Cancer
Development of algorithms to identify screen-detected lung, breast, and colorectal cancers in cancer registry-linked claims data Sarah Soppe* Sarah Soppe Sharon Hinton Peacock Allison Verbyla Walker Redd Caroline A. Thompson
Population-level screening is a critical tool to reduce cancer burden through earlier detection and improved survival. While trends in screening rates in the general population can be documented through national surveys, less is known about the proportion of cancer patients whose cancer was diagnosed as a result of screening. The prevalence of “screen-detected cancers” is an underused but important metric that could be abstracted from routine data to evaluate screening effectiveness over time, yet validated algorithms for classifying screen-detected cancers in claims have not been established. Thus, this project aimed to develop cancer site-specific algorithms to classify screen-detected lung, breast, and colorectal cancers in SEER-Medicare data to estimate the proportion of screen-detected cancers. The study population included patients at least 66 years old diagnosed with their first invasive cancer between 2008 and 2017, excluding those not of a screening-eligible age and those without at least 12 months of pre-diagnosis continuous enrollment in fee-for-service Medicare Parts A/B. Algorithmic classification depended on the type and order of procedure codes in the pre-diagnostic lookback period (e.g., codes for low-dose computed tomography in the 3 months before lung cancer diagnosis), considering the commonly used diagnostic investigations for each site. In our representative sample of older adults with cancer, less than 1% of lung cancer diagnoses were screen-detected compared to 83% of breast and 22% of colorectal cancers. Screen-detected cancer prevalence was also examined by calendar year, cancer stage, age at diagnosis, race and ethnicity, and sex, among other clinical and demographic variables. These algorithms will be applied in statewide multi-payer claims linked to registry data to assess their implementation among younger age groups, while validation using electronic health records as a reference standard will enable further algorithm refinement.