Estimating county-level lung cancer screening prevalence using linked claims and multi-source data.

Abstract

158 Background: Accurate estimation of county-level lung cancer screening prevalence at geographic levels smaller than states is limited by the lack of individual-level smoking history data to assess eligibility, with existing publicly available smoking data lacking geographic and/or demographic granularity. We developed the Smoking history and County-level Age-Stratified Lung cancer Screening Eligibility (SCALE) framework to estimate county-level screening prevalence using insurance claims and population-based data. Methods: We used 2016-2020 statewide multi-payer claims data from the UNC Lineberger Cancer Information and Population Health Resource (CIPHR) to construct analytic cohorts of cancer-free, insured North Carolinians aged 55-79 years, with 15 months of enrollment coverage, to assess receipt of low-dose computed tomography (LDCT) screening. Two approaches were evaluated to estimate each county’s screen-eligible population based on either Behavioral Risk Factor Surveillance System (BRFSS) data or National Cancer Institute Small Area smoking estimates (NCI SAE), combined with county-level United States Census data. SCALE-A, used age-specific state-level smoking prevalence from BRFSS, and SCALE-C used NCI SAE county-level adult smoking prevalence without age stratification. We assessed agreement between methods using Pearson correlation, intraclass correlation coefficient (ICC), Lin’s concordance correlation coefficient (CCC), and Bland-Altman analysis. Results: The estimated statewide prevalence of lung cancer screening among the insured increased from 2016 through 2019 and declined in 2020. In 2019, SCALE-A estimated 11.95% of eligible individuals were up to date with screening, compared with 16.51% from SCALE-C; the 2019 North Carolina BRFSS estimate among insured adults fell between these at 12.39%. Across counties, both approaches showed similar spatial patterns, but SCALE-A consistently produced lower prevalence estimates than SCALE-C. These estimates were highly correlated (Pearson’s r=0.97) but agreement was modest (ICC=0.62; CCC=0.67). Bland-Altman analysis showed a systematic difference with SCALE-A lower than SCALE-C. Conclusions: The SCALE framework enables county-level estimation of lung cancer screening prevalence based on synthesis of publicly available but incomplete smoking data. Although distinct data sources yielded highly correlated county-level estimates, they differed in magnitude, with SCALE-A producing consistently lower estimates than SCALE-C. Our findings highlight the importance of methodologic assumptions in studies of cancer screening prevalence and suggest that the SCALE framework can support small area lung cancer screening surveillance even when granular smoking history data are unavailable.