Detecting biomarkers associated with antipsychotic-induced extrapyramidal syndromes by using machine learning techniques

Abstract

Background: Antipsychotic-associated extrapyramidal syndromes (EPS) are a common side effect that may result in discontinuation of treatment. Although some clinical features of individuals who develop specific EPSs are well defined, no specific laboratory parameter has been identified to predict the risk of developing EPS.

Methods: Three hundred and ninety hospitalizations of patients under antipsychotic medication were evaluated. Machine learning techniques were applied to laboratory parameters routinely collected at admission.

Results: Random forests classifier gave the most promising results to show the importance of parameters in developing EPS. Albumin has the maximum importance in the model with 4.28% followed by folate with 4.09%. The mean albumin levels of EPS and non-EPS group was 4,06 ± 0,40 and 4,24 ± 0,37 (p = 0,027) and folate level was 6,42 ± 3,44 and 7,95 ± 4,16 (p = 0,05) respectively. Both parameters showed lower levels in EPS group.

Conclusions: Our results suggest that relatively low albumin and folate levels may be associated with developing EPS. Further research is needed to determine cut-off levels for these candidate markers to predict EPS.

Keywords: Biomarkers; Extrapyramidal syndrome; Machine learning.