A Machine Learning Approach Using Effective Connectivity to Predict Response to Clozapine Treatment

Area of Science:

• Neuroscience
• Psychiatry
• Machine Learning

Background:

• Clozapine is effective for treatment-resistant schizophrenia (TRS-SCZ), but side effects delay its use.
• Predictive biomarkers for clozapine response are needed for timely treatment initiation.
• Schizophrenia affects approximately one-third of patients with chronic treatment resistance.

Purpose of the Study:

• To develop a machine learning (ML) algorithm using pre-treatment electroencephalogram (EEG) data.
• To predict clozapine treatment response in patients with TRS-SCZ.
• To identify biological markers for distinguishing high and low responders.

Main Methods:

• Utilized pre-treatment EEG data from 57 TRS-SCZ patients.
• Applied brain source localization (BSL) with LCMV beamforming to extract source waveforms.
• Calculated effective connectivity using symbolic transfer entropy (STE) on source waveforms.
• Trained an ML algorithm on the STE matrix to differentiate most-responders (MR) from least-responders (LR).

Main Results:

• The ML algorithm achieved a prediction accuracy of 95.83%.
• Symbolic transfer entropy (STE) features effectively discriminated between MR and LR schizophrenia patients.
• The source STE matrix shows promise for predicting clinical response to clozapine.

Conclusions:

• Pre-treatment EEG analysis, particularly source STE, can predict clozapine response in TRS-SCZ.
• This approach aids in distinguishing patient subgroups for optimized treatment strategies.
• The findings support the use of EEG-based biomarkers for personalized clozapine therapy.