Mathematical modeling strategies for the analysis of epidemiologic research.

In the inaugural volume of the Annual Review of Public Health, Reuel Stallones suggested the following central axiom of epidemiology (27): "Disease does not distribute randomly in human populations." As a corollary to this axiom, Professor Stallones stated: "Variations in the frequency of human disease occur in response to variations in the intensity of exposure to etiologic agents or other more remote causes, or to variations in the susceptibility of individuals to the operation of these causes." The principal objective of many epidemiologic studies is to evaluate the association between exposure to a single risk factor and the occurrence of a specific disease. In this type of investigation, it is essential to isolate the effect of interest from the effects of other risk factors. The effects of relevant covariates may be controlled in the design or analysis of epidemiologic studies. Two approaches to the analytical control of covariates are stratification and mathematical modeling. The technique of stratified analysis was introduced in a seminal paper by Mantel & Haenszel (21). This procedure involves the categorization of the study population into discrete levels of the relevant