Assessing the Role of Cytokines in Psychosis Prognosis Prediction: A Machine Learning Approach

Abstract

Abstract In this study, the role of cytokines in predicting treatment outcome of first-episode psychosis (FEP) patients will be assessed. Background: Schizophrenia is a chronic mental disorder in which early response to treatment is associated with improved prognosis. However, accurate prediction of treatment response is still a problem for modern psychiatry. Aims: Investigate the predictive value of aggregate cytokine data in the prediction of FEP patients’ clinical remission. Methods: Data from the OPTiMiSE cohort was used to predict clinical remission as a binary outcome. Using a deep neural network, remission was predicted for patients (n=309) undergoing amisulpride treatment for 4 weeks (phase 1). In addition, remission was predicted for patients (n=57) not in remission after phase 1, who then underwent 6 weeks of either amisulpride or olanzapine treatment (phase 2). Results: Cytokines performed better than chance in predicting treatment response for phase 1 (AUC = 0.58, 95% CI = 0.56-0.60, p = 0.024, permutation n = 1000) and phase 2 (AUC = 0.67, 95% CI = 0.59-0.75). Conclusions: A data modality consisting of 39 cytokines performed better than chance in predicting FEP patients’ clinical remission. Although these findings are modest, they suggest that cytokines should be included in a multimodal approach to predict FEP patients’ treatment response.