An empirical staging model for schizophrenia using machine learning

Introduction One of the great challenges still to be achieved in schizophrenia is development a staging model that reflects progression disorder. The previous models suggested have been developed from theoretical point view and do not include objective variables such as biomarkers, physical comorbidities, or self-reported subjective (Martinez-Cao et al. Transl Psychiatry 2022; 12(1) 1-11). Objectives Develop multidimensional for based on empirical data. Methods Naturalistic, cross-sectional study. Sample: 212 stable patients with Schizophrenia (F20). Assessments: ad hoc questionnaire (demographic clinical information); psychopathology: PANSS, CDS, OSQ, CGI-S; functioning: PSP; cognition: MATRICS; laboratory tests: C-Reactive Protein (CRP), IL-1RA, IL-6, Platelets/Lymphocytes (PLR), Neutrophils/Lymphocytes (NLR), Monocytes/Lymphocytes (MLR) ratios. Statistical analysis: Variables selection was performed an algorithm this research. referred makes use genetic algorithms (GA) select those show best performance classification according their global CGI-S. function GA maximizes individuals correct support vector machines (SVM) employs input given by (Díez-Díaz Mathematics 2021; 9(6) 654). Models assessed help 3-fold cross-validation these process repeated 10,000 times each one assessed. Results Mean age(SD): 39.5(13.54); men: 63.5%; secondary education: 59.50%. Most our sample had never married (74.10%), more than third received disability benefits due (37.70%). mean length disease 11.98(12.02) years. SVM included following variables: 1)Clinical: number hospitalizations, positive, negative, depressive symptoms general psychopathology; 2)Cognition: speed processing, visual learning social cognition; 3)Functioning: PSP total score; 4)Biomarkers: PLR, NLR MLR. This executed again 100,000 applying cross-validation. In 95% executions 53.52% were classfied right CGI-S category. On average 61.93%. About specificity sensitivity values obtained 0.85 0.64 respectively. Conclusions Our robust method appropriately distributes severity Highlights importance clinical, functional cognitive factors classify patients. Finally, inflammatory parameters MLR also emerged potential biomarkers schizophrenia. Disclosure Interest None Declared