Generalizability of national survival estimates to adolescent and young adults with cancer at an urban safety-net institution.

Abstract

140 Background: Survival estimates for adolescent and young adults (AYAs) with cancer in the US are largely derived from the Surveillance, Epidemiology, and End Results (SEER) program, which does not include certain geographic areas with diverse populations such as Texas. The generalizability of these estimates to other populations is unknown but may be useful for identifying high-risk populations. Therefore, we aimed to assess the generalizability of survival estimates for AYAs with cancer in SEER to AYAs with cancer at an urban safety-net institution in Texas. Methods: We used data from the SEER program and the JPS Oncology and Infusion Center, an accredited Comprehensive Community Cancer Program. Eligible individuals in both populations were aged 18 – 39 years and diagnosed with a first primary cancer between 2008 and 2015. We computed entropy-balanced weights to adjust the baseline distribution of age, gender, race/ethnicity, insurance status, marital status, cancer type, and stage of the SEER population to our safety-net population. We used these weights to compute Kaplan-Meier estimates of expected survival for our safety-net population. We subsequently estimated the absolute survival difference between expected and observed (i.e. actual) survival and corresponding 95% confidence limits (CL). Results: Our safety-net population comprised 833 AYAs with cancer, of whom 60% were female, 62% were racial/ethnic minorities, and 66% were uninsured at diagnosis. The most common cancers were hematologic (17%), female genital (16%), thyroid (11%), and breast (11%). Metastasis was observed at diagnosis for 24% of AYAs. The expected 5-year survival for the safety-net population based on SEER data was 77%, whereas observed survival was 68%, which equated to an absolute survival difference of 9.5% (95% CL: 5.3%, 14%) after 5 years. Conclusions: Our results suggest that survival estimates of AYAs with cancer in SEER may not be generalizable to AYAs with cancer at an urban safety-net institution in Texas. Lack of generalizability may be relevant to other populations not represented in SEER. The reasons for poor survival of AYAs with cancer at safety-net institutions should be explored.