Objective: Most of the previous risk prediction models for lung cancer were developed from smokers, with discriminatory power ranging from 0.57 to 0.72. We constructed an individual risk prediction model for lung cancer among the male general population of Hong Kong. Methods: Epidemiological data of 1,069 histology confirmed male lung cancer cases and 1,208 community controls were included in this analysis. Residential radon exposure was retrospectively reconstructed based on individual lifetime residential information. Multivariable logistic regression with repeated cross-validation method was used to select optimal risk predictors for each prediction model for different smoking strata. Individual absolute risk for lung cancer was estimated by Gail model. Receiver-operator characteristic curves, area under the curve (AUC) and confusion matrix were evaluated to demonstrate the model performance and ability to differentiate cases from non-cases. Results: Smoking and smoking cessation, education, lung disease history, family history of cancer, residential radon exposure, dietary habits, carcinogens exposure, mask use and dust control in workplace were selected as the risk predictors for lung cancer. The AUC of estimated absolute risk for all lung cancers was 0.735 (95% CI: 0.714–0.756). Using 2.83% as the cutoff point of absolute risk, the predictive accuracy, positive predictive value and negative predictive value were 0.715, 0.818 and 0.674, respectively. Conclusion: We developed a risk prediction model with moderate discrimination for lung cancer among Hong Kong males. External validation in other populations is warranted for this model in future studies.
Loading....