Methods to identify disease risk factors from a series of cases are considered. These include methods that compare risk factor levels among diagnostic categories and methods that relate risk factor levels to age at diagnosis, with a single diagnostic category. Statistical aspects considered include modelling assumptions, parameter identifiability, hypothesis-testing efficiency, assumptions concerning unsampled diagnostic categories and requirements for risk factor data and confounding factor data. It is argued that methods to identify risk factors using data on a single diagnostic category involve such strong assumptions that they have limited usefulness. Analyses that compare risk factor levels among diagnostic categories, on the other hand, should continue to play an important role in epidemiologic research, though there are important limitations in relation to analyses involving disease-free controls.

In this paper, situations in which individuals move through a finite set of states according to a continuous-time Markov process are considered. Only aggregate data are available: these consist of the number of individuals in each state at specified observation times. We develop conditional least squares and approximate maximum-likelihood-estimation procedures for time-homogeneous models, and extend the methods so that they can handle immigration of individuals into the system during observation. Asymptotic covariance estimates are presented, and some problems for future study are noted.

Female residents of six counties in Washington and Utah in whom a diagnosis of ovarian cancer was made during 1975-79 were interviewed concerning prior use of oral contraceptives. Interviews with a random sample of women drawn from these same counties were obtained for comparison. A smaller proportion of women with epithelial ovarian tumors had taken oral contraceptives than had controls: the estimated risk of ovarian cancer in uses relative to that in non-users was 0.57 (p = 0.04). The magnitude of the decreased risk depended on the duration over which oral contraceptives had been taken, as the negative association was present only after 4 or more years of use. In many ways, the use of oral contraceptives mimics the physiologic effects of pregnancy. The results of the present investigation, when interpreted together with those from other epidemiologic studies and from experiments in animals, suggest that, like pregnancy, the use of oral contraceptives has a protective effect against the development of ovarian cancer.