Introduction:  Current eligibility criteria for lung cancer screening by LDCT exclude approximately 50% of lung cancer cases whilst including many individuals at low risk who may not benefit from screening. Biomarkers have the potential to enhance smoking-based risk assessment tools. We previously developed a fit-for-purpose protein biomarker panel (INTEGRAL panel) comprising 21 proteins selected to inform lung cancer screening. We aimed to develop a biomarker-based risk model for use in LDCT-screening using a robust study design in the target population.
Methods: The Lung Cancer Cohort Consortium (LC3) involves 25 population cohorts from around the world with pre-diagnostic blood and risk factor information on up to 3,000,000 study participants, including 70,000 incident lung cancer cases. We designed 14 independent case-cohorts in LC3, including 2,305 randomly sampled sub-cohort participants and 1,390 incident lung cancer cases. All participants had a history of smoking, and cases were diagnosed within three years of blood collection. We developed the protein-based INTEGRAL-Risk model using data from seven case-cohorts, incorporating age, smoking history, and absolute concentrations of 13 proteins. Model testing was performed in seven independent cohorts by assessing model calibration and discrimination and comparing performance with existing risk assessment tools.
Results: In the testing set (1,161 sub-cohort participants and 583 lung cancer cases), the INTEGRAL-Risk model demonstrated superior discrimination for lung cancer incidence over one year compared to the smoking-based PLCOm2012 model (AUC: 0.88 [95% CI: 0.85–0.90] vs. 0.79 [95% CI: 0.75–0.83]; p<0.001). The model was well-calibrated for one-year lung cancer risk prediction (expected/observed ratio: 0.86 [95% CI: 0.61–1.25]). The USPSTF screening criteria identified 25% of individuals with smoking exposure as eligible for LDCT screening, capturing 63% of lung cancer cases within one year. Using a risk threshold that selected an equivalent 25% of participants, the INTEGRAL-Risk model identified 85% of lung cancers as eligible for screening. Stratified analyses by race/ethnicity demonstrated consistent model performance across white (AUC: 0.88 [95% CI: 0.84–0.92]), black (AUC: 0.90 [95% CI: 0.81–0.97]), and Asian (AUC: 0.88 [95% CI: 0.82–0.93]) participants.
Conclusion: This study represents a major effort from an international consortium to develop a biomarker-based risk prediction tool for use in LDCT screening. The protein-based INTEGRAL-Risk model improves upon existing risk assessment tools for identifying high-risk individuals with a history of smoking. These findings underscore the potential of biomarker-based risk models to enhance LDCT screening efficiency by better targeting individuals likely to benefit.