Personalized prediction of treatment response during the acute phase of schizophrenia based on the connectome-based predictive model.

BACKGROUND: Schizophrenia shows great variability in symptoms and treatment response. Integrating neuroimaging techniques with data-driven models may help to predict patient responses and develop personalized therapies. METHODS: Using functional connectivity (FC) changes before and after treatment as the neuroimaging feature and reduction rate (RR) of different symptom dimensions as the clinical feature, this study established a prediction model for the acute-phase treatment response of patients with schizophrenia using the connectome-based predictive modeling (CPM) method. The model was then validated in independent samples. Transcriptome-neuroimaging correlation analysis was conducted to identify genes associated with FC changes. RESULTS: Prediction models were established, which involved the total score of PANSS and scores of positive and affective symptoms. The models were Y = 0.029Xpos + 0.500 (r = 0.407, P = 0.040) for the RR of the total score of PANSS, Y = 0.010Xpos - 0.010Xneg + 0.542 (r = 0.421, P = 0.038) for the RR of positive symptoms, and Y = -0.037Xneg + 0.475 (r = 0.486, P = 0.040) for the RR of affective symptoms. The positive and negative network models for predicting the RR of positive symptoms (P = 0.011, R² = 0.184) and the negative network model for the RR of affective symptoms (P = 0.041, R² = 0.125) were validated. Gene enrichment analysis linked the models to synaptic structures and cell channel activation. CONCLUSIONS: Our study established and validated a three-network predictive model for predicting treatment response regarding positive/affective symptoms, which may help individualized treatment monitoring and efficacy prediction.