A Low-Cost Method for Designing and Updating a DRGs Classifier Based on Machine Learning

Abstract

Diagnosis-related groups(DRGs) is a payment system that can effectively solve the problem of excessive increases in health care costs. When DRGs was implemented in China, due to the complex medical environment, the design and update cost of traditional rules-based DRGs classifier became extremely high. In this paper, we proposed a low-cost method for designing and updating a DRGs classifier based on machine learning. This method first uses a rule-based classifier to classify cases roughly according to their major clinical features. With the assistance of the decision tree algorithm, this rule-based classifier can be easily designed and updated by experts. Then, an XGBoost(Extreme Gradient Boosting) classifier based on the one-vs-all(OVR) strategy is trained by a large number of cases labeled by experts or existing DRGs classifier, which will classify cases to each DRG. In the experiments, we proved that the method can utilize cases generated and labeled by China Healthcare Security Diagnosis Related Groups(CHS-DRG) classifier to design a classifier with the performance similar to the CHS-DRG classifier.Updated by low cost, the classifier performance can constantly improve after putting into use.