Artificial intelligence-enabled prediction of chemotherapy-induced cardiotoxicity from baseline electrocardiograms

Ryuichiro Yagi,Shinichi Goto,Yukihiro Himeno,Yoshinori Katsumata,Masahiro Hashimoto,Calum A. MacRae,Rahul C. Deo

DOI: https://doi.org/10.1038/s41467-024-45733-x

IF: 16.6

2024-03-21

Nature Communications

Abstract:Abstract Anthracyclines can cause cancer therapy-related cardiac dysfunction (CTRCD) that adversely affects prognosis. Despite guideline recommendations, only half of the patients undergo surveillance echocardiograms. An AI model detecting reduced left ventricular ejection fraction from 12-lead electrocardiograms (ECG) (AI-EF model) suggests ECG features reflect left ventricular pathophysiology. We hypothesized that AI could predict CTRCD from baseline ECG, leveraging the AI-EF model’s insights, and developed the AI-CTRCD model using transfer learning on the AI-EF model. In 1011 anthracycline-treated patients, 8.7% experienced CTRCD. High AI-CTRCD scores indicated elevated CTRCD risk (hazard ratio (HR), 2.66; 95% CI 1.73–4.10; log-rank p < 0.001). This remained consistent after adjusting for risk factors (adjusted HR, 2.57; 95% CI 1.62–4.10; p < 0.001). AI-CTRCD score enhanced prediction beyond known factors (time-dependent AUC for 2 years: 0.78 with AI-CTRCD score vs. 0.74 without; p = 0.005). In conclusion, the AI model robustly stratified CTRCD risk from baseline ECG.

multidisciplinary sciences

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: how to use artificial intelligence (AI) models to predict chemotherapy - induced cardiotoxicity, especially cancer - treatment - related cardiac dysfunction (CTRCD), from the baseline electrocardiogram (ECG). Specifically, the research aims to develop an AI - based model (AI - CTRCD model) to predict the risk of CTRCD in patients receiving anthracycline treatment by analyzing the baseline ECG. This will help optimize cardiac monitoring strategies in resource - limited situations, thereby detecting and preventing CTRCD more effectively.

### Background and Problem Statement

1. **Problem Background**:
- Anthracycline drugs (such as doxorubicin) can lead to cancer - treatment - related cardiac dysfunction (CTRCD), which seriously affects the prognosis of patients.
- Although guidelines recommend cardiac monitoring (such as echocardiography) for all patients receiving cardiotoxic chemotherapy, in practice, only about half of the patients receive such monitoring.
- Early detection and treatment of a decrease in left ventricular ejection fraction (LVEF) can reverse or improve clinical outcomes, so accurate risk assessment is crucial.

2. **Existing Challenges**:
- The accuracy of existing risk assessment methods is limited and they are difficult to be widely applied.
- Resource limitations make it impractical to provide regular echocardiography monitoring for all patients.

3. **Research Objectives**:
- Develop an AI - based model to predict the CTRCD risk using baseline ECG data.
- Improve the existing AI - EF model (used to detect low ejection fraction) through transfer learning to adapt to CTRCD risk prediction.

### Main Contributions

1. **Model Performance**:
- The AI - CTRCD model can significantly distinguish between high - risk and low - risk patients (hazard ratio (HR) = 2.66; 95% confidence interval (CI) 1.73–4.10; log - rank p < 0.001).
- After adjusting for other risk factors, the model still shows strong predictive ability (adjusted HR = 2.57; 95% CI 1.62–4.10; p < 0.001).

2. **Clinical Application Potential**:
- The AI - CTRCD model performs consistently in different subgroups (such as different cancer types, genders, baseline LVEF, and anthracycline drug doses).
- The model significantly improves the prediction performance (the area under the curve (AUC) within 2 years is increased from 0.74 to 0.78, p = 0.005).
- Simulation results show that using the AI - CTRCD model can increase the detection rate of CTRCD under the same echocardiography resources (increasing 32 cases).

### Conclusion

This study shows that the AI - CTRCD model can accurately predict the CTRCD risk from the baseline ECG, surpassing traditional risk factors. This provides a new tool for optimizing cardiac monitoring strategies, especially in resource - limited situations, enabling more efficient identification of high - risk patients, thereby improving their management and prognosis.