2C-PARF). The derivation of the 2C-PARF and its variance are included in Appendix 4. For these analyses, the joint distribution of the risk factors was estimated using the 1987 U.S. NHIS (described below) and a 1989 Chinese ecologic survey of 69 rural counties in China 32 (Dr. Banoo Parpia, personal communication).

The relative attributable risk (RAR) 33 was used to estimate the proportion of the difference between the Chinese and U.S. breast cancer incidence rates that was due to differences in the distributions of these five risk factors between the two countries. The RAR was estimated as

[pic]

where IL is the incidence rate in China (the low-risk population) and IH is the incidence rate in the U.S. (the high-risk population). The 95% confidence interval was estimated using the multivariate delta method 34.

We also estimated the percent reduction in breast cancer risk associated with a shift in the median value in each risk factor in four high-risk countries (the U.S., Canada, Netherlands, and Sweden) compared to the median value in a low-risk country (China) 35. The median value for each risk factor was estimated using the 1987 U.S. NHIS 29, the sub-cohort of the Canadian National Breast Screening Study 36, the sub-cohort in the Netherlands Cohort Study 37, and the baseline population of the Sweden Mammography Cohort 38. Participants in the three cohort studies were recruited from population registries 37 or mammography screening clinics 36, 38. The risk factor distribution in rural China was estimated from a 1989 Chinese ecologic survey of 69 rural counties in China 32 (Banoo Parpia, personal communication). The relative risks of developing breast cancer were estimated from the Pooling Project. The risk factors were modelled as continuous variables in the regression analyses with linearity assumed; nulliparous women were assigned a value of zero for age at first birth. Sensitivity analyses were conducted using the lower and upper 95% CI bounds for each risk factor 39.

Pooling Project

For calculating the PAR, we estimated the relative risks for postmenopausal, invasive breast cancer from the Pooling Project 23, 28, 40. Briefly, seven prospective studies 37, 38, 41-45 (Table 1) were identified that met the following pre-defined criteria: 1) identified at least 200 incident cases of breast cancer; 2) assessed usual intake of foods and nutrients, and 3) completed a validation study of the dietary assessment instrument or closely related instrument. The New York State Cohort was excluded from these analyses because age at menarche was not ascertained in that study. The Nurses' Health Study was divided into two studies (1980-1986 and 1986-1995 follow-up periods) because it had repeated assessments of breast cancer risk factors and a longer follow-up period than the other studies. Self-administered questionnaires were used to assess reproductive factors, anthropometric factors, diet, medical history, and family history in each study. Incident breast cancers were ascertained using follow-up questionnaires, inspection of medical records and/or tumor-registry linkage. In all cohorts, follow-up was estimated to be more than 90 percent complete.

For each study, after applying the exclusion criteria used by that study, we excluded participants who reported energy intakes greater or less than three standard deviations from the study-specific loge-transformed mean energy intake of the baseline population, reported a history of cancer except non-melanoma skin cancer at baseline, or were premenopausal at baseline. Participants also were excluded from these analyses if they had missing data on age at menarche, parity, age at first birth, body weight, or height.

For the Adventist Health Study, Iowa Women’s Health Study, Nurses’ Health Study (a), Nurses’ Health Study (b) and Sweden Mammography Cohort, nested case-control datasets were formed for each study with a 1:10 ratio of cases diagnosed with invasive breast cancer to controls free of diagnosed cancer. A nested case-control design also was used for the Canadian National Breast Screening Study; the investigators of that study selected two controls for each case 41. Analyzing prospective studies as nested case-control studies has been shown to be an efficient and unbiased alternative to full cohort analysis 46. The Netherlands Cohort Study used a case-cohort design 47; cases were identified from the total cohort and compared with a subcohort of 1812 women randomly sampled at baseline.

For these analyses, the data for all the studies were combined into a single dataset. Using the combined dataset allowed for more stable relative risk estimates due to the enhanced sample size; in previous analyses we observed no between-study heterogeneity in any of the risk factors considered here 23, 40. Relative risks for each variable were estimated by a conditional logistic regression model stratified by study using Epicure software 48 for the multivariate model including only the variables of interest.

1987 U.S. National Health Interview Survey (NHIS)

Estimates of the U.S. prevalences of the risk factors were obtained from the 1987 NHIS, a national household survey of noninstitutionalized U.S. civilians 29. A total of 4,800 women, representing 35,924,766 individuals, were identified that were 35-64 years (the same age range covered by the 1989 ecologic survey in China 32), did not have a personal history of cancer, and had no missing data for age at menarche, parity, age at first birth, body mass index, and height. To remove outlying values, NHIS participants were excluded from the analysis if any of the six variables of interest exceeded the range among the control or sub-cohort participants in the Pooling Project dataset. Prevalences and their variances and covariances were estimated using SUDAAN, version 7.0 49, to account for the stratified sampling design of this study 29.

RESULTS

As reported previously 23, 40, age at menarche was inversely associated with postmenopausal breast cancer risk; body mass index, and height, and were each positively associated with postmenopausal breast cancer risk (Table 2). A 47% elevation in breast cancer risk was observed for nulliparous women compared to women who had parity of at least 4 and an age at first birth of 22 years or younger.

We estimated the proportion of breast cancer cases in the U.S. that could be eliminated if all U.S. women had values for age at menarche, parity, age at first birth, body mass index, and height found in the low-risk half of women living in rural China. The cutpoint for the reference category of each risk factor was determined using the median value of that risk factor from the 1989 ecologic survey of 69 rural Chinese counties 32 (Banoo Parpia, personal communication). The partial PARs for the individual risk factors ranged from 10% for body mass index to 32% for age at menarche. Based on this model, 43% (95% CI 27-59%) of the breast cancer cases in the U.S. would be avoided if women had an age at menarche of at least 17 years, parity of at least 4, and an age at first birth of 22 years or younger. Twenty-five percent (95% CI 12-38%) of breast cancer cases would be avoided if women had a body mass index < 21 kg/m2 and height < 1.54 m. The PAR for the low-risk combination of the five reproductive and anthropometric factors was 57% (95% CI 44-71%). No women in the NHIS were in the low-risk group for the combination of all five risk factors.

To assess the impact of a switch in the distribution of these five risk factors in U.S. women to the distribution seen in women living in rural China, we calculated the summary estimates for the 2C-PARF and RAR. The 2C-PARF was 31% (95% CI 23-41%). This means that 31% of the breast cancer cases in the U.S. could be avoided if the risk factor distributions for age at menarche, parity, age at first birth, body mass index, and height in the U.S. were changed to the distributions in rural China (rather than to the low-risk half). Using the breast cancer incidence rates for the U.S. (90.7/100,000/year) and Qidong, China (11.2/100,000/year) 2, differences in the distribution of these five risk factors accounted for 35% of the difference in breast cancer incidence rates between the U.S. and rural China (RAR=35%, 95% CI 25-45%). Using the breast incidence rate for Shanghai, one of the largest cities in China (26.5/100,000/year) 2, the RAR was 41% (95% CI 29-53%).

We also estimated the proportional reduction in breast cancer risk if the median values of each of the reproductive and anthropometric risk factors in women living in four high-risk countries (the U.S., Canada, the Netherlands, and Sweden) were changed to the median levels of women living in rural China. Women in the United States, Canada, the Netherlands, and Sweden, on average, had an earlier age at menarche, had fewer children, were heavier, and were taller compared to women in rural China (Table 3). Mean age at first birth was older in Canada, the Netherlands and Sweden compared to the U.S. and rural China. Based on the multivariate model with each risk factor as a continuous variable, the estimated reduction in breast cancer risk for the individual risk factors comparing the medians in the U.S. and rural China ranged from -0.4% for age at first birth to 24% for age at menarche (Table 4). The risk reduction for age at first birth was negative because the median age at first birth in the U.S. was slightly younger than that observed in rural China. The summary risk reductions were 34% (95% CI 24-43%) when age at menarche, parity, and age at first birth were considered together and 15% (95% CI 10-21%) for body mass index and height. When all five factors were considered simultaneously, breast cancer risk would be decreased by 44% (95% CI 31-55%) by a change from the median of each risk factor in the U.S. population to the median among women living in rural China. Although the risk reductions for individual risk factors in Canada, the Netherlands, and Sweden were slightly different from those observed in the U.S., results were similar when all five risk factors were considered (range: 37-44%).

DISCUSSION

Breast cancer continues to be a leading cause of cancer death among women worldwide 1. We estimated that 57% (95% CI 44-71%) of postmenopausal breast cancer cases in the U.S. would be eliminated by a simultaneous shift in age at menarche, parity, age at first birth, body mass index, and height from the current U.S. risk factor distribution to the low-risk half of women living in rural China. This PAR would correspond to a reduction in the U.S. breast cancer age-adjusted incidence rate from 90/100,000/yr 50 to 38/100,000/yr. If the U.S. distribution for these same five reproductive and anthropometric risk factors shifted to the distribution of women living in rural China (rather than to the low-risk half of the rural female Chinese population), the estimated reduction in breast cancer incidence in the U.S. was 31% (95% CI 23-41%) or a new rate of 62/100,000/yr. The estimated reduction in risk comparing women with the median values for the 5 risk factors was 44% (95% CI 31-55%).

Comparison of summary PARs across studies is difficult because the risk factors and cutpoints used to define the reference groups vary, the strengths of the associations for each risk factor vary across studies, and the distribution of the prevalences of the risk factors vary across populations. Nevertheless, the summary PAR of 57% that we observed for the simultaneous shift of age at menarche, parity, age at first birth, body mass index, and height in the U.S. distribution to the lower risk half of the rural Chinese population was greater than previously reported summary PARs using different combinations of risk factors and different cutpoint for defining the high risk group 6, 8-11, 51-54. In the Carolina Breast Cancer Study 8 and the Breast Cancer Detection Demonstration Project 6, 10, the PAR for the low-risk combination of an age at menarche > 14 years, age at first birth < 20 years, no history of benign breast disease or < 2 breast biopsies, and no family history of breast cancer was determined. The PAR due to all four risk factors was 25% (95% CI 6-48%) in the Carolina Breast Cancer Study 8 and 52-55% in the Breast Cancer Detection Demonstration Project 6, 10. Because the two studies used virtually identical classifications for the risk factors, the discrepancy in the PARs between the studies is due to differences in the risk factor distributions between the populations and in the strengths of the associations observed. In a subsequent analysis of the Breast Cancer Detection Demonstration Project, the PAR for the low-risk combination of having an age at first birth of less than 20 years, no family history of breast cancer, and an income in the lower third of the population was 40.8% (95% CI 1.6-80.0%) 53. In an Italian case-control study, a PAR of 50% was estimated for the low-risk combination of education < 7 years, age at first birth < 20 years, and no family history 11. In a 1959 American Cancer Society cohort study which followed women for six years, the PAR for women aged 55-84 years for the low-risk group having none of the ten risk factors studied (family history of breast cancer, history of breast surgery, being Jewish, menopause after age 50, age at menarche under 12 years, never married, age at first birth over age 30 or nulliparity, college graduate, daily alcohol consumption, or 110% of ideal body weight) was 29% 54. This study frequently has been misinterpreted as meaning that most breast cancer cases occur in women with no known risk factors (for a review of examples, see Rockhill et al. 7).

Another method that has been used to estimate the potential impact of risk factors on disease is to correlate disease rates with population-based risk factor data. In a study of sixteen European countries, 80% of the variation in breast cancer incidence rates across these countries was suggested to be due to differences in average age at first birth and height 55. Our study supports the findings that differences in age at menarche and height are main contributors to differences in breast cancer incidence rates between countries with different breast cancer incidence rates. Changes in breast cancer risk factors over time also have been assessed within countries and compared to changes in breast cancer incidence rates. In Japan, for example breast cancer incidence rates have been increasing which is probably due in part to changes in the distribution of lifestyle factors such as altered childhood nutrition leading to increases in height and decreases in age at menarche 56-58.

Ecologic studies have suggested that differences in fat consumption might account for much of the international variation in breast cancer rates 59, 60. However, prospective studies in North America and Europe have not established associations between dietary fat in adult life and breast cancer risk 28, 61-63. Diet during childhood, on the other hand, may have an important influence on future breast cancer risk, due, in part, to the influence of energy intake on both age at menarche and height 17.

A limitation of our analyses is that the effect of factors other than age at menarche, parity, age at first birth, body mass index, and height on breast cancer risk was not determined. Potential examples include lactation history 64, timing between pregnancies 65, physical inactivity 66, postmenopausal hormone use 67 and age at menopause []. These variables were not included in our analyses because associations with these risk factors have been inconsistent or because we did not have uniform prevalence and relative risk data. Further, we may have underestimated the effect of some of the variables included in our analysis. For example, because several of the studies in the Pooling Project had a measure of body mass index only at baseline, we were unable to assess the effect of adult weight gain, which has been associated with breast cancer risk in some 51, 68, but not all 69, studies. In addition, body mass index is only an indirect measure of fat mass 70, 71, and for the same body mass index, Chinese women are probably leaner due to a less sedentary lifestyle [Walt-reference?]. For these reasons, the contribution of adiposity to breast cancer risk is probably substantially underestimated. In addition, differences in height between populations may be primarily due to nutritional factors whereas height within western populations is largely a function of genetic factors. Thus height may only be a poorly measured surrogate of the true nutritionally related variables. Likewise, we did not capture the effects of menstrual cycle length and age at onset of regular menstrual cycles (as opposed to remote recall of the first cycle) which also have been associated with breast cancer risk 72, 73.

There are also limitations to using PARs. The PAR estimate is sensitive to the cutpoints used to define the reference category for each risk factor because changes in the reference category affect both the relative risk and prevalence estimates 7, 8.

The PAR is also not equivalent to the proportion of exposed cases 7. In the analyses which defined the reference group for each risk factor using the median value of the rural Chinese population, the PAR was 57%. This PAR should not be interpreted to mean that a large population of breast cancer in the U.S. occurs in women with no known risk factors; in this analysis, all of the postmenopausal women who participated in the NHIS had at least one of the five risk factors considered.

In our continuous analyses, we used an approach that has been used previously to estimate the contributions of disease prevention and improvements in medical treatment to the decrease in coronary heart disease mortality rates over time 35, 39. Using a similar method, the breast cancer incidence rate for a 65 year old woman in the U.S. has been estimated to be approximately three times higher than the rate in China based on differences between the two populations in age at menarche, parity, age at first birth, and the spacing between births 74. Our results which evaluated a change from the median values of each of five reproductive and anthropometric risk factors in the U.S., Canada, the Netherlands, or Sweden to the medians observed in rural China yielded an individual risk reduction of 37-44%. A limitation of these types of analyses is that they can only be used to describe the expected reduction in individual risk for a switch from one risk factor profile to another specific risk factor profile. Correlations in the distribution of these risk factors are not considered in these analyses; thus, these analyses cannot be used to infer changes in rates between populations. Finally, these analyses assume that the association between each risk factor and the risk of breast cancer is linear. However, in the Pooling Project, we found that the associations between body mass index 40 and parity (unpublished data) and breast cancer risk were nonlinear.

These results demonstrate that a moderate proportion of the differences in breast cancer incidence rates between the U.S. and China is due to differences in the distributions of age at menarche, parity, age at first birth, body mass index, and height. However, because these variables may be measured with considerable error, or are only indirect surrogates of the biologically relevant variables, our findings probably represent a serious underestimate of the contribution of these factors to international differences in breast cancer rates.

Appendix 1. Derivation of [pic]

The formula for the PAR applicable to cohort studies 30 can be used only when the all risk factors for the disease of interest are considered in the PAR calculation. That is, in the formula

[pic]

the S levels correspond to all possible levels of risk known for the disease of interest. This would be the case if there was only one risk factor and S levels of this risk factor, or if the disease were multifactorial but there were S unique combinations of levels of the several risk factors. No variance formula was previously available for the PAR estimated using prevalences (ps, s=1,...,S) obtained from a population external to the population used to estimate the relative risks (RRs, s=1,...,S); hence, we derive it here.

Let s=1,...,S unique combinations of levels of all risk factors of the outcome of interest, let ps, s=1,...,S, denote the prevalences of risk factor combination s, where s=1 denotes the reference level, let RRs=exp{β(xx}, the relative risk for combination s under the proportional hazards model where βp, p=1,...,P, is the log relative risk for corresponding to the pth risk factor of P risk factors in all and xs is a p-dimensional vector in which each element contains the appropriate value for each risk factor in the sth combination. That is, xps is the value of the pth covariate at the sth combination.

By (A1),

[pic]

and by the multivariate delta method 34,

[pic]

where G1 is an S-dimensional vector in which the sth element is [pic]and G2 is an S-dimensional vector in which the sth element is [pic].

By the multivariate delta method, under the proportional hazards model, [pic], where V is the p(p variance-covariance matrix of [pic]obtained from the Cox, Poisson or conditional logistic regression model run in the data set used to obtain the relative risks and B is a S(P-dimentional matrix in which [pic].To estimate the variance, estimates of the parameters upon which the formula depends are substituted for the true values. In this study, a conditional logistic regression model was run in the Pooling Project of Prospective Studies of Diet and Cancer to obtain [pic]and [pic]are national prevalences obtained from NHIS and Var([pic]) is obtained from the 1987 National Health Interview Survey data using SUDAAN software 49 to account for the complex sample survey design of that study29. To improve the asymptotic behavior of the 95% confidence intervals of [pic], it is useful to transform to the logit scale, i.e.

[pic]

Then, the 95% confidence intervals for the PAR would be obtained as [pic] .

Appendix 2. Derivation of the PARP, the partial PAR

[pic]

Define the partial PAR, PARP, as the proportion of the total cases which would be avoided if members of the target population in the higher risk categories of one or more risk factors of interest switched to the lowest, reference level, while one or more other risk factors are unchanged 6. Let t indicate a strata of unique combinations of levels of all risk factors other than the risk factors of interest, t=1,...,T. Let RR2t indicate the relative risk in combination t, relative to the lowest risk level, where RR21=1. As previously, s indicates an exposure group defined by each of the unique combinations of the levels of the index risk factors, i.e. those risk factors to which the PARP applies, s=1,...,S, and RR1s is the relative risk corresponding to combination s, relative to the lowest risk combination, RR11. Let pst be the joint prevalence of exposure group s and strata t in the target population to whom the PARP is to be applied. Then, the number of cases in exposure group s in strata t is NpstI0RR21sRR2t , where N is the total amount of person-time in the target population and I0 is the baseline incidence rate in the target population. The number of cases which would occur in this same strata if all of the members of this strata switched to the lowest risk exposure level is equal to NpstI0RR2t. Hence, the proportion of cases which would be avoided in the target population, the PARP, is thus

where p.t is the marginal prevalence of strata t. PARP is estimated by substituting consistent estimates for all parameters in the formula.

Appendix 3. Derivation of [pic]

[pic]

By the multivariate delta method 34,

[pic]

where [pic][pic]

[pic][pic][pic]

RR1=(RR11,RR12,...,RR1S)(, and RR2=(RR21,RR22,...,RR2T)(.

Under the proportional hazards model,[pic] , where xs is the vector of values of the exposure variables at the sth level, and dim(x)=P, and [pic], where yt is the vector of exposure variables at the tth level. Then, [pic] where [pic], and is obtained from the conditional logistic regression model run in the Pooling Project, and D=[(Dup),u=1,...,S+T, p=1,...,P], where

[pic]

References

1. World Cancer Research Fund, American Institute for Cancer Research Expert Panel (J.D. Potter, Chair). Food, Nutrition and the Prevention of Cancer: a Global Perspective. Washington DC: American Institute for Cancer Research, 1997.

2. Cancer Incidence in Five Continents, vol. VII. Lyon: International Agency for Research on Cancer, 1997.

3. Ziegler RG, Hoover RN, Pike MC, Hildesheim A, Nomura AMY, West DW, et al. Migration patterns and breast cancer risk in Asian-American women. Journal of the National Cancer Institute 1993;85:1819-27.

4. McMichael AJ, Giles GG. Cancer in migrants to Australia: extending the descriptive epidemiological data. Cancer Research 1988;48:751-56.

5. Kelsey JL, Horn-Ross PL. Breast cancer: magnitude of the problem and descriptive epidemiology. Epidemiologic Reviews 1993;15:7-16.

6. Bruzzi P, Green SB, Byar DP, Brinton LA, Schairer C. Estimating the population attributable risk for multiple risk factors using case-control data. American Journal of Epidemiology 1985;122:904-14.

7. Rockhill B. Use and misuse of population attributable fractions. American Journal of Public Health 1998;88:15-19.

8. Rockhill B, Weinberg CR, Newman B. Population attributable fraction estimation for established breast cancer risk factors: considering the issues of high prevalence and unmodifiability. American Journal of Epidemiology 1998;147:826-33.

9. Spiegelman D, Hunter D, Hertzmark E, Colditz G. Re: Validation of the Gail et al. model for predicting individual breast cancer risk. Journal of the National Cancer Institute 1994;86:1350-51.

10. Gail MH, Brinton LA, Byar DP, Corle DK, Green SB, Schairer C, et al. Projecting individualized probabilities of developing breast cancer for white females who are being examined annually. Journal of the National Cancer Institute 1989;81:1879-86.

11. Tavani A, Braga C, La Vecchia C, Negri E, Russo A, Franceschi S. Attributable risks for breast cancer in Italy: education, family history and reproductive and hormonal factors. International Journal of Cancer 1997;70:159-63.

12. Longnecker MP. Alcoholic beverage consumption in relation to risk of breast cancer: meta-analysis and review. Cancer Causes and Control 1994;5:73-82.

13. Levi F, Pasche C, Lucchini F, La Vecchia C. Alcohol and breast cancer in the Swiss Canton of Vaud. European Journal of Cancer 1996;32A:2108-13.

14. Viel J-F, Perarnau J-M, Challier B, Faivre-Nappez I. Alcoholic calories, red wine consumption and breast cancer among premenopausal women. European Journal of Epidemiology 1997;13:639-43.

15. Mezzetti M, La Vecchia C, Decarli A, Boyle P, Talamini R, Franceschi S. Population attributable risk for breast cancer: diet, nutrition, and physical exercise. Journal of the National Cancer Institute 1998;90:389-94.

16. Hankey BF, Miller B, Curtis R, Kosary C. Trends in breast cancer in younger women in contrast to older women. J Natl Cancer Inst Monogr 1994(16):7-14. Available from .

17. Hunter DJ, Willett WC. Diet, body size, and breast cancer. Epidemiologic Reviews 1993;15:110-32.

18. Holmberg L, Baron JA, Byers T, Wolk A, Ohlander E-M, Zack M, et al. Alcohol intake and breast cancer risk: effect of exposure from 15 years of age. Cancer Epidemiology, Biomarkers and Prevention 1995;4:843-47.

19. Schatzkin A, Carter CL, Green SB, Kreger BE, Splansky GL, Anderson KM, et al. Is alcohol consumption related to breast cancer? Results from the Framingham Heart Study. Journal of the National Cancer Institute 1989;81:31-35.

20. Byers T, Graham S, Rzepka T, Marshall J. Lactation and breast cancer. Evidence for a negative association in premenopausal women. American Journal of Epidemiology 1985;121:664-74.

21. Kelsey JL, Gammon MD, John EM. Reproductive factors and breast cancer. Epidemiologic Reviews 1993;15:36-47.

22. Smith-Warner SA, Spiegelman D, Yaun S-S, van den Brandt PA, Folsom AR, Goldbohm RA, et al. Alcohol and breast cancer in women: a pooled analysis of cohort studies. Journal of the American Medical Association 1998;279:535-40.

23. Hunter DJ, Spiegelman D, Adami H-O, van den Brandt PA, Folsom AR, Goldbohm RA, et al. Non-dietary factors as risk factors for breast cancer, and as effect modifiers of the association of fat intake and risk of breast cancer. Cancer Causes and Control 1997;8:49-56.

24. Bernstein L, Ross RK. Endogenous hormones and breast cancer risk. Epidemiologic Reviews 1993;15:48-65.

25. Key TJA, Pike MC. The role of oestrogens and progestagens in the epidemiology and prevention of breast cancer. European Journal of Cancer and Clinical Oncology 1988;24:29-43.

26. Tseng M, Weinberg CR, Umbach DM, Longnecker MP. Calculation of population attributable risk for alcohol and breast cancer (United States). Cancer Causes Control 1999;10(2):119-23. Available from .

27. Greenland S, Robins JM. Conceptual problems in the definition and interpretation of attributable fractions. American Journal of Epidemiology 1988;128:1185-97.

28. Hunter DJ, Spiegelman D, Adami H-O, Beeson L, van den Brandt PA, Folsom AR, et al. Cohort studies of fat intake and the risk of breast cancer -- a pooled analysis. New England Journal of Medicine 1996;334:356-61.

29. National Center for Health Statistics: 1987 National Health Interview Survey (database on CDROM). CD-ROM Series 10, No. 1. SETS Version 1.21. Washington: U.S. Government Printing Office, 1993, 1993.

30. Walter SD. The estimation and interpretation of attributable risk in health research. Biometrics 1976;32:829-49.

31. Wacholder S, Benichou J, Heineman EF, Hartge P, Hoover RN. Attributable risk: advantages of a broad definition of exposure. American Journal of Epidemiology 1994;140:303-09.

32. Hu G, Zhang X, Chen J, Peto R, Campbell TC, Cassano PA. Dietary vitamin C intake and lung function in rural China. American Journal of Epidemiology 1998;148:594-99.

33. Lele C, Whittemore AS. Different disease rates in two populations: how much is due to differences in risk factors? Statistics in Medicine 1997;16:2543-54.

34. Bishop Y, Fienberg S, Holland P. Discrete Multivariate Analysis. Cambridge: MIT Press, 1975.

35. Weinstein MC, Coxson PG, Williams LW, Pass TM, Stason WB, Goldman L. Forecasting coronary heart disease incidence, mortality, and cost: The Coronary Heart Disease Policy Model. American Journal of Public Health 1987;77:1417-26.

36. Rohan TE, Jain M, Howe GR, Miller AB. Alcohol consumption and risk of breast cancer: a cohort study. Cancer Causes Control 2000;11(3):239-47.

37. van den Brandt PA, Goldbohm RA, van't Veer P. Alcohol and breast cancer: results from the Netherlands Cohort Study. American Journal of Epidemiology 1995;141:907-15.

38. Wolk A, Bergström R, Hunter D, Willett W, Ljung H, Holmberg L, et al. A prospective study of association of monounsaturated fat and other types of fat with risk of breast cancer. Archives of Internal Medicine 1998;158:41-45.

39. Hunink MGM, Goldman L, Tosteson ANA, Mittleman MA, Goldman PA, Williams LW, et al. The recent decline in mortality from coronary heart disease, 1980-1990: the effect of secular trends in risk factors and treatment. Journal of the American Medical Association 1997;277:535-42.

40. van den Brandt PA, Spiegelman D, Yaun S-S, Adami H-O, Beeson L, Folsom AR, et al. Pooled analysis of Prospective Cohort Studies on Height, Weight and Breast Cancer Risk. American Journal of Epidemiology 2000;152(5):514-27.

41. Friedenreich CM, Howe GR, Miller AB, Jain MG. A cohort study of alcohol consumption and risk of breast cancer. American Journal of Epidemiology 1993;137:512-20.

42. Gapstur SM, Potter JD, Sellers TA, Folsom AR. Increased risk of breast cancer with alcohol consumption in postmenopausal women. American Journal of Epidemiology 1992;136:1221-31.

43. Graham S, Zielezny M, Marshall J, Priore R, Freudenheim J, Brasure J, et al. Diet in the epidemiology of postmenopausal breast cancer in the New York State Cohort. American Journal of Epidemiology 1992;136:1327-37.

44. Mills PK, Beeson WL, Phillips RL, Fraser GE. Dietary habits and breast cancer incidence among Seventh-day Adventists. Cancer 1989;64:582-90.

45. Willett WC, Stampfer MJ, Colditz GA, Rosner BA, Hennekens CH, Speizer FE. Moderate alcohol consumption and the risk of breast cancer. New England Journal of Medicine 1987;316:1174-80.

46. Langholz B, Thomas DC. Nested case-control and case-cohort methods of sampling from a cohort: a critical comparison. American Journal of Epidemiology 1990;131:169-76.

47. Prentice RL. A case-cohort design for epidemiologic cohort studies and disease prevention trials. Biometrika 1986;73:1-11.

48. EPICURE user's guide: the PEANUTS program. Seattle: Hirosoft, 1993.

49. Shah BV, Barnwell BG, Bieler GS. SUDAAN User's Manual, Release 7.0. Research Triangle Park, NC: Research Triangle Institute, 1996.

50. Cancer Incidence in Five Continents, vol. VI. Lyon: International Agency for Research on Cancer, 1992.

51. Huang Z, Hankinson SE, Colditz GA, Stampfer MJ, Hunter DJ, Manson JE, et al. Dual effects of weight and weight gain on breast cancer risk. Journal of the American Medical Association 1997;278:1407-11.

52. La Vecchia C, Negri E, Franceschi S, Talamini R, Bruzzi P, Palli D, et al. Body mass index and post-menopausal breast cancer: an age-specific analysis. British Journal of Cancer 1997;75:441-44.

53. Madigan MP, Ziegler RG, Benichou J, Byrne C, Hoover RN. Proportion of breast cancer cases in the United States explained by well-established risk factors. Journal of the National Cancer Institute 1987;87:1681-85.

54. Seidman H, Stellman SD, Mushinski MH. A different perspective on breast cancer risk factors: some implications of the nonattributable risk. CA - A Cancer Journal for Clinicians 1982;32:301-13.

55. Baanders AN, de Waard F. Breast cancer in Europe: the importance of factors operating at an early age. European Journal of Cancer Prevention 1992;1:285-91.

56. Hoel DG, Wakabayashi T, Pike MC. Secular trends in the distribution of the breast cancer risk factors - menarche, first birth, menopause, and weight - in Hiroshima and Nagasaki, Japan. American Journal of Epidemiology 1983;118:78-89.

57. Insull WJ, Oiso T, Tsuchiya K. Diet and nutritional status of Japanese. American Journal of Clinical Nutrition 1968;21:753-77.

58. Hirayama T. Epidemiology of breast cancer with special reference to the role of diet. Preventive Medicine 1978;7:173-95.

59. Gray GE, Pike MC, Henderson BE. Breast-cancer incidence and mortality rates in different countries in relation to known risk factors and dietary practices. British Journal of Cancer 1979;39:1-7.

60. Prentice RL, Sheppard L. Dietary fat and cancer: consistency of the epidemiologic data, and disease prevention that may follow from a practical reduction in fat consumption. Cancer Causes and Control 1990;1:81-97.

61. Hunter DJ, Willett WC. Nutrition and breast cancer. Cancer Causes and Control 1996;7:56-68.

62. Boyd NF, Martin LJ, Noffel M, Lockwood GA, Tritchler DL. A meta-analysis of studies of dietary fat and breast cancer risk. British Journal of Cancer 1993;68:627-36.

63. Smith-Warner S, Spiegelman D, Adami H, Beeson L, van den Brandt P, Folsom A, et al. Types of Dietary Fat and Breast Cancer: A Pooled Analysis of Cohort Studies. International Journal of Cancer 2001.

64. Freudenheim JL, Marshall JR, Vena JE, Moysich KB, Muti P, Laughlin R, et al. Lactation history and breast cancer risk. American Journal of Epidemiology 1997;146:932-38.

65. Pathak DR, Whittemore AS. Combined effects of body size, parity, and menstrual events on breast cancer incidence in seven countries. American Journal of Epidemiology 1992;135:153-68.

66. Thune I, Brenn T, Lund E, Gaard M. Physical activity and the risk of breast cancer. New England Journal of Medicine 1997;336:1269-75.

67. Bergkvist L, Persson I. Hormone replacement therapy and breast cancer - a review of current knowledge. Drug Safety 1996;15(360-370):360-70.

68. Barnes-Josiah D, Potter JD, Sellers TA, Himes JH. Early body size and subsequent weight gain as predictors of breast cancer (Iowa, United States). Cancer Causes and Control 1995;6:112-18.

69. van den Brandt PA, Dirx MJM, Ronckers CM, van den Hoogen P, Goldbohm RA. Height, weight, weight change, and postmenopausal breast cancer risk: the Netherlands Cohort Study. Cancer Causes and Control 1997;8:39-47.

70. Spiegelman D, Israel RG, Bouchard C, Willett WC. Absolute fat mass, percent body fat, and body-fat distribution: which is the real determinant of blood pressure and serum glucose? American Journal of Clinical Nutrition 1992;55:1033-44.

71. World Health Organization Expert Committee on Physical Status, Physical status: The use and interpretation of anthropometry. World Health Organization, Geneva, 1995.

72. Henderson BE, Ross RK, Pike MC. Toward the primary prevention of cancer. Science 1991;254:1131-38.

73. Garland M, Hunter DJ, Colditz GA, Manson JE, Stampfer MJ, Spiegelman D, et al. Menstrual cycle characteristics and history of ovulatory infertility in relation to breast cancer risk and a large cohort of US women. American Journal of Epidemiology 1998;147:636-43.

74. Colditz GA, Frazier AL. Models of breast cancer show that risk is set by events of early life: prevention efforts must shift focus. Cancer Epidemiology, Biomarkers and Prevention 1995;4:567-71.

Table 1. Characteristics of the Pooling Project studies included in the analyses of the population attributable risk of postmenopausal breast cancer due to reproductive and anthropometric factors

Study Years of Baseline Age Range No. of

Follow-up Cohort (yrs) Cases*

Adventist Health Study 1976-1982 15,172 28-90 100

Canadian National Breast Screening Study 1982-1987 56,837 40-59 242

Iowa Women’s Health Study 1986-1995 34,406 55-70 1103

Netherlands Cohort Study 1986-1992 62,412 55-69 843

Nurses’ Health Study (a) 1980-1986 89,046 35-60 557

Nurses’ Health Study (b) 1986-1995 68,817 40-66 1240

Sweden Mammography Cohort 1987-1993 61,471 38-76 274

TOTAL 319,344 4,359

*Cases consisted of women diagnosed with postmenopausal invasive breast cancer after exclusion of women with missing data on age at menarche, parity, age at first birth, body mass index and height.

Table 2. Partial population attributable risks for five reproductive and anthropometric factors which define the high-risk group using the median of a rural Chinese population

Pooling Project U.S. (1987 NHIS)

Risk Factor* RR 95% CI Prevalence % PAR‡ 95% CI

Age at Menarche (years)

17 1.00 2.1

Parity, Age at First Birth (no., years)

0 1.47 1.28 - 1.68 11.8

1-22 1.09 0.96 - 1.23 5.0 16 9 - 24

Body Mass Index (kg/m2)

21.0 1.14 1.03 - 1.26 82.1 10 2 - 18

Height (m)

1.54 1.22 1.04 - 1.43 92.7 17 5 - 29

* The reference category for each risk factor was defined using the median value of that risk factor (Banoo Parpia, personal communication) from a 1989 ecologic survey of 69 rural Chinese counties [Hu, 1998 #1177].

‡ The PAR is calculated for a shift in the U.S. distribution for each risk factor to the low-risk half of a rural Chinese population.

Table 3. Descriptive data for five breast cancer risk factors for the United States, Canada, the Netherlands, Sweden, and China

United States* Canada† Netherlands‡ Sweden§ China**

Variable Mean (SE) Median Mean (SD) Median Mean (SD) Median Mean (SD) Median Mean (SD) Median

Age at menarche (years) 12.8 (0.0) 12 12.8 (1.9) 13 13.7 (1.8) 14 13.3 (1.4) 13 16.8 (2.1) 17

Parity 2.6 (0.0) 2 2.6 (1.7) 2 2.8 (2.1) 3 2.1 (1.2) 2 4.1 (1.9) 4

Age at first birth (years) 22.4 (0.1) 21 24.4 (4.4) 24 26.6 (4.0) 26 24.1 (4.6) 23 22.7†† (3.9) 22

Body mass index (kg/m2) 25.3 (0.1) 24.3 25.5 (4.5) 24.6 25.2 (3.5) 24.8 24.4 (3.9) 23.7 21.3 (3.1) 21.0

Height (m) 1.63 (0.00) 1.62 1.62 (0.06) 1.62 1.65 (0.06) 1.65 1.65 (0.06) 1.65 1.54 (0.06) 1.54

* 1987 National Health Interview Survey (n = 4,800 35-64 year old women)

† Sub-cohort of the Canadian National Breast Cancer Screening Study, Tom Rohan, personal communication (n = 5,681 women)

‡ Sub-cohort of the Netherlands Cohort Study (n = 1,107 55-64 year old women)

§ Baseline cohort of the Sweden Mammography Cohort (n = 24,336 38-64 year old women)

** 1989 ecologic survey of 35-64 year old women residing in 69 rural Chinese counties, Dr. Banoo Parpia, personal communication

†† Data are provided for age at first pregnancy, not age at first birth.

Table 4. Predicted reduction in individual risk for changes in the medians of five breast cancer risk factors in the U.S., Canada, Netherlands, and Sweden to the medians in rural China

Pooling Project Individual Risk Reduction (Sensitivity Analysis**)

Risk Factor Increment RR (95% CI) U.S. Canada Netherlands Sweden

Age at Menarche 1 year decrease 1.06 (1.03-1.08) 24 (15-32) 19 (12-26) 15 (9-20) l9 (12-26)

Parity 1 birth decrease 1.08 (1.06-1.10) 14 (10-17) 14 (10-17) 7 (5-9) 14 (10-17)

Age at First Birth 1 year increase 1.00 (1.00-1.01) -0.4 (-1-0) 1 (0-1) 1 (0-3) 0.4 (0-1)

Body Mass Index 1 kg/m2 increase 1.02 (1.01-1.02) 5 (2-7) 5 (3-7) 5 (3-8) 4 (2-7)

Height 1 m increase 1.02 (1.01-1.02) 11 (7-15) 11 (7-15) 15 (10-20) 15 (10-20)

Reproductive factors* 34 (24-43) 31 (21-40) 22 (14-30) 31 (21-39)

Anthropometric factors† 15 (10-21) 16 (10-21) 20 (13-26) 18 (12-24)

All 44 (31-55) 42 (29-53) 37 (25-48) 44 (30-54)

* Reproductive factors = age at menarche, parity, and age at first birth

† Anthropometric factors = body mass index and height

** As a sensitivity analysis, the individual risk reduction was recalculated using the lower and upper bound of the relative risks for the risk factor(s) involved.

................
................

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

Google Online Preview   Download