Abstract
BACKGROUND: To evaluate the predictive performance of three previously established risk models for deep vein thrombosis (DVT) in patients with acute stroke and to support model selection in clinical practice. METHODS: In this single-center spatial validation cohort study, patients diagnosed with acute stroke and admitted to the Stroke Center of South China Hospital of Shenzhen University between January 2023 and January 2025 were consecutively enrolled. Three DVT prediction models, previously identified by a systematic review conducted by our research team, were selected for external validation. Model performance was assessed by the area under the receiver operating characteristic curve (AUC), Brier score, bootstrap calibration curves, and decision curve analysis (DCA). RESULTS: A total of 1,270 patients with acute stroke were included, among whom 217 developed DVT, yielding an incidence of 17.08%. The AUCs were 0.699, 0.804, and 0.753 for the Shen Xiaofang, Lu Qiufang, and Xi Pan models, respectively. The Lu Qiufang model achieved the highest positive predictive value (48.4%), specificity (84.9%), accuracy (82.1%), and Youden index (0.536). All three models had negative predictive values exceeding 90%. The Brier scores were 0.182, 0.154, and 0.245. Calibration curves indicated that the Lu Qiufang model demonstrated the best goodness-of-fit, whereas the Shen Xiaofang and Xi Pan models exhibited systematic bias in certain risk intervals. DCA curves showed that the Lu Qiufang model provided greater net benefit within the threshold probability range of approximately 0.20-0.70, indicating superior clinical decision value. CONCLUSION: All three DVT risk prediction models demonstrated acceptable predictive performance in patients with acute stroke. Among them, the Lu Qiufang model showed comparatively superior discrimination, calibration, and clinical net benefit. However, given the single-center design of this study, further multicenter and cross-regional validation studies are warranted to confirm model transportability and generalizability across diverse healthcare settings.