05/12/2025

Comparison of Demographics and Clinical Characteristics Using Real World Data from Tempus Multimodal Database and SEER Cancer Registry Across 17 Solid Cancer Cohorts

ISPOR 2025 PRESENTATION
Authors Katie Mo, Xifeng Wang, Gary Grad, Emilie Scherrer, Chithra Sangli

OBJECTIVES: The Tempus database consists of de-identified longitudinal clinical data collected and abstracted from electronic health records of cancer patients who undergo sequencing at Tempus from ~65% of US academic medical centers and several hundred community institutions. This is a valuable resource for real world evidence generation especially when both clinical and biomarker data are required. The Surveillance, Epidemiology, and End Results Program (SEER) database is a well-known source for population cancer surveillance and research in the US. We aim to benchmark the Tempus database against the SEER cancer registry.

METHODS: Patients diagnosed with solid tumors between 2016 and 2021 were included in this analysis. Patients were required to have at least one additional clinical event 90 days following diagnosis. Baseline demographic, clinical and treatment characteristics were compared to the SEER 17 database.

RESULTS: A total of 63,520 patients across 17 unique cancer types were analyzed from the Tempus database. Cancer incidence was similar in both databases. The Tempus database had larger representation from clinical sites in the Midwest, whereas the West was heavily represented in the SEER database. Patients in the Tempus database were on average younger, but had more advanced cancer staging as compared to patients from the SEER database. In addition, Tempus patients were more diverse based on self-reported race, and treatment data was more complete in the Tempus database.

CONCLUSIONS: Our analysis finds that the Tempus and SEER databases are generally comparable with respect to demographics and clinical characteristics among cancer patients, while the Tempus database has more granularity on treatment data. There is greater representation of late stage disease in the Tempus database attributable to sequencing patterns in clinical care. These similarities and differences arise from the respective data generation mechanisms and should be considered in the design of real world data studies.

VIEW THE PUBLICATION