Highlights
- Lung cancer screening (LCS) uptake remains critically low despite robust evidence of mortality reduction; personalized approaches are needed to reach those who benefit most.
- Prediction-augmented shared decision-making (SDM) tools enable clinicians to tailor the strength of their recommendations based on individualised lung cancer risk and life expectancy.
- A large-scale study within the Veterans Affairs (VA) system demonstrated that implementing these tools significantly increases LCS uptake, particularly among high-benefit individuals (9.0 percentage point difference).
- Transitioning from binary eligibility criteria to benefit-based screening represents a paradigm shift toward precision population health.
Background
Lung cancer remains the leading cause of cancer-related mortality globally. The National Lung Screening Trial (NLST) and the NELSON trial have firmly established that annual screening with low-dose computed tomography (LDCT) can reduce lung cancer mortality by 20% to 24%. Despite these findings and the subsequent expansion of US Preventive Services Task Force (USPSTF) eligibility criteria in 2021, the real-world uptake of LCS remains suboptimal, often hovering below 15% among eligible populations.
A significant challenge in LCS implementation is the “screening paradox”: individuals at the highest risk for lung cancer, who stand to gain the most from screening, are often the least likely to undergo it due to socioeconomic barriers, limited healthcare access, or poor communication regarding the personalized benefits of the procedure. Furthermore, shared decision-making (SDM), while mandated for Medicare reimbursement, is frequently performed perfunctorily. Clinicians often lack the time or tools to distinguish between patients for whom screening is a “close call” (preference-sensitive) and those for whom the mortality benefit far outweighs the risks of false positives and overdiagnosis (high-benefit).
Key Content
The Evolution of Risk-Based Screening
Early LCS protocols relied on broad categorical criteria, such as age and pack-year smoking history. However, evidence-based medicine has shifted toward the use of multivariable risk-prediction models, such as the PLCOm2012 or the Lung Cancer Risk Assessment Tool (LCRAT). These models incorporate additional factors—including ethnicity, education level, body mass index, and personal history of respiratory disease—to more accurately predict an individual’s 6-year lung cancer risk. The integration of these models into clinical workflows allows for “benefit-based” screening, where the intensity of the clinical recommendation is proportional to the individual’s predicted absolute risk reduction.
Methodological Advancement: Prediction-Augmented SDM
The core innovation addressed in recent literature, specifically the study by Caverly et al. (2024), is the application of a decision support tool augmented by a prediction model. Unlike standard clinical reminders that merely flag an eligible patient, this tool provides clinicians with a categorized assessment of the patient’s predicted benefit from screening. Patients are classified into groups, such as “intermediate benefit” (where the decision is highly preference-sensitive) and “high benefit” (where the evidence strongly favors screening). This allows clinicians to use “nudges” or stronger recommendations for those in the high-benefit tier, ensuring that limited clinical resources are directed where they will have the greatest impact.
Analysis of the VA Interrupted Time Series Study
A pivotal quality improvement study was conducted at six Veterans Affairs sites between 2017 and 2019, involving 9,904 LCS-eligible veterans. The study employed an interrupted time series design to evaluate the impact of a prediction-augmented SDM tool on screening uptake.
Study Cohort and Demographics
The cohort was predominantly male (94%) and White (82%), with a median age of 64 years. Crucially, the population was almost evenly split between those predicted to have intermediate benefit (52%) and those with high benefit (48%). This distribution underscores the importance of being able to distinguish between these two groups in a clinical setting.
Primary Outcomes and Statistical Significance
Prior to the tool’s implementation, screening rates were low across both groups. Following implementation, there was a statistically significant increase in overall uptake. Most importantly, the tool successfully drove “benefit-based” uptake. High-benefit individuals had a predicted probability of screening of 24.8%, compared to 15.8% for intermediate-benefit individuals. The mean absolute difference of 9.0 percentage points (95% CI, 1.6%-16.5%) demonstrates that the tool helped clinicians prioritize the right patients.
Clinical Workflow Integration
The intervention utilized clinical reminders within the electronic health record (EHR) to prompt primary care clinicians and screening coordinators. This integration is vital; for prediction models to work in the real world, they must be embedded into the point-of-care workflow rather than existing as external calculators. The VA study suggests that when clinicians are presented with a clear indication of a patient’s high-benefit status, they are more likely to successfully navigate the SDM process and reach a decision to screen.
Expert Commentary
From a clinical and policy perspective, these findings are highly relevant. The move toward prediction-augmented SDM addresses several current gaps in LCS programs. First, it mitigates the “one-size-fits-all” approach to screening. For a 55-year-old with exactly 30 pack-years, the benefit-risk ratio is vastly different from a 70-year-old with 80 pack-years. The use of a prediction tool acknowledges this heterogeneity.
However, there are controversies and limitations to consider. The study population (Veterans) may not be fully representative of the general population, particularly women and non-veteran minority groups. Additionally, while the study showed an increase in uptake, a 24.8% uptake in the high-benefit group still leaves 75% of those most likely to benefit unscreened. This suggests that while decision support tools are necessary, they are not a panacea. Barriers such as patient skepticism, logistical hurdles, and the “stigma” of smoking-related illness require multi-faceted interventions.
There is also an ethical dimension to “tailoring the strength of encouragement.” Critics might argue that this could lead to paternalism, where clinicians push screening too hard on high-risk individuals, potentially infringing on the “shared” nature of the decision. Nevertheless, the consensus among experts is that providing more accurate, personalized information actually empowers the patient to make a more informed choice.
Conclusion
The integration of prediction-augmented shared decision-making tools represents a significant advancement in the effort to improve lung cancer screening uptake. By moving beyond binary eligibility and toward a benefit-based framework, healthcare systems can ensure that those at the highest risk receive the strongest encouragement to undergo life-saving LDCT. Future research should focus on expanding these tools to more diverse populations and examining long-term outcomes, such as stage-at-diagnosis and lung cancer-specific mortality, following the implementation of risk-based SDM.
