Receiver operating characteristic analysis for validating EEG sleep discrimination of elderly depressed and demented patients.

Receiver operating characteristic analysis is a useful tool in validating a diagnostic marker. This technique was applied to a discriminant function model using selected electroencephalographic sleep measures (sleep maintenance, percentage of rapid-eye-movement sleep, and percentage of indeterminate non-rapid-eye-movement sleep) in elderly patients with major depression or dementia of the Alzheimer type. The diagnostic model correctly classified 79.8% of elderly depressed and demented patients. The area under the curve was 0.86, statistically different from 0.5, which would be expected by chance (P less than .0001) and indicating good predictive power of the model. A logistic regression analysis demonstrated the reliability and stability of the diagnostic equation.