基于决策树的DRGs制度研究——以鼻咽癌为例

Diagnosis Related Groups Study Based on Decision Tree—Using Data of Nasopharyngeal Carcinoma (NPC)

摘要: 根据影响疾病治疗费用的主要因素，本文基于决策树构建了鼻咽癌(NPC)诊断费用的预测模型，得到疾病诊断相关分组(DRGs)下费用更为精细的划分，为医疗保险资源的合理使用和分配提供合理的建议。数据提取自广东省某肿瘤医院，共计2064个鼻咽癌病案。本文通过以下三步对数据进行了处理和分析。首先，我们选取了病人的年龄、性别、TNM诊断分期以及住院天数等特征为预测变量。然后，使用Box-Cox变换方法提取接近正态分布的因变量。其后，使用分类回归树(CART)对不同因变量建立决策树，并得到分组，对DRGs分组下的单病种预付费得到了更为准确的预测模型。

Abstract: According to the major factors affecting the treatment expenses, this passage constructs a predic-tion model of treatment expenses of nasopharyngeal carcinoma(NPC) based on decision tree, and obtains its Diagnosis Related Groups, thus providing scientific basis for the allocation of resources and supervision of the health care insurance fund. Data were extracted from a tumor hospital in Guangdong Province, amounted to 2064 cases of nasopharyngeal carcinoma. We divide this analysis into 3 steps. Firstly, we choose the main factors of hospitalization expenses (HE), such as age, gender, TNM and duration of hospitalization. And then, we construct proper dependent variables that are approximately subject to normal distribution, by using Box-Cox Transformation. Additionally, we build up regression model using Classification and Regression Tree (CART) algorithm and get its modified DRGs. The grouping results, also as DRGs, of NPC cases based on CART and other Boosting algorithm are more reasonable and easier than those before.