A Novel Clinical Prediction Model for Pulmonary Hypertension Based on Computed Tomography Angiography, Laboratory Data, and basic demographic information

Pulmonary hypertension (PH) is defined as mean pulmonary artery pressure (mPAP) > 20 mmHg and diagnosed invasively via right heart catheterization (RHC). In this study, we developed a noninvasive PH prediction model. Recursive feature elimination (RFE) selected 11 CTA indicators (RESV, LPA, PA, DHDAD, REDV, RSV, RPA, RCO, AAD, REF, and DAD) as predictors. Among ten machine learning models, XGBoost performed best, with SHAP analysis highlighting MPA as the most influential variable. The CTA model achieved high accuracy (training: 97.9%, validation: 90.9%) and robust metrics (AUC > 0.875). Univariate logistic regression identified additional predictors (sex, age, platelet volume, fibrinogen), which, combined with CTA data, improved performance (AUC: training 0.998, validation 0.909). The final logistic regression model with L1 regularization was visualized as a nomogram. Decision curve analysis confirmed clinical utility. This noninvasive approach, integrating CTA, lab tests, and demographics, aids PH diagnosis in RHC-contraindicated or resource-limited settings.