Optimal thresholds of early response to atypical antipsychotics: Application of signal detection methods

Objective: Identify the optimal magnitude of response to antipsychotic medication at various early time points that best predicts subsequent non-response at 8 weeks. Methods: Data were pooled from 5 randomized, double-blind clinical trials of atypical antipsychotics in the treatment of schizophrenia and related disorders (n = 1137 moderately-to-severely ill; n = 300 less than moderately ill). Signal detection methods (receiver-operating characteristic curves) were used to identify the optimal response threshold based on percent change from baseline on the PANSS total score at different early time points (Weeks 1-4) to predict subsequent 'non-response' at 8 weeks (i.e., not 'minimally improved', 'much improved' or 'remitted') while holding the false positive rate to a level of 30% or less. Analyses were implemented separately for patients with schizophrenia who differed on baseline illness severity. Results: Using Area Under the Curve (AUC) >= 0.8 to define optimal discriminative ability at the earliest time point, the early response threshold in moderately-to-severely ill patients for predicting not 'minimally improved' was <15% reduction in PANSS total at Week 2, not 'much improved' was <23% at Week 2, and not 'remitted' was <26% at Week 4. Similarly, in less than moderately ill patients, the optimal early response threshold for predicting not 'minimally improved' was < 12% reduction in PANSS total at Week 2, and not 'much improved' was < 14% at Week 1. Conclusion: Specific thresholds of response were identified at early time points for predicting subsequent non-response. Not attaining these early response thresholds may serve as important clinical markers of subsequent non-response to antipsychotic therapy. (C) 2009 Elsevier B.V. All rights reserved.