Front Med (Lausanne). 2026 Aug 6;13:1899303. doi: 10.3389/fmed.2026.1899303. eCollection 2026.
ABSTRACT
BACKGROUND: Postoperative deep vein thrombosis (DVT) is an important complication in older patients undergoing transurethral resection of the prostate (TURP). Although associations between geriatric vulnerability and thrombosis are established, these domains are not routinely integrated into TURP-specific risk assessment. We therefore evaluated whether preoperative geriatric assessment (GA) domains add predictive information to age-adjusted D-dimer in this setting.
METHODS: In this retrospective single-center cohort (n = 420; 294 development and 126 internal validation), an exploratory two-stage procedure comprising univariable screening followed by LASSO regression identified four predictors: preoperative age-adjusted D-dimer positivity, serum albumin, Activities of Daily Living (ADL) score, and Mini-Mental State Examination (MMSE) score. An unweighted logistic regression model was used as the primary integrated model. Discrimination, calibration, precision-recall performance, and decision-curve analysis were assessed. Logistic regression (LR), random forest (RF), and support vector machine (SVM) were additionally compared using the same development-set cross-validated sigmoid calibration framework.
RESULTS: The perioperative DVT incidence was 11.19% (47/420). In the internal validation cohort, the primary integrated LR model achieved an AUC of 0.750 (95% CI: 0.557-0.903), a precision-recall AUC of 0.457, and a Brier score of 0.081. Compared with the D-dimer-only clinical model, the integrated model showed improvements in NRI (0.893, 95% CI: 0.360-1.345; P < 0.001) and IDI (0.078, 95% CI: 0.020-0.145; P = 0.015). At a prespecified probability threshold of 0.20, sensitivity was 0.571 and specificity was 0.911. In the uniformly calibrated algorithm comparison, SVM had the highest AUC point estimate (0.796), but its difference from LR (0.751) was not statistically conclusive (bootstrap difference 0.045, 95% CI: -0.004 to 0.102; P = 0.071).
CONCLUSIONS: A TURP-specific model combining preoperative age-adjusted D-dimer with geriatric assessment domains showed preliminary value for postoperative DVT risk stratification. Because the study was retrospective, single-center, and based on only 47 events, the findings should be considered exploratory and require prospective external validation before clinical implementation.
PMID:42625599 | PMC:PMC13489928 | DOI:10.3389/fmed.2026.1899303
