Development of a revisit prediction model for the outpatient in a hospital / 대한의료정보학회지

ABSTRACT

OBJECTIVES: This study is for developing the prediction model of outpatient's revisit in target hospital. Using this model, hospital managers can make efficient customer relationship. METHODS: This is based on the medical record data of patients in target hospital (with 967 beds). They are divided into two groups, which are used each for different purpose. One(raw data) is used to make the prediction model of revisit and the other(test data) is used to evaluate the model. For raw data were used the 4,273 outpatient cases, where patients visited the first time between august and september in 2000, and visited till december in 2003. For the test data were used 9,392 outpatient cases, where patients visited the first time between august and september in 2003, and visited till december in 2006. That is, each data was selected from the outpatient's medical records for three-years. RESULTS: The decision tree model is better than the logistic regression model as prediction model of outpatient's revisit in target hospital. The decision tree model is evaluated more excellent in ROC curve and classification accuracy in test data. For predicting the outpatient's revisit, it is more useful to have 4 variables - non-insured expenses, special medical service, cooperation service with oriental medicine and visit via ER. We can predict the revisit of outpatients over 39.5% rate by these variables. CONCLUSIONS: By using decision tree model, target hospital can make more accurate prediction of outpatient's revisit and make good customer relation management. So, target hospital can use some CRM program including 4 variables. To make more useful model for other hospitals in Korea, each hospital managers need to understand more their hospital environment and patient's characteristics.