External validation and model updating of the Cleveland Clinic Model for risk assessment of venous thromboembolism in medical inpatients

Abstract

Background

Venous thromboembolism (VTE) is a leading cause of morbidity and mortality in medical inpatients. The Cleveland Clinic Model (CCM), a novel risk assessment model (RAM) to guide thromboprophylaxis, has never been independently validated.

Objective

To validate and update the CCM, and compare its prognostic accuracy with the Padua Prediction Score (PPS) and the International Medical Prevention Registry on Venous Thromboembolism (IMPROVE) score.

Methods

We used data from a prospective cohort of medical inpatients admitted >24 h to general internal medicine wards at 3 Swiss university hospitals (06/2020–01/2022). The outcome was objectively confirmed symptomatic VTE within 90 days. Patients were stratified into higher and low VTE-risk groups by RAM. We compared test characteristics across 3 versions of the CCM (original CCM and 2 updated versions, 1 with a derived cut-off, and 1 modified derived cut-off version with 45-day follow-up and without missing variables), the PPS, and the IMPROVE score and drew calibration plots for the original and modified derived cut-off CCM.

Results

Of 1352 medical inpatients, 28 (2.1%) experienced VTE within 90 days. The 90-day VTE cumulative incidence was 2.48% in higher-risk vs. 1.91% in low-risk patients (P = 0.52) according to the original CCM. Discrimination was poor across all RAMs (areas under the receiver operating characteristics curve ≤60%). In calibration plots, the original and modified derived cut-off CCMs underestimated the VTE risk.

Conclusion

In this non-U.S., non-ICU cohort, all RAMs showed comparable but limited accuracy for predicting VTE up to 90 days. Updating the original CCM did not improve predictive performance.