Oral Contraceptives and Survival in Breast Cancer Patients Aged 20 to 54 Years

Use of oral contraceptive (OC) pills is common. In the 2002 National Survey of Family Growth, OCs were the leading method of contraception, with 82% of U.S. women aged 15 to 44 years reporting ever using them (1). Worldwide, more than 300 million women have ever used OCs (2). Thus, understanding the full range of health effects associated with their use is important.

Among younger women, current or recent OC use and use of higher dose and potency pills have been associated with modestly higher breast cancer incidence (3, 4), but effects on survival are unclear (5-22). To date, no studies have examined pill characteristics and survival.

The number of female breast cancer survivors was estimated to be 2.36 million in 2003 (23). Obesity at diagnosis has been associated with increased mortality, possibly due to estrogen-mediated mechanisms (24). Thus, other hormone-related characteristics (e.g., OC use) may influence survival, possibly contributing to the higher mortality among younger patients (23).

This large, population-based study of breast cancer patients aged 20 to 54 years with detailed data on OC use investigated associations between survival and OC use before diagnosis, including ever use, duration, age at first use and last use, duration of use before age 25, duration of use before a first birth, and time since first and last use. We also examined whether survival systematically differed according to OC dose, potency, or formulation.

A cohort of patients with invasive breast cancer previously enrolled in a population-based case-control study (25) was followed to investigate predictors of survival. The parent case-control study included 2,203 patients, aged 20 to 54 years, diagnosed with primary in situ or invasive breast cancer between May 1, 1990, and December 31, 1992, who were residents of central New Jersey, metropolitan Atlanta, GA, and metropolitan Seattle, WA (25). Structured in-person interviews were conducted by trained interviewers shortly after diagnosis (median, 4.2 months). Interviews were completed for 86% of eligible in situ and invasive cases (25). Reasons for patients not being interviewed included refusal (5.4% physician refusal and 6.4% patient refusal), death (0.4%), and illness (0.6%; ref. 25).

The cohort of women in this report includes only the invasive breast cancer cases from New Jersey and metropolitan Atlanta, GA. Of the 1,283 patients in the cohort, 19 patients with missing vital status were excluded; 1,264 (98.5%) patients were included in this analysis and had follow-up for 8 to 10 years. Institutional Review Boards at collaborating institutions approved this study.

Interviews from the case-control study provided data on OC use before diagnosis and most of the covariate information. A reproductive history calendar was used to help women recall OC use by relating it to key life events. For each episode of OC use, participants were asked the starting and stopping dates and to name the pill used. Color photographs and listings of all marketed pills were used to aid brand recognition.

To analyze OC dose, potency, and formulation, those who could not recall dates of use or pill types (202 women, 22.5% of users) or used progestin-only pills (6 women, <1% of users) were excluded, similar to our previous analyses (4). Because classification of pill content was complicated for use of multiple pills at different times for varying durations, components of combination OCs used during various periods were analyzed (4). This classification included information on the pill used during (a) the most recent period, (b) the longest period within 5 years before diagnosis, (c) the longest period within 10 years before diagnosis, and (d) the longest period. For each period, the corresponding OC was classified by formulation (estrogen and progestin type), potency (cross-classification of estrogen and progestin potency), and estrogen dose. Potency was calculated as previously described (4), according to the scheme of Piper and Kennedy (26), which classifies OCs by pharmacologic effects on target organs. Pills with progestins of intermediate potency were ultimately classified as low progestin potency. Pills with estrogens of intermediate potency were defined as low estrogen potency if they contained ≤35 μg ethinyl estradiol or ≤50 μg mestranol; otherwise, they were defined as high potency.

Summary disease stage (local, regional, or distant; ref. 27), tumor grade, and estrogen and progesterone receptor status were abstracted from medical records in the case-control study. In the interview, all patients were queried about previous treatment. In Atlanta only, more detailed information on disease stage according to the American Joint Committee on Cancer (28) and the first course of breast cancer treatment (radiation, chemotherapy, hormonal therapy, and surgery) were available from the Surveillance, Epidemiology, and End Results program and additional medical record review.

Follow-up time was calculated from diagnosis to death or study completion (median, 8.5 years; range, 3 months to 9.8 years). Linkage to the National Death Index provided data on vital status and date and cause of death. By the end of follow-up (January 1, 2000), there were 292 deaths; 248 (85%) attributed to breast cancer, 15 (5%) to cardiovascular disease, 13 (4.5%) to other causes (infections, accidents, other cancers, diabetes, and liver disease), and 16 (5.5%) had no cause of death listed.

The proportional hazards assumption was assessed through log(−log(survival)) plots and including interactions with follow-up time. Household income and time since last OC use violated this assumption. Time-dependent covariates were introduced to address the violation. The nonproportionality for income was addressed by including an interaction term between income and follow-up time. For time since last OC use, the log(−log(survival)) plot revealed that there was a cross-over at 24 months (i.e., the effect of time since last OC use was different before and after 24 months of follow-up). Therefore, we added an interaction term between time since last OC use and an indicator function for follow-up time being >24 months, which allowed for the effect of time since last use to differ before and after 24 months of follow-up. Variables were categorized based on exploratory analyses of the data and were modeled by indicator variables.

Adjusted hazard ratios (HR) and 95% confidence intervals (95% CI) were estimated by Cox regression methods (29). Because the quality of cause of death information on death certificates has been questioned in at least one study (30), all-cause mortality was the main end point, and patients alive at the end of follow-up were censored. Analyses of breast cancer–specific mortality were also conducted, and participants who died of causes other than breast cancer were censored. Modeling was done separately for each attribute of OC use: ever use for ≥6 months, duration, age at first use, age at last use, duration of use before age 25, duration of use before a first birth, time since first use, and time since last use. To be consistent with previous analyses based on the same group of breast cancer patients (4, 25), the reference group for all analyses reported here is non-OC users, defined as those who used OCs for less than 6 months (including never users). Results were similar when never users were used as the reference (data not shown).

Covariates were initially included in multivariable models if they were associated both with the particular OC variable and mortality in bivariate analyses. Potential confounders included age at diagnosis, education, household income, physical activity (at age 20 and the year before diagnosis), body mass index (at age 20 and the year before diagnosis), family history of breast cancer, race, smoking status, alcohol consumption, parity, age at menarche, recency of giving birth, study site, comorbidity (thyroid disease, diabetes, high cholesterol, high blood pressure, and cancers other than breast cancer), method of cancer detection (e.g., accidental, routine self-examination, or mammogram), chemotherapy or radiation before interview, summary stage, tumor grade, and hormone receptor status. Model building proceeded through backward elimination, and variables were retained if inclusion caused >10% change in the estimate for the OC characteristic. All results were adjusted for age (<35, 35-44, or 45-54 years) and annual household income (<$15,000, $15,000 to <$25,000, $25,000 to <$90,000, or ≥$90,000) because these were the only consistent confounders. Each OC use and mortality estimate was also adjusted for additional confounders specific to that association (listed in the footnotes to Tables 2–4). Estimates adjusted for summary stage, tumor grade, and hormone receptor status were similar to estimates without adjustment. In the Atlanta subgroup, adjustment for the detailed treatment information or stage according to the American Joint Committee on Cancer did not alter any estimates.

Effect modification by age (<45 or ≥45 years), menopausal status, estrogen receptor status, summary stage, family history, smoking status, body mass index the year before diagnosis, and method of cancer detection were evaluated by examining stratum-specific HRs and including product interaction terms. Effect modification was considered statistically significant if the P value for the likelihood ratio test comparing models with and without the interaction term(s) was <0.05.

The majority of women were 35 to 44 years of age, had local disease, and were premenopausal (Table 1). Of the 1,264 participants, 367 (29.0%) were non-OC users, 122 (9.7%) started use at or before 18 years of age, and 97 (7.7%) were using OCs or had recently stopped use at diagnosis of breast cancer (data not shown).

Ever use (≥6 months) of OCs was not associated with all-cause mortality (HR, 1.00; 95% CI, 0.77-1.29), nor was total duration of use or age at first use, relative to non-use (Table 2). Age at last use, duration of use before age 25 years, and duration of use before a first birth were also not associated with survival (data not shown).

Timing of OC use before breast cancer diagnosis seemed to affect survival, although estimates were not statistically significant (Table 2). Users at diagnosis or those who had stopped OC use in the previous year tended to have higher all-cause mortality versus nonusers (HR, 1.57; 95% CI, 0.95-2.61), but only after 24 months of follow-up (Table 2). A similar point estimate (HR, 1.60) was observed for having recently started OCs (<10 years before diagnosis). When we included both time since first use and time since last use in the same model, the estimates for each were unchanged.

Women who used high-dose estrogen pills (>35 μg of ethinyl estradiol or >50 μg of mestranol) in the 5 years before diagnosis were more than twice as likely to die than nonusers (HR, 2.39; 95% CI, 1.29-4.41; Table 3). Pill potency was not independently associated with mortality. Across all examined periods, use of the progestin levonorgestrel was associated with higher mortality. For example, the HR for use of levonorgestrel in the most recently used pill was 2.01 (95% CI, 1.03-3.91), relative to nonusers. However, use of levonorgestrel in multiple periods was highly correlated, so it was difficult to determine which period was most important. After examining all progestins separately (data not shown), levonorgestrel was the only progestin that seemed to affect survival; therefore, all other progestins were grouped together. Adjustment for dose and potency did not alter time since first and last use estimates.

No statistically significant effect modification was observed. HRs for mortality specific to breast cancer (Table 4) were higher than those for all-cause mortality.

Results of previous studies on OCs and mortality among women with breast cancer have been inconsistent; a few reported lower mortality among OC users (6, 17), and some reported higher mortality (12, 13, 15, 16, 20, 21) or no effect (5, 7-11, 14, 18, 19, 22). Our null results for duration of use are consistent with most previous work (9, 14, 16, 18). Two studies observed that although recent OC users had a similar risk of death compared with never users, mortality decreased with increasing time since last OC use (therefore, recent OC users had increased mortality relative to non–recent users; refs. 16, 21). For example, the HR for more than 10 years since last OC use relative to nonusers was 0.85 (95% CI, 0.61-1.18), with a significant trend among users (21). Other studies have not observed associations between recent OC use and mortality (9, 14, 17, 18). Previous research was limited by the use of non-population–based samples and broad classification of exposure to OCs (6, 8, 10, 11, 13, 19). The earliest studies were limited by small sample sizes of women exposed to OCs (5, 6, 8, 11, 13).

In this study, time since first and last use was associated with mortality, although confidence intervals were wide; however, duration was not associated with mortality. This is similar to the results observed in the parent study for recency of use and risk of developing breast cancer (25). Despite the modest correlation between time since first and last use and total duration, duration and timing may represent two aspects of use. Whereas long duration of use can span different periods relative to a breast cancer diagnosis, it seems that only OC use close to diagnosis affects survival. A lack of observed trend in our results may result from small numbers of recent OC users, particularly in the 1- to 4-year range. Our results for time since first and last use do not seem to be explained by unusual patient characteristics. Recent OC users were more likely to be younger, be nulliparous or have fewer children, and were more likely to have had a recent birth and to have tumors that were discovered by self-examination rather than mammography (data not shown). Adjustment for parity, time since last birth, or mammography did not alter estimates for time since first and last OC use.

Higher mortality was associated with the progestin levonorgestrel and high-dose estrogen formulations. These results are noteworthy but should be interpreted cautiously because a small proportion of women used these formulations. Furthermore, pill types are not prescribed randomly, and no data were available on the reasons women used certain pill formulations. However, our results for high-dose estrogen pills mirror results observed in the parent case-control study for incidence (4). Mortality was higher among recent OC users, regardless of dose and formulation. For OCs taken in less recent periods (>5 to 10 years before diagnosis), content was not related to mortality, suggesting that only pills used recently in relation to diagnosis may affect mortality. Therefore, time since first and last use and pill content seem to be independent constructs of use. High-dose estrogen pills are not regularly dispensed currently, but many recently available pills contain levonorgestrel (31). Of OCs available in 2004, 21% contained levonorgestrel (32) versus 9% of pills used by women in this study. Levonorgestrel has higher potency and exhibits the highest level of androgenic activity relative to other progestins currently on the market (33).

Limitations include possible residual confounding by lifestyle changes after diagnosis. Although we did not collect postdiagnostic exposure and lifestyle data, OCs are contraindicated for women diagnosed with breast cancer (33). However, factors like income and obesity may have changed after diagnosis, but it remains uncertain whether postdiagnosis changes influence survival (34, 35).

Accurate recall of OC use in observational studies may also be problematic, but previous studies showed that recall is high when detailed questionnaires, reproductive calendars, and pill photographs are used (36, 37). Recall of specific brands is typically less accurate (36, 37), and in this study, 22.5% of OC users could not recall pill brand. Compared with participants who had complete content data, those with incomplete content data tended to be older (26% of those with complete data were ≥45 years versus 41% of those with incomplete data) and non–recent OC users (13.5% of those with complete data used OCs <1 year before diagnosis versus 1.5% of those with incomplete data). However, because women missing content data had similar mortality and income levels, bias was likely minimal, especially for recent pill use. Any bias was most likely nondifferential because completeness of the National Death Index is well validated (38). The accuracy of cause of death on death certificates remains equivocal (30, 39). We reached similar conclusions when using all-cause mortality (the primary end point of interest) or breast cancer–specific mortality.

Our results may not be generalizable to all breast cancer patients. Nonparticipation in the parent case-control study was often due to patient refusal or illness (25), so subsequent mortality was likely higher among nonresponders than among responders. Participation in the parent study was satisfactory (86%), although 83% of responders were ever OC users versus 76% of nonresponders (40). No data are available on the proportion of nonresponders who were recent users. It is unknown how inclusion of all potential participants in the study would have influenced the effect estimates. We had no follow-up information on the Seattle patients who participated in the original study. However, the parent study observed minimal differences in the OC results when stratifying by study site (25). In this present analysis, adjustment for study site did not affect the results. Therefore, it seems unlikely that the exclusion of the Seattle cases has materially biased the present results, although this cannot be completely ruled out.

Study strengths include use of data from a population-based study of young (<55 years of age) breast cancer patients with long-term follow-up and a comprehensive assessment of OC use. The focus on young women, many of reproductive age, is important because this population had ample opportunity for exposure to OCs. Due to comparatively recent diagnosis of breast cancer, exposure to OCs in this study population is more similar to that for current breast cancer survivors than in previous cohorts. To our knowledge, there is no published literature on pill attributes such as dose and potency in relation to breast cancer survival. The detailed nature of the OC data in this study allowed examination of these relationships.

Many women take OCs, often for years. Therefore, a greater understanding of their specific health effects is important. Mortality among OC users is lower than mortality associated with childbirth until age 40 for nonsmokers (33). Thus, comprehensive clinical recommendations about OC use at various ages would depend on a complex risk-benefit ratio.

In this population-based study of young breast cancer patients, most aspects of OC use were not associated with survival. However, there were provocative, but limited, indications that recent OC use may be associated with modest increased risk of mortality. The 2-fold increase in mortality associated with recent use of high-dose pills or levonorgestrel raises concern and deserves further investigation.