Epidemiology of cancer-associated bleeding in a nationwide electronic health record database

Abstract Background People with cancer frequently experience hemostatic imbalances that result in excessive clotting or bleeding. While the risk of thrombosis is well-studied, the lack of epidemiological data on cancer-associated bleeding limits thromboprophylaxis implementation. We examined the incidence and risks for bleeding events, and its association with mortality in patients with newly diagnosed cancers in a nationwide, contemporary electronic health record (EHR) cohort representative of the US population. Methods Data used in this study was derived from Epic Cosmos, a dataset created in collaboration with a community of Epic health systems representing more than 300 million patient records from over 1700 hospitals and 40k clinics as of July 2025. The community represents patients from all 50 states and D.C. (cosmos.epic.com). We included adults with newly diagnosed solid and hematologic neoplasms from 1/1/2018 to 12/31/2023. Patients were followed until the earliest of first outcome event, censor date (a gap of 6 months without face-to-face encounter), death, or 7/9/2025. Active cancer diagnosis was ascertained using validated computable phenotype algorithm, whereas bleeding events were defined using previously validated ICD-10-CM codes from inpatient billing diagnosis only (93% precision; 86% recall). We assessed cumulative incidence with death as a competing risk. We used multivariable Cox regression to assess associations between baseline variables and bleeding risk, where candidate variables were selected via LASSO. Cancer therapy and anticoagulant (AC) initiations were modeled as time-varying covariates (TVC). We also assessed the association between bleeding events (TVC) and mortality within the first year after adjusting for other confounders. Results Bleeding occurred within 1 month prior to cancer diagnosis in 7.3% (n=131,813) patients (mostly colorectal, bladder, uterine, and lung cancers). We excluded them from subsequent analysis because they represented bleeding that led to cancer diagnosis than treatment complication. The remaining analytic cohort included 1,668,574 patients with newly diagnosed cancer from 173 health systems (median age 67 [58-74]; 52.7% female; 81.3% White; 12.1% Black; Hispanic 5.4%). The most frequent cancers were breast (22.9%), prostate (17.2%), and lung (9.7%); 28.7% had metastasis; 34.6% (n=577,233) received systemic therapy within 90 days (treated cohort). In the overall cohort, the bleeding incidence at 6 and 12 months after cancer diagnosis was 2.6% and 3.9%, respectively. The majority in the first year were gastrointestinal (GI) bleeds (2.1%), while intracranial hemorrhage (ICH) occurred in 0.5%. After exclusion for baseline AC, the bleeding incidence was 2.4% and 3.6%, respectively. In the treated cohort, the bleeding incidence was 2.9% and 4.5%, respectively. In the multivariable analysis, age, sex, cancer type, stage, systemic therapy, AC, antiplatelet, hospitalization, prior bleeding, NCI comorbidity, and selective labs were independently associated with bleeding by 12 months. Compared with breast cancer, acute leukemia (adjusted hazard ratio [HR] 5.1 [4.6-5.8], hepatocellular carcinoma (HR 5.1 [4.8-5.4]), and upper GI cancer (HR 5.1 [4.9-5.4] were associated with the greatest risk. Other notable associations included brain metastasis (HR 2.6 [2.4-2.7]), non-brain metastasis (HR 1.7 [1.6-1.7]), prior bleeding (HR 1.5 [1.5-1.5]), and higher NCI comorbidity (HR 1.4 [1.4-1.4]). Baseline hemoglobin <10 g/dL, platelet <50 X 109/L, bilirubin >2.4 mg/dL, albumin <3.5 g/dL, and eGFR <30 were independently associated with bleeding risk (HRs between 1.3-1.7). When examining the TVCs, AC initiation was associated with modest risk (HR 1.7 [1.7-1.8]), while systemic therapy initiation was associated with minimal risk (HR <1.3). The overall 1-year mortality rate was 8.2%. After adjusting for confounders, both ICH (HR 5.5 [5.3-5.7] and GI bleeding (HR 3.5 [3.4-3.5]) within the first year were independently associated with increased mortality. Conclusion In this EHR cohort of 1.6+ million adults with newly diagnosed cancer across all stages and therapies, the annual bleeding incidence was 3.9% in overall and 4.5% in systemic therapy cohort, with GI bleeding being the most common. Compared with thrombosis, cancer-associated bleeding shares similar yet distinct risk factors. To date, this study provides the largest contemporary epidemiological evaluation of bleeding risk in cancer patients in the US.