Oral contraceptives and breast cancer review and meta-analysis

Share Embed

Descrição do Produto

Oral Contraceptives and Breast Cancer Review and Meta-Analysis Isabelle Romieu, MD,* Jesse A. Berlin, DSc,t$ and Graham Colditz, MBBSt.5

To evaluate the relation between use of oral contraceptives and the incidence of breast cancer, the authors reviewed the epidemiologic literature and used quantitative methods to summarize the data. Study results for any use of oral contraceptives were pooled using a model that accounted for both interstudy and intrastudy variability. The authors also explored interstudy variability and modeled a duration-effect relation between oral contraceptive use and breast cancer. Case-control and follow-up studies were considered separately. Overall, the authors observed no increase in the risk of breast cancer for women who had ever used oral contraceptives, even after a long duration of use. These results were consistent across study design. However, data combined from case-control studies revealed a statistically significant positive trend (P= 0.001) in the risk of premenopausal breast cancer for women exposed to oral contraceptives for longer duration. This risk was predominant among women who used oral contraceptives for at least 4 years before their first term pregnancy (relative risk = 1.72; 95% confidence interval = 1.36 to 2.19). Additional study is required to determine whether this finding in a subgroup of exposed women is confirmed and whether the risk remains increased with advancing age. Cancer 66:2253-2263,1990.


the approval of oral contraceptives for general use, concern was raised that use of these exogenous female hormones may result in an increased risk of breast cancer.' Many epidemiologic studies have been performed to address this a s s ~ c i a t i o n , however, ~-~~ the controversy persists and has recently been reinforced by two studies showing a positive association.25226 An adOON AFTER

From the Harvard School of Public Health, Boston, Massachusetts; and the Channing Laboratory, Brigham and Women's Hospital and Harvard Medical School, Boston, Massachusetts. Supported by research grants CA 40356, CA 46475, and CA 50385 from the National Institutes of Health and by grant HS 05936 from the National Centers for Health Services Research and Health Care Technology Assessment. * Department of Epidemiology, Harvard School of Public Health, Boston, Massachusetts. t Technology Assessment Group, Department of Health Policy and Management, Harvard School of Public Health, Boston, Massachusetts. 4 Clinical Epidemiology Unit, Section of General Internal Medicine, University of Pennsylvania, Philadelphia, Pennsylvania. 8 Channing Laboratory, Department of Medicine, Brigham and Women's Hospital and Harvard Medical School, Boston, Massachusetts. Address for reprints: Graham Colditz, MBBS, 180 Longwood Avenue, Boston, MA 021 15-5899. Accepted for publication April 26, 1990.

ditional analysis of existing data seems worthy for several reasons: First, breast cancer is the most common cancer among women in the United States and many developed countries and a leading cause of cancer mortality. Second, oral contraceptives are widely used,38therefore, even a small effect on the risk of breast cancer will have an important public health impact. Third, use patterns have changed over time,39and recent studies must be included with earlier studies in an analysis of time trends. Finally, in view of the wide media coverage after the report of an elevated risk of premenopausal breast cancer in relation to oral contraceptive use,26it seemed important to place this finding in the context of all published studies of oral contraceptives and breast cancer. In a comprehensive review of the health effects of oral contraceptives, Prentice and Thomas4' concluded that there was no association between ever having used oral contraceptives and risk of breast cancer (relative risk [RR] = 1.0). Others have drawn similar conclusions of no association between use of oral contraceptives and risk of breast ~ a n c e r . ~In' ,this ~ ~article we combine data to evaluate quantitatively the possible association of oral con-


CANCERDecember I 1990


traceptive use and the risk of breast cancer, taking account of disparities in populations and time frames among studies.

Vol. 66

lated to breast cancer survival, introducing a potential bias.45-47

Statistical Met hods Methods

By using a computerized literature search (MEDLINE) from 1966 through 1989, and by checking references of previous manuscripts, we identified all available epidemiologic studies of oral contraceptive use and breast cancer published in English and French. When a study had been reported in different articles, only the most recent report was included in the analysis (unless the necessary data had appeared only in an earlier article). Similarly, we omitted investigations that were only a subset of a larger study for which separate reports were available. Several issues may present obstacles to the pooling of different studies and must be borne in mind when evaluating the relevant literature. I First, the latency period of the carcinogenic process may require prolonged exposure and an extended follow-up period to be assessed. The earliest investigations of oral contraceptives and breast cancer may have been conducted too close to the introduction of oral contraceptives to detect any adverse association. Second, the formulation of the pill must be considered, because the dosage and constituents have both changed significantly since the introduction of oral contraceptives. Third, epidemiologic evidence suggests that risk factors for breast cancer may differ according to menopausal stat ~ . Different ~ ~ ,proportions ~ ~ of premenopausal and postmenopausal patients among studies could therefore obscure a potential relation between oral contraceptive use and breast cancer risk. Finally, time of exposure with regard to a woman’s reproductive life may influence the carcinogenic process. To address these issues, both the age range of the patients and the years during which cases were accrued were accounted for in the analysis. We combined the available data to determine the risk of breast cancer associated with ever using oral contraceptives, duration of oral contraceptive use, time since first oral contraceptive use, and duration of oral contraceptive use before first full-term pregnancy. We also explored the possible adverse effect of oral contraceptive use among women with a family history of breast cancer. The two different types of study design, case-control and follow-up, were considered separately because methodologic issues may be more likely to bias the results in studies using a case-control design. From the available case-control studies, we excluded one study2’ that did not provide sufficient information to compute an overall RR and confidence limits and a study based on prevalent cancer cases3’ because oral contraceptive use may be re-

Although the measure of association obtained in casecontrol studies is the odds ratio, and it is the RR in followup studies, we use the term relative risk (RR) throughout this article to refer to both estimates of association. To combine RR, we used a method proposed by DerSimonian and Laird.48The method was derived for rate differences in randomized trials, however, the authors suggest that their random effects approach might be adapted to the log odds ratio or log RR scales. Differences among study populations in age structures, and other factors, and differences among authors in analytic approaches should lead to variability among studies in the estimates of the RR for breast cancer associated with oral contraceptive use. The DerSimonian and Laird method allows the true underlying RR for each population to vary among studies and accounts for both the within-study and amongstudy variance by incorporating both components of the variance into weights assigned to each study when taking the weighted average of within-study results. DerSimonian and Laird provide a test (based on Cochran’s Q statistic4’) of the hypothesis that the within-study estimates of the log, (RR)[ln(RR)] are homogeneous across studies. In the presence of any heterogeneity, even when the formal test accepts the null hypothesis of homogeneity, the DerSimonian and Laird method should lead to wider confidence intervals than the methods assuming constancy of the ln(RR) across studies. We used the DerSimonian and Laird method to summarize data from studies that provided comparable exposure categorization (any use of oral contraceptives, 10 or more years duration of oral contraceptive use, time since first use of 10 or more years, 4 or more years duration of oral contraceptive use before first full-term pregnancy, family history of breast cancer). In a second approach, we used a weighted least-squares regression model to test for trend in the relation between duration of oral contraceptive use and breast cancer. In this model, the In(RR) of breast cancer (as adjusted for covariates by the original authors) was expressed as a function of oral contraceptive use, controlling for study. In other words, the slope of the duration-response curve (on a log scale) was assumed to be the same for each study, but the RR of breast cancer at a given duration of oral contraceptive use was allowed to vary. Models were estimated with the use of the SAS General Linear Models program (SAS Institute, Cary, NC).” The weight given to each ln(RR) was the inverse of the variance calculated from the confidence interval. Thus, findings from larger studies were given greater weight in estimating the overall

No. I 1



Romieu et al.


effect of oral contraceptives. Similar models were used to dence interval, 0.91 to 1.63) for I0 or more years of use compared with those who never used oral contraceptives assess the effects of latency and duration before first fullterm pregnancy. Heterogeneity of slopes across studies (Table 2). We then looked for a trend in breast cancer risk with was tested statistically by fitting models with an interaction term for study by duration and by conducting F-tests." increasing duration of oral contraceptive use, combining those studies with the majority ofcases accrued after 1975 On the basis of our a priori knowledge of the potential and that provided estimates for duration of oral contrasources of variability in the relation between oral contraceptive use.739,11-16,18-28 Overall, we observed no trend in ceptive use and breast cancer, we also fitted separate risk with increasing duration of oral contraceptive use. models for case-control studies in which accrual ended However, when testing for heterogeneity in slopes, we obbefore 19752-6and those in which accrual started in 1980 served an interaction of borderline significance between or more r e ~ e n t l y . ' ~Menopausal -~~ status was not clearly study and duration of oral contraceptive use. We explored defined in many of the studies, hence we considered that this apparent heterogeneity by regressing the In(RR) on women younger than 46 years of age were premenopausal various study attributes and observed that this effect esunless otherwise specified. We fitted separate models for sentially resulted from differences in the proportion of studies including only premenopausal cases13,14,18,19,25,27,28 premenopausal patients among the studies. Adjusting for and those with less than 50% premenopausal cases.7.9,11,12,16,20,22,24,25 the midpoint year of case accrual, the model showed that the ln(RR) increased significantly with an increasing proportion of premenopausal patients ( P = 0.007). We next Results limited the analysis to data available on women younger than 46 years of age (i.e., predominantly premenopausal Case-Control Studies patients).~3,14,~8,~9,25-28 The slope and standard error of the coefficient for years of oral contraceptive use indicate that We identified 29 case-control studies; two were exthe duration-response relation was statistically significant cluded from this analysis (see Methods). Details of the 27 ( P = 0.008), corresponding to a 46% increase in risk for remaining studies-including age range of the cases, pe10 years of oral contraceptive use (Table 3). riod of case accrual, number of cases and controls inSince the publication of the study by Pike ec ~ l . ,the '~ cluded, factors of adjustment, and RR of breast cancer role of oral contraceptive use before first full-term pregfor any use of oral contraceptives and different durations nancy has received much epidemiologic attention, esof use-are presented in Table 1. In 24 studies--7,9-25,21 pecially within investigations of young women (younger any use of oral contraceptives did not alter the risk of than 46 years of age at diagnosis). The results are conflictbreast cancer. In the three remaining studies,8326.28 any ing but suggest a possible adverse effect from long duration use of oral contraceptives and breast cancer were signifof use before first full-term pregnancy (Table 4). The RR icantly associated. In the Swedish study,28both duration range from 0.6 for 6 or more years of oral contraceptive of use before 25 years of age and commencement of use use22to 8.0 for 5 or more years (this result was based on at a young age were associated with an increase in the risk of breast cancer. In a case-control study from Toronto,26 one case)". When we combined the results from studies with data available on 4 years or more of oral contracepMiller et al. used community controls and found an RR tive use before first full-term pregnancy among women of 3 for use of less than 1 year among nulliparous women and an RR ranging from 1.8 to 2.2 among parous women. younger than 46 years of age,14,18,19,25,26,27,28,32 the pooled RR estimate for breast cancer was 1.72 (95% confidence These high RR for very short durations of use suggest that interval, 1.36 to 2.19). To address the issue of change in control participation may have been biased. pill composition, we also pooled studies with cases accrued Overall, the RR estimates reported from the case-conafter 1980,19,25,26,32but the estimates remained similar trol studies ranged from 0.717 to 2.0.26When we consid(Table 4). ered the studies that provided details on duration of oral We next explored the possibility of a trend in breast contraceptive use, the RR of breast cancer for the longest cancer risk with increasing duration of oral contraceptive duration of use ranged from 0.49 to 2.8.26 When sumuse before first term pregnancy among premenopausal marizing studies with similar exposure categorization, and women and observed a small increase in risk with duration accounting for interstudy variability, the RR of breast of oral contraceptive use. However, this relation did not cancer for 10 years or more of oral contraceptive use was reach statistical significance (Table 3). 1.14 (95% confidence interval, 0.90 to 1.42). Because The carcinogenic process may require prolonged excomposition of oral contraceptives has changed over time, posure and an extended period of follow-up to assess any we next repeated the analyses including studies with cases latent effect. Eleven studies addressed the issue of accrued after 1980, to reflect the effect of the most recent latency7.1 1,12,14,16,19,22-25,31 (Table 5 ) . Overall, the RR of pill composition. The RR increased to 1.22 (95% confi-

CANCERDecember 1 1990


Vol. 66

TABLE1. Oral Contraceptive (OC) Use and Breast Cancer: Case-Control Studies Age range

Time cases diagnosed

No. of cases

No. of controls

Henderson ef a/. ( 1974)’ Paffenbarger et al. (1977)’(1979)’





A, R, P

4-6 6+
Lihat lebih banyak...


Copyright © 2017 DADOSPDF Inc.