Developing a screening tool for in-hospital deep venous thrombosis risk in severe pneumonia patients admitted to respiratory intensive care unit

The incidence of deep venous thrombosis (DVT) is higher for severe pneumonia patients hospitalized in respiratory intensive care unit (RICU). Currently, no validated risk assessment tool specifically exists for evaluating DVT risk in this population. This study aimed to develop a screening tool to estimate the risk of in-hospital DVT in patients with severe pneumonia admitted to the RICU. In this retrospective cohort study, least absolute shrinkage and selection operator (LASSO) regression was used to identify candidate predictors of in-hospital DVT. We then developed two prediction models, a support vector machine (SVM) algorithm and a nomogram model, to predict DVT risk in patients with severe pneumonia in the RICU. We included 368 patients with severe pneumonia admitted to the RICU. DVT occurred in 124 patients (33.7%) during hospitalization. The area under the Receiver operating characteristic (ROC) curve (AUC) of Padua scores, SVM algorithm and nomogram model to predict the risk of DVT were 0.504 [95% confidence interval (CI): 0.442–0.566], 0.526 (95%CI: 0.466–0.587) and 0.754 (95%CI: 0.703–0.805) respectively. The radar chart also indicated that nomogram model had better performance in multiple metrics. For nomogram model, the calibration plot displayed good concordance between predicted and actual outcomes. Clinical utility was assessed using decision curve analysis (DCA). With a range of threshold probability values between 0.1 and 0.8, the nomogram model provided greater net clinical benefit to predict the risk of DVT than Padua scores and SVM algorithm. We developed a nomogram to predict the risk of in-hospital DVT in severe pneumonia patients admitted to RICU. The nomogram showed moderate discriminatory performance and may help identify patients who warrant intensified surveillance or confirmatory ultrasonographic assessment.