Clinical Report: Deep Learning-Based Integration of EHR and ECG for Anticipating New-Onset Coronary Heart Disease in Patients with Cancer
Overview
This study developed a multimodal deep learning model to predict new-onset coronary heart disease (CHD) in cancer patients, achieving an AUC of 0.975. The model outperformed conventional machine learning methods and identified key predictive features.
Background
Cancer patients are at an increased risk of developing coronary heart disease (CHD), which significantly impacts morbidity and mortality rates. Accurate risk stratification is essential for improving patient outcomes, yet existing tools have limitations. This study explores the integration of electronic health records (EHRs) and electrocardiogram (ECG) data using deep learning to enhance prediction capabilities for new-onset CHD in this vulnerable population.
Data Highlights
Metric
Value
Total Patients
1262
Developed New-Onset CHD
722 (57.2%)
Early-Stage Patients
696 (55.2%)
AUC of CNN-LSTM Model
0.975 (95% CI: 0.962–0.988)
AUC in Early-Stage Patients
0.924
AUC in Late-Stage Patients
0.888
Key Findings
The CNN-LSTM model achieved an AUC of 0.975, outperforming seven conventional machine learning models.
Subgroup analysis showed an AUC of 0.924 for early-stage patients and 0.888 for late-stage patients.
Key predictive features included PAB, CRP, age, cTnI, PT, and ECG markers such as HR, QT, and QTc intervals.
Of the 1262 patients analyzed, 57.2% developed new-onset CHD during hospitalization or follow-up.
Clinical utility and calibration were assessed using ROC, decision curve analysis, and calibration curves.
Clinical Implications
The multimodal CNN-LSTM model provides a robust tool for predicting new-onset CHD in cancer patients, integrating various clinical features and biomarkers. This approach may facilitate early identification and management strategies for at-risk patients.
Conclusion
The study demonstrates the potential of deep learning models in enhancing the prediction of new-onset CHD in cancer patients, highlighting the importance of integrating diverse clinical data.
