Parallels, convergences, and departures in case-control studies and clinical trials.

In studies of the etiology of disease one usually must rely on nature's experiments, observing populations both with and without a disease, then determining how these populations differ with respect to characteristics that some people think could be related to the disease. These unplanned natural experiments can provide meaningful information, or they can mislead. The key to unlocking the useful information the experiments may contain lies in the intelligent, informed choice of “controls,” the nondiseased individuals, for comparison with diseased cases. Some principles for the choice of controls are given in this paper. We show how some of the principles and problems of the controlled trial parallel the principles and problems of epidemiological studies. Examples of matching and cautions concerning possible overmatching are given. Relationships are drawn with controlled and uncontrolled clinical trials and between prospective and retrospective comparison. The two major principles that can be laid down are: ( a ) know the subject matter, but do not choose controls in a given way merely because other people have done it that way before; ( b ) know the specific question(s) the study is designed to answer. This helps avoid overmatching.