PREDICTING ANTIPSYCHOTIC TREATMENT OUTCOMES USING BRAIN CONNECTIVITY AND NEUROCOGNITION IN EARLY PHASES OF PSYCHOSIS

Background: Prognostic biomarkers of clinical remission in individuals with psychotic disorders treated with antipsychotic drugs have been lacking. The development of biomarkers for prediction of clinical improvement may be helpful to guide treatment strategies. The aim of this study was to develop and validate machine learning models based on brain functional connectivity and neurocognition to predict clinical remissions in individuals at Recent Onset of Psychosis (ROP) treated with antipsychotics. Methods: ROP patients treated with antipsychotics were classified as remitters or as non-remitters based on the adapted remission criteria of the Positive and Negative Syndrome (PANSS) Scale after a follow-up period of 3, 6 and 9 months. Baseline resting-state fMRI measures of brain connectivity and a neuropsychological test battery were evaluated on their ability to predict symptom remission using machine learning algorithms. Results: Machine learning models using resting-state fMRI connectivity data predicted clinical remission in 51 ROP individuals treated with antipsychotics after 3 months with the best overall performances (BAC: 71.2%, AUC: 0.77). In order to evaluate the generalizability of our methods, we applied machine learning models to an independent replication sample of 41 ROP patients (BAC: 66.7%, AUC 0.77). Conclusions: These results suggest that functional brain connectivity data at baseline could represent potential biomarkers of symptomatic improvement prediction in ROP individuals treated with antipsychotics. The methods and findings in this study could provide a critical step toward fMRI-based personalized patient treatment in early psychosis. Future work is needed to improve prediction performance to be clinically useful.