PERSONALIZING LUNG CANCER SCREENING RECOMMENDATIONS ACROSS THE SPECTRUM OF BENEFIT

Purpose:

Clinicians need guidance on how to communicate about lung cancer screening (LCS) with eligible patients who have different levels of lung cancer risk and differing preferences. We developed a natural history microsimulation model that provides evidence on how to personalize LCS recommendations by taking into account both: 1) individual-specific health gain estimates with LCS and 2) Heterogeneity in patient preferences about the benefits and harms.

Method:

We estimated lifetime quality-adjusted life year (QALY) gains comparing 3 annual screens with no screening, then examined the effect of different assumptions about the disutilities associated with the screening test itself, follow-up diagnostic evaluation (e.g., invasive procedures), and treatment. Model probabilities were based on published lung cancer incidence, other-cause mortality, and lung cancer detectability models, individual-level data from 2 large randomized trials on lung cancer screening (NLST and PLCO), and the Surveillance, Epidemiology and End-Results (SEER) cancer registry. We quantified preferences using literature-derived utilities, varying utilities across a plausible range in our primary analyses to understand the effect of preferences on the net benefit of screening.

Result: 

Model predicted lung cancer incidence and mortality outcomes were consistent with those observed in the trials. As expected, QALY gains varied substantially across eligible smokers with different levels of lung cancer risk (range in base-case analysis: 2.8 QALYs lost per 100 screened for patients older than 70 in the lowest risk quintile to 6.5 QALYs gained per 100 screened among patients 61-70 in the highest risk quintile). In our primary analysis, NLST patients with lower lung cancer risk  (i.e., < 3rd decile of baseline risk) experienced a smaller benefit that was highly sensitive to assumptions about preferences -- while those at higher lung cancer risk (i.e., > 3rd decile) experienced net benefit even with unfavorable preference assumptions.

Conclusion:

If these findings are confirmed in nationally representative samples of eligible smokers (in progress) our study can provide important guidance about tailoring LCS recommendations to both individual risk and preferences. Clinicians might suggest screening as the best option for most eligible patients at higher lung cancer risk (i.e., > 3rd decile of lung cancer risk). On the other hand, formulating default personalized recommendations for eligible patients at lower risk is harder to justify, because the benefit is so sensitive to preferences and other patient characteristics.