(158) Predictors Of The Utilization Of Lung Cancer Screening With Low-Dose Computed Tomography (LDCT) In The United States (US) Commercially Insured Population

Background: Lung cancer is one of the deadliest cancers as it is mostly detected at a late stage. Lung cancer screening reduces lung cancer mortality and is recommended by the US Preventive Services Task Force, yet adoption is low. Intervention trials to improve screening require large sample sizes since screening is a rare endpoint so proxy measures that are intermediates to screening may be useful to advance evidence-based interventions to increase screening.

Objectives: To identify the predictors of lung cancer screening in a large administrative claims database that could be used as secondary, better-powered endpoints in lung cancer prevention trials.

Methods: This retrospective cohort study included 383,045 smokers aged 50-65 who were free from a pulmonary neoplasm before or at the time of the first outpatient encounter during 2021 (index) and had continuous enrollment for ≥1 year before index from the US Merative MarketScan® Commercial Claims/Encounters database. The cumulative incidence of lung cancer screening with LDCT within 1 year post index was estimated using the Kaplan-Meier method. An ensemble model, AutoGluon-tabular classifier, which stacks multiple different machine learning models, was used to identify the predictors using all patients with a screening (n=14,535) and 10% of random sampling of those without screening (n=36,822). The cohort was divided into training (70%), validation (15%), and test (15%) sets. The model input features included patients’ age, sex, location state, and insurance plan type at index, history of Elixhauser comorbid conditions, and outpatient visits, provider types, service categories, and procedure groups within 1 year before index. A simplified model with the top 10 important features was developed. The area under the receiver operating characteristic (AUROC) curve was calculated on the test set.

Results: The 1-year cumulative incidence of lung cancer screening with LDCT was 4.19% (95% confidence interval: 4.12%, 4.25%) overall and 0.96% in those aged 50-54. Massachusetts had the highest incidence (9.30%). AUROCs were similar in the models with top 10 features and all features (0.77 vs. 0.78). The top predictors included age, thoracic chest CT scan procedures, state, chronic pulmonary disease history, obesity history (negative predictor), and use of other preventive medical services and pulmonary disease providers. In the subpopulation without lung cancer screening with LDCT before index, the top predictors were similar, but the screening incidence (2.39%) and AUROCs in the models with top 10 features (0.68) and all features (0.69) were lower.

Conclusions: Lung cancer screening with LDCT is low in the US commercially insured patients. Increasing preventive medical service use may improve the lung cancer screening adoption.