Beyond Ovulation: Oral Contraceptives and Epithelial Ovarian Cancer

Abstract

In a case-control study in three Australian states that included 794 women with epithelial ovarian cancer and 853 community controls for whom we had adequate contraceptive and reproductive histories, we examined the effects of oral contraceptive use after controlling for estimated number of ovulatory cycles. Other covariates included in the multiple logistic regression analysis were parity, smoking, and history of pelvic surgery. The protective effect of duration of oral contraceptive use appeared to be multiplicative, with a 7% decrease in relative risk per year [95% confidence interval (CI) = 4–9%], persisting beyond 15 years of exposure. Use for up to 1 year may have a greater effect than predicted (odds ratio = 0.57; 95% CI = 0.40–0.82), whereas use before the first pregnancy may be additionally beneficial (odds ratio = 0.95; 95% CI = 0.87–1.03, adjusted for overall duration of use). Better control for ovulatory life might attenuate these estimates somewhat. There was little evidence of waning protection with time since last exposure or of extra benefit with early commencement of oral contraceptive use. We found no convincing evidence of effect modification in any factor examined or differences in effect among the three main histologic cancer types or between borderline and malignant tumors. Oral contraceptives may act by both suppressing ovulation and altering the tumor-promoting milieu.

Exogenous female hormones, including oral contraceptives (OCs), are used regularly by a substantial fraction of women in the developed world at some time in their lives. Their side effects, beneficial or otherwise, are thus of major public health interest, and their association with gynecologic cancers in particular has occupied the attention of many epidemiologists over several decades. The effects of OC use on ovarian cancer are of particular interest, because the limited scope for early detection and consequent poor prognosis of this neoplasm make prevention especially important. Over and above the public health aspect, evaluation of such associations has the potential to throw light on the biology of carcinogenesis.

Not surprisingly then, this topic has been addressed in numerous research papers, reviews, and editorials. ^1–3 Most if not all studies have found OC use to be associated with a reduced risk of ovarian carcinoma. Because the great majority of OC preparations suppress ovulation, any protection might be thought, on the incessant ovulation hypothesis, ^4,5 to be ascribable solely or principally to this property.

To our knowledge, no authors have attempted to ascertain whether OC use has additional, independent effects on ovarian cancer risk by controlling in their analyses for length both of ovulatory life and of the other major suppressor of ovulation, pregnancy. We were able to do so in a large Australian case-control study of ovarian cancer in which fertility, contraceptive practices, and reproductive history were a central focus. ⁶ A particular strength of the setting is that in Australia, hormonal OCs were both readily available and widely prescribed from their earliest days. Thus, former users are to be found even among the oldest women in our study. We were also able to evaluate effect modification by a wide variety of factors, all of which have been investigated in one or more published papers, often in samples of modest size and with little biologic underpinning.

Subjects and Methods

We ascertained all histologically confirmed incident cases of primary epithelial ovarian cancer registered in all major gynecologic-oncology treatment centers in three Australian states. Women diagnosed in 1991 and 1992 were recruited in New South Wales and Victoria; in Queensland, where the cancer registry was an additional source, women diagnosed in these 2 years and also in 1993 and the latter part of 1990 were eligible for enrollment. A specialist gynecologic pathologist in each state conducted an independent histologic review of all specimens. A detailed description of the study has been published. ⁶ Briefly, after consent from the attending doctor, research assistants invited all eligible cases—those between ages 18 and 79 years and competent to complete a questionnaire—to participate either before discharge or while attending clinic for follow-up. We selected controls from the electoral roll by a random procedure designed to yield an age and regional distribution similar to that anticipated among the cases; enrollment to vote is compulsory in Australia. Women with a history of ovarian cancer or bilateral oophorectomy, or who were incapable of completing the questionnaire, were ineligible for inclusion in the control series.

Trained interviewers administered a standard questionnaire in person either in the clinic (cases) or in the woman’s home (some cases, all controls). Topics covered included demography, medical and surgical history, and family history of cancer. We collected details of each woman’s reproductive experience by means of a pregnancy and lactation record, including time to return of menses, and a month-by-month calendar dealing with contraception and attempts to become pregnant during the reproductive years. Information on OC composition proved impracticable to obtain. Marketing and prescribing of OC formulations in Australia followed closely those in the United States. ⁷ Subjects completed a separate food-frequency questionnaire ⁸ at their convenience.

Several women reported OC use before 1960, when the first formal marketing of an OC commenced in Australia. ⁷ Some may have been reporting use of progestogen preparations, the first of which appeared as early as 1939. Anecdotal information from sources we considered reliable suggested that hormonal preparations that suppressed ovulation were available in Australia from the beginning of 1957 onward. Regulation of pharmaceutical products was poorly organized in Australia before 1962. We therefore took January 1957 as the earliest acceptable date for OC use and ignored all prior reported use. Three cases and 15 controls were affected by this decision. Analyses with later starting dates (January 1958 and January 1959) gave virtually identical results.

Data Analysis

The principal aspect of OC use considered here is total duration of use. Preliminary analyses suggested a multiplicative effect on ovarian cancer relative risk; that is that the logarithm of the odds ratio (OR) was linear in duration of use. We tested this by including indicator variables for each year of total use up to 20 years in a multiple logistic model with the covariates described below, and fitting a straight line with zero intercept to the resulting coefficients by least squares, weighted by the relevant portion of the total estimated covariance matrix. For comparison, we fitted a fractional polynomial to the coefficients ⁹ (a cubic in the square root of duration) and a straight line in the logarithm of years of use plus 1.0. ¹⁰ The latter model has an element of arbitrariness in that abscissae for small to moderate durations are not invariant under a change of time unit from, say, months to years. The goodness-of-fit criterion was residual sum of squares.

Further measures of exposure we considered were duration of use before first pregnancy, among ever-pregnant women; age at commencement of use; and time since last use, the latter two analyses being restricted to ever-users of OCs. We adjusted the ORs for each measure for total duration of use and other covariates by conditional multiple logistic regression analysis. Duration of prepregnancy use was treated as a quantitative variable in the same way as total duration, on similar grounds. In the analysis of duration of OC use before the first pregnancy, we included also age at first pregnancy (nulligravid, <20, 20–24, 25–29, or ≥30 years); in the case of time since last use, age at last birth or pregnancy (<25, 25–29, 30–34, or ≥35 years) was substituted; for age at first use the model contained both age at first pregnancy and duration of prepregnancy OC exposure. Other covariates appearing in all models were age in years, average daily alcohol consumption (none, 0.1–5.0 gm, or >5.0 gm), smoking (current, past, or never), history of tubal ligation, prior hysterectomy, and years of ovulatory life (in decades).

Ovulatory life is a measure of a woman’s total number of ovulations. For each woman, we calculated her ovulatory life variable within each decade of life between age at menarche and age at menopause, ¹¹ diagnosis, or interview, as appropriate. From the length of each full or partial decade, we subtracted the total reported anovulatory period in that decade due to pregnancy; postpartum amenorrhea; OC use; and/or amenorrhea due to illness, underweight, or other causes. ¹² We then weighted the results from each decade by the intensity of ovulation in that decade of life applicable to this population ¹³ and summed. Finally, we adjusted this sum upward or downward depending on the extent to which each participant’s reported average menstrual interval was proportionately lower or higher, respectively, than a standard 4 weeks. To allay concerns over residual confounding of ovulatory life by age, ¹⁴ we used in most analyses a conditional multiple logistic regression model with strata of age, in single years, at the cost of a small loss in statistical efficiency. We express all findings in terms of estimated ORs, with their 95% confidence interval (CIs).

We considered the following factors as potential modifiers of the relation of duration of OC use to ovarian cancer risk: age, parity, menopausal status, education, breastfeeding history, alcohol consumption, family history of breast or ovarian cancer, history of smoking, tubal ligation, hysterectomy , regularity of menstrual periods, lactose intake, body mass index, and state of residence. We also assessed the effect of OC use on the three main histologic types (serous, mucinous, and clear cell/endometrioid) and the three classes (peritoneal, borderline, and malignant) of ovarian cancer, using the same pool of controls for each.

Results

The ascertainment procedures identified 1,116 women with epithelial ovarian cancer; 201 (18%) were ineligible on grounds of age, language problems, mental incapacity, or histology. Of the remaining 915, 824 (90%) were interviewed, 50 (5.5%) died too early, and 41 (4.5%) patients or doctors refused. ⁶ We excluded 28 known not to be on the electoral roll and 2 whose calendars contained insufficient information on contraceptive practices, leaving 794 cases for analysis. To form the control series, we chose a total of 1,527 names of women at random from the electoral rolls; 192 (13%) could not be traced, and 162 (11%) were ineligible or physically unable to participate. Among the remaining 1,173, 855 (73%) agreed to be interviewed ⁶; of these, 853 provided adequate contraceptive histories. Mean and median ages of included cases were 55.7 and 56 years, respectively, with a standard deviation of 13.4 years. Among included controls, the corresponding values were 54.8, 56, and 14.2 years.

We tested the assumption of log linearity of relative risk in duration of use by fitting various models to the logistic coefficients of single years of use. No alternative to log linearity was superior in terms of residual error.

Table 1 shows the case and control distributions of total duration of OC use, use before the first pregnancy, age at first use, and time since last use, as well as the percentage of ever-users and median duration of use among ever-users by decade of age. Women in the control series had used OCs more often and for longer than had women with ovarian cancer, overall and at all ages. Within each series, median duration of use differed little by age apart from women in their 30s. Prepregnancy use was somewhat higher among controls, but among ever-users, the differences between case and control women in the distributions of age at first use and time since last use are less striking.

Table 1: Distributions of Cases and Controls by Duration of Oral Contraceptive (OC) Use, Overall and Before the First Pregnancy, Age at First Use, and Time since Last Use, with Crude and Adjusted Odds Ratios (OR) and 95% Confidence Intervals (95% CI)

Table 1 also contains unadjusted and adjusted estimates of relative risk of epithelial ovarian cancer in each category of the above measures, with their 95% CIs. Adjustment was by multiple logistic regression controlling for age; alcohol consumption; ovulatory life; smoking; pelvic surgery; and, where appropriate, total duration of OC use and age at first or last birth or pregnancy. In addition, we give estimates of relative risk due to a year of total OC use and of use before first pregnancy, similarly adjusted. According to these results, each year of OC use reduces the relative risk of epithelial ovarian cancer by some 7% (95% CI = 4–9%). The decline in risk appears to persist beyond 15 years of exposure (OR = 0.25; 95% CI = 0.13–0.49). There are indications that even short-term use—up to 1 year—has a protective effect of similar magnitude to use of from 1 to 5 years [OR = 0.57 (95% CIs = 0.40–0.82); OR = 0.73 (95% CI = 0.52–1.03), respectively]. ¹⁵ Use before the first pregnancy may be protective over and above the effect of total duration of use (OR = 0.95; 95% CI = 0.87–1.03). There are no obvious trends in risk with increasing time since last use.

Estimated relative risks of the three main histologic types of epithelial ovarian cancer per year of OC use are very similar; the same is true for borderline and malignant tumors, but not for peritoneal cancers (Table 2). We present results in Table 2 only for those potential modifying factors with an OR per year of OC exposure lying outside the CI for total duration in the entire sample (0.91–0.96) in at least one stratum. There is, however, no implication that the observed heterogeneity is beyond what might be expected by chance. Other factors examined, namely age (<40, 40–59, or ≥60 years), menopausal status, education (secondary vs postsecondary), average daily alcohol consumption (none, 0.1–5.0 gm, or >5.0 gm), smoking history (never, current, previous, or additionally controlled for alcohol consumption), history of tubal ligation, regularity of menstrual periods, and state of residence did not meet this criterion. ORs for OC use in strata of lactose intake divided at the 15th, 35th, 65th, and 85th sample percentiles had a wider range, but without trend (data not shown).

Table 2: Numbers of Cases and Controls within Strata of Certain Potential Modifying Factors, with Odds Ratios (OR) for Ovarian Cancer per Year of Total Oral Contraceptive Use by Histological Type and Cancer Class, and 95% Confidence Intervals (CI)

Discussion

This study estimates that after controlling for number of lifetime ovulations, a year of OC use reduces ovarian cancer relative risk by 7%, and by 9% in the absence of such control. This latter figure is within the range of values from ten case-control studies of ovarian cancer in relation to duration of OC use ³; none of these studies adjusted for number of lifetime ovulations.

Imprecision in the calculated number of lifetime ovulations could have led to a smaller than appropriate adjustment. We have explored this possibility approximately in a simple mathematical model based on linear regression, in which misclassification in the ovulatory life variable was modeled as an additive normal variable with variance proportional to the “true” value. Consideration of the proportion of observed values that depart from the actual value by at least 50%, as an example, suggests that our estimate of 7% annual decrease in relative risk might be too large by (at most) one or two percentage points.

OCs appear, therefore, to have a protective influence on epithelial ovarian cancer beyond their anovulatory action. Little waning of effect with time since last use was apparent. We also found suggestions that use before first pregnancy was additionally protective. No convincing evidence of important effect modification came to light. Whereas we were unable to collect usable information on dosage or composition, Ness et al¹⁶ found low-dose OCs to be no less protective against ovarian cancer than the older higher-dose formulations.

Issues of bias have been addressed in some detail elsewhere. ^6,9 Selection bias is not likely to be strongly present. Ascertainment of cases was virtually complete in Queensland and largely so in Victoria; in New South Wales, cases were ascertained essentially on the basis of their choice of surgeon, which is unlikely to be related directly to OC usage. Refusals among the women with ovarian cancer were infrequent and usually related to stage of illness. Nonresponse bias due to refusals from women selected as controls is possible if women who declined to participate or could not be contacted differed markedly in their contraceptive experiences from those who were successfully interviewed. Compared with participants in a national health survey in 1989–1990, ¹⁷ women more than 39 years of age in our control series had a slightly higher rate of current OC use. Our exclusion criteria—bilateral oophorectomy and inability to complete a questionnaire in English—would probably be sufficient to account for the discrepancy.

Recall or interviewer bias is also unlikely. The association between ovarian cancer and OC use has not been widely reported outside the epidemiologic literature. Moreover, information on contraception was collected by means of a rigidly structured monthly calendar as only one aspect of a detailed interview schedule. ⁶ Imprecision of recall, particularly at older ages, could act to reduce the strength of associations, although no weakening of effect was apparent among older women. We controlled for confounding by the variables of which the omission altered the OR to a noticeable degree. In particular, the adjustment for number of lifetime ovulations was by conditional estimation within yearly strata of age, with control for ovulatory life in the model. The converse approach, with yearly strata of ovulatory life and close control of age in the model, gave almost identical results. No evidence of uncontrolled confounding came to light in our examination of many potential modifying factors. Confounding by unsuspected factors, however, is always a possibility.

The notion that suppression of ovulation is not the sole route by which OCs protect against epithelial ovarian cancer is implicit in the etiologic hypothesis proposed by Cramer and Welch in 1983. ¹⁸ They suggested that high gonadotropin levels, by acting on steroid-producing stroma of the ovary, stimulate ovarian epithelial cells and promote neoplastic changes, especially in entrapped epithelium (inclusion cysts). OCs are known to suppress gonadotropin production. ² Furthermore, although ovulation can produce inclusion cysts, it is not the sole mechanism. ¹⁹ More recently, Risch, ²⁰ after a detailed review of the epidemiologic evidence, explicitly concluded that repeated ovulation cannot be the sole risk factor for epithelial ovarian cancer and postulates that hormonal factors are equally important. In particular, he postulates that high androgen levels and/or low progesterone levels may increase ovarian cancer risk. Again, OCs suppress the former, ²⁰ and all OC formulations contain progestogens. These effects could explain, in part, the relatively strong protective effect of short-term OC use we observed in that altering the tumor-stimulating hormonal milieu shortly after the initiation of neoplastic changes may significantly delay or inhibit cancer development.

Acknowledgment

The contribution made by the other members of the Survey of Women’s Health Study Group has been acknowledged previously. ^6,11