Individualized risk prediction model for lung cancer in Korean men

Abstract

Purpose: Lung cancer is the leading cause of cancer deaths in Korea. The objective of the present study was to develop an individualized risk prediction model for lung cancer in Korean men using population-based cohort data.

Methods: From a population-based cohort study of 1,324,804 Korean men free of cancer at baseline, the individualized absolute risk of developing lung cancer was estimated using the Cox proportional hazards model. We checked the validity of the model using C statistics and the Hosmer-Lemeshow chi-square test on an external validation dataset.

Results: The risk prediction model for lung cancer in Korean men included smoking exposure, age at smoking initiation, body mass index, physical activity, and fasting glucose levels. The model showed excellent performance (C statistic = 0.871, 95% CI = 0.867-0.876). Smoking was significantly associated with the risk of lung cancer in Korean men, with a four-fold increased risk in current smokers consuming more than one pack a day relative to non-smokers. Age at smoking initiation was also a significant predictor for developing lung cancer; a younger age at initiation was associated with a higher risk of developing lung cancer.

Conclusion: This is the first study to provide an individualized risk prediction model for lung cancer in an Asian population with very good model performance. In addition to current smoking status, earlier exposure to smoking was a very important factor for developing lung cancer. Since most of the risk factors are modifiable, this model can be used to identify those who are at a higher risk and who can subsequently modify their lifestyle choices to lower their risk of lung cancer.

Figure 1. Association between age at smoking initiation and the smoking amount per day among current smokers.

Figure 2. Discrimination and calibration of the lung cancer prediction model (A: discrimination, B: calibration).

Figure 3. Discrimination and calibration of the lung cancer prediction model tested on an external validation set* (A: discrimination, B: calibration, n = 507,046, event = 4,539).