A model for handling missing items on two depression rating scales

A problem in clinical trials for antidepressants is that patient rating scales are not always completed properly. If one or more items are missing, a patient evaluation may be prevented from contributing towards an efficacy analysis. This can be very wasteful, particularly if data are missing at a baseline evaluation. In this study, we present a novel approach to estimating missing item scores on both the Montgomery-Asberg Depression Rating Scale and the Beck Depression Inventory. Our approach works on the assumption that patient response is non-uniform across items. By capturing the average between-item variation, we have developed a set of weights for each scale which can be used to predict missing item scores with high levels of accuracy. These weights were tested on several sets of patient data when one item was missing. The proposed weights are stable and will be of practical use to researchers in the field of Major Depression.