Clinical Usefulness of Artificial Intelligence-Enabled Electrocardiography for Pulmonary Hypertension Detection: A Meta-analysis Based on Pre- and Post-Test Probability
In the authors' words
Pulmonary hypertension is frequently diagnosed late because its early manifestations are nonspecific. Artificial intelligence-enabled electrocardiography may provide a scalable method for identifying patients who warrant further evaluation. We searched eight databases from inception through May 2026 without language restrictions for studies evaluating artificial intelligence or machine-learning algorithms applied to standard 12-lead electrocardiograms against invasive hemodynamic or echocardiographic reference standards. Two reviewers independently selected studies, extracted data, and assessed risk of bias. Sensitivity and specificity were pooled using a bivariate random-effects model. The protocol was prospectively registered in PROSPERO (CRD420251149532). Five retrospective studies comprising 103,884 individuals were included. Pooled sensitivity was 0.81 (95% CI, 0.79-0.82), pooled specificity was 0.80 (95% CI, 0.76-0.83), and the summary area under the curve was 0.85 (95% CI, 0.81-0.88). The positive likelihood ratio was 4.03 (95% CI, 3.38-4.80), and the negative likelihood ratio was 0.24 (95% CI, 0.23-0.26). At a pretest probability of 20%, positive and negative results corresponded to posttest probabilities of approximately 50% and 6%, respectively. Heterogeneity was substantial, but leave-one-out analyses did not materially alter pooled estimates. Artificial intelligence-enabled electrocardiography showed good performance for detecting pulmonary hypertension and may support targeted case finding. Prospective studies should determine whether its use improves diagnostic pathways and patient outcomes.
Appeared: Tuesday, October 27. medRxiv. Preprint, not yet peer-reviewed.