Applying causal inference methods to improve estimation of the real-world benefits and harms of lung cancer screening

ABSTRACT Randomized controlled trials have demonstrated that low-dose computed tomography can substantially reduce lung cancer mortality, albeit at the cost of relatively high rates of false positives and complications from downstream procedures. However, systematic differences between trial and general populations eligible for lung cancer screening raise concerns about the relevance of trial findings for guiding the development and dissemination of lung cancer screening programs in clinical practice. Despite clear recommendations from the United States Preventative Services Task Force, lung cancer screening uptake and adherence remain low. Several studies have documented dramatic and selective attrition across the screening continuum – where about 10-20% of eligible individuals undergo lung cancer screening and of those, only about 40-60% are up-to- date with their annual screening at 15 months. When the benefits and harms of an intervention vary across subgroups and there is selective attrition, the balance of population-level benefits and harms is expected to change. As such, there is an urgent need to better characterize the effectiveness of lung cancer screening with low-dose computed tomography when applied to individuals outside of clinical trial settings. The primary objective of this proposal is to generate real-world evidence of the benefits and harms of lung cancer screening with low-dose computed tomography that explicitly considers the characteristics of populations at each step of the screening continuum, from the screening-eligible, to the screened, to the adherent. To address this objective, we will use cutting-edge causal inference methods, including trial transport and target trial emulation using real-world data, which can avoid the potential for time-related biases. To carry-out our proposed analyses, we will draw upon individual-level data from the randomized National Lung Screening Trial, as well as four real-world datasets including the National Health and Interview Survey, the Behavioral Risk Factors Surveillance Survey (Lung Cancer Screening Module), a 20% nationwide sample of Medicare claims, and the North Carolina Lung Screening Registry. Findings from this study will generate information about the effectiveness of lung cancer screening in real-world settings that can be used by patients, providers, and policymakers. This work will enhance the evidence base used by policymakers to update screening recommendations and refine decision aids to support communication with patients about screening net- benefits during shared decision-making. Ultimately, this work will support efforts to improve the delivery of lung cancer screening at the population level.