Critical Appraisal of the Exposure-Potential Restriction Rule1

Schlesselman's rule to restrict all case-control studies to those individuals potentially at risk of exposure was designed to reduce a bias toward the null. Found to be without basis, the need for the rule came into question, and Schlesselman and Stadel now suggest several other arguments on behalf of the rule and extend its jurisdiction from case-control to all epidemiologic studies. It is not necessary to decide on a single view of fairness on the contrast of case-control designs in order to assess the impact on precision of the exposure-potential restriction rule. There exists a class of counter-examples for which the rule reduces precision from either perspective of a fair contrast. This class includes all studies in which the exposure prevalence is closer to its optimal value for precision prior to applying the rule than it is after the rule is applied. Raising the exposure prevalence diminishes precision in such examples, even if the number of cases and the number of controls are held constant between the 2 studies. Concerning fairness in the contrast of study designs, enhancement of precision does not justify a rule to restrict every study to people not known to lack the opportunity for exposure. The other argument in favor of the exposure-potential restriction rule is that adherence to the rule incidentally will result in control of confounding, if the indicator of exposure potential also happens to be a confounder. The fault with this argument is that restriction is an irrevocable way of controlling possible confounding. The mere possibility of inadvertently controlling confounding does not provide a good reason to use every indicator of exposure opportunity as a restriction criteria in every study. This is what the proposed rule requires. Schlesselman and Stadel introduce 3 new arguments to defend the exposure-potential restruction rule. 1 of the 3 arguments draws an analogy between the proposed rule and theuse of randomization in experimental studies. The 2nd argument draws an analogy between potential for exposure and potential for disease. The last argument invokes a requirement for epidemiologic studies to respond to questions of relevance and importance. Regarding the latter argument, they provide only illustrations, among them that "the real issue" in a study of the effects of oral contraceptives (OCs) "is the risk of disease in women who choose this contraceptive method." The authors identify the experience of individuals who are not OC users as "irrelevant to the purpose of the study." None of the arguments Schlesselman and Stadel propose answer the question of why every study should be free of individuals who were barred by some nonconfounder from becoming exposed. This is the question if one accepts the experience of unexposed individuals as essential to studying the effects of any exposure, and if one acknowledges the long-standing use of restriction to control confounding.