OUTCOME-ORIENTED CUTPOINTS IN ANALYSIS OF QUANTITATIVE EXPOSURES

In the analysis of epidemiologic data in which exposure has been measu red on a continuous scale, cutpoints can be defined to delineate categ ories or exposure can be modeled as a continuous covariate by assuming a special functional shape of the effect on disease status. Rules for classifying exposure into two or more categories range from a priori selection of cutpoints to data-oriented rules. The risk estimates may vary, however, with the choice of cutpoint. If the cutpoint selected i s that for which the most impressive effect of exposure on outcome is observed, the final result must be qualified by adjustment. In this pa per, the authors propose a method for adjusting results which are deri ved by varying the cutpoint on a specified selection interval. Adjustm ent is derived from the null distribution of the maximally selected te st statistic. The method should be applied to correct p values if the cutpoint used to define different levels of exposure is selected in su ch a way that the measure of difference between two risk groups, such as the odds ratio or relative risk, is maximized. No method is yet ava ilable for adjusting the resulting risk estimate and the corresponding confidence limits. The authors illustrate the statistical method by a pplying it to data from a case-control study of the association betwee n exposure to magnetic fields and risk of cancer in children which was conducted recently in Denmark.