Six-year Performance of Risk-based Selection for Lung Cancer Screening in the Manchester Lung Health Check Cohort

Objective: Risk prediction models are used to determine eligibility for targeted lung cancer screening. However, prospective data regarding model performance in this setting are limited. Here we report the performance of the PLCO risk model, which calculates 6 year lung cancer risk, in a cohort invited for lung cancer screening in a socioeconomically deprived area.

Methods And Analysis: Calibration (expected/observed (E/O) lung cancer diagnoses over 6 years) and discrimination (area under the receiver operating characteristic curve) of PLCO and other models was performed in Manchester Lung Health Check (M-LHC) participants, where PLCO ≥1.51% was used prospectively to determine screening eligibility. Lung cancers diagnosed by any route were captured within 6 years of risk assessment, for both screened and non-screened participants. Performance of a range of models was evaluated.

Results: Out of 2541 attendees, 56% were high-risk (n=1430/2541) and offered screening; 44% were low-risk (n=1111/2541) and not screened. Over 6 years, 7.3% (n=105/1430) and 0.9% (n=10/1111) were diagnosed with lung cancer in the high and low-risk cohorts, respectively (p<0.0001). Risk was underestimated in both high-risk, screened (E/O 0.68 (0.57-0.82)) and low-risk, unscreened groups (E/O 0.61 (0.33-1.14)). Most other models also underestimated risk.

Conclusion: Risk-based eligibility using PLCO successfully classified most eventual lung cancer cases in the high-risk, screened group. Prediction models generally underestimated risk in this socioeconomically deprived cohort, irrespective of screening status. The effect of screening on increasing the probability of lung cancer diagnosis should be considered when interpreting measures of prediction model performance.