New Techniques for the Analysis of Cohort Studies

A review of the basic approaches to the analysis of the different types of cohort data, which focuses on empirical and mechanistic approaches to model specification, and some special problems, such as measurement error, dependent outcomes, and the unique problems of reproductive data, are addressed.

Abstract

Cohort studies involve the key element of follow-up of individuals over time to study an outcome in relation to some earlier exposure factor or a fixed host characteristic (such as genotype). While the outcome under study could be the change in some continuous variable, we shall restrict this review to studies of disease incidence. Subjects can be randomly assigned to different exposures, as in a clinical or prevention trial, or the exposure histories of free-living individuals can be passively observed, as in most epidemiologic cohort studies. Although the issues of confounding and comparability are very different in randomized and observational studies, the basic analysis methods are similar (except, perhaps, for a greater emphasis on adjustment for covariates in observational studies) and the distinction will be ignored. Similarly, follow-up can conducted prospectively or retrospectively in real time, but this too has no significance for methods of analysis. We begin with a brief description of several cohort studies with different types of data structures and different analysis problems that will be used to illustrate the statistical issues. Following a review of the basic approaches to the analysis of the different types of cohort data, we focus on empirical and mechanistic approaches to model specification. Some special problems, such as measurement error, dependent outcomes, and the unique problems of reproductive data, are addressed. We conclude with a more in-depth treatment of approaches to the analysis based on cohort sampling methods—the nested case-control and case-cohort designs, and variants thereof.