基于决策树模型的急性髓系白血病DRG分组研究

Study on DRG grouping of acute myeloid leukemia based on decision tree model

目的:探讨急性髓系白血病患者住院费用的影响因素,基于决策树模型进行病例分组,为完善该病种的DRG管理提供参考。方法:提取2020年至2021年福建省血液病研究所临床血液内科出院的以急性髓系白血病为主要诊断(ICD编码前4位为C92.0、C92.4、C92.5、C93.0)的病案首页数据,通过Wilcoxon秩和检验或Kruskal-Wallis秩和检验以及多元线性逐步回归分析确定住院费用的影响因素,并将其作为分类节点,运用χ2自动交互检验法决策树模型对纳入病例进行分组。同时,按福建省试运行的C-DRG版本对纳入病例进行分组,比较两种分组方法的区别。结果:住院天数、治疗类型、是否伴有并发症及年龄是急性髓系白血病患者住院费用的影响因素,将其纳入决策树模型形成9个病例组合;该模型的方差减少量为75.77%,组间异质性较高,变异系数为0.33~0.61,组内差异较小。按福建省试运行的C-DRG版本共分为2组,其方差减少量为27.57%,组间异质性不高,变异系数为0.59和1.25,组内差异较大。结论:根据住院天数、治疗类型、是否伴有并发症及年龄进行急性髓系白血病病例分组,分组结果较为合理,可为该病种的DRG管理和成本管控提供参考。

Objective To explore the influencing factors of hospitalization cost of acute myeloid leukemia,to group the cases based on decision tree model and to provide reference for improving the DRG management in this regard.Methods Homepage data were retrieved from the medical records with acute myeloid leukemia as the main diagnosis(the top four ICD codes were C92.0,C92.4,C92.5,and C93.0).These patients were discharged from the clinical hematology department of the Fujian Institute of Hematology from January 2020 to December 2021.Then the influencing factors of hospitalization expenses were identified using Wilcoxon rank sum test or Kruskal-Wallis rank sum test and multiple linear stepwise regression analysis,with such factors used as classification nodes.The decision tree model ofχ2 automatic interactive testing method was used to group the cases so included.At the same time,the included cases were grouped according to the trial run C-DRG version in Fujian province,for comparison of the differences between the two grouping methods.Results The length of stay,the type of treatment,whether associated complications and age of patients were found as the influencing factors for the hospitalization costs of patients with acute myeloid leukemia,and such factors were included in the decision tree model to form 9 case mixes.The variance reduction of this model was 75.77%,featuring a high inter-group heterogeneity,and the coefficient of variation was 0.33-0.61,featuring a low in-group difference.The patients were divided into two groups according to the C-DRG version in Fujian province.The variance reduction of this method was 27.57%,featuring a low inter-group heterogeneity,and the coefficients of variation were 0.59 and 1.25,featuring high in-group difference.Conclusions The cases of acute myeloid leukemia were grouped based on length of stay,type of treatment,whether accompanied by complications,and age proved reasonable enough to serve as reference for DRG management and cost control of this disease.