Psychometrics and event-related potentials in the diagnosis of dementia.

This study attempted to increase the reliability of predicting the normal latency of the P3 component in elderly adults. It was hypothesized this would allow a more accurate distinction to be made between normal adults (n = 20) and patients with dementia (n = 19) on the basis of observed P3 latencies. Two models for predicting latency were assessed. The first, which used age alone as a predictor, explained 18% of the normal variation in latency and misclassified 40% of the patients. The second model, which incorporated scores on psychometric tests, explained 74% of latency variation and misclassified 20% of the patients. Scores on the Similarities subtest of the WAIS and the Rey Auditory Verbal Learning Test, however, correctly identified 92% of the patients. This finding casts doubt on the justification of the P3 latency measures to diagnose dementia.