[Application of machine learning for predicting the outcome of treatment of patients with schizophrenia according to the indicators of «Thrombodynamics» test].

OBJECTIVE

To identify relationships between thrombodynamic values and the severity of the condition in patients with schizophrenia spectrum disorders (SSD) before and after treatment.

MATERIAL AND METHODS

The study included 92 patients in an acute state of schizophrenia or schizotypal disorder, aged 16 to 57 years (median age [Q1; Q3] - 25 years). All patients received complex psychopharmacotherapy adequate to their psychopathological state. The PANSS was used to assess the severity of symptoms in patients. The coagulation parameters were determined by the thrombodynamics test, in which the growth of fibrin clots in platelet free plasma are observed from special activator. The patient population was divided into two groups with weak and strong response to treatment. Data analysis included machine learning (ML) techniques: logistic regression, random forests, decision trees, support vector machines with radial basis functions, statistically weighted syndromes, permutation method.

RESULTS

An analysis using permutation method revealed statistically significant different thrombodynamics values between groups of patients with weak and strong responses. There are significant differences between thrombodynamics values: T1D, T2D, T2Tlag and DTlag, and values characterizing the severity of positive symptoms before and after treatment (T1PposTot, T2PposTot), severity of psychopathological symptoms before treatment (T1Ppsy1, T1Ppsy6, T1Ppsy13). All ML techniques showed the relationship between thrombodynamics values and response to treatment. The best statistical significance was for statistically weighted syndromes method.

CONCLUSION

The combination of the results of different ML techniques at a high level of statistical significance identifies the thrombodynamic predictors of weak effect of treatment of SSD.