Explainable and externally validated machine learning for neurocognitive diagnosis via electrocardiograms

TL;DR

This study demonstrates that ECG features, coupled with basic demographics, can accurately predict neurocognitive disorders such as dementia, delirium, and Parkinson's disease across diverse populations. Using separate XGBoost classifiers for each ICD-10-CM code, the authors achieve robust internal and external validation (e.g., F03 dementia AUROC 0.848 internal and 0.865 external; G30 Alzheimer's AUROC 0.809 internal and 0.863 external) and provide explainable insights via SHAP, identifying age as a dominant predictor and highlighting disorder-specific ECG patterns. The work emphasizes the potential of non-invasive, explainable ECG biomarkers for early detection, risk stratification, and personalized monitoring, while acknowledging limitations related to confounding by treatments and generalizability. Overall, this approach lays groundwork for integrating ECG-based diagnostics into neurocognitive care, with implications for screening, monitoring, and longitudinal management using wearable ECG data.

Abstract

Background: Electrocardiogram (ECG) analysis has emerged as a promising tool for detecting physiological changes linked to non-cardiac disorders. Given the close connection between cardiovascular and neurocognitive health, ECG abnormalities may be present in individuals with co-occurring neurocognitive conditions. This highlights the potential of ECG as a biomarker to improve detection, therapy monitoring, and risk stratification in patients with neurocognitive disorders, an area that remains underexplored. Methods: We aim to demonstrate the feasibility to predict neurocognitive disorders from ECG features across diverse patient populations. We utilized ECG features and demographic data to predict neurocognitive disorders defined by ICD-10 codes, focusing on dementia, delirium, and Parkinson's disease. Internal and external validations were performed using the MIMIC-IV and ECG-View datasets. Predictive performance was assessed using AUROC scores, and Shapley values were used to interpret feature contributions. Results: Significant predictive performance was observed for disorders within the neurcognitive disorders. Significantly, the disorders with the highest predictive performance is F03: Dementia, with an internal AUROC of 0.848 (95% CI: 0.848-0.848) and an external AUROC of 0.865 (0.864-0.965), followed by G30: Alzheimer's, with an internal AUROC of 0.809 (95% CI: 0.808-0.810) and an external AUROC of 0.863 (95% CI: 0.863-0.864). Feature importance analysis revealed both known and novel ECG correlates. ECGs hold promise as non-invasive, explainable biomarkers for selected neurocognitive disorders. This study demonstrates robust performance across cohorts and lays the groundwork for future clinical applications, including early detection and personalized monitoring.

Figures (3)

• Figure 1: Diagrammatic illustration of our proposed methodology. The MIMIC-IV-ECG dataset serves as our internal dataset, providing demographic and ECG features used as input for training a tree-based model to diagnose various neurocognitive disorders. For external validation, we utilize a second patient cohort from the ECG-View II dataset, extracting the same set of features and neurocognitive targets. The disorders are defined based on ICD-10-CM codes.
• Figure 2: Model evaluation results across five neurocognitive disorders. Columns represent: (1) Discriminative performance (AUROC with 95% CI), (2) Calibration plots, (3) Net benefit decision curves, and (4) SHAP-based feature importance. Rows correspond to the following ICD-10 codes; G30: Alzheimer's disease, G20: Parkinson's disease, F01: Vascular dementia, F03: Unspecified dementia, and F05: Delirium due to physiological condition.
• Figure 2: Performance in terms of AUROC, calibration curve, net benefit, as well as explainability for G931: Anoxic brain damage.