Development, temporal validation and external validation of a pre-operative risk prediction model for in-hospital venous thromboembolism in patients with traumatic fractures

Venous thromboembolism (VTE) is a common and potentially fatal complication of traumatic fracture, yet existing risk tools discriminate poorly in this population. We developed and validated a logistic regression model for in-hospital VTE from routinely available clinical data. We studied 1,700 patients with traumatic fractures at two tertiary centres: a development cohort (n = 800; 277 VTE events), a temporal validation cohort (n = 500; 163 events) and an external validation cohort (n = 400; 128 events). Thirteen prespecified predictors were examined, twelve available within 24 h of admission, including age, fracture site, Injury Severity Score, time to surgery, D-dimer, albumin and inflammatory indices. The model was applied without recalibration. Discrimination was assessed by the area under the receiver operating characteristic curve (AUC), calibration by slope, intercept and the observed-to-expected (O:E) ratio, and clinical utility by decision curve analysis. AUCs were 0.904 (95% CI 0.881–0.927), 0.907 (0.880–0.933) and 0.927 (0.898–0.955). Risk was over-predicted in the external cohort (O:E 0.867, 95% CI 0.781–0.952). Eight predictors were independently associated with VTE, and D-dimer contributed substantially: its removal reduced external AUC to 0.842 (0.800–0.884). Discrimination transported well between centres, but calibration was imperfect; recalibration and prospective evaluation are required before clinical use.