Abstract: Objective To predict the risk of lung cancer in individuals with missing covariates by combining a Bayesian network with a Cox model. Methods Data were obtained from the UK Biobank. Predictors associated with lung cancer were screened with univariate Cox regression analysis. Based on the predictors identified, the individual risk prediction model of lung cancer was established. The identification and calibration of the model were determined to evaluate its predictive performance. Results The prediction model had good identification and calibration ability, and the AUC of the training and validation cohorts were 0.854(95%CI: 0.836-0.870)and 0.885(95%CI: 0.871-0.897), respectively. Conclusion A lung cancer risk prediction model based on Bayesian network and Cox model was constructed. The model has good identification and calibration ability, and can effectively predict the high-risk population of lung caner. Combined model provides an effective risk prediction method in the presence of missing predictors, which can provide theoretical reference for the prevention and control of lung cancer.

Key words: Lung cancer, Risk prediction model, Bayesian network, Cox model, Missing data