Investigating risk factors and developing a predictive model for venous thromboembolism in patients with sepsis

Patients with sepsis are at high risk for venous thromboembolism (VTE). However, existing assessment tools designed for general medical or surgical populations may inadequately capture the specific risk profile of septic patients. This study aimed to identify independent predictors of VTE in sepsis and develop a sepsis-specific prediction model.

In this retrospective cohort study, adult patients with sepsis diagnosed at a single center (2015-2024) were enrolled. Candidate predictors and treatments were restricted to data available within the first 24 h of diagnosis. After exclusions, 157 VTE cases and 703 matched non-VTE controls formed the prediction cohort, which was split into training (n = 602) and validation (n = 258) sets. The Synthetic Minority Oversampling Technique was applied to the training set. LASSO regression followed by multivariable logistic regression identified independent factors for the nomogram. Model performance was assessed via discrimination, calibration, and decision curve analysis and compared with the Padua and Caprini scores in the validation cohort.

Among 3,228 sepsis patients, 157 (4.86%) developed VTE. Multivariable analysis identified four independent predictors: surgery within 3 months, central venous catheterization within 24 h, history of VTE ≥1 year prior, and absence of early anticoagulant therapy (all P < 0.001). The resulting nomogram showed excellent discrimination (AUC: 0.905, 95% CI: 0.875-0.935 in training; 0.909, 95% CI: 0.867-0.951 in validation) and good calibration. Decision curve analysis supported its clinical utility. In the validation cohort, the nomogram significantly outperformed the Padua (AUC: 0.712) and Caprini (AUC: 0.738) scores, demonstrating superior incremental predictive value (P < 0.001 for net reclassification improvement and integrated discrimination improvement).

A sepsis-specific nomogram incorporating recent surgery, early catheterization, previous VTE, and absence of early anticoagulation demonstrated good discrimination and calibration. It outperformed the Padua and Caprini tools in internal validation. External validation is required before clinical implementation.