背景：人口老龄化的加剧使得神经系统疾病的患病率在1990-2016年间高速增长，增长率高达117%，使医保基金池承受着较大压力，医保支付改革迫在必行。医疗保险给付模式变革同时涉及分组和付费，分组是统一付费的基础，同时四川省医保局鼓励各市区在CHS-DRG分组的基础上，积极探索DRG细分组。因此对神经系统疾病（MDCB）患者的住院费用影响因素和DRG细分组进行研究有助于减轻患者的经济负担并推进医保改革。目的：本文以某医院MDCB住院患者为研究对象，旨在探讨其费用构成、影响因素和DRG细分组方案，并确定最优影响因素模型、分组模型，为确定医保合理超支因素提供建议。方法：收集2018-2020年四川省某三级医院MDCB住院患者的基本数据，数据清理后共获得52057例数据。结合CV和样本例数，选取4组ADRG进行细分组研究：外科组：BD1，BJ1；内科组：BR2，BV1。利用非参数检验、Gamma、ANN、RF模型探讨费用影响因素。以影响因素为预测变量，住院费用为目标变量，通过E-CHAID模型进行ADRG组的重新细分。通过CV和RIV评价分组合理性。结果：（1）费用构成：2018-2020年，MDCB共52057例住院患者，次均住院日为9.81天，次均住院费用为31631元。外科病组中，材料费和治疗费占比最高（34.20%，22.87%）；内科病组中，诊断费和药品费占比最高，（38.01%，20.00%）。（2）费用预测模型：从MAE结果来看，三种模型中，Gamma模型的预测准确性最高，其次是ANN模型。（3）ADRG分组：依据CHS-DRG分组方案评估30个ADRG组的组内变异，46.67%的病组CV大于1。4组分组中，癫痫病所有细分组的CV均大于1。合并发症是4个病组中都包括的分组节点。（4）影响因素：住院天数、合并发症、年龄对住院费用影响较大。结论：对于MDCB患者而言，费用构成占比与内外科类型有关；Gamma分布是最合适的费用预测模型，其预测性能优于RF和ANN模型；分组节点基本和CHS-DRG方案一致，包括合并发症、年龄、离院方式，结果显示入院途径也可作为潜在分组节点。住院天数、合并发症、年龄、入院途径、离院方式对费用的影响较大。

Background: The aggravation of population aging caused the prevalence of neurological diseases to increase rapidly from 1990 to 2016, with a growth rate of 117%, which put the medical insurance fund pool under great pressure, and the reform of medical insurance payment is imperative. The reform of the medical insurance payment model involves both grouping and payment. Grouping is the basis for unified payment. At the same time, the Sichuan Provincial Healthcare Security Adiministration encourages cities to actively explore DRG subgroups on the basis of CHS-DRG grouping. Research on the influencing factors of patients’ hospitalization costs and DRG subgroups can help reduce the financial burden of patients and promote medical insurance reform.Objective: This paper takes MDCB inpatients as the research object, and aims to explore the cost composition, influencing factors and DRG subdivision grouping scheme, and determine the optimal influencing factor model and grouping model, so as to provide suggestions for determining the reasonable overexpenditure factors of medical insurance.Methods: Basic data of MDCB inpatients in a tertiary hospital in Sichuan Province from 2018 to 2020 were collected, and a total of 52,057 cases were obtained after data cleaning. Combined with CV and the number of samples, 4 groups of ADRG were selected for subdivided study: surgical groups (BD1, BJ1); internal medicine groups (BR2, BV1). Using nonparametric test, Gamma, ANN, RF model to explore the influencing factors of hospitalization costs. Taking the influencing factors as the predictor variables and the hospitalization expenses as the target variables, the ADRG group was re-subdivided by the E-CHAID model. The rationality of grouping was evaluated by CV and RIV.Results: (1) Cost composition: From 2018 to 2020, there were a total of 52,057 hospitalized patients in MDCB, the average hospitalization days was 9.81 days, and the average hospitalization cost was 31631 yuan. In the surgical group, the material cost and treatment cost accounted for the highest proportion (34.20%, 22.87%); in the internal medicine group, the diagnosis cost and drug cost accounted for the highest proportion (38.01%, 20.00%). (2) Expense prediction model: From the MAE results, among the three models, the Gamma model has the highest prediction accuracy, followed by the ANN model. (3) ADRG grouping: According to the CHS-DRG grouping scheme, the intragroup variation of 30 ADRG groups was evaluated, and 46.67% of the disease groups had a CV greater than 1. In the 4 groups, the CVs of all epilepsy subgroups were greater than 1. Commorbities and complications are the grouping nodes included in all 4 patient groups. (4) Influencing factors: days of hospitalization, commorbities and complications, and age have a greater impact on hospitalization costs.Conclusion: For MDCB patients, the proportion of cost components is related to the type of medical and surgical procedures; Gamma distribution is the most suitable cost prediction model, and its predictive performance is better than that of RF and ANN models; the grouping nodes are basically the same as the CHS-DRG scheme, including commorbities and complications, age, discharge patterns, the results show that admission route can also serve as a potential grouping node. The number of days of hospitalization, commorbities and complications, age, admission route, and discharge patterns have a greater impact on costs.