Quantitative biomarkers to predict response to clozapine treatment using resting EEG data

Abstract

Clozapine is an anti-psychotic drug that is known to be effective in the treatment of patients with chronic treatment-resistant schizophrenia (TRS-SCZ), commonly estimated to be around one third of all cases. However, clinicians sometimes delay the initiation of this drug because of its adverse side-effects. Therefore, identification of predictive biomarkers of clozapine response is extremely valuable to aid on-time initiation of clozapine treatment. In this study, we develop a machine learning (ML) algorithm based on the pre-treatment electroencephalogram (EEG) data sets to predict response to clozapine treatment in TRS-SCZs, where the treatment outcome, after at least one-year follow-up is determined using the Positive and Negative Syndrome Scale (PANSS). The ML algorithm has two steps: 1) an effective connectivity named symbolic transfer entropy (STE) is applied to resting state EEG waveforms, 2) the ML algorithm is applied to STE matrix to determine whether a set of features can be found to discriminate most responder (MR) SCZ patients from least responder (LR) ones. The findings of this study revealed that the STE features could achieve an accuracy of 89.90%. This finding implies that analysis of pre-treatment EEG could contribute to our ability to distinguish MR from LR SCZs, and that the STE matrix may prove to be a promising tool for the prediction of the clinical response to clozapine.

Keywords: Clozapine treatment; Effective connectivity; Machine learning; Resting state electroencephalography (EEG); Schizophrenia; Symbolic transfer entropy.