Explainable machine learning for neoplasms diagnosis via electrocardiograms: an externally validated study

TL;DR

This work tackles the need for non-invasive, accessible neoplasm diagnostics by leveraging ECG-derived features combined with tree-based machine learning and SHAP explainability. Using internal MIMIC-IV-ECG data and external ECG-VIEW-II validation, the study trains separate XGBoost classifiers for multiple ICD-10-CM neoplasm codes, achieving strong discrimination and calibration across cohorts. SHAP analyses reveal age and key ECG intervals as important predictors, providing interpretable insights into cardio-neoplasm interactions and potential biomarkers. The results demonstrate robust external validity and hold promise for scalable deployment in resource-limited settings, contributing to cardio-oncology by linking electrical cardiac signals with oncologic diagnoses.

Abstract

Background: Neoplasms are a major cause of mortality globally, where early diagnosis is essential for improving outcomes. Current diagnostic methods are often invasive, expensive, and inaccessible in resource-limited settings. This study explores the potential of electrocardiogram (ECG) data, a widely available and non-invasive tool for diagnosing neoplasms through cardiovascular changes linked to neoplastic presence. Methods: A diagnostic pipeline combining tree-based machine learning models with Shapley value analysis for explainability was developed. The model was trained and internally validated on a large dataset and externally validated on an independent cohort to ensure robustness and generalizability. Key ECG features contributing to predictions were identified and analyzed. Results: The model achieved high diagnostic accuracy in both internal testing and external validation cohorts. Shapley value analysis highlighted significant ECG features, including novel predictors. The approach is cost-effective, scalable, and suitable for resource-limited settings, offering insights into cardiovascular changes associated with neoplasms and their therapies. Conclusions: This study demonstrates the feasibility of using ECG signals and machine learning for non-invasive neoplasm diagnosis. By providing interpretable insights into cardio-neoplasm interactions, this method addresses gaps in diagnostics and supports integration into broader diagnostic and therapeutic frameworks.

Figures (6)

• Figure 1: Schematic representation of our proposed approach. We use as internal dataset the MIMIC-IV-ECG dataset from which we use as input features demographics and ECG features to train a tree-based model and diagnose diverse neoplasms. For external validation we take a second cohort of patients from the ECG-View II dataet from which we collect the same set of features and neoplasms targets. The definition of neoplasms are represented by ICD10-CM codes.
• Figure 2: Exemplary performance analysis for the condition "C34: Lung cancer" condition, showing the model's performance across three key evaluation metrics: AUROC curves (discrimination), calibration curves (agreement between predicted and observed risks), decision curve analysis (net benefit compared to "refer all" and "refer none" strategies). Corresponding plots for all other considered conditions can be found in the appendix.
• Figure 3: Explainability results for the investigated neoplasms. The beeswarm plot visualizes through a single dot per feature and sample if the feature contributes positively (right hand side) or negatively (left hand side) to the model prediction. In addition, the color-coding allows to infer if a point is associated with high (red) or low (blue) feature values.
• Figure 4: AUROC curves for all investigated labels, evaluating the model's ability to discriminate between positive and negative cases.
• Figure 5: Calibration curves for each label, assessing the agreement between predicted probabilities and observed outcomes.