The development of the methods of cancer control epidemiology has been focused in three areas: 1) studies in causal and preventive inference, 2) studies in causal, preventive and 'mixed' interactions, and 3) studies in selection bias. One inferential issue being examined is the use of inductive and deductive logic in epidemiologic explanation. When four distinct properties of causal and preventive explanations are considered -- their origin, consistency, testability, and permanence -- deductive logic proves superior except for their origin which is not a logical process. The implications for cancer control epidemiology are important: competing hypotheses should be proposed before any study begins; and the choice of the best hypothesis is equivalent to that which has been most rigorously tested. Another inferential issue is that of the relationship of the current causal criteria to explanatory progress and to public health decisions. For purposes of explanation, two categories of criteria emerge: those dependent upon the form of the hypothesis being tested and those independent of it. Studies of causal, preventive and mixed interactions have revealed that the links between biological and statistical modles are necessary for scientific progress and that in instances in which multistage and initiation-promotion models of carcinogenesis have been examined, the multiplicative model of statistical interaction is deduced. Nevertheless, the use of the multiplicative model of preventive interaction ( and the use of the additive model of causal interaction) as thresholds for public health action is not justified, primarily due to ethical considerations. Studies of selection processes in unexposed occupational populations have revealed these findings: the absence of a primary selective effect on mortality at young ages and the presence of an increased secondary selection effect on cancer mortality for workers with long service durations.
