Artificial intelligence-enhanced risk stratification of cancer therapeutics-related cardiac dysfunction using electrocardiographic images

Abstract

BACKGROUND: Risk stratification strategies for cancer therapeutics–related cardiac dysfunction (CTRCD) rely on serial monitoring by specialized imaging, limiting their scalability. We aimed to examine an application of artificial intelligence (AI) to ECG images as a surrogate for imaging risk biomarkers and its association with early CTRCD. METHODS: Across a US-based health system (2013–2023), we identified 1550 patients (aged, 60 [interquartile range, 51–69] years, 1223 [78.9%] women) without cardiomyopathy who received anthracyclines or trastuzumab for breast cancer or non-Hodgkin lymphoma and had ECG performed ≤12 months before treatment. We deployed a validated AI model of left ventricular systolic dysfunction to baseline ECG images and defined low-, intermediate-, and high-risk groups based on AI-ECG left ventricular systolic dysfunction probabilities of <0.01, 0.01 to 0.1, and ≥0.1 (positive screen), respectively. We explored the association with early CTRCD (new cardiomyopathy, heart failure, or left ventricular ejection fraction <50%), or left ventricular ejection fraction <40%, up to 12 months after treatment. In a mechanistic analysis, we assessed the association between global longitudinal strain and AI-ECG left ventricular systolic dysfunction probabilities in studies performed within 15 days of each other. RESULTS: Among 1550 patients without known cardiomyopathy (median follow-up, 14.1 [interquartile range, 13.4–17.1] months), 83 (5.4%), 562 (36.3%), and 905 (58.4%) were classified as high, intermediate, and low risk, respectively, by baseline AI-ECG. A high-risk versus low-risk AI-ECG screen (≥0.1 versus <0.01) was associated with a 3.4-fold and 13.5-fold higher incidence of CTRCD (adjusted hazard ratio, 3.35 [95% CI, 2.25–4.99]) and left ventricular ejection fraction <40% (adjusted hazard ratio, 13.52 [95% CI, 5.06–36.10]), respectively. Post hoc analyses supported longitudinal increases in AI-ECG probabilities within 6 to 12 months of a CTRCD event. Among 1428 temporally linked echocardiograms and ECGs, AI-ECG left ventricular systolic dysfunction probabilities were associated with worse global longitudinal strain (global longitudinal strain, −19% [interquartile range, −21% to −17%] for probabilities <0.1, to −15% [interquartile range, −15% to −9%] for ≥0.5 [ P <0.001]). CONCLUSIONS: AI applied to baseline ECG images can stratify the risk of early CTRCD associated with anthracycline or trastuzumab exposure in the setting of breast cancer and non-Hodgkin lymphoma therapy.