BACKGROUND: The optimal risk assessment model (RAM) for venous thromboembolism in hospitalized medical patients remains controversial. This study aimed to assess the prognostic performance of Padua, IMPROVE and NICE guidelines' RAMs.

METHODS: A multicenter prospective observational study was conducted in 15 Spanish hospitals, monitoring consecutive medical inpatients for symptomatic venous thromboembolism over a 90-day follow-up period. The discriminative performance was evaluated using time-to-event analyses and competing risk models accounting for non- venous thromboembolism-related mortality. Sensitivity, specificity, and area under the ROC curve (AUC) were calculated to assess predictive accuracy.

RESULTS: Among 1,273 patients, the 90-day cumulative venous thromboembolism incidence was 1.0%. After adjusting for pharmacological thromboprophylaxis, high-risk patients did not exhibit a significantly increased venous thromboembolism risk compared to low-risk patients according to Padua (aSHR 5.71; 95% CI 0.70-46.86), IMPROVE (aSHR 3.72; 95% CI 1.00-13.87), and NICE RAM (aSHR 0.97; 95% CI 0.30-3.18). Padua had the highest sensitivity (92.3% [95% CI, 62.1%-99.6%]) but lowest specificity (32.3% [95% CI, 29.7%-35.0%]), whereas IMPROVE exhibited the highest specificity (52.9% [95% CI, 50.1%-55.7%]) with moderate sensitivity (76.9% [95% CI, 46.0%-93.8%]). Discriminative performance was suboptimal for all RAMs (AUC: Padua 62.3%, IMPROVE 64.9%, NICE 50.1%).

CONCLUSIONS: Padua, IMPROVE, and NICE RAMs demonstrated poor predictive accuracy in stratifying venous thromboembolism risk among hospitalized medical patients. These findings underscore the need for more precise, dynamic RAMs tailored to real-world clinical settings to enhance thromboprophylaxis and patient outcomes.