Abstract
A common problem in psychiatric epidemiology is that of finding appropriate statistical methods to determine the specificity of an association of two disorders (A and B), when disorder A is also associated with a third disorder C. This paper discusses this problem, which is particularly frequent in mental disorders due to substantial comorbidity between disorders. It is suggested that a measure of specificity of association (MSA) can be applied to address this problem. By building on mutually exclusive categories, classes, this measure indicates a high specificity of association A–B whenever two conditions are fulfilled: (a) there is a considerable difference in strength of association of A with pure B (not C) on the one hand and A with pure C (not B) on the other hand, and (b) the association of A with comorbid cases (cases that have both disorders B and C) is at least not stronger than the association of A with pure B cases. The measure is based on a logistic regression model with the probability of having disorder A as the outcome and mutually exclusive categories of B and C as explanatories as well as possibly other confounding variables.
Statistical inference in MSA is based on a bias‐corrected bootstrap confidence interval. This paper exemplifies the use of this measure with an example from a longitudinal prospective study on the relationship of nicotine dependence with ‘affective’ and ‘anxiety’ disorders. Copyright © 1999 Whurr Publishers Ltd.