Artificial Intelligence in Electrocardiography and Echocardiography for Early Diagnosing Pulmonary Hypertension: A Systematic Review and Meta-Analysis (Preprint)

Abstract

<sec> <title>BACKGROUND</title> Early detection of pulmonary hypertension (PH) remains challenging, and right heart catheterization (RHC) is invasive. Artificial intelligence (AI) applied to electrocardiography (ECG) and echocardiography (Echo) offers a potential non-invasive triage strategy, but its overall diagnostic accuracy and clinical readiness remain unclear. </sec> <sec> <title>OBJECTIVE</title> We aimed to systematically evaluate the diagnostic performance of ECG- and Echo-based AI models for detecting PH and compare their accuracy with clinician interpretation. </sec> <sec> <title>METHODS</title> We searched PubMed, Embase, and Web of Science from inception to Jan 1, 2026, for studies evaluating AI models applied to ECG or Echo for PH detection, using RHC or guideline-directed clinical criteria as reference standards. Risk of bias and applicability were assessed with PROBAST+AI, and certainty of evidence was graded using GRADE. Pooled sensitivity and specificity were estimated using a bivariate random-effects model. </sec> <sec> <title>RESULTS</title> Fifteen retrospective studies comprising 504,108 participants (96,890 patients with PH) were included. Echo-based AI demonstrated a pooled sensitivity of 0.91 (95% CI [0.85-0.94]; I2 = 94.41%) and specificity of 0.81 (95% CI [0.69-0.89]; I2 = 98.66%), with a summary area under the curve (AUC) of 0.93 (95% CI [0.91-0.95]). ECG-based AI yielded a pooled sensitivity of 0.81 (95% CI [0.79-0.82]; I2 = 96.91%) and specificity of 0.75 (95% CI [0.68-0.81]; I2 = 99.92%), with an AUC of 0.84 (95% CI [0.81-0.87). In two head-to-head comparative studies, diagnostic performance did not differ significantly between Echo-based AI and clinicians (P =0.80). </sec> <sec> <title>CONCLUSIONS</title> AI models based on ECG and Echo demonstrate moderate-to-high diagnostic accuracy for PH detection, suggesting a potential role in non-invasive clinical triage. However, based on GRADE assessment, the current certainty of evidence is low due to substantial heterogeneity, retrospective study designs, and inconsistent reference standards. Implementation into clinical pathways cannot be recommended until validated by prospective, multicenter studies using standardized RHC definitions. </sec> <sec> <title>CLINICALTRIAL</title> CRD42024590680 </sec>

Ähnliche Arbeiten