Early prediction of relapse in schizophrenia: an application of receiver operating characteristic (ROC) methods

We compared different methods of identifying prodromal periods with regard to their ability to predict relapse in schizophrenia. Fifty stabilized schizophrenic patients, who received a low dose of schizophrenic patients, who received a low dose of fluphenazine decanoate (5 to 10 mg every 2 weeks) were monitored with weekly evaluations to determine whether they met criteria for nonpsychotic prodromal episodes. We evaluated three different scales: (1) the Anxious-Depression subscale of the Brief Psychiatric Rating Scale (BPRS); (2) a modification of the patient self-report Early Signs Questionnaire and (3) the Idiosyncratic Prodromal Scale (IPS). We used receiver operating characteristic (ROC) methods for comparing the different instruments as methods for predicting whether patients would or would not demonstrate a psychotic exacerbation in the 4 weeks following the assessment. Both the IPS and the BPRS cluster scores were better than chance at correctly identifying periods of vulnerability to psychotic exacerbation. The ROC analyses suggest that relatively small changes in the signs and symptoms of chronic schizophrenic patients in maintenance treatment may be clinically meaningful.