Artificial intelligence enabled electrocardiography and digital auscultation for detection of peripartum cardiomyopathy and pregnancy-associated left ventricular systolic dysfunction: A systematic review and meta-analysis of diagnostic accuracy studies.

BACKGROUND: Peripartum cardiomyopathy (PPCM) is an important cause of maternal morbidity and mortality. Delayed recognition remains common because symptoms may overlap with normal pregnancy-related physiological changes. Artificial intelligence (AI)-enabled electrocardiography (ECG) and digital auscultation have emerged as scalable, noninvasive tools that may support case finding and referral for echocardiography, but their diagnostic performance in peripartum populations has not been systematically synthesized. OBJECTIVE: To evaluate the diagnostic accuracy of AI-enabled ECG and digital auscultation algorithms for detecting PPCM and pregnancy-associated left ventricular systolic dysfunction (LVSD), using echocardiography as the reference standard. DATA SOURCES: PubMed, Embase, and the Cochrane Central Register of Controlled Trials were searched from inception through January 18, 2026. STUDY ELIGIBILITY CRITERIA: Diagnostic accuracy studies evaluating AI-enabled ECG or digital auscultation models for detecting PPCM or pregnancy-associated LVSD in pregnant or postpartum women aged 18 years or older were included. RESULTS: Six studies, including 3,506 women, were included. For AI-enabled 12-lead ECG models targeting LVSD defined as left ventricular ejection fraction (LVEF) < 45%, pooled sensitivity was 82.9% and specificity was 93.6%, with an area under the curve (AUC) of 0.90 and negative predictive value (NPV) of 98.2%. Digital auscultation evaluated against LVEF < 50% showed sensitivity of 91.4%, specificity of 86.2%, AUC of 0.93, and NPV of 99.3%. Evidence was limited and heterogeneous. CONCLUSIONS: AI-enabled ECG and digital auscultation showed promising rule-out performance for peripartum LVSD, but current evidence remains limited and heterogeneous and does not yet support routine implementation.