Comparative benchmarking of U.S. oncology populations and survival using a multi-health system EHR data.

Abstract

e23401 Background: Real-world oncology research relies on electronic health record (EHR) data that are representative of the U.S. cancer population and enable reliable survival estimates. Truveta receives EHR data from U.S. health care systems that are refreshed daily and used for oncology real-world evidence generation. This study benchmarks oncology populations in Truveta Data against the Surveillance, Epidemiology, and End Results (SEER) Program to assess demographic representativeness and concordance of overall survival (OS) estimates across six solid tumor types (bladder, breast, cervical, colorectal, lung, and prostate). Methods: 3.8M patients with a qualifying cancer diagnosis were identified in Truveta Data. Patients included in the study were diagnosed between January 1, 2017 and December 22, 2025 and followed from diagnosis until death or last encounter. Patients included had at least one encounter in the year prior to diagnosis and data characterizing their tumor (stage, grade, or performance status). SEER data were accessed on December 22, 2025. All patients were aged ≥18 years and had one primary tumor. Demographics were summarized using counts and percentages. OS was estimated at prespecified time points (1–5 years) by cancer type and subgroup (age group, sex, and race/ethnicity). Results: The study included 368,774 Truveta patients (bladder n = 10,350; breast n = 143,883; cervical n = 7,023; colorectal n = 42,797; lung n = 100,190; prostate n = 64,531) and 3,563,896 SEER patients. Overall, 58% of Truveta and 55% of SEER patients were aged ≥65 years. Across all cancers, demographic distributions from Truveta were broadly concordant with SEER. The proportion of non-Hispanic White patients was higher in Truveta (63–85%) than in SEER (51–85%), with differences of 4–12% across cancer types. Conversely, the proportion of Hispanic patients was lower in Truveta (3–16%) than in SEER (7–25%), with differences of −3% to −9%. Distributions for other race/ethnicity categories, age groups, and sex differed minimally (≤5 absolute percentage points). Estimated 1-year OS in Truveta was 81% for bladder cancer, 97% for breast, 89% for cervical, 96% for prostate, 86% for colorectal, and 68% for lung. Absolute differences in 1-year OS relative to SEER were ≤2 percentage points for bladder, breast, cervical, and prostate cancers. Truveta estimates were 8% higher for colorectal and 26% higher for lung. Conclusions: OS estimates from Truveta were broadly concordant with SEER across multiple cancer types. Differences varied by cancer type, underscoring the importance of reporting time-specific survival estimates with demographic distributions when interpreting results from real-world and registry-based data sources. These findings support the use of real-world EHR data to examine associations between risk factors, treatment patterns, outcomes, and in-depth longitudinal care of cancer patients.