Development and Validation of Risk Prediction Models for Residual Pulmonary Vascular Obstruction Using Clinical Data and Metabolomics

Background Previous research identifies residual pulmonary vascular obstruction (RPVO) as an independent risk factor for venous thromboembolism (VTE) recurrence and a predictor of poor outcomes, yet its risk factors and prognostic impact remain unclear. Purpose This study developed predictive models combining clinical and metabolic biomarkers to identify high-risk RPVO patients, offering clinically actionable guidance for optimizing anticoagulation therapy. Patients and Methods This retrospective study analyzed 363 acute pulmonary embolism (APE) patients (2018.1-2024.12) with ≥3-month follow-up. We developed comprehensive and simplified RPVO predictive models by identifying risk factors and assessing long-term outcomes. The models, incorporating metabolomic biomarkers from baseline blood samples, were validated in a prospective APE cohort (2024.12-2025.2). Results Multivariable analysis identified five independent RPVO predictors: 1) affected lobes on V/Q scan, 2) sPESI score, 3) pulmonary artery pressure, 4) recent surgery/immobilization, and 5) active cancer. Both the comprehensive and simplified predictive models showed excellent discrimination (kappa > 0.6) in training, validation, and prospective cohorts. Metabolomic analysis revealed azelaic acid and L-3-phenyl lactate as key differentiating metabolites, whose inclusion enhanced model performance. Notably, RPVO presence and extensive lung involvement (≥6 lobes) independently predicted adverse outcomes (recurrent VTE, cardiopulmonary failure, or mortality). Conclusion We developed comprehensive and simplified RPVO prediction models incorporating five clinical predictors and two metabolic biomarkers (azelaic acid and L-3-phenyl lactic acid), significantly improving model performance. RPVO independently predicted adverse outcomes, highlighting the clinical value of combined clinical and molecular profiling.