A screening algorithm to identify clinically significant changes in neuropsychological functions in the diabetes control and complications trial

Abstract

Neuropsychological (NP) evaluations provide an accepted means of monitoring safety in multi-center long-term medical trials. However, using neuropsychologists to review test protocols and rate level of clinical impairment can be a costly and logistically complex undertaking. To facilitate that process, the DCCT Research Group developed a computerized screening strategy that utilized statistical models to identify individuals with possible cognitive deterioration. Two hundred and eight subjects with insulin-dependent diabetes mellitus were assessed twice, 2 years apart, with an extensive battery of NP tests, and the results were rated by expert clinicians. Multiple logistic regression was used to develop a statistical model to predict clinically significant NP worsening (as determined by clinical raters) on the basis of changes in scores (year 2 - baseline) derived from the actual tests. A subsequent performance evaluation with an additional 1087 subjects demonstrated that the computerized algorithm was highly successful in identifying individuals with significantly worsened NP performance. Despite a high false positive rate, the algorithm can achieve an 80-90% reduction in the number of cases requiring evaluation by expert neuropsychologists.