A 5-year mortality-prediction model for patients with stomach cancer, based on the Korean nationwide health insurance claim database

Stomach cancer remains a significant health concern in Korea despite a declining trend in incidence and mortality rates. The aim of this study was to develop a predictive model for the 5-year survival of patients with stomach cancer that is particularly useful in the primary-care setting. This retrospective cohort study included 4156 individuals diagnosed with stomach cancer from 2005 to 2016 and registered in the Korean National Health Insurance Database. The outcome variable was the probability of survival for at least 5 years. Potential confounders affecting mortality were considered, including age, body mass index, systolic blood pressure, lifestyle factors, the Charlson comorbidity index (CCI) score, and laboratory test results. A Cox proportional-hazards regression model was employed to develop the prediction model. The predictive performance was measured using the mean area under the receiver operating characteristic curve after 10-fold cross-validation. The mean 5-year survival probabilities were 0.888 for men and 0.892 for women. For men with stomach cancer, age, body mass index, glucose concentration, smoking status, alcohol consumption, physical activity, total cholesterol concentration, gamma-glutamyl transferase concentration, and the CCI score were significant predictors of mortality. In contrast, for women, only age and the CCI score were significant predictors. The mean areas under the receiver operating characteristic curve after 10-fold cross-validation were 0.699 for men and 0.717 for women. The 5-year survival-probability model had a moderately good predictive performance. It may be used to predict the probability of death for individuals who are diagnosed with stomach cancer.