Defining "disease" in epidemiologic studies of pulmonary function: percent of predicted or difference from predicted?

Epidemiologic studies often seek to determine the presence and magnitude of association between a trait and a disease, and spirometry test results are often used to define "disease". We evaluated two methods of comparing observed to predicted FEV1. The percent of predicted method is the ratio of a subject's observed to predicted result, and the residual method is the difference between the predicted and the observed result. FEV1 data from 541 men were used for computer simulated sampling for epidemiologic study of groups with different age structures. The residual method was superior because it produced more consistent measures of association, was less likely to produce an apparent association when one did not exist, and is more consistent with the statistical basis of the commonly employed prediction equations.