基于医保数据的疾病预测和转诊行为分析

发布时间：2018-08-25 09:45

【摘要】：医疗保险在人们的医疗行为中扮演着非常重要的作用,医保数据记录了病人疾病、费用、就诊时间、就诊地点及人口统计学等方面的诸多信息。通过对医保数据的分析和挖掘,我们可以做很多有意义的研究,如发现疾病的发病模式,预测疾病的发展趋势,分析病人的就诊模式,评估医疗政策在实际中的效果等等。这一方面可以为医疗专业领域的人提供进一步研究的启示,另一方面可以提示人们在面对疾病时的注意事项,也可以从政策层面,为医疗管理者提供有借鉴意义的参考。本文针对医保数据,主要从两方面来进行研究。一个是疾病之间的潜在关系,通过数据中相似病人的发病历程,寻找疾病之间的关联性,从而为疾病的预防控制和治疗提供指导。这是从疾病的防控方面来进行论述。另一方面,由于该研究使用的数据是全样本数据,可以对病人在观察期内的所有就诊记录进行分析,从而找到影响病人就诊行为的因素。这是从病人就诊治疗的过程方面来进行论述。传统的疾病预测主要是集中研究某一种或某几种疾病将来的预后状况,而且是宽泛的研究,并不能针对病人自身,进行个性化的预测。另一方面,传统的研究方法需要进行长期的观察和实验,对病人的状况进行跟踪和记录,需要消耗大量的人力和物力成本。本研究提出的CAC方法,将数据挖掘中的几种算法结合起来,仅仅利用医保数据中人口统计学和疾病相关的信息,可以对多病种进行分析,还可为病人进行个性化的预测。本文根据慢性病人和急性病人的不同特征,分别进行了预测,对急性疾病的预测准确率达到了 71%,对慢性疾病的预测准确率则达到了 82%。这种对多病种个性化的预测方法在文献中很少出现,本文的结果比仅有的少数研究都有所提高。针对本研究的预测方法,本文利用测试集数据做了案例分析,病人的实际状况与本研究的预测结果吻合度非常高。将预测结果与病人的不同特征相结合,本研究发现病人的医疗费用并没有反映其病情的真实状况,由此对医疗政策给出相关的建议,并引出第二个研究点。本文这里研究的病人就诊行为主要是病人就诊时的转诊行为。我们国家的医疗体系是一个庞大而复杂的层级系统。理论上来讲,病人对医疗机构的选择是不受限制的。但是,在实际就医的过程中,针对不同的医疗群体,又存在政策、医院和病人自身因素等多方面的影响。在多种因素的综合作用下,病人会如何选择,哪些因素对病人的选择会产生怎样的影响,这是政府层面、医院方面以及病人都非常关注的问题。本研究化繁为简,对病人就诊时的不同选择行为进行归类,对病人的就诊行为(本文称之为广义转诊,简称转诊)进行了清晰的定义,并以此为基础,提出细致化的转诊模型,考虑到了多次转诊的情况,对转诊模式进行了全面分析。提出换医院和是否再住院的回归模型,通过对比,分析了不同因素对病人转诊模式的影响,发现了一些非常具有实际指导意义的转诊规律。提出有时间限制的和无时间限制的转诊模型,根据对比的结果,对相关政策进行评判,并提出相应建议。由于转诊行为极其复杂,加之我国的医疗体系庞大而独特,本研究在相关领域具有一定的启发意义,可为以后的研究提供新的思路。
[Abstract]:Medical insurance plays a very important role in people's medical behavior. Medical insurance data record a lot of information about patient's disease, cost, time, place and demography. On the one hand, it can provide enlightenment for people in the medical profession to further study, on the other hand, it can remind people to pay attention when facing diseases, and it can also provide reference for medical managers from the policy level. Reference. In this paper, medical insurance data, mainly from two aspects to study. One is the potential relationship between diseases, through the data similar to the course of disease, to find the relationship between diseases, so as to provide guidance for disease prevention and control and treatment. The data used in this study are full-sample data, which can be used to analyze all patient visits during the observation period to identify the factors affecting patient behavior. This is discussed from the process of patient visits. On the other hand, the traditional research methods need long-term observation and experiments to track and record the patient's condition, which requires a lot of manpower and material costs. Combining with the data of medical insurance, demography and disease-related information can be used to analyze multiple diseases and personalized prediction for patients. According to the different characteristics of chronic patients and acute patients, the prediction accuracy of acute diseases is 71% and chronic diseases are predicted. The accuracy of this method is 82%. This method rarely appears in the literature, and the results of this paper are higher than those of only a few studies. This study found that the patient's medical expenses did not reflect the true state of the patient's illness, thus giving relevant suggestions to the medical policy, and leading to the second research point. A large and complex hierarchical system. Theoretically, patients'choice of medical institutions is unrestricted. However, in the actual process of seeking medical treatment, according to different medical groups, there are policies, hospitals and patients' own factors and other factors. Under the combined effect of a variety of factors, patients will choose which factors. The government, hospitals and patients are all concerned about the impact of patient selection. This study simplifies the complexity of the study, classifies the different choices of patients, and defines the patient's behavior (referred to as general referral, referred to as referral) clearly. A meticulous referral model was developed, and the referral mode was comprehensively analyzed considering the situation of multiple referrals. A regression model was proposed for the change of hospital and re-hospitalization. By comparison, the influence of different factors on the referral mode was analyzed, and some referral rules with practical significance were found. According to the results of the comparison, the relevant policies are judged and corresponding suggestions are put forward. Because of the extremely complex referral behavior and the huge and unique medical system in our country, this study has some inspiration in the relevant fields and can provide new ideas for future research.
【学位授予单位】：中国科学技术大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：R197.1;F842.684

【参考文献】