Abstract
IMPORTANCE: Venous thromboembolism (VTE) is associated with increased mortality and morbidity in patients with cancer. Existing risk prediction models are typically validated within individual sites, a fragmented approach that limits clinical adoption. OBJECTIVE: To validate the electronic health record cancer-associated thrombosis (EHR-CAT) score compared with the benchmark Khorana score in a contemporary cohort of patients with cancer across the nation, before and after treatment, excluding those at high risk of bleeding. DESIGN, SETTING, AND PARTICIPANTS: This prognostic study included patients in a nationwide longitudinal EHR database from January 2018 to December 2023 with follow-up continuing to April 2025. Patients with newly diagnosed, invasive, solid, or hematologic malignant neoplasms (defined using validated International Statistical Classification of Diseases, Tenth Revision, Clinical Modification [ICD-10-CM] algorithms) receiving systemic therapy (defined using the first antineoplastic medication) were included. Those with recent history of acute VTE diagnosis or anticoagulant prescription were excluded. EXPOSURES: Demographics, risk model variables, and common anticoagulant trial exclusion criteria (as a proxy for identifying people at high risk of bleeding) were extracted on or before index therapy initiation date. MAIN OUTCOMES: Incident VTE and bleeding outcomes at 6 months were defined using validated ICD-10-CM algorithms. RESULTS: A total of 732 594 patients (median [IQR] age, 65.0 [56.9-73.0] years; 425 124 female [58.0%]; 25 634 Asian [3.5%], 94 269 Black [12.9%], 48 266 Hispanic [6.6%], 583 047 White [76.9%]) with active cancer receiving systemic therapy between 2018 and 2023 from 184 health systems were identified. With a median (IQR) follow-up of 676 (340-1151) days, the incidence of 6-month VTE, bleeding, and mortality was 4.7% (34 499 patients), 3.7% (26 993 patients), and 8.4% (60 239 patients), respectively. Bleeding risk was higher in the 26.0% of patients (190 413) meeting anticoagulant trial exclusion criteria (7.2% vs 2.4%; hazard ratio, 2.5 [95% CI, 2.5-2.5]). The EHR-CAT score stratified patients into discriminative risk groups (C statistic, 0.70-0.71) both before and after exclusion for bleeding risk. When compared with the benchmark Khorana score (C statistic, 0.63), EHR-CAT reclassified 20% of patients into revised categories with improved prediction accuracy. Furthermore, EHR-CAT had consistent calibration in subgroups by age, sex, race, ethnicity, and individual health system sites. CONCLUSIONS: This prognostic study of the EHR-CAT risk score demonstrated the external validity and feasibility of using readily available structured EHR data to estimate VTE risk in patients with cancer.