基于健康保险数据的医疗费用共病网络分析及深度学习预测  

Comorbidity network analysis and deep learning prediction of medical expenses based on health insurance data

作　　者：曹毓文 梅好 孙佳仪 胡炯宇 徐雅晴 CAO Yuwen;MEI Hao;SUN Jiayi;HU Jiongyu;XU Yaqing(Scientific Research Center for Applied Statistics,Renmin University of China,Beijing 100872,China;School of Statistics,Renmin University of China,Beijing 100872,China;School of Humanities,Tsinghua University,Beijing 100084,China;Department of Endocrinology,First Affiliated Hospital of Army Military Medical University,Chongqing 400038,China;School of Public Health,Shanghai Jiao Tong University,Shanghai 200025,China)

机构地区：[1]中国人民大学应用统计科学研究中心,北京100872 [2]中国人民大学统计学院,北京100872 [3]清华大学人文学院,北京100084 [4]陆军军医大学第一附属医院内分泌科,重庆400038 [5]上海交通大学公共卫生学院,上海200025

出　　处：《重庆医学》2024年第24期3686-3691,共6页Chongqing Medical Journal

基　　金：国家自然科学基金项目(72301283,82204153);教育部人文社会科学重点研究基地重大项目(22JJD910001);重庆市技术创新与应用发展专项重点项目(CSTB2023TIAD-KPX0047);中国人民大学新教师启动金项目(23XNKJ07)。

摘　　要：目的通过分析相关医疗记录构建医疗费用共病网络,并结合疾病网络与长短期记忆神经网络构建深度学习预测模型,以提升个体医疗费用预测的准确度,同时为优化医疗政策、提升患者健康管理水平提供助力。方法基于中国台湾健康保险研究数据库2000-2013年的医疗记录,分析9963例患者的584万条就诊数据,构建包含104种常见疾病的医疗费用共病网络,分析网络结构并预测潜在共病,结合患者的性别、年龄、病史等信息输入构建深度学习模型个体医疗费用。结果构建的医疗费用共病网络包含104个节点、3390条边和6个模块,是一个节点高度相连的网络,表示疾病间医疗费用具有高度相关性。构建的深度学习预测模型较传统回归模型及未充分考虑共病信息的深度学习模型相比,显著提高了预测精度。结论构建的模型为理解疾病共病性提供了全新的理论视角,还为精准预测医疗费用、优化医疗资源配置以及实现个性化医疗服务提供了有效工具。Objective To construct a comorbidity network for medical expenses by analyzing the relevant medical records,and to construct a deep learning prediction model by combining with the disease networks and long short-term memory neural networks in order to improve the accuracy of individual medical expense prediction and provide the assistance for optimizing the medical policies and enhancing the patient health management level.Methods Based on the medical records of Taiwan,China Health Insurance Research Database during 2000-2013,the data of 5.84 million visits from 9963 patients were analyzed,and a comorbidity network of medical expenses for 104 common diseases was constructed.The network structure was analyzed and the potential comorbidity was predicted,and the deep learning model of individual medical cost was constructed by combining the input of patient’s gender,age,medical history and other information.Results The constructed medical cost comorbidity network consists of 104 nodes,3390 edges and 6 modules,and is a highly connected network with nodes,indicating that the medical costs possesses the high correlation between diseases.The constructed deep learning prediction model significantly improves the prediction accuracy compared to the traditional regression models and deep learning models that do not fully consider the comorbidity information.Conclusion The constructed model provides a new theoretical perspective for understanding the comorbidity of diseases,as well as an effective tool for accurately predicting medical costs,optimizing medical resource allocation and achieving the personalized medical services.

关 键 词：慢病管理 健康保险数据 疾病网络 深度学习 医疗费用预测 

分 类 号：R112[医药卫生—公共卫生与预防医学]