When cancer survival decreases



Why cancer survival may worsen

(Or, alternative: Explanations for worsening cancer survival)

Esther de Vries,1,2 Henrike E Karim-Kos,1 Maryska LG Janssen-Heijnen,2 Isabelle Soerjomataram,1,2 Lambertus A Kiemeney,3,4,5 Jan Willem W Coebergh1,2

1 Department of Public Health, Erasmus MC University Medical Centre, Rotterdam;

2 Eindhoven Cancer Registry, Comprehensive Cancer Centre South, Eindhoven;

3 Department of Epidemiology, Biostatistics and Health Technology Assessment,

Radboud University Nijmegen Medical Centre, Nijmegen;

4 Department of Urology, Radboud University Nijmegen Medical Centre, Nijmegen;

5 Comprehensive Cancer Centre East, Nijmegen

Correspondence:

Dr Esther de Vries

Department of Public Health, Erasmus MC

P.O. Box 2040, 3000 CA Rotterdam

The Netherlands

Tel: +31 10 704 3730

Fax: +31 10 704 8475

E-mail: e.devries@erasmusmc.nl

Abstract

When cancer survival is reported to be worsening over time or inferior compared to other countries, politicians and healthcare workers may get blamed for presumed suboptimal care. Yet, there is a variety of reasons for cancer survival statistics to change for the worse, deterioration of (access to) care being only one. Other explanations include improved diagnosis of premalignant lesions, causing only the most aggressive cancers, with a worse prognosis, to remain. Deleterious changes in the distribution of prognostic factors, and changes in the distribution of socio-demographic characteristics may also negatively affect survival proportions. In this overview article, we identify the pitfalls of comparisons of published population-based survival data from different time periods or populations.

Introduction

Cancer survival statistics attract a lot of attention, particularly when comparisons between calendar periods or countries show that survival decreases over time or is lower than expected based on the average of surrounding countries. Population-based cancer survival proportions generally tend to remain stable or increase over time in most industrialized countries and for most cancer types.1,2 Such increases in survival proportions, however, do not necessarily reflect true improvements in cancer treatment. For example, early detection and screening practices have artificial effects on survival due to lead time or length bias (Box 1).3,4 Decreasing survival proportions can sometimes be observed over time. Such deteriorating survival proportions have a variety of causes, even after adjustment for age and all-cause mortality.1,5 We briefly explain the principles of cancer survival and illustrate that a decrease in survival proportion can be attributed to four factors: deterioration of (access to) care, improved diagnosis of premalignant lesions, deleterious changes in the distribution of prognostic factors, and changes in the distribution of socio-demographic characteristics that negatively affect survival proportions. [AU: Please confirm edit - OK] In this Perspectives article, we identify possible pitfalls when comparing published survival data from different time periods or in different populations.

Short explanation of cancer survival

Cancer survival is estimated for cohorts of newly diagnosed patients, based on cytological or histological criteria and sometimes clinical criteria of patients when entered into a cohort study. Cancer patient survival is measured as the time from cancer diagnosis until death; a 5-year follow-up period is most frequently used as an indicator of outcome, although survival at 10 years would be more suitable for many cancer types, such as those amenable to screening, for example breast and prostate cancer. Endpoints for calculating survival can vary. Death due to any cause allows the overall (all-cause) survival proportion to be calculated, that is, the proportion of cancer patients alive at a certain point in time after diagnosis. Death due to the cancer under study or its treatment is reflected in the disease-specific survival proportion.

Many practical problems are encountered in correctly determining and registering cause of death. For example, it can often be difficult to determine the proper underlying cause of death.3 Relative survival circumvents the need to determine the cause of death, because it represents the ratio of the overall survival proportion for a cohort of cancer patients and the expected overall survival proportion for the general population with the same sex and age distribution as the cancer patient cohort. Relative survival therefore measures mortality associated directly and/or indirectly with the diagnosis of a cancer and thus includes deaths due to complications of cancer treatment. In order for relative survival to be interpreted as a measure of excess mortality among cancer patients it is important to have accurately estimated expected mortality data for the general population. It is assumed that cancer mortality and non-cancer mortality are independent. Moreover, in theory, relative survival is dependent on trends in other causes of death in the general population. These assumptions, however, should sometimes be questioned. For example, a markedly decreased incidence and mortality from cardiovascular diseases, which results in an increase in the expected overall survival proportion for the general population, would lead to a decrease in the relative survival ratio of cancer, even when the observed disease-specific survival proportion remains stable. Also, it is assumed that cancer patients would have had an average life expectancy if they had not been diagnosed with cancer. This is of course an arbitrary assumption for some types of cancer. For example, patients with lung cancer may have a lower life expectancy because the cause of their cancer (smoking) is also a strong risk factor for mortality from other diseases.

Calculation of relative survival ratios (also referred to as relative survival rates) requires mortality data by sex, age, and calendar year from the general population.3 Relative survival ratios can be compared over time or between geographical regions, despite the ageing of populations, but the results are meaningful only when they are age-adjusted. The most reliable survival comparisons are based on age-specific risks in case of differential periods of observation.6 The relative survival ratio is very useful for specific cancer (sub)sites, but it is not suitable when it is calculated for all cancers combined because all cancers combined do not represent a small proportion of deaths due to all causes of death combined. The "expected" survival figures for relative survival calculations in this scenario are based on competing risk mortality data that are too high and so the estimates become less valid. Moreover, survival statistics for all-sites should generally not be compared directly because the case-mixes can differ greatly between different populations and time periods.

In order to judge the quality of survival estimates given by a registry, the inclusion and exclusion rules of the registry database should be clear. If these rules are changed, for example, in coding systems such as the ICD-O international classification of diseases or TNM, survival estimates might change rapidly. When previously non-invasive ‘cancers’ were included as ‘new’ cancer cases (for example, the change of classification of bladder papillomas into papillocarcinomas in 1978), survival estimates increased markedly.7 Moreover, completeness of study follow-up is essential for accurate survival estimates. Unfortunately, administrative completeness can vary with time across and between countries. Survival proportions tend to become lower when the completeness of follow-up improves, because patients who were initially lost to follow-up usually turn out to have died. The number of death certificate only (DCO) cases in cancer registries depends on the quality of the registry and access to death certificates. It is common practice to exclude these cases for survival analysis because the date of diagnosis (and hence survival time) of DCO cases is unknown.8 Registries with a high proportion of DCO cases will overestimate their survival. Therefore, when the proportion of DCO cases decreases over time survival estimates may seem to worsen.9

Deterioration of access to care

The most obvious, but uncommon, reason for decreasing survival proportions is less aggressive or substandard cancer care, resulting in failing early detection or less effective treatment. For example, decreasing relative survival ratios for patients with laryngeal cancer were observed in the mid-1990s in the US compared with the 1980s.10 [During the mid-1990s, many clinicians preferred irradiation above laryngectomies. Detailed analyses over time revealed shorter survival for patients with laryngeal squamous-cell carcinoma following non-surgical treatment.11

Likewise, a comparison of patients with high-grade T1 bladder cancer who underwent radical surgery in the US showed a 5-year disease-free survival of 70% for patients who had radical surgery before 1998 versus only 40% after 1998. During the 1990s, intravesical therapy (for example, immunotherapy and/or chemotherapy) facilitated bladder-sparing strategies for these patients. Before 1998, 74% of these patients underwent radical surgery without prior intravesical therapy versus 43% after 1998. The observed decrease in survival was attributed to delayed radical surgery after the increased use of intravesical therapy.12

The economic collapse of the former socialistic countries of Central Europe coincided with decreasing survival proportions during the transition period; survival proportions of ovarian, cervical and uterine cancer, childhood soft-tissue sarcomas, and hepatic and germ-cell tumours temporarily decreased between 1988–1992 and 1993–1997.13,14 These temporary decreases were likely due to disintegrating health care systems and infrastructures.

When cancers are detected at a later stage or mistakenly staged as a low-stage cancer because of deteriorating screening or staging availability (for example, due to poor imaging capacities), treatment will be less effective and survival proportions will decrease. It is possible to adjust for stage at diagnosis,15 although improvements in staging methods and changes in stage-coding could still be problematic when comparing stage-adjusted survival estimates over time.

Improved diagnosis of precancers

Survival proportions may decrease while therapeutic options remain stable or improve over time and/or early detection remains unchanged or improve. This occurred for cervical cancer in many European countries, where large-scale and high quality population-based programs for cervical screening gradually became available (Table 1).1 The same phenomenon might be observed in the future for colorectal cancer.

The aim of screening for cervical and colorectal cancers is not only the detection of cancers in early stages but also the detection of pre-malignant lesions, thereby preventing the development of ‘invasive’ cancer. However, the cancers that still occur, despite screening, may consist of a selected group of rapidly growing, aggressive tumors that may be more difficult to treat and thus might result in decreasing survival proportions, preceded by a lower incidence.

Screening for most other types of cancer (for example, breast or prostate) will detect early stages of cancer rather than pre-malignant lesions. Survival increases due to lead time bias and even length bias (Box 1), and possibly as a result of more-effective treatment. Survival proportions, however, may decrease again when the awareness or willingness to participate in screening programs decreases, possibly leading to higher stage disease at diagnosis. If the date of death is not postponed by treatment this will result in worsening survival proportions, that is, an inverse ‘lead time’ effect.

New and improved methods for diagnosing cancer frequently result in improved survival proportions because these methods are more precise and/or detect the cancers earlier, hopefully resulting in better therapeutic options. Conversely, the introduction of a new diagnostic technique may temporarily decrease cancer survival proportions. Pancreatic cancer is difficult to diagnose and treat and is associated with a poor prognosis. New diagnostic technologies have resulted in more diagnoses during the patients’ lifetime, whereas previously the diagnosis would have been made at autopsy and recorded as a DCO. These tumors with a very bad prognosis will now be included in the survival statistics resulting in lower survival estimates.3

Changes for the worse in the distribution of prognostic factors

Subtypes and subsites

Changes in risk factor exposure may lead to shifts in the distribution of cancer subtypes. In the Netherlands, relative survival ratios for adenocarcinomas of the lung decreased during the 1980s despite increased application of better endoscopic techniques by more lung physicians accompanied by better access to specialized care. This decrease in relative survival ratios was partly attributed to the termination of mass screening for tuberculosis in the early 1980s, which sometimes detected slow-growing peripheral adenocarcinomas. The higher concentration of carcinogens in the peripheral lung zone as a result of the increased use of filter cigarettes and deep inhalation may also have caused tumors to more metastasize rapidly.16 Similar decreases in overall lung cancer survival were observed in Malta, with stable overall incidence and mortality rates but a relative rise in the incidence rate of adenocarcinomas.1,17

Shifts in cancer subtype and subsite distribution may need to be studied over time as a determinant of survival. This is illustrated by a study of changes in incidence and survival of gastric cancer in the southeastern part of the Netherlands. Despite marked improvements in endoscopic early detection, staging, surgery, and peri-operative mortality, no improvement in survival occurred during the period 1982 to 1995 because of the increasing proportion of cardiac and diffuse cancers with a poor prognosis.18 Another example of the negative influence of shifts in cancer subsite on survival is laryngeal cancers.8 Laryngeal cancers of the glottis exhibit a 5-year relative survival of around 60–80%, and supraglottic cancers around 40%.19 Besides earlier detection of cancer of the glottis, alcohol consumption and tobacco smoking have different etiological effects on tumor subsite, with alcohol being more important for supraglottic tumors and smoking for glottic cancers. With the decreasing smoking prevalence and stable alcohol consumption rates in many European countries, the proportion of cancers of the poor prognostic subsite (supraglottis) will increase and will negatively affect the relative survival ratios of laryngeal cancer over time.

Co-morbidity

Changes in prevalence of risk factors can also affect overall survival because of the presence of concomitant diseases (co-morbidity). If a concomitant (co-morbid) condition becomes more prevalent in the population as a whole, the relative survival ratio is corrected for this change. However, relative survival ratios are likely to be affected when the co-morbid condition is strongly associated with the tumor (for example, COPD in lung or laryngeal cancer patients) or when the co-morbid condition is not only life-threatening in itself, but also influences the eligibility of patients to receive aggressive cancer treatments such as surgery or chemotherapy, as is the case for example for diabetes.20

Changes in (behavioural) risk factors, such as an increase in alcohol consumption, cause on the long term rises in alcohol-related cancer incidence, mortality and co-morbidity, such as ischemic heart disease, stroke, hypertension, diabetes, liver cirrhosis and depression. An increased prevalence of such co-morbid conditions will cause lower survival proportions.21 Since alcohol consumption is an important risk factor, particularly for squamous-cell cancer of the esophagus and supraglottic cancer of the larynx, the above-mentioned co-morbid conditions are likely to be more prevalent among esophageal and laryngeal cancer patients resulting in less optimal treatment and more complications. This phenomenon may also have contributed to the observed declines in survival of these cancers during the transition period in the former socialist countries in central Europe.1,22,23

In a similar manner, the increased prevalence of infectious diseases related to cancer, such as the human immunodeficiency virus and the hepatitis B and C viruses,24 may thus not only cause a higher incidence of tumors such as lymphomas and liver cancer, but may also have a negative effect on survival of patients with these cancers.

Changes in socio-demographic characteristics

When patient populations change rapidly, survival proportions can also be affected; for example, when certain subgroups comprise a larger part of the cancer burden due to changes in the distribution of socio-economic status. Cancer survival proportions among people of low socio-economic classes are generally lower than those of the mid or higher socio-economic classes, with relative risks of death within 5 years of diagnosis in the most deprived groups being 1.3–1.5-fold higher compared with the most affluent group.25,26 Underlying causes of the worse prognosis are related to lower awareness, unfavourable tumor characteristics (for example, later stage at diagnosis due to diagnostic delay), patient characteristics such as race, screening participation rates, psychosocial factors, co-morbidity or health care factors including treatment, screening, and quality of medical care.26 Changes in the distribution of socio-economic classes, for example by selective emigration of the more healthy and enterprising, may cause decreasing survival proportions.

Conclusions

In conclusion, a variety of factors can cause decreasing cancer survival proportions, most of which do not necessarily represent deteriorating care. Unfavourable changes in underlying risk factors, and early detection or screening practices that lead to detection of relatively less aggressive lesions are often the cause of this phenomenon as well as changing demographic profiles of the populations. Worsening survival proportions merit in-depth investigation, taking into account preceding trends in incidence, particularly for tumor subtype or subsite.

Acknowledgement

The work on this research was performed within the framework of the project ‘Progress against cancer in the Netherlands since the 1970s?’ (Dutch Cancer Society grant 715401) and co-funded by the Comprehensive Cancer Centre South in Eindhoven, the Netherlands.

References

1. Karim-Kos, H.E. et al. Recent trends of cancer in Europe: a combined approach of incidence, survival and mortality for 17 cancer sites since the 1990s. Eur. J. Cancer 44, 1345–1389 (2008).

2. Jemal, A. et al. Cancer statistics, 2007. CA Cancer J Clin 57, 43-66 (2007).

3. Dickman, P.W. & Adami, H.O. Interpreting trends in cancer patient survival. J. Intern. Med. 260, 103–117 (2006).

4. Welch, H.G., Schwartz, L.M. & Woloshin, S. Are increasing 5-year survival rates evidence of success against cancer? JAMA 283, 2975–2978 (2000).

5. Brenner, H., Gondos, A. & Arndt, V. Recent major progress in long-term cancer patient survival disclosed by modeled period analysis. J. Clin. Oncol. 25, 3274–3280 (2007).

6. Pokhrel, A. & Hakulinen, T. How to interpret the relative survival ratios of cancer patients. Eur. J. Cancer (2008).

7. Kiemeney, L.A. et al. Bladder cancer incidence and survival in the south-eastern part of The Netherlands, 1975-1989. Eur. J. Cancer 30A, 1134–1137 (1994).

8. Berrino, F. The EUROCARE Study: strengths, limitations and perspectives of population-based, comparative survival studies. Ann. Oncol. 14 Suppl 5, v9–13 (2003).

9. Robinson, D., Sankila, R., Hakulinen, T. & Moller, H. Interpreting international comparisons of cancer survival: the effects of incomplete registration and the presence of death certificate only cases on survival estimates. Eur. J. Cancer 43, 909–913 (2007).

10. Induction chemotherapy plus radiation compared with surgery plus radiation in patients with advanced laryngeal cancer. The Department of Veterans Affairs Laryngeal Cancer Study Group. N. Engl. J. Med. 324, 1685–1690 (1991).

11. Hoffman, H.T. et al. Laryngeal cancer in the United States: changes in demographics, patterns of care, and survival. Laryngoscope 116, 1–13 (2006).

12. Lambert, E.H. et al. The increasing use of intravesical therapies for stage T1 bladder cancer coincides with decreasing survival after cystectomy. BJU Int 100, 33–36 (2007).

13. Coebergh, J.W. et al. Leukaemia incidence and survival in children and adolescents in Europe during 1978-1997. Report from the Automated Childhood Cancer Information System project. Eur. J. Cancer 42, 2019–2036 (2006).

14. Aareleid, T. & Brenner, H. Trends in cancer patient survival in Estonia before and after the transition from a Soviet republic to an open-market economy. Int. J. Cancer 102, 45–50 (2002).

15. Ciccolallo, L. et al. Survival differences between European and US patients with colorectal cancer: role of stage at diagnosis and surgery. Gut 54, 268–273 (2005).

16. Janssen-Heijnen, M.L. & Coebergh, J.W. Trends in incidence and prognosis of the histological subtypes of lung cancer in North America, Australia, New Zealand and Europe. Lung Cancer 31, 123–137 (2001).

17. Micallef, R. Incidence of lung cancer in Malta for the years 1992-2005 with all morphology subtypes. Personal communication. [AU: Is this an unpublished reference? Is so, this does not get cited in the reference list but as unpublished data within the text and includes the author’s initial and surname] – OK – yes this is a personal communication – not published

18. Pinheiro, P.S., van der Heijden, L.H. & Coebergh, J.W. Unchanged survival of gastric cancer in the southeastern Netherlands since 1982: result of differential trends in incidence according to Lauren type and subsite. Int. J. Cancer 84, 28–32 (1999).

19. Berrino, F. & Gatta, G. Variation in survival of patients with head and neck cancer in Europe by the site of origin of the tumours. Eur. J. Cancer 34, 2154–2161 (1998).

20. van de Poll-Franse, L.V. et al. Less aggressive treatment and worse overall survival in cancer patients with diabetes: a large population based analysis. Int. J. Cancer 120, 1986–1992 (2007).

21. Janssen-Heijnen, M.L. et al. Prognostic impact of increasing age and co-morbidity in cancer patients: a population-based approach. Crit. Rev. Oncol. Hematol 55, 231–240 (2005).

22. Zaborskis, A., Sumskas, L., Maser, M. & Pudule, I. Trends in drinking habits among adolescents in the Baltic countries over the period of transition: HBSC survey results, 1993–2002. BMC Public Health 6, 67 (2006).

23. Zatonski, W. & Jha, P. (Cancer Center and Institute of Oncology, Warsaw, 2000). [AU: Please can you specify if this is a conference and provide the date and details] No this is a report from a project. The title somehow disappeared from the reference list and I did not notice. My apologies, the title of the report is: The Health Transformation in Eastern Europe after 1990: A Second Look and the report can be found through this link:

24. Parkin, D.M. The global health burden of infection-associated cancers in the year 2002. Int. J. Cancer 118, 3030–3044 (2006).

25. Coleman, M.P., Babb, P., Sloggett, A., Quinn, M. & De Stavola, B. Socioeconomic inequalities in cancer survival in England and Wales. Cancer 91, 208–216 (2001).

26. Woods, L.M., Rachet, B. & Coleman, M.P. Origins of socio-economic inequalities in cancer survival: a review. Ann. Oncol. 17, 5–19 (2006).

Table 1 Relative survival for cervical cancer in Europe

|Region |Start of screening |Type of screening invitation |5-year relative survival |Trends in survival |

| |program | |ratios | |

| | | |1990–1994a |2000–2002b, c, | |

| | | | |d | |

|Austria (Tirol) |1970 |PB/OP (regional) |63.6 |64.2 |Increased |

|Iceland |1964 |NRS |68.6 |70.6 |Increased |

|Finland |1963 |PB |66 |65.8 |No change |

|Italy |1982–1998 |PB/OP (regional) |66.6 |67 |No change |

|Netherlands (3 regions) |1980 |PB |69.4 |69.2 |No change |

|Scotland |1988 |NRS |60.6 |61 |No change |

|Norway |1995, pilot 1992 |NRS |69 |67.5 |Decreased |

|Sweden |1967–1977 |NRS |69.6 |66.7 |Decreased |

|Switzerland |No data |OP |68.7 |66.8 |Decreased |

|Czech republic |1966 |OP |65.2 |59.8 |Large decrease |

|England |1988 |NRS |63.8 |58.6 |Large decrease |

|Germany (Saarland) |1971 |OP |63.5 |55.5 |Large decrease |

|Malta |No data |No data [AU: okay? – yes] |64.4 |46.5 |Large decrease |

|Poland |No data |No data [AU: okay? - yes] |48.2 |56 |Large increase |

|Slovenia |2003 (1995 |NRS |59.9 |65.2 |Large increase |

| |opportunistic) | | | | |

|Spain |1986 (regional) |PB/OP |68.7 |60.4 |Large decrease |

|Wales |1988 |NRS |58.7 |52.6 |Large decrease |

|PB: population-based: invitational programs |

|NRS: only invitation to women that did not recently have an (opportunistic) smear |

|OP: opportunistic only |

|=: 5% point difference |

|aSant M, et al. EUROCARE-3: survival of cancer patients diagnosed 1990-94-results and commentary. Ann Oncol. 2003;14 Suppl 5:V61-V118. |

|bVerdecchia A, et al. Recent cancer survival in Europe: a 2000-02 period analysis of EUROCARE-4 data. Lancet Oncol. 2007;8:784-96. |

|c Linos A, Riza E. Comparisons of cervical cancer screening programmes in the European Union. Eur J Cancer. 2000;36:2260-65. |

|d Anttila A, et al. Cervical cancer screening programmes and policies in 18 European countries. Br J Cancer. 2004;91:935-41. [AU: Please |

|clarify what these two references pertain to – hope this is better now?] |

Box 1. Definitions of lead time and length bias.Lead time: early detection or screening aims to diagnose a disease in an earlier stage than it would be without screening. When the moment of death is not postponed by screening, and therefore no additional life-time has been gained, the survival time since diagnosis is longer for a screened person compared to unscreened people. In this case, screening appears to increase survival time, and this gain is called lead time.

Length bias: slow growing tumours often have better prognosis than tumours with high growth rates. Early detection or screening is more likely to detect slow growing tumours as they have a longer time-period of existing without clinical symptoms, some of them might actually never have caused clinical disease. In such cases, screen-detected tumours appear to be associated with better survival statistics, because they actually represent a group of tumours which inherently already had a better prognosis. This effect is called length bias.

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

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

Google Online Preview   Download