e24123 Background: The risk factors for bleeding during anticoagulation in patients with cancer-associated thrombosis (CAT) remain largely unexplored. The recently developed PredictAI model ( Muñoz Martín AJ et al. JCO 40, e18744-e18744(2022) was aimed to predict major bleeding (MB) in anticoagulant-treated cancer patients within the first 6 months following venous thromboembolism (VTE) diagnosis. The goal was to validate the PredictAI model using an independent cohort of patients from the TESEO international, observational, prospective registry. Methods: The performance of the three predictive models developed in PredictAI were assessed in terms of accuracy, precision, recall, F1-score, and the receiver operating characteristic (ROC) area under the curve (AUC): PredictAI decision tree classifier ROC-AUC 0.66 (confidence interval 95% [CI 95%], 0.63-0.69; P < .0001), random forest classifier ROC-AUC 0.61 (CI 95%, 0.58-0.64; P < .0001), and logistic regression ROC-AUC 0.70 (CI 95%, 0.67-0.73; P < .0001). The outcome was MB within 6 months following VTE diagnosis, whereas the analyzed predictors for MB at baseline comprised those from PredictAI, namely patient’s age, presence of metastasis, hemoglobin levels, platelet count, leukocyte count, and serum creatinine levels. Results: The TESEO cohort used for the external validation comprised 2,179 patients (51.9% male, mean age = 64.72 ± 11.34 years) with active cancer and VTE under anticoagulant treatment. The median (Q1, Q3) follow-up duration across patients was 6 (2, 14) months. Considering the ROC-AUC values for the three tested models (table 1), the results confirm a statistically significant validation and best performance of the logistic regression (ROC-AUC = 0.59; 95% CI = 0.53, 0.65; P = .002) approach. Conclusions: Our results support the first external validation of a risk assessment model of bleeding specifically developed for CAT. This model may guide objective decisions regarding the extension and dosing of anticoagulant therapy in this population. Future steps should involve further validation of the present model in other patient cohorts and the generation of a clinical tool to facilitate its use in healthcare settings. [Table: see text]