Abstract
Background Prior cancer-associated thrombosis (CAT) epidemiological studies focused on specific populations or single healthcare systems, limiting their generalizability. We assessed the incidence and risk factors for acute venous thromboembolism (VTE) and its association with mortality in patients with new cancer diagnosis in a nationwide, contemporary electronic health record (EHR) cohort representative of the US population.
Methods Study data came from Epic Cosmos, a collaborative dataset of Epic health systems including more than 300 million patient records from over 1,748 hospitals and 40.7k clinics from all 50 states and D.C. as of July 2025 (cosmos.epic.com).
We included adults with newly diagnosed solid and hematologic neoplasms treated at eligible health systems with oncology departments and longitudinal records from 1/1/2018 to 12/31/2023. Patients were followed until the earliest onset of outcome event, censor date (a gap of 6 months without face-to-face [F2F] encounters), death, or 7/9/2025. We excluded individuals with acute VTE in the last 6 months or on active anticoagulation at index date. Cancer diagnosis and acute VTE were identified using validated computable phenotype algorithms that required repeated ICD-CM codes from F2F visits. VTE included pulmonary embolism (PE), deep vein thrombosis (LE-DVT), and upper-extremity deep vein thrombosis (UE-DVT). Relevant baseline and treatment variables were extracted from the Cosmos database.
We used cumulative incidence with death as a competing risk for VTE incidence. We used multivariable Cox regression to assess associations between baseline variables and VTE risk; candidate variables were selected via LASSO. Cancer treatment onset was modeled as a time-varying covariate (TVC). We then modeled the association between VTE onset (TVC) and mortality within the first year after adjusting for other confounders.Results There were 1,628,626 patients (median age 66 [58-74]; 52.7% female; 80.7% White; 12.3% Black; 5.6% Hispanic) from 173 healthcare systems. The most frequent cancers were breast (22.5%), prostate (16.7%), lung (9.3%), and colorectal (6.6%); 28.7% were metastatic; and 34.8% received systemic therapy within 3 months of cancer diagnosis (treated cohort).
Before cancer diagnosis, 2.8% of patients had a remote VTE history (>6 months) and 3.0% had a recent event (<6 months). After excluding recent events, the incidence of new VTE at 6 and 12 months was 2.7% and 3.7%, respectively. In the treated cohort, the incidence at 6 and 12 months after therapy was 4.4% and 5.7%, respectively. The trend in annual VTE incidence did not increase during the 6 years. Individual VTE subtypes followed similar patterns, except UE-DVT had notably higher incidence in acute leukemias.
In the multivariable analysis, cancer type, treatment, stage, age, sex, race, body mass index, hospitalization, history of paralysis, history of VTE, blood count, and albumin were independently associated with VTE risk. Social vulnerability index, rurality, geographic region, marital status, language, renal, and hepatic function did not have significant associations. Compared with breast cancer (1.8%), pancreatic (12.0%; adjusted hazard ratio [HR] 3.3 [3.1-3.4]), bile/gallbladder (10.8%; HR 2.6 [2.5-2.6]), and upper gastrointestinal (GI) cancer (9.5%; HR 2.5 [2.4-2.6]) were associated with the greatest risk of VTE at 12 months. Compared with those untreated, patients receiving cytotoxic chemotherapy (HR 1.9 [1.9-1.9]), immune checkpoint inhibitor (HR 1.2 [1.2-1.3]), and targeted therapy (HR 1.2 [1.2-1.2]) had higher risk, while those on endocrine therapy (HR 0.8 [0.8-0.8]) had lower risk. Other notable associations included metastatic disease (HR 3.2 [3.1-3.3]) and remote VTE history (HR 2.3 [2.2-2.4]).
After adjusting for baseline confounders and systemic therapy initiation, the occurrence of PE (HR 2.4 [2.4-2.5]), LE-DVT (HR 2.1 [2.1-2.2]), and UE-DVT (HR 2.3 [2.2-2.4]) within the first year of cancer diagnosis were each independently associated with increased mortality.
Conclusions In this nationwide cohort of 1.6 million people with newly diagnosed cancer, patients with pancreatic and upper GI cancers, advanced stage, prior history of VTE, and those receiving cytotoxic chemotherapy had the highest risk of VTE. This study demonstrates the feasibility of using a nationwide EHR database to conduct a wide-scale epidemiological study and provides the largest evaluation of CAT in US patients to date.
