Artificial intelligence for identification of patients with increased risk of severe cancer therapy-related cardiac dysfunction following anthracycline therapy.

Abstract

Background: Early detection of cancer therapy-related cardiac dysfunction (CTRCD) after anthracycline exposure is critically important in minimizing morbidity and mortality. Artificial intelligence models applied to electrocardiograms (ECG-AI) may allow for early identification of CTRCD and improved outcomes.

Methods: Patients treated with anthracyclines between 2002 and 2022 across three tertiary centers were evaluated. Characteristics, echocardiograms pre- and post-chemotherapy, and outcomes were reviewed. ECG-AI predictive scores for systolic dysfunction within one year following treatment were collected. ROC analysis was conducted for accuracy of ECG-AI score to detect severe CTRCD (left ventricular ejection fraction <40%).

Results: Overall, 3439 patients were included, mean age 60.2 ± 14.1 years, 53.6% male. Severe CTRCD was present in 114 patients. ROC analysis of ECG-AI scores post-initiation of therapy was superior to that of pre-anthracycline ECG-AI scores and had moderate accuracy for detection of severe CTRCD (AUC 0.761). An ECG-AI score >3.0% was independently associated with significantly poorer survival at one year (HR 2.19, 95%CI 1.92-2.51) and five years (HR 1.69, 95%CI 1.54-1.87) post-anthracycline therapy.

Conclusions: ECG-AI indicating increased likelihood for LVSD post-anthracycline therapy accurately detected severe CTRCD. Clinically, this tool may allow early diagnosis and treatment of high-risk patients and may reduce unnecessary surveillance in those with lower risk.