Artificial intelligence–enabled electrocardiography as a pre-treatment predictor of chemotherapy-related cardiac dysfunction

Background

Chemotherapy-related cardiac dysfunction (CTRCD) remains a major limitation of contemporary breast cancer therapy.1 Early identification of patients at high risk for left ventricular ejection fraction (LVEF) decline may enable closer surveillance and timely intervention. We investigated whether artificial intelligence–enabled electrocardiography (AI-ECG) performed prior to chemotherapy initiation predicts subsequent LVEF reduction and incident heart failure (HF).

Purpose

To evaluate the association between pre-treatment baseline AI-ECG scores and post-treatment LVEF decline or new-onset HF in breast cancer patients undergoing chemotherapy.

Methods

In this single-center retrospective study, 1,380 breast cancer patients who underwent 12-lead ECG within 6 months prior to first chemotherapy were analyzed. Patients with prior HF or baseline LVEF <50% were excluded.

From baseline ECGs, AI-derived left ventricular systolic dysfunction (LVSD) and diastolic dysfunction (LVDD) scores were calculated. Each score is a continuous metric ranging from 0 to 100, trained using surrogate markers of EF ≤40% (LVSD) and E/e′ >15 (LVDD).2,3

The primary outcome was time to first post-treatment LVEF <50% or new-onset HF/cardiomyopathy. Median follow-up was 431 days (IQR 331–609). Cox proportional hazards models were used to estimate hazard ratios (HRs) per standard deviation (SD) increase in AI-ECG scores, adjusted for age, hypertension, diabetes mellitus, ECG parameters, and chemotherapy regimen.

Results

Among 1,380 patients, 75 (5.4%) experienced the primary outcome during follow-up. Patients who developed CTRCD had a higher prevalence of diabetes mellitus and were more frequently treated with anthracycline-only regimens.

In multivariable-adjusted Cox analysis, higher baseline AI-ECG LVSD score was independently associated with increased risk of LVEF decline or incident HF (adjusted HR 1.26 per SD; 95% CI 1.13–1.40; p<0.001). The LVDD score was also an independent predictor (adjusted HR 1.20 per SD; 95% CI 1.01–1.42; p=0.035). (Table 1) Kaplan–Meier analysis demonstrated significantly higher cumulative event rates in the top quartile of AI-ECG scores compared with lower quartiles (log-rank p=0.020). (Figure 1)

Conclusion

Pre-treatment AI-ECG scores were independently associated with subsequent LVEF decline and new-onset HF in breast cancer patients undergoing chemotherapy. AI-ECG may serve as a scalable and non-invasive tool for early risk stratification of CTRCD in cardio-oncology.Figure 1.KM analysis