Epidemiology: Information Needs and Data Gaps
With limited observational data about the influence of widespread use of oral contraceptives on breast cancer incidence and mortality, almost the entire body of scientific information about the influence of oral contraceptive use on the relative risk of breast cancer can be found in about a dozen research reports. These studies were all carried out in developed nations and are either population-based, hospital-based case-control, or cohort designs (see Appendix B). There are also a few case-control studies from developing nations. The present system of small, ad hoc studies is unsatisfactory, and there is a need for more, long-term, systematically coordinated epidemiological studies with sufficient statistical power.
Epidemiology is the scientific study of the distribution and determinants of disease in populations. As a science of public health it employs research methods derived from many other fields such as survey research, biology, toxicology, and statistics. It attempts to integrate the available data so as to answer important questions about health. Although the experimental method can sometimes be employed, observational methods are often the only possible research strategy, especially when adverse effects are under investigation. The principal methods used by epidemiologists are the case-control (retrospective) method, the cohort (prospective) method, and ecological
methods in which trends of exposure and disease in populations are correlated. The application of these methods in efforts to shed light on the question of the possible association of oral contraceptives and breast cancer is discussed in the following sections.
The case-control method compares the previous use of birth control methods by women with breast cancer and a comparison group. This method is the most efficient in terms of both time and cost, but it is limited in the number of adverse or beneficial effects it can measure. One of its greatest disadvantages is the poor documentation of prescriptions for oral contraceptives and the likelihood that women who do not use the pill have other characteristics or conditions that make them inappropriate for purposes of comparison.
The number of cases of breast cancer and of controls needed to obtain statistically sound results depends on the prevalence of exposure to oral contraceptives and not on the rarity of the disease. It is not surprising, therefore, that much of what is known about oral contraceptives comes from case-control studies. Given the disadvantages of the case-control method, it is also not surprising that there is little or no scientific information about the results of long-term exposure.
Case-control studies have been proven reliable guides when the detected relative risks are moderate or large, but a problem emerges when the relative risks are small or marginal. In these cases, the possibility that bias of some sort may have led to the finding of increased risk is a more tenable alternative hypothesis than when the relative risks are large.
In the matter of the relation of oral contraceptives to breast cancer the relative risks have been either modest or small, and they have not been found consistently. This situation has created uncertainty about the meaning of some of the “positive” studies. One approach to sorting out inconsistent replications of small risk measurements is to initiate a large, case-control study with sufficient statistical power to resolve the matter. The National Cancer Institute's new, large (2,000 cases), case-control study is just such an effort (Brinton, 1990). In addition to this large study, it is highly likely that future case-control studies will be required as oral contraceptive formulations and use patterns change.
Seven strategies are likely to improve future case-control studies:
standardized history-taking using pictures of the oral contraceptive formulations so that analysis by formulation is possible;
attempts to validate complete contraceptive history;
information-gathering about histopathology of tumors, if possible;
use of biological markers (when and if these become available) and other laboratory methods to help categorize risk group;
studies in populations that use oral contraceptives but that have much lower incidence rates of breast cancer than most Western countries;
regular studies that employ comparable methods of information-gathering (because of the problem of changing formulations, changing use patterns, and long length of time from tumor induction or promotion to tumor diagnosis); and
use of information about screening and earlier diagnosis to correct for possible “early diagnostic bias” arising from selective screening exams (and more sensitive screening devices) applied to oral contraceptive users now that the possible link between oral contraceptives and cancer has been widely publicized.
There have been few cohort investigations of the relationship between oral contraceptives and breast cancer because of the large number of women who must be followed for so many years. Even when such longitudinal studies are possible, their statistical power is unlikely to be large; thus, the hope of identifying small risks is quite low. Nevertheless, the cohort design has the advantages of allowing better evaluation of the role of bias and easier validation of exposure histories. The use of telephone and mail to follow up large cohorts has been shown to be practical in the Nurses Health Studies I and II in Boston (Romieu et al., 1990), the Family Planning Study at Oxford (McPherson, 1990; Vessey, 1990), and the cohort assembled by the Royal College of General Practitioners (Kay, 1990).
Other opportunities for cohort-type research, either prospective or “historical,” arise through prepaid health plans and data linkage systems that include information about drug prescription and cancer incidence. Examples include the Saskatchewan data base in Canada, the population-based data systems of the Uppsala region of Sweden, and large, closed health maintenance organizations (HMOs) in the United States (e.g., Kaiser Permanente in California; Group Health Cooperative of Puget Sound).
A special category of cohort studies is postmarketing surveillance following the introduction of new contraceptives. This surveillance should be required for oral contraceptives that contain new hormonal
constituents or for those with significant alterations in the existing formulations. The plan for such an effort must be based on sound epidemiological principles and would bear little resemblance to current postmarketing surveillance, which often amounts to little more than sporadic reporting of adverse drug reactions or company-conducted marketing studies.
Case-control and cohort studies share a significant methodological problem: uncertainty as to how to choose the comparison group. The usual strategy is to compare disease occurrence among the exposed population relative to the nonexposed population. With oral contraceptives, the issue is complicated because the majority of women in this country have used the pill; those who have not may not be representative of the general population of women (for additional discussion, see Appendix A). Furthermore, the ways in which non-users are unrepresentative will vary, so that the direction of the bias is not easily known in advance. Nonusers may include women of both high and low social class, health-conscious women, women with medically diagnosed illnesses, women who abstain from sexual activity, infertile women, and women with family histories of cancer and cardiovascular disease. These factors and others, such as frequency of mammography, may relate to risk of breast cancer and rates of detection; consequently, their variable distribution in non-oral-contraceptive-using women could seriously distort the estimates of effect—that is, breast cancer risk estimates—if nonusers are the comparison group.
Short-term users of oral contraceptives may also be a poor choice as a comparison group because they probably include a large segment of women who for medical reasons or perceived symptoms discontinued oral contraceptive use soon after starting it. In sum, there is not an obvious solution to the problem of choice of a comparison group in case-control and cohort studies.
Ecological studies correlating trends in breast cancer incidence rates with trends in oral contraceptive use have not been informative in the past and are unlikely to be so in the future, except possibly in developing countries where the underlying breast cancer rates are low. (On the other hand, in developing countries the cancer information systems are unreliable, so useful data generally will be lacking.)
The reason for the probable lack of utility of this method is that, in developed countries, it is unlikely that any influence on breast cancer rates would be large and clear enough to be distinguished from other influences. This indistinguishability is in contrast to the situation of pulmonary embolism in young women, for which the effect of oral contraceptives was detected. In that instance, the risk was fairly large, immediate, and led to sudden death, and the underlying rates were low. Cancer of the endometrium was another such case: there, the widespread use of replacement hormones was age-specific and increased rapidly in a short time.
Women Exposed to Hormones Other than Oral Contraceptives
Opportunities exist to study women who have been heavily exposed to hormones. Some are endogenously exposed (e.g., women with polycystic ovaries), and others are exogenously exposed (e.g., women with Turner 's syndrome who have been treated with steroid hormones, girls given estrogens to stop their long bone growth, or women given diethylstilbestrol (DES) during pregnancy). In the case of DES, both the exposed women and their progeny, male and female, can be followed. Such studies can employ traditional epidemiological designs, comparing exposed subjects with nonexposed subjects for subsequent risk of breast cancer. Controls can be internal (e.g., pregnant women who were not exposed to DES from the same time period and hospital as those who were exposed), or external, which is often a statistical estimate based on the breast cancer experience of a large population. In these heavily exposed populations, investigators can evaluate the direct effects of particular hormones on breast cancer risk, but the issue of inferring that these results pertain to oral contraceptive use remains.
Women Genetically Susceptible to Breast Cancer
Are there differences in susceptibility among oral contraceptive users who do and do not get breast cancer? Or, alternatively, among women with a high risk for breast cancer, does oral contraceptive use further increase that risk?
These questions arise from an inference that seems logical, given the existing data. Breast cancer occurrence before the age of 40 or even 45 is a rare event. Only 15 percent of all breast cancers occur before age 45. Let us suppose that there is a small increase in the
frequency of these cancers in association with long-term oral contraceptive use, as some studies have suggested. This small increase is occurring in conjunction with a high use prevalence of oral contraceptives, 50 to 70 percent of all women currently less than age 45, the exact percentage depending on the exact age. Given a small increase in the occurrence of a rare event resulting from an extremely common exposure, what is the likelihood that the excess events are randomly distributed in the exposed population versus the likelihood that only a small segment of the exposed population is at high risk of incurring breast cancer at an early age? Irrespective of oral contraceptive use, the occurrence of breast cancer at young ages is more likely to have a strong family history of the disease, to be present in breast cancer families (i.e., pedigrees), and to be associated with known genetic syndromes (e.g., ataxia telangiectasia—a DNA repair deficiency). Thus, genetic susceptibility factors are thought to be much more strongly associated with early-onset than with late-onset disease. The possibility of an interaction between oral contraceptive use and other biological phenomena, which are most likely genetically determined, necessitates some additional types of studies.
To characterize genetic susceptibility and study its interaction with oral contraceptives will require specialized populations and innovative methodologies. For example, one might identify patients and pedigrees with genetic syndromes that are known to confer a high risk of breast cancer, such as a DNA repair deficiency, or heterozygotes and ask: Does oral contraceptive use increase this risk further? What is the breast cancer risk in oral contraceptive users versus nonusers in families with known DNA repair deficiencies (e.g., ataxia telangiectasia families)? To the extent that assays are available to detect molecular alterations that increase breast cancer risk or closely linked molecular markers, four groups of women could be formed and followed for breast cancer occurrence: women with and without the marker, divided into oral contraceptive users and nonusers. If the marker assays are complex, limiting the number that can be performed, studies of oral contraceptive users only would be informative.
It is important to recognize that the current capability of molecular technology may not be adequate to characterize susceptibility (see Chapter 3). In that event, other options are needed for characterizing susceptible women. High-risk phenotypes may be characterized using more traditional methods. As an example, the ongoing National Cancer Institute study of breast cancer and oral contraceptives in women under 45 years of age could serve as a source of subjects. One could identify very young cases and controls with one or two first-degree relatives with breast cancer diagnosed at a young age (specific criteria to be
defined). If oral contraceptives interact with susceptibility, the increased risk should be observable in this restricted population.
An alternative strategy derived from a large case-control study might use the same criteria for identifying susceptible cases but use phenotypically normal sisters as controls. Ideally, a genetic analysis (e.g., restriction fragment length polymorphism) of both case and control subjects would be performed to detect specific molecular alterations. In addition, or alternatively, linkage analyses could be done.
Molecular markers for mutational effects (i.e., oncogene activation), loss of heterozygosity, or polymorphisms (i.e., alleles) can be sought, given the availability of appropriate assays. One such study of traditional case-control design is currently in progress, funded by the National Cancer Institute. Rare alleles of Ha-ras are being sought as susceptibility markers with oral contraceptives being considered as effect modifiers (Garrett and Hulka, personal communication).
One innovative strategy (Swift et al., 1990) tests hypothesized associations between a candidate allele, for which there is a specific laboratory test, and a common chronic disease, such as breast cancer. Families in which this allele is segregating are identified through index individuals who are heterozygous or homozygous for the allele. One relative with the disease of interest (e.g., breast cancer) must be available for each index case. The proportion of heterozygotes observed in the diseased sample is compared with the expected proportion, based on each diseased relative's null probability. The advantage of this strategy over a more traditional case-control approach is the reduction in sample size required to test the hypothesis of a specific allele rendering susceptibility to breast cancer.
To enhance the capability of molecular epidemiological research, the molecular characterization of breast cancer in oral contraceptive users and nonusers is needed. If polymerase chain reaction can be used and restriction fragment length polymorphisms can be studied using formalin-fixed tissues, existing repositories of tumor samples (i.e., clinical pathology laboratories) could provide expanded opportunities for these studies (Frye et al., 1989; Resnick et al., 1990). Whatever study design and epidemiological strategy are used, every attempt should be made to observe the influence of oral contraceptive use on specific types of malignant breast tumors, using the best generally available technology.
Estrogen metabolism may figure prominently in future epidemiological research. Fishman and colleagues (1984) have proposed that breast cancer patients, in comparison with normal controls, have higher
levels of the 16α-hydroxyestrone metabolite and lower levels of the 2-OH metabolite of estrone. These results need to be replicated in larger studies and the effect of oral contraceptive use on the levels of these metabolites explored. An alternative to the existing radiometric assay is required to evaluate these estrogen metabolites in epidemiological studies using blood or urine samples as the biological medium. Furthermore, it has been stated that the 16α-hydroxyestrone metabolite can form permanent, covalent adducts with the E2 receptor, turning on the receptor indefinitely. This metabolite is also said to form adducts with albumin and hemoglobin. Because a variety of assays for adduct formation are available, additional clarification of these issues would be useful for future epidemiological research. For example, levels of estrogen-metabolite adduct formation in breast tumor and normal tissue from oral contraceptive-using and non-oral contraceptive-using patients could provide suggestive information on the carcinogenic potential of these metabolites.
Further studies are needed of the biological effects of the progestin component of oral contraceptives and its interaction with the estrogen component. This effort should be tackled from an interdisciplinary perspective using more combined, in vivo/in vitro studies of endogenous and exogenous hormone effects on human breast epithelial cells. Through use of the thymidine labeling index (TLI), epithelial proliferation of normal breast lobules has been shown to change over the menstrual cycle (Anderson et al., 1990). Proliferation is increased in the second half of the monthly cycle—when progesterone levels are elevated. Furthermore, breast epithelium from young women was shown to be more responsive than that of older women. Nulliparous oral contraceptive users had a significantly greater increase in TLI than their parous counterparts during the last week of their “cycle” (i.e., the equivalent of the late secretory phase of a normal cycle). High-dose estrogen oral contraceptives may cause more proliferation than oral contraceptives with lower estrogen dosages, but there were no significant differences among progestins. This avenue of investigation needs to be extended more broadly in the future to further document the differences between parous and nulliparous women, and to include more data on a variety of oral contraceptive formulations. Such studies should be designed to gather additional data such as blood and tissue levels of endogenous or exogenous hormones (at the time of diagnostic biopsy) that are critical to the interpretation of these findings.
Changing Oral Contraceptive Formulations
In recent reviews of the existing case-control and cohort studies of the influence of oral contraceptives on the relative risk of breast cancer, a great deal of concern has been expressed about the constraints placed on investigators by changes in oral contraceptive formulations and inconsistent methods of recording ages and patterns of use. (The evolving formulations of oral contraceptives are described in Appendix C.) These concerns have centered on oral contraceptives themselves; investigators now need to ask whether subsequent hormone replacement therapy with differing duration of use and differing formulations (in the presence or absence of oophorectomy) modifies the long-term oral contraceptive effects. Attention must be paid to the sample sizes needed in cohort or case-control studies to measure a specified effect against differential background trends and sequential confounding or modifying exposures.
SUMMARY AND CONCLUSIONS
There is a continuing need for well-designed observational epidemiological studies of the relationship of oral contraceptives to breast cancer. As new formulations are introduced, continued postmarketing surveillance must be carried out; this surveillance is particularly important because of emerging patterns of long-term use and increasing use of hormone replacement therapy during menopause. It is possible that newer techniques derived from the explosion of knowledge in molecular biology may become useful in epidemiological investigations.
The lack of precision surrounding small risks can be overcome by large-scale studies. The possibly long latency periods can be addressed by following cohorts for substantial periods of time and periodically repeating case-control studies. In addition, new cohort studies must be initiated to gather information about new formulations and use patterns.
The problem of the relationship between oral contraceptives and breast cancer illustrates both the complexity of biological interrelationships and the difficulties inherent in monitoring long-term exposures of human populations in a modern, mobile society that can quickly change its contraceptive practices and patterns of use of exogenous hormones.
Current knowledge of the putative relationship comes from complementary studies of exposures in Northern Europe, the United States, and developing countries. Future studies should continue to capital-
ize on these complementary opportunities (e.g., the advanced use of new formulations in Europe and variations in background breast cancer incidence rates). The potential to maximize future knowledge by planned coordination of international research should be explored most immediately in the United Kingdom and the United States.
The cost of long-term prospective studies is driven by the number of people who must be tracked and the number of years tracking must be maintained. For the purposes of these studies, breast cancer is relatively uncommon; thus, large numbers of women must be followed —making studies of breast-cancer incidence unavoidably expensive. Research costs can be contained by careful selection of study populations and sources of information about outcome measures. National, regional, or institutionally defined populations with linked record systems offer the best opportunity for relatively low-cost collection of information about exposure and, in some cases, outcomes. Population-based cancer registries such as the Surveillance, Epidemiology, and End Results (SEER) program provide an important resource for identification of cancer cases. It may be more cost-efficient to provide support to maintain the integrity of these sources of subjects, information, and events than to establish entirely new, dedicated recruitment, collection, and follow-up systems. In the United States, particular consideration should be given to the populations of large, comprehensive, closed health maintenance organizations within SEER collection areas.