Comparative Performance of Clinical Risk Assessment Models for Hospital-Acquired Venous Thromboembolism in Medical Patients

Background Improved thromboprophylaxis for acutely ill medical patients relies on valid predictions of thrombotic risks. Our aim was to compare the performance of the Improve and Geneva clinical risk assessment models (RAMs), and to simplify the current Geneva RAM. Methods Medical inpatients from eight Swiss hospitals were prospectively followed during 90 days, for symptomatic venous thromboembolism (VTE) or VTE-related death. We compared discriminative performance and calibration of the RAMs, using time-to-event methods with competing risk modelling of non-VTE death. Results In 1,478 patients, the 90-day VTE cumulative incidence was 1.6%. Discrimination of the Improve and Geneva RAM was similar, with a 30-day AUC (areas under the curve) of 0.78 (95% CI [confidence interval]: 0.65-0.92) and 0.81 (0.73-0.89), respectively. According to the Improve RAM, 68% of participants were at low risk (0.8% VTE at 90 days), and 32% were at high risk (4.7% VTE), with a sensitivity of 73%. According to the Geneva RAM, 35% were at low risk (0.6% VTE) and 65% were at high risk (2.8% VTE), with a sensitivity of 90%. Among patients without thromboprophylaxis, the sensitivity was numerically greater in the Geneva RAM (85%) than in the Improve RAM (54%). We derived a simplified Geneva RAM with comparable discrimination and calibration as the original Geneva RAM. Conclusions We found comparably good discrimination of the Improve and Geneva RAMs. The Improve RAM classified more patients as low risk, but with possibly lower sensitivity and greater VTE risks, suggesting that a lower threshold for low risk (< 2) should be used. The simplified Geneva RAM may represent an alternative to the Geneva RAM with enhanced usability.