Personalised screening intensity based on existing lung cancer risk and spirometry

Background Existing targeted lung cancer screening programmes are either annual or biennial. We tested the hypothesis that existing risk models can inform personalised screening intervals to improve cost effectiveness and reduce interval cancers. Spirometric airflow obstruction is a biomarker of lung cancer risk available in some cohorts, so we also explored whether this could add predictive value for stratification. Methods We tested the performance of risk models to predict next-round risk among baseline-negative participants in annual screening in Manchester and National Lung Screening Trial (NLST) cohorts. Models were adapted in the ECOG-ACRIN substudy of the NLST to investigate whether incorporating the spirometric percentage of predicted forced expiratory volume in one second (FEV1%) improved discrimination and screening efficiency at next-round screens. Results In Manchester, where a PLCO m2012 score >1.51% was used to determine eligibility, cancer was detected in 1.1% of those with negative baseline scans after a 1y interval (n=45/4,136), and 2.5% among those with risk ≥6% (n=27/1,082). The same trend was seen in NLST, but with lower risk overall. PLCO m2012 discriminated next-round risk with AUC 0.72 (0.65–0.79), with potential superiority of the LCRAT+CT neg model with AUC 0.75 (0.68–0.82) on external validation. The addition of FEV1% as a predictor variable modestly improved discrimination (ΔAUC +0.01 to +0.04). Discussion Baseline risk could be used to extend intervals for substantial numbers of lower-risk participants, reducing cost and screening harms, while focusing resources on those at highest risk to avoid delayed detection. Adding FEV1% is unlikely to meaningfully contribute to next-round risk prediction.