Development and Validation of a 30-Day Mortality Prediction Model for Patients with Cancer-Associated Thrombosis (CAT) in the Riete Database

Abstract

Introduction:Cancer-associated thrombosis (CAT) remains a leading cause of morbidity and mortality. The current scoring systems underperform identifying patients with worse survival after CAT. Proper risk stratification of patients for early mortality after a VTE event may lead to customized anticoagulation protocols.Aim:-To derive and validate a scoring system to predict 30-day mortality after CAT.Methods:We selected patients with active cancer and thrombosis from the Computerized Registry of Patients with Venous Thromboembolism (RIETE) database. The main outcome was all cause mortality within the month following a CAT event. We abstracted demographics, biochemical, treatment and VTE-related variables. We used descriptive statistics to summarize the distribution of baseline parameters. We used a simple random selection to split are data in a derivation (2/3 of the dataset) and a validation (1/3 of the dataset) cohort. In the derivation cohort, we used recursive partitioning to identify groups at risk for our primary outcome. To determine pathways recursively, the variable most associated with the outcome (F-statistic) was used as a splitting point to segregate the sample into groups at the next tree level. After identification of the predictive variables, we used standard binary logistic regression model to determine the likelihood of the primary outcome. The derived risk score was then applied in the validation cohort. Statistical analysis was conducted with IBM SPSS version 24 (Armonk, New York)Results:From a total 10,025 eligible patients with active cancer and newly diagnosed thrombosis, 5.343 (53 %) patients were men, with a median age of 67 (IQR:59-77), most had metastatic disease (63%) and the predominant tumor types were: lung cancer 1.658 (16.4%), breast 1.418 (14%) and colorectal cancer 1.392 (13.8%). We identified 6 predictors of mortality using recursive partitioning: white blood cell count ≥ 11.5 × 109/L, platelet counts ≤ 160 × 109/L, metastasis, immobility, any pulmonary embolism and BMI < 18.5. A risk score was developed based on regression coefficients from the final multivariate model and divided the population into 3 risk categories based on the score from our risk model: low (score 0-3), moderate (score 4-6), and high (score ≥7) (Fig. 1A). In the validation cohort, among the patients classified as high risk, 72% had the primary outcome compared to 18.6% in the moderate and 7.3% in the low risk group (Fig. 1B). The low-risk group had a specificity of 71% and a negative predictive value of 84%; whereas the high-risk group had a sensitivity of 72% and a positive predictive value of 23%.Conclusion:The risk of death following CAT can be categorized using this scoring system. Our validated mortality risk model, may assist physicians regarding anticoagulant choices considering expanding anticoagulant options. [Display omitted] DisclosuresNo relevant conflicts of interest to declare.