The evaluation of epidemiologic evidence for policy-making.

Received for publication July 20, 2000, and accepted for publication November 3, 2000. Abbreviations: ARIC, Atherosclerosis Risk in Communities. From the Department of Epidemiology, Bloomberg School of Public Health, Johns Hopkins University, Baltimore, MD. Reprint requests to Dr. Moyses Szklo, Johns Hopkins Bloomberg School of Public Health, 615 North Wolfe Street, Suite W6009, Baltimore, MD 21205 (e-mail: [email protected]). For many factors affecting human health, experimental epidemiologic evidence is often unavailable or missing; thus policy decisions must frequently be made on the basis of observational data. Whereas precision is of importance in both well designed experimental and observational studies, in observational research consideration of confounding and bias becomes of paramount importance, particularly when associations between individual risk factors or interventions and disease outcomes are not strong, e.g., those characterized by relative risks of less than 2.0. Thus, a small degree of negative confounding or bias may lead to such an underestimation of a true weak association that the observed value of the association measure becomes null; on the other hand, positive confounding or bias of a small magnitude may easily result in an observed weak association. Weak associations in observational epidemiologic research have been extensively discussed in the literature and were the focus of a recent symposium (1) and a pointcounterpoint exchange (2, 3). Approaches to deal with weak associations, referred to by Rothman and Poole (4) as “a strengthening” program, include focusing on low risk subjects, and preventing bias and confounding. Focusing on especially susceptible groups is yet another strategy to “strengthen” an association when the average (“main”) effect is small (5). The possibility of confounding and bias should be carefully considered in observational research when deciding whether a weak association reflects the true level of effect. When confounding and bias are not regarded as plausible reasons for the association, reliance on the consistency of findings from different studies to infer causality (6) assumes special importance when evaluating weak associations. As a corollary, both weak associations and lack of consistency among studies pose important challenges for translating epidemiologic data into policy. Selected issues that must be considered in assessing a weak association are succinctly reviewed next, including its definition, consideration of confounding and bias, focusing on especially susceptible groups, studying associations in low risk groups, and consistency among studies.