AI-enabled electrocardiography as a screening tool for peripartum cardiomyopathy: a meta-analysis of diagnostic accuracy
R R A Santos Silva, P G Batista, M V Montenegro, G C Gaelzer, L A Lucena, M E Molinari, R Huntermann, J Peres de Oliveira, M Malik, M Dandamudi, C Fischer Bacca, J GiorgiAbstract
Background
Peripartum cardiomyopathy (PPCM) remains an important cause of maternal morbidity and mortality, in part due to delayed recognition and limited feasibility of routine echocardiographic screening. Artificial intelligence (AI)–enabled electrocardiography (ECG) and digital auscultation tools have emerged as scalable approaches that may support earlier identification and targeted imaging referral, but their diagnostic performance in the peripartum setting has not been systematically quantified.
Purpose
This study aimed to evaluate the diagnostic accuracy of AI-based ECG and digital stethoscope algorithms for the detection of PPCM and pregnancy-associated left ventricular systolic dysfunction (LVSD), using echocardiography as the reference standard.
Methods
We searched MEDLINE, Embase, and the Cochrane Central Register of Controlled Trials from inception to January 18, 2026 for studies evaluating AI-enabled ECG and digital stethoscope models to detect PPCM or LVSD during pregnancy and the postpartum period. For AI-ECG, the primary definition of LVSD used left ventricular ejection fraction (LVEF) <45%, with secondary analyses at LVEF <50% when available, and LVEF <50% for digital stethoscope. We estimated pooled sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) using random-effects models and assessed area under the receiver-operating characteristic curve (AUC) using parametric bootstrapping.
Results
Six studies (n=3,506 women) were included. For AI-based ECG models detecting PPCM with LVEF <45%, pooled sensitivity was 82.9% (95% CI 62.0-93.5) and specificity 93.6% (85.6-97.3), PPV was 58.7% (37.9-76.9) and NPV 98.2% (94.7-99.4), the overall diagnostic performance demonstrated an AUC of 0.90 (0.73-0.94). In secondary analyses using LVEF <50%, AI-ECG models showed sensitivity of 80.1% (60.7-91.3) and specificity of 95.2% (88.3-98.1), with PPV 61.7% (40.4-79.3), NPV 98.2% (95.3-99.3), and AUC 0.91 (0.75-0.94). For PPCM detection with LVEF <50%, digital stethoscope models revealed a pooled sensitivity of 91.4% (80.9-96.4%) and specificity of 86.2% (81.2-90.1%), with PPV 40.4% (18.5-66.9%), NPV 99.3% (96.4-99.9%), and AUC of 0.93 (0.83-0.94). Limitations include between-study heterogeneity and a small evidence base.
Conclusions
AI-enabled ECG and digital stethoscope algorithms demonstrate moderate-to-good diagnostic discrimination for peripartum LVSD, with high rule-out performance. These technologies may serve as adjunctive screening tools to prioritize echocardiography, although prospective validation and implementation studies are needed.Graphical Abstract