General Health Questionnaire (GHQ) in general hospitals: selecting a set of items using a stepwise hierarchical procedure

Abstract

Abstract Several studies have shown that, among non‐psychiatric in‐patients, validity indices of the General Health Questionnaire (GHQ) are relatively low when compared to those among primary healthcare attenders. We made an attempt to improve validity by excluding items from the 30‐item GHQ. This exclusion was performed by means of logistic regression analysis, on the one hand, and by using correlations of GHQ sum‐scores with the results of a psychiatric criterion interview (Clinical Interview Schedule) on the other. For this purpose we used a hierarchical approach based on subsets of items. In order to describe the overall discriminating ability of the 30‐, 20‐, 12‐item GHQ and of the two shortened GHQ versions, Receiver Operating Characteristics were used. The discriminating performance of the eight items gathered by Stepwise Hierarchical Variable Selection was significantly better than that of the six items yielded from logistic regression and of the above mentioned three traditional GHQ versions. These analyses are based on 511 participants of a survey in Austrian general hospitals. Overall, Stepwise Hierarchical Variable Selection procedure suggested by the authors for item selection based on sum‐scores might be a useful tool in future two‐phase surveys. Copyright © 2001 Whurr Publishers Ltd.