Abstract
BACKGROUND AND HYPOTHESIS: Accurate prediction of treatment response is essential for optimizing therapeutic strategies in patients with schizophrenia. Compared to neuroimaging or genetic biomarkers, clinical symptom patterns have received relatively little attention as predictors of treatment outcome. This study aimed to address this gap by comprehensively analyzing early symptom trajectories to predict long-term treatment outcomes. STUDY DESIGN: A cohort of 387 inpatients with schizophrenia during acute episodes was followed for 8 weeks of standardized antipsychotic treatment. Clinical symptom severity was assessed by Positive and Negative Syndrome Scale (PANSS) at baseline, week 2, and week 8. Using network analysis and machine learning model, we evaluated symptom patterns associated with treatment outcome and the predictive value of early clinical symptom trajectories. STUDY RESULTS: (1) Effective treatment responders (ETR) and poor treatment responders (PTR) exhibited distinct clinical symptom profiles at baseline and early treatment response. (2) At week 2, ETR patients showed a denser PANSS change network compared to PTR, indicating more coordinated symptom changes. (3) Early symptom change was significantly correlated with 8-week treatment outcome. (4) Although the absence of early treatment response had limited predictive value, a machine learning model based on early %PANSS change achieved 76% balanced accuracy, with changes in the negative domain emerging as key predictors. CONCLUSIONS: These findings highlight the distinctive symptom profiles associated with different treatment outcomes and underscore the importance of early symptom patterns in predicting 8-week responses in patients with acute schizophrenia.