Abstracts. (2024). The Journal of ECT, 40(3), e20–e30. https://doi.org/10.1097/yct.0000000000001056
Abstract:
Background: Electroconvulsive therapy (ECT) is a swift and effective intervention for drug-resistant depression and schizophrenia. Nonetheless, there is substantial variability in individual responses and potential for cognitive side effects.
Objective: To develop machine learning models to predict ECT treatment response and cognitive side effects and identify features important for prediction.
Methods: We conducted a retrospective analysis of 2155 ECT courses from 1499 patients treated at the Institute of Mental Health Singapore between 2017 and 2022. We used extreme gradient boosting (XGBoost) to predict treatment outcomes—the Clinical Global Impression Scale (CGI-S), Brief Psychiatric Rating Scale (BPRS), and Montgomery-Asberg Depression Rating Scale (MADRS)—and cognitive side effects—Montreal Cognitive Assessment (MoCA). Model inputs included demographic information, medical history, ECT dose settings, and ictal electroencephalography (EEG) data acquired during ECT administration. We compared the prediction performance of the model using different combinations of input features and from different ECT sessions.
Results: The XGBoost model predicted treatment non-response with an AUC of 0.611 for CGI-S, 0.656 for schizophrenia subscale of BPRS and 0.543 for MADRS. Cognitive side effects (more than 2 point decline in MoCA) was predicted with an AUC of 0.673. While MoCA decline was predicted well using pre-treatment MoCA score, prediction of other clinical outcomes benefited from the addition of ECT dose and EEG features.
Conclusion: The developed models can be used to identify features important for ECT treatment response and to anticipate cognitive side effects and inform treatment decisions.
License type:
Publisher Copyright
Funding Info:
This research / project is supported by the Agency for Science, Technology and Research, Singapore - A*STAR Strategic Programmatic Fund - Brain Body Initiative
Grant Reference no. : A18A2b0046