Performance of Western HA-VTE Risk Models in East Asian Populations: A Prospective Multicenter Study

Hospital-associated venous thromboembolism (HA-VTE) causes significant morbidity, but risk prediction models are inconsistently validated, especially in non-Western populations. We prospectively validated five HA-VTE risk scores (Caprini, Padua, IMPROVE, Kucher, and Intermountain) in a prospective multicenter cohort of 2,948 adult inpatients from 17 Chinese hospitals. Symptomatic VTE events were recorded through 3-months follow-up. The 90-day HA-VTE incidence was 2.54%. All five models showed moderate discrimination (AUC 0.726-0.759). Caprini, Padua, and Intermountain did not differ significantly (DeLong test, P > 0.05). Caprini had the highest sensitivity (0.83) but classified 42.6% of patients as high-risk. Padua and Intermountain favored specificity (0.75-0.81) with lower sensitivity (0.64-0.67). Clinical model selection should align with specific balances between sensitivity and utility. In surgical inpatients, Caprini performed best (AUC 0.804, sensitivity 0.90, NPV 1.00), though its more than 40 input variables limit practical use. Calibration was generally acceptable. Performance varied across subgroups. The prophylaxis subgroup showed attenuated performance, with reduced discrimination (e.g., Caprini AUC decreased to 0.682) and less stable calibration. Western-derived HA-VTE models have moderate predictive value in East Asian Populations inpatients. Performance differs across subgroups. Local recalibration and development of simpler, population-specific scores are needed.