A multimodal prediction model for heart failure with pulmonary hypertension based on right heart catheterization and echocardiography

Objective Right heart catheterization (RHC) and echocardiography (ECHO) are essential for evaluating heart failure (HF) with pulmonary hypertension (PH). We developed a multimodal model and validated its diagnostic performance. Methods This study enrolled 217 HF patients, including 151 diagnosed with HF combined with PH via RHC and ECHO, of whom 31 had isolated pre-capillary PH (pre-capillary PH), 78 had isolated post-capillary PH (Ipc-PH), and 42 had combined pre- and post-capillary PH (Cpc-PH). The consistency between RHC and ECHO in PH diagnosis was evaluated. A multimodal prediction model was developed by integrating clinical characteristics, biomarkers, and ECHO data. Then, 50 prospectively enrolled patients were assessed using ROC curves to determine the model's diagnostic efficacy. Results The prevalence of HF with PH was 69.6%, significantly higher than the traditional diagnostic standard of 52.1%. ECHO assessment of pulmonary artery systolic pressure (PASP) had a false-positive rate of 69.7%, and at the optimal cutoff value of 49.5 mmHg, sensitivity was 41.9%. In ROC curve analysis, the multimodal prediction model(0.238 × (TAPSE/PASP ≤ 0.330) + 0.037 × (BNP > 307.5) + 0.081 × (AF) + 0.110 × (LVEF < 52.5%) + 0.357 × (ESRA > 21.5) + 0.401 × (RVDd > 3.25)) demonstrated significantly superior diagnostic performance, compared with single ECHO parameters (AUC = 0.951 vs. 0.804, ΔAUC = 18.3%). Additionally, the model showed a significant association with adverse events (heart failure rehospitalization, respiratory failure, pulmonary encephalopathy, and sudden cardiac death, OR = 5.052). In the validation cohort, the model achieved an AUC = 0.943 (95% CI: 0.880–0.991), with a sensitivity of 0.917 and a specificity of 0.857. Conclusion The prevalence of HF with PH was notably higher under the new diagnostic criteria. Relying solely on ECHO for diagnosis has limitations, whereas the non-invasive multimodal prediction model significantly improves diagnostic accuracy, enabling risk stratification and subtype differentiation of PH, thereby offering a novel strategy for clinical management.