To examine the predictive validity of early improvement in a naturalistic sample of inpatients and to identify the criterion that best defines early improvement.

Two hundred and forty-seven inpatients who fulfilled ICD-10 criteria for schizophrenia were assessed with the Positive And Negative Syndrome Scale (PANSS) at admission and at biweekly intervals until discharge from hospital. Remission was defined according to the recently proposed consensus criteria, response as a reduction of at least 40% in the PANNS total score from admission to discharge.

Receiver operating characteristic (ROC) analyses showed that early improvement (reduction of the PANSS total score within the first 2 weeks of treatment) predicts remission (AUC=0.659) and response (AUC=0.737) at discharge. A 20% reduction in the PANSS total score within the first 2 weeks was the most accurate cut-off for the prediction of remission (total accuracy: 65%; sensitivity: 53%; specificity: 76%), and a 30% reduction the most accurate cut-off for the prediction of response (total accuracy: 76%; sensitivity: 47%; specificity: 90%).

The findings of clinical drug trials that early improvement is a predictor of subsequent treatment response were replicated in a naturalistic sample. Further studies should examine whether patients without early improvement benefit from an early change of antipsychotic medication.