Development and Validation of a Prediction Model for Venous Thrombus Embolism (VTE) in Patients With Colorectal Cancer

Cancer patients are at high risk of developing venous thromboembolism (VTE). The risk of VTE could be mitigated with the administration of prophylactic anticoagulants. Therefore, risk assessment models would be a useful tool in order to identify those patients who are at higher risk and will be benefited more by prophylactic anticoagulants. This study retrospectively examined 528 newly diagnosed colorectal cancer patients from January 2019 to January 2021. Specified logistic regression models were employed to screen the factors and establish prediction tools based on nomograms according to the final included variables. Discrimination, calibration, and clinical applicability were used to assess the performance of screening tools. In addition, internal verifications were conducted through 10-fold cross-verification, leave-one-out cross-validation, and Bootstrap verification. Four risk factors, closely related to the occurrence of VTE in colorectal cancer patients, were identified after univariate and multivariate logistic regression, including age, body mass index, activated partial thromboplastin time, and D-Dimer value. Besides, the risk assessment model named ABAD was built on the basis, displaying good discriminations and calibrations. The area under the curve was 0.705 (95% confidence interval [CI], 0.644 to 0.766). According to Hosmer–Lemeshow goodness-of-fit test, a good agreement between the predicted and observed VTE events in patients with newly-diagnosed gastrointestinal cancer was observed for χ2 = 6.864, P = .551. Internal validation was applied with a C-index of 0.669 in the 10-fold cross-verification, 0.658 in the leave-one-out cross verification and 0.684 in the bootstrap verification. We developed a prediction model called ABAD for newly diagnosed colorectal cancer patients, which can be used to predict the risk of VTE. After evaluation and internal verification, we believe that ABAD exhibited high predictive performance and availability and could be recommended.