Derivation of a novel risk assessment model of venous thromboembolism in hospitalized patients
Zhiliang Zhou, Jie Weng, Fan Fei, Yaqi Xu, Jiaze Song, Chen Liu, Wenyi Jin, Fengyu Chen, Liang Wang, Chan Chen, Zhiyi Wang, Zhe XuBackground
Venous thromboembolism (VTE), which includes deep vein thrombosis (DVT) and pulmonary embolism (PE), significantly contributes to morbidity and mortality among hospitalized patients. Despite the existence of various VTE risk assessment models (RAMs), their performance in accuracy, sensitivity and specificity remain suboptimal, highlighting opportunities to improve predictive accuracy for clinical decision-making.
Methods
We conducted a retrospective multicenter study involving three hospitals, which enrolled patients with VTE from January 1, 2021, to December 30, 2023. A novel RAM (Weng score) was developed through three different strategies: clinical knowledge-driven model (Model A), data-driven model (Model B), and decision tree-based model (Model C). The primary outcome was in-hospital VTE. Prediction of PE alone was examined as a secondary outcome. Model performance was evaluated through discrimination, calibration, precision, and decision curve analysis (DCA).
Results
A total of 1,791 patients were analyzed, with 680 VTE events recorded during hospitalization. The Weng score, derived from Model A, demonstrated superior predictive performance for VTE and PE compared to existing RAMs, with an area under the receiver operating characteristic curve (AUROC) of 0.895 (95% confidence interval (CI) [0.880–0.909]) for VTE and 0.877 (95% CI [0.851–0.903]) for PE. In comparison, the AUROCs for existing RAMs (Caprini, Padua, Wells, Geneva, and Autar scores) ranged from 0.687 to 0.789 for VTE prediction and from 0.682 to 0.769 for PE prediction. The Weng score also demonstrated excellent calibration and discrimination, outperforming the Caprini, Padua, Wells, Geneva, and Autar scores in hospitalized patients. The Weng score’s clinical utility for relative risk stratification was further supported by DCA within this case-control sampled cohort, showing a higher net benefit in predicting VTE and PE than existing RAMs.
Conclusions
We developed and internally validated the Weng score using retrospective data from Chinese hospitals. While it showed more favorable calibration and discrimination than existing RAMs in our cohort, external validation in diverse settings and prospective studies accounting for anticoagulation management are essential before clinical adoption. The Weng score is intended for VTE risk stratification only; clinical decisions regarding thromboprophylaxis should integrate both VTE and bleeding risk assessments using validated tools. Because the current model was derived from a case-control sampled cohort (oversampled VTE events), the absolute risk estimates and net benefit findings reflect relative risk ranking and require external calibration in a representative prospective cohort before any clinical implementation.