Can an AI-powered smartphone app estimate moderate pulmonary hypertension by taking a 12-lead ECG of Japanese patients?

Abstract Background The integration of artificial intelligence (AI) and 12-lead ECGs is an important focus in digital cardiology, and the evidence is a growing focus. Recently, a smartphone app has enabled the capture and analysis of 12-lead ECGs. In this study, our AI-based app captures 12-lead ECGs and extracts ECG rhythms and digital biomarkers. Purpose This study will evaluate the accuracy of detecting moderate pulmonary hypertension (PH) by 12-lead ECG imaging using this app in Japanese patients. Methods A cross-sectional study was conducted on patients who underwent Swan-Ganz catheterization (SGC) from January 2020 to August 2024 at a medical institution in Japan. The app was used to extract 10 digital biomarkers, including the "PH score" (0-100 points), which estimates electrocardiographic rhythm and pulmonary arterial pressure (PAP) elevation. Right ventricular systolic pressure (RVSP) estimated by transthoracic echocardiogram (TTE) was also analysed in the same patients. Results Among 726 patients, 457 met inclusion criteria (exclusion criteria: cardiac surgery, pacemaker rhythm, no ECG within 3 days after SGC, etc.). The mean age was 65 ± 16 years, and 59% were male; moderate PH (mean PAP 40 mmHg) diagnosed at SGC was 7.2%. The AUC-ROC (Area Under the Receiver Operating Characteristic Curve) of the "PH score" calculated by the app was 0.84 (95% CI: 0.77-0.91, p 0.001) (Figure). The "PH score" threshold of 35 points (maximum Youden index) resulted in a sensitivity of 79% and a specificity of 81%; the AUC-PR (Area Under the Precision-Recall Curve) was 0.407 (95% CI: 0.171-0.662) The AUC-ROC of RVSP by TTE was 0.94 (95% CI: 0.89-0.99, p 0.001) (Figure), and the RVSP threshold of 49 mmHg (maximum Youden index) achieved a sensitivity of 87% and specificity of 86%. Finally, PH estimation by RVSP outperformed the "PH score" calculated by the app (DeLong's p = 0.010). Conclusion Moderate PH can be predicted from a 12-lead ECG using an AI-powered smartphone app. Although RVSP is superior to "PH Score" in estimating moderate PH, the app has the potential to identify potentially lethal PH with good AUC-ROC in settings where cardiologists and echocardiography are not available. Additionally, it could also be integrated with electronic medical records.