J Chron Dis Vol. 32. pp. 83 to 88 Perpamon Press Ltd 1979. Printed in Great

Britain

CASE-CONTROL STUDIES SEX STEROIDS ON WOMEN DWIGHT

ON THE EFFECT OF AND THEIR OFFSPRING

T. JANERICH,* DONNA GLEBATIS,* ELLEN FLINK*

and

MARGARET B. HOFF*

BACKGROUND

presentation will be directed to a general discussion of the case-control method for studies of birth defects and breast tumors in relation to use of female sex hormones. Sex steroids are being prescribed to sick women and healthy women as well as young women and middle-aged women. Indirectly, children are being exposed to these agents prior to birth by maternal use during pregnancy. In order to evaluate claims of putative risks in these women and their offspring, a cost-and-time efficient study method is needed. The case-control method seems perfectly suited to fill this need. We use the term case history or case-control for any investigation which seeks to compare exposure frequencies between diseased and nondiseased groups, and the term cohort for any investigation of disease frequencies among population subsets with and without some putative environmental exposure or other risk factors. Individual study designs will occasionally fall outside this simple dichotomy, but in general, this classification system is comprehensive and useful. The term exposure is used in an identical way for either case-control or cohort studies, and can refer to any characteristic, including; possession of a gene, use of a drug, or being nulliparous as opposed to parous. Ascertainment of conditions of exposure in a cohort study are often thought of as unbiased because these conditions are usually ascertained without prior knowledge of whether the person eventually will develop the disease. The direct measure of risk and the opportunity to minimize bias in exposure data are the foremost advantages of the cohort method. The time sequence of the cohort study is also closely related to the way causual inferences are made in everyday experience, at least in the sense that a cause is observed before its effect is known. Case history studies, on the other hand, have two important handicaps; first, relative risk is expressed in terms of exposure among the diseased-just opposite to the usual direction of reasoning to a causal inference; and second, exposure is ascertained after the disease has occurred-which can increase the possibility of obtaining biased exposure information. Unfortunately, these inherent handicaps of the case-control study have led to an all too general insinuation that cohort studies are legitimate studies, while case history studies are something less than legitimate. This attitude is unwarranted by the facts. The first of the two handicaps inherent in the case-control method is conceptual and has been examined and successfully resolved by a number of investigators [l, 21. Unfortunately, the second handicap of case history studies cannot be so readily resolved. In order to eliminate the problem of bias each study must be designed to assure procedural excellence and absolute equality in the ascertainment of exposure in cases and controls. Careful evaluation of all sources of bias and their implications for the study results is an integral part of the case-control study procedure. Nevertheless, since it THIS

*Cancer

Controi

Bureau.

N.Y.S. Department

of Health. 83

Albany.

NY 12237. U.S.A.

84

DWIGHT T. JANERICH ef d.

is nearly impossible to prove that something does not exist, the assertion of bias is difficult to dismiss. The possibility of bias in exposure information will almost always be an Achilles’ heel of the case history study. Despite denigrating criticism by many apparent opponents, it seems reasonable to suggest that the cost efficiency which is characteristic of the case history study will cause the case history study method to become the basic tool of epidemiologic investigation in the future, if it has not already achieved that status. Case-control studies are the strongest analytic epidemiological tools for the study of causal mechanisms of diseases that are rare, or that have long latency periods. In studies of rare diseases their strength lies in their ability to allow the investigator to concentrate his efforts to obtain information on diseased individuals without needing to consume research resources in studying large numbers of nondiseased individuals. Teratogens always have latency periods that are short, nine months or less, but, the case-control method has an advantage over the cohort method in studies of putative teratogenic drugs, because the background rate of disease is low and the risk from exposure is usually small. Case-control studies also have an advantage over cohort studies for investigations of potentially carcinogenic agents even with a disease as common as breast cancer, because the latency period is long. Birth defects have lifetime rates in the range of h-f of 1%. Breast cancer, on the other hand, has an eventual incidence rate of approximately 5% in the United States. A causal relationship between drug use and a disease as common as breast cancer would be easily recognized by a casual clinical observer provided the latency period was not long or the risk from drug use was not small. Iatrogenic disease is also more likely to be recognized by a casual clinical observer if the disease caused by the drug is unique, or rare. However, in most instances, this is not the case, and the iatrogenic disease will closely resemble its non-iatrogenic counterpart. When actually facing the task of conducting a case-control study, the most immediate problem the investigator encounters is selection of the case and control popuiation. Control selection often receives the lion’s share of attention in design of case-control studies. MacMahon and Pugh [3] expressed the view that the weighing of the various factors in control selection for case history studies represents one of the most difficult processes in epidemiologic research. Speaking of the various steps involved in control selection, Cochran [4] stated, “ . . . the selection of the control population is a more crucial step than the selection of a method of matching or the covariables on which to match.” Thus, although the cases have the condition of concern, it is the controls which constitute the key methodological part of the case-control method. As it most often happens, however, an investigator must choose, not the ideal control group, but the best control group available. Still another important matter of control selection is the information which an investigator uses to design control selection procedures. When an investigator struggles with the problem of which controls to select, he considers all known epidemiological facts about the disease and how these might influence exposure. These facts form the basis of the eventual control selection procedure. Once the study starts, these cannot be changed. If a study is properly done, the investigator will always learn something new about the disease, and will most likely learn something new about the relationship between the disease and the exposure under study. Because this new information can never be available until the study data has been analyzed, it is never available when the decisions on control selection are made. Therefore, control selection procedures can only be an approximation of the ideal, and, although control selection is one of the most important aspects of the case history studies, whatever process is finally used will always be judged as less than ‘ideal’ by the time the study is reported. The ‘perfect’ control group for a case history study will never exist, except as an intellectual construct. because the results of a study will always provide new information that can improve control selection procedures.

Effect of Sex Steroids on Women and their Offspring BREAST

DISEASE

AND

ORAL

85

CONTRACEPTIVES

One of the most important current issues in epidemiological studies on drug safety is the concern that oral contraceptives may cause breast cancer. In early 1977 we started a case-control study to investigate the history of oral contraceptive use among young women with breast cancer. Our data gathering is complete, but data analysis is at a preliminary stage and not yet ready for presentation. However, a discussion of our study design may be useful as an example of the problems and concerns that we had to deal with in control selection. The case and control selection procedure that we used reflects our location in a state health department. We attempted to tailor our design to the types of data that are readily available to us, namely cancer registry data and vital records data. We felt that two specific characteristics about oral contraceptive use and breast cancer risk were especially important because they are potential confounding factors in case history studies. First, oral contraceptive use increased dramatically in the mid-19603 following low introductory levels of use in the first part of that decade. Utilization rates for oral contraceptives grew unequally in each age group. These rapidly changing utilization rates could be important confounders in case-control studies unless the controls were related to cases in a way which allowed precisely similar likelihoods that a case or control would choose to use the pill in a given calendar year. Second, we were concerned about the well-known protection against breast cancer which is afforded by an early first pregnancy. Specifically, we wanted to know whether young women ( < 45) experienced pregnancy related protection. Together, these factors could create a potential confounding relationship in case-control comparisons. Let’s say, for example, that two women (one who will eventually develop breast cancer by 35 yr of age, and one who will not) were each born in 1940, each got married in 1965, and each had two children between 1965 and 1975. If the early pregnancy-related risk factor worked as hypothesized, the woman who would eventually get breast cancer is more likely to have had her first child later in the 1965-1975 period than the woman who would not get breast cancer. Although the woman who is destined to get breast cancer was born in the same year, and married in the same year as her noncancer counterpart, she is likely to have had her first child later. This delay is important because it increases the likelihood that the eventual breast cancer patient would have used the pill or some other contraceptive before getting breast cancer. In this case the risk factor of late first pregnancy among women with breast cancer is a confounding factor. Since this delay in childbirth would be occurring at a time when oral contraceptive use was increasing rapidly (1965-1969), we thought it important to assure the comparability of cases and controls with regard to the age at diagnosis of case and the calendar year of birth of the first child. We had the opportunity to deal with the problem by matching and used it in our study design. We also considered several other possible potential confounding issues. The first possible problem is benign breast disease. Some breast disease lesions are precancerous, or at least they increase the risk of breast cancer [6]. Several studies have shown that women who use the pill have less benign breast disease than women who are non-users. However, it is not clear whether the pill actually prevents, or merely masks, the development of benign breast tumours. The issue is also complicated by the fact that some benign breast tumours cari be common and recurring. In addition, many physicians consider benign breast tumors as contraindications for prescribing the pill [7], which produces a bias which is difficult, but not impossible, to deal with. It would have been difficult for us to match cases and controls on benign breast disease, so we elected to deal with the influence of benign breast disease in the analysis of the data. Our final major consideration was ‘the descriptive epidemiology, or rather the lack of descriptive epidemiology of breast cancer in young women. At present, any study of oral contraceptive use and breast cancer will necessarily be a study of breast cancer in young women. Age at first pregnancy, for example, is a very important risk factor for breast cancer. Although there has been substantial effort to do the descriptive epi-

86

DWIGHT

T.

JANERICH

Yr al.

demiology of breast cancer, we were aware of no reported study which concentrated on the breast cancer in younger women. Therefore, it was unknown whether early pregnancy is protective in young women, as it is in older women. To clarify this point, and as a preliminary step for our case-control studies of oral contraceptives and breast cancer, we conducted a separate case-control investigation, to determine whether early pregnancy does protect against early breast cancer. The new information was useful in two ways. First, it confirmed the importance of considering age at first pregnancy as a risk factor in young women. Second, the study suggested that the sex of the first child was a previously unsuspected risk factor for breast cancer in young women. Since mother’s age at the time of birth does have a slight effect on the sex ratio of the offspring [8] we took steps to control this factor in control selection and in the analysis of our data. The data will be presented in detail at another meeting but we will briefly describe the results of our preliminary study on breast cancer. TABLE 1. CASE-CONTROL COMPARRON PREGNANCY

Age at diagnosis (Y0

BY AGE

OF

MEAN

Case Mean/SD/No.

Control (means)

24.50 4.49 163 22.49 2.85 154

23.18 4.77 326 21.73 3.23 308

35-44

Case-control studies on the effect of sex steroids on women and their offspring.

J Chron Dis Vol. 32. pp. 83 to 88 Perpamon Press Ltd 1979. Printed in Great Britain CASE-CONTROL STUDIES SEX STEROIDS ON WOMEN DWIGHT ON THE EFFECT...
627KB Sizes 0 Downloads 0 Views