Prediction Augmented Screening Initiative

Overview

Lung cancer is responsible for more deaths in the United States than breast, prostate and colon cancer combined and is the number one cancer killer of Veterans. This is because lung cancer is usually diagnosed when the disease has spread, and cure is less likely. Lung cancer screening (LCS) finds cancer at an earlier stage when it is curable, yet only 20% of eligible Veterans have been screened. Uptake is even lower among Black Veterans despite higher lung cancer risk. Using prediction models to identify high-benefit people for whom LCS should be encouraged improves efficiency and reduces disparities. Moreover, it is more patient-centered as shared decision-making conversations can be tailored with personalized information. The US Preventive Services Task Force has called for research to demonstrate that prediction-augmented LCS can be feasibly implemented at the point-of-care. The investigators propose for VA to lead this effort with a large-scale pragmatic clinical trial to show that prediction-augmented LCS is both feasible and improves LCS uptake.

Background: Despite large-randomized trials demonstrating the mortality benefit from lung cancer screening (LCS) and a recommendation from the US Preventive Services Task Force (USPSTF) and VHA since 2013, only 20% of eligible Veterans have received LCS. Uptake is even lower among Black Veterans despite higher lung cancer risk. Current USPSTF eligibility criteria of age and smoking history are simple, but do not incorporate the heterogeneity in lung cancer risk and life expectancy across people and leads to exclusion of some persons, especially Blacks, with potential for high benefit from LCS. While the USPSTF acknowledged that using prediction models to augment simple eligibility criteria is more efficient and equitable, they stopped short of recommending prediction-augmented LCS, noting that a pragmatic trial was needed to demonstrate that prediction-augmented LCS can be feasibly implemented in real-world settings and assess its impact on LCS uptake. Significance: By demonstrating the real-world feasibility of prediction-augmented LCS, and its ability to improve LCS uptake especially in those of high-benefit, the VA as a learning healthcare system will influence national LCS guidelines and improve LCS outcomes both inside and outside the VA. Innovation \& Impact: Prediction-augmented LCS is based on strong evidence, yet implementing this approach would represent paradigm shift from typical preventive cancer screening. The proposed work is a unique opportunity for the VA to advance implementation of more equitable, personalized LCS by improving on the status quo of making broad 'one-size-fits-all' recommendations. The innovation is the advancement of primary care-facing and population management informatics tools that present individualized information on how strongly to encourage LCS, with the potential to be expanded to other cancer screenings. Specific Aims: 1. Conduct a pragmatic stepped wedge (site-level) factorial trial comparing usual care (USPSTF criteria) versus prediction-augmented LCS (supported by primary care-facing informatics tools, LCS team population management tools, external facilitation) on effectiveness at increasing LCS uptake. 2. Determine what drives implementation success of prediction-augmented LCS in various contexts, using mixed methods. Methodology: The investigators will utilize a factorial stepped wedge design in the Lung Precision Oncology Program network to establish the effectiveness and evaluate the implementation of precision-augmented LCS. Veterans assigned a PCP at a participating site and who meet inclusion criteria based on Clinical Data Warehouse data will be passively enrolled in the study: 1) patients meeting USPSTF criteria for LCS; OR 2) patients whose predicted LCS benefit exceeds a stringent high-benefit threshold of life gained with annual LCS (to capture high-benefit Veterans currently excluded by USPSTF criteria). The primary outcome is the probability of completion of an initial LDCT scan among eligible subjects defined as high-benefit during each quarter of the study (quarter assigned by date of the LDCT order as this event is most proximal to the LCS decision). The factorial design allows us to discern the effect on LCS uptake of primary care-facing vs LCS team population management tools. Secondary outcomes include 1) LDCT orders and completion among all eligible Veterans; 2) LCS orders and completion among high-benefit Black Veterans and among all eligible Black Veterans; 3) LDCT orders and completion among eligible but non-high-benefit Veterans and among those with high comorbidity; 4) Effects on lung cancer detection rates; 5) LCS outcomes (invasive procedures and complications) and projected cancer outcomes (e.g., lung cancer deaths avoided) among additional people screened due to the interventions, and care gaps in which LCS was ordered but not completed. Implementation evaluation will be guided by the i-PARIHS framework using mixed methods. This approach combines data drawn from across all trial sites (tool usage data, clinician surveys, facilitation data, patient interviews) to gain a broad view of implementation, with in-depth ethnographic assessments from 6 selected sites that will provide deeper insights into how and why implementation succeeded (or faced challenges) in different contexts. Our findings will inform creation of an implementation playbook to support enterprise-wide spread of prediction-augmented LCS. Next Steps / Implementation: The investigators will work with our operational partners to spread implementation of prediction-augmented LCS across the VA enterprise.