Clinical Decisions for Psychiatric Inpatients and Their Evaluation by a Trained Neural Network

Abstract

AbstractNinety-two consecutive treatment decisions regarding psychotic and depressed inpatients were evaluated by a trained neural network. Simultaneous evaluations according to the Brief Psychiatric Rating Scale (BPRS), the Hamilton scale, and the neural network were performed. A 15-point decrease in the BPRS for psychotic patients or a 10-point decrease in the Hamilton scale for depressed patients was the cut-off point for treatment success. The neural network performed similarly to the clinicians. The combined clinician-network success rate reached 79% and was significantly higher than that for each alone. Clinicians and neural network disagreed on 62% of decisions made for depressed patients and on 55% of decisions made for psychotic patients. The proportion of agreement was lower in both diagnostic groups than expected by chance. However, the high success of the combined clinician and neural network decisions and the high rate of mutual disagreement may imply that a combined decision is fruitful.