What question did this study set out to answer?

This research aims to develop a novel classification of comorbidities to enhance treatment and outcomes for patients with acute pulmonary embolism.

May 20, 2026

C18-02 A Novel Comorbidity Phenotype Classification and Its Impact on Outcomes in Patients With Acute Pulmonary Embolism: Insights From a Large Prospective Registry and Machine Learning Analysis

Key Points

This research aims to develop a novel classification of comorbidities to enhance treatment and outcomes for patients with acute pulmonary embolism.
Analyzed two large datasets from the China pulmonary thromboembolism registry study, with Stage II for derivation and Stage I for validation.
Used latent class analysis to identify comorbidity phenotypes and performed survival analyses to evaluate mortality.
Developed an XGBoost classifier for predicting comorbidity phenotypes and validated model performance.
Identified four phenotypes: Low-morbid, Respiratory, Cardiometabolic, and Cardiorenal.
Cardiorenal phenotype had the highest in-hospital mortality, with persistent higher mortality in Respiratory and Cardiorenal groups up to 12 months.
Patients on direct oral anticoagulants had a significantly lower bleeding risk in Cardiometabolic and Cardiorenal phenotypes compared to warfarin.

Abstract

Abstract Background Pulmonary embolism (PE) occurs across diverse comorbidity profiles that heterogeneously influence thrombotic pathophysiology, hemodynamic response, treatment efficacy, and long-term outcomes. Yet, current clinical guidelines and existing comorbidity indices capture this complexity crudely. Therefore, developing a comorbidity-based phenotypic classification may enable precision management and improve patient outcomes. Methods Two independent, large-scale real-world datasets from the China pUlmonary thromboembolism REgistry Study (CURES) were analyzed, with Stage II used for derivation and Stage I for external validation. Patients with malignant tumors were excluded. Twenty comorbidities with a prevalence greater than 2% and established associations with PE were selected for unsupervised clustering using latent class analysis (LCA). Survival analyses assessed in-hospital and 3-, 6-, and 12-month mortality across comorbidity phenotypes. A multivariate logistic regression model was applied to evaluate associations between anticoagulant type and bleeding events. Additionally, an XGBoost-based classifier was developed to predict comorbidity phenotypes, and both the clustering reproducibility and model performance were validated in the CURES Stage I cohort. Results Derivation cohort included 6639 patients, while 6381 patients were included in the validation cohort. From the derivation cohort, four distinct comorbidity phenotypes were identified: Low-morbid, Respiratory, Cardiometabolic, and Cardiorenal. The Low-morbid phenotype included patients with minimal comorbidity burden. The Respiratory phenotype was characterized by a high prevalence of chronic obstructive pulmonary disease (71%) and bronchiectasis (20%), whereas the Cardiometabolic phenotype showed a predominance of hypertension (92%) and higher body mass index. The Cardiorenal phenotype comprised patients with congestive heart failure (69%) and impaired renal function. The Cardiorenal phenotype exhibited the highest in-hospital mortality, and both Respiratory and Cardiorenal phenotypes demonstrated persistently higher mortality at 3, 6, and 12 months compared with the Cardiometabolic and Low-morbid groups. These comorbidity phenotypes were not aligned with the Charlson Comorbidity Index (CCI) or the simplified Pulmonary Embolism Severity Index (sPESI), suggesting they captured distinct biological and clinical heterogeneity. Multivariate analysis showed that among Cardiometabolic and Cardiorenal phenotypes, patients receiving direct oral anticoagulants (DOACs) had a significantly lower bleeding risk compared with those treated with warfarin (Cardiorenal: OR 0.30, 95% CI 0.12-0.77, p = 0.012; Cardiometabolic: OR 0.43, 95% CI 0.26-0.71, p 0.001). External validation confirmed consistent phenotype structures and excellent classifier performance, with micro-averaged AUC of 0.96. Conclusion This novel comorbidity-oriented phenotypic classification provides a comprehensive framework for integrating comorbidity patterns into short- and long-term risk stratification and treatment decision-making in PE, supporting a paradigm shift from uniform management toward comorbidity-guided precision care. This abstract is funded by: the Fund of The National Key Research and Development Program of China (2023YFC2507201)

AI से पूछें

Bookmark