Risk-based prediction model for selecting eligible population for lung cancer screening among ever smokers in Korea

This study developed a new lung cancer risk prediction model for the Korean population and evaluated the performance, compared to the previously reported risk models developed in Western countries.

Among the 6,811,893 people who received health examinations from the Korean National Health Insurance Service, 969,351 ever-smokers (40-79 years) were included. Performance of Bach, Lung Cancer Risk Models for Screening, PLCOM2012, Pittsburgh, and Liverpool Lung Project models were evaluated. The ever-smokers were divided into the training and validation datasets by random sampling. The lung cancer risk model was developed and validated in the Korean population. The efficiency of model-based selection for lung cancer screening was compared with the eligible criteria of the National Lung Screening Trial (NLST).

The Korean lung cancer risk model showed the area under the curve and expected/observed (E/O) ratio of 0.816 and 0.983 in the training dataset and 0.816 and 0.988 in the validation dataset. The Korean lung cancer risk model included age-mean of age, square of age-mean of age, sex, square root of pack-years of smoking, years since cessation, physical activity, alcohol consumption, body mass index, and medical history of chronic pulmonary obstructive disease, emphysema, pneumoconiosis, and interstitial pulmonary disease. Compared with the NLST criteria, the Korean lung cancer risk model's cut-off criteria (>2.1%) had more improved sensitivity (61.4% vs. 44.3%) and positive predictive value (4.1% vs. 2.9%). The Korean lung cancer risk model showed better discrimination and calibration than previously developed models in Western population.

The Korean lung cancer risk model can select eligible population for low-dose computed tomography screening among the Asian population. The efficiency of risk model-based selection for lung cancer screening is superior to that of fixed criteria-based selection.