earticle

영어

BACKGROUNDS To find early liver cancer, the ministry of health and welfare has conducted surveillance targeting high-risk patients. In 2017, the incidence rate of liver cancer in surveillance was 0.9%, suggesting that a broad patient group was included in surveillance. In this study, to reduce surveillance patients, a prediction model with zero-falsenegative was developed using a machine learning. METHODS To develop the model, we used 2016 Health Insurance Review & Assessment Service-National Patients Sample utilized to the Common Data Model (CDM). This study targeted patients who did not have a severe condition of liver cancer in surveillance. The number of the target was 13,703 cases. The covariates for the model were identified by a chi-square test conducted on gender, age group, condition between a case and control group. LASSO was performed to develop the model. RESULTS Gender, age group, forty diseases were selected as a covariate. The model has an AUC of 0.745, a negative rate of 4.0%, a specificity of 4.5%, and a PPV of 11.8% with zerofalse- negative. CONCLUSION It might be possible to refine surveillance and save the budget of the National Health Insurance Service, and governments.