A comparative analysis of heterogeneity in lung cancer screening effectiveness in two randomised controlled trials

Clinical trials demonstrate that screening can reduce lung cancer mortality by over 20%. However, lung cancer screening effectiveness (reduction in lung cancer specific mortality) may vary by personal risk-factors. Here we evaluate heterogeneity in lung cancer screening effectiveness through traditional sub-group analyses, predictive modelling approaches and machine-learning in individual-level data from the Dutch-Belgian lung cancer screening trial (NELSON; 14,808 participants, 12,429 men, 2377 women, 2 persons with an unknown sex) and the National Lung Screening Trial (NLST; 53,405 participants, 31,501 men, 21,904 women). We find that screening effectiveness varies by pack-years (screening effectiveness ranges across trials: lowest groups = 26.8-50.9%, highest groups = 5.5-9.5%), smoking status (screening effectiveness ranges across trials: former smokers = 37.8-39.1%, current smokers = 16.1-22.7%) and sex (screening effectiveness ranges across trials: women = 24.6-25.3%; men = 8.3-24.9%). Furthermore, screening effectiveness varies by histology (screening effectiveness ranges across trials: adenocarcinoma = 17.8-23.0%, other lung cancers = 24.5-35.5%, small-cell carcinoma = 9.7%-11.3%). Screening is ineffective for squamous-cell carcinoma in NLST (screening effectiveness = 27.9% (95% confidence interval: 69.8% increase to 4.5% decrease) mortality increase) but effective in NELSON (screening effectiveness = 52.2% (95% confidence interval: 25.7-69.1% decrease) mortality reduction). We find that variations in screening effectiveness across pack-years, smoking status, and sex are primarily explained by a greater prevalence of histologies with favourable screening effectiveness in these groups. Our study shows that heterogeneity in lung screening effectiveness is primarily driven by histology and that relaxing smoking-related screening eligibility criteria may enhance screening effectiveness.