Background: Venous thromboembolism (VTE) is the leading cause of preventable hospital death. National guidelinesrecommend VTE prophylaxis for all medical patients not at low risk. Several VTE risk prediction models have been
developed, but no US models have been validated in a medical population. The ACCP recommends use of the Padua
risk assessment model with a cutoff of 4 points. We developed a prediction model and compared it to Padua.
Methods: We identified all medical patients admitted to Cleveland Clinic hospitals between 2011 and 2016. The first 5
years comprised the training set and 2016 was reserved as a validation set. Potential VTE events were identified by
combination of ICD-9 codes and diagnostic testing. All events were verified through chart review. Potential predictors
included 30 clinical risk factors identified from previous studies. Multiple logistic regression analysis with step-down
variable selection method was used to select the best model. The model was validated both internally and externally
using bootstrapping. The final model was tested against the Padua model using the validation set.
Please note: This study was funded by AHRQ (R01HS022883)
Results: Our training set included 98,661 patients, of whom 226(0.2%) had a VTE develop in the hospital. The final
model included 16 variables, including age, sex, smoking status, history of VTE, thrombophilia, respiratory failure,
chronic kidney disease, inflammatory arthritis, decubitus ulcer, cancer, acute infection, activity level, peripherally
inserted central catheter, central line, mechanical ventilation, and steroids. C-statistic for the training set was 0.79.
Individual predicted risks ranged from 0.03% to >28%. The validation set included 10,753 patients, of whom 55 (0.5%)
developed VTE. The C-statistic for our model was higher than that of the Padua model (0.79 vs. 0.63). At our optimal
treatment threshold of 0.3%, our model identified 2680 (25%) patients as high risk (average risk 1.42%) and 8080
(75%) as low risk (average risk 0.21%), while Padua at a threshold of 4 identified 8011 (75%) as high risk (average
risk 0.69%) and 3102 (25%) as low risk (average risk 0.06%).
Conclusions: Our VTE risk model had better discrimination than the Padua model when applied to the validation set.
Use of the Cleveland Clinic model could substantially reduce the number of patients who qualify for
chemoprophylaxis, while more appropriately targeting high-risk patients.