Development and Validation of a Blood Test Informed Risk Prediction Model for Lung Cancer—The Copenhagen Lung Cancer Risk Model (CPH-Lung)

Abstract

Purpose Lung cancer screening reduces lung cancer mortality, but identifying high-risk individuals remains difficult. We wanted to develop a blood test–based risk model (CPH-Lung) to improve risk stratification before image-based screening. Materials and Methods We included individuals from two prospective general population cohorts to model 6-year lung cancer risk; the Copenhagen General Population Study (CGPS) for training and the Copenhagen City Heart Study (CCHS) for validation. We analyzed clinical and biochemical variables from ever-smokers (689 cases and 1,924 controls) nested within CGPS to train a 6-year lung cancer risk model. CPH-Lung was validated in a cohort of ever-smokers age ≥50 years in the Copenhagen City Heart Study (CCHS, 3,195 individuals) and compared with the validated PLCOm2012 model. Results There were 689 and 68 events of 6-year lung cancer in the CGPS and the CCHS. In the CCHS validation study, the CPH-Lung versus the PLCOm2012 yielded AUCs of 0.783 (95% CI, 0.731 to 0.835) versus 0.720 (95% CI, 0.669 to 0.770) for detection of all 6-year lung cancers ( P = .0059). The discrimination was similar in the CGPS. At specificity of 77.7% (as would have been achieved using currently applied screening criteria, eg, NELSON or NLST), sensitivities were 57.6% versus 39.2% in women ( P < .0001) and 61.4% versus 54.7% in men ( P = .004) for CPH-Lung versus PLCOm2012. The mean age at diagnosis was 68.2 years (standard deviation [SD], 7.50) in the 120 individuals with lung cancer predicted exclusively by CPH-Lung compared with 76.5 years (SD, 8.04) in the 37 individuals predicted exclusively by PLCOm2012 ( P < .0001). Conclusion After inclusion of blood test information, the CPH-Lung model was more sensitive in the Danish population than PLCOm2012 for prediction of 6-year lung cancer risk particularly among women and younger individuals.