WO2022082908A1 - 一种基于两阶段筛查流程的医保异常购药的检测方法及装置 - Google Patents

一种基于两阶段筛查流程的医保异常购药的检测方法及装置 Download PDF

Info

Publication number
WO2022082908A1
WO2022082908A1 PCT/CN2020/129363 CN2020129363W WO2022082908A1 WO 2022082908 A1 WO2022082908 A1 WO 2022082908A1 CN 2020129363 W CN2020129363 W CN 2020129363W WO 2022082908 A1 WO2022082908 A1 WO 2022082908A1
Authority
WO
WIPO (PCT)
Prior art keywords
stage
abnormal
medical insurance
model
data
Prior art date
Application number
PCT/CN2020/129363
Other languages
English (en)
French (fr)
Inventor
吴健
姜晓红
应豪超
徐黎明
胡荷萍
Original Assignee
浙江大学山东工业技术研究院
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 浙江大学山东工业技术研究院 filed Critical 浙江大学山东工业技术研究院
Publication of WO2022082908A1 publication Critical patent/WO2022082908A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H20/00—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance
    • G16H20/10—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance relating to drugs or medications, e.g. for ensuring correct administration to patients
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/067—Enterprise or organisation modelling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/08—Insurance

Definitions

  • the invention belongs to the technical field of medical data processing, and in particular relates to a method and device for detecting abnormal drug purchases by medical insurance based on a two-stage screening process.
  • the medical insurance system occupies an important position in my country's social security system with its indispensable importance. Since the implementation of the medical insurance system, the medical insurance fund has played an important role in solving the problem of medical treatment for the general public, and medical insurance is also considered to be a game process between the insured, medical institutions and medical insurance institutions, driven by the interests of the three parties. It is often easy to induce fraudulent activities. Under this circumstance, the intensifying medical insurance fraud has caused huge losses to the medical insurance fund and threatened the normal operation of the medical insurance fund. Therefore, in order to realize the long-term normal operation of the medical insurance fund, it is very important to increase the detection of medical insurance fraud.
  • the technical problem to be solved by the present invention is to provide a method to solve the problems of low accuracy and slow speed in the abnormal detection of excessive drug purchases by insured persons in actual medical insurance, and to detect abnormal drug purchases by medical insurance based on a two-stage screening process. method and apparatus. For this reason, the present invention adopts the following technical solutions:
  • a detection method for abnormal drug purchase by medical insurance based on a two-stage screening process comprising the following steps:
  • step S2 carry out grouping statistics to the data in the model of step S1 according to the individual number of the insured person, and input the analysis data of each insured person obtained by the statistics into the one-stage model;
  • step S3 performing abnormal value detection on the data in the one-stage model obtained in step S2, and performing threshold processing on the detected abnormal insured persons;
  • step S4 screen out the prescription records of the abnormal insured person detected in step S3 from all prescription records, count the prescription frequency of the drug, and regard the drug at the upper quartile of the prescription frequency as a high-frequency drug;
  • step S5 screening the drug purchase records of each of the high-frequency drugs obtained in step S4 by the abnormal insured person, and inputting the analysis data of the abnormal insured person into the two-stage model;
  • step S6 combine the input data of the two-stage model in step S5 with each drug in the high-frequency drugs obtained in step S4, carry out two-stage abnormality detection, and obtain the abnormal value of the insured person predicted to be abnormal;
  • step S7 Perform union and deduplication on the insured persons predicted to be abnormal obtained in step S6 to obtain a detection result.
  • the quartile is the quartile, which means that in statistics, all the values are arranged from small to large and divided into four equal parts, and the values at the position of the three dividing points are a set of data sorting. values at the 25% and 75% positions. Quartiles are divided into 4 equal parts by 3 points, where each part contains 25% of the data. Obviously, the middle quartile is the median, so the quartile usually refers to the value in the 25% position (called the lower quartile) and the value in the 75% position. Numerical value (called the upper quartile).
  • the drug purchase data records extracted in the step S1 at least include transaction serial number, insured individual number, transaction item amount, transaction item medical insurance reimbursement amount, transaction item code, transaction item name, item unit price, item quantity, item price, Project self-pay amount, project self-pay amount and transaction time.
  • the grouping statistical method in the described step S2 adopts the Groupby method
  • the analysis data is the one-stage abnormality detection model analysis data
  • the analysis data includes the number of visits per year, the annual medical insurance reimbursement amount, the number of types of drugs prescribed in the year, and the total amount of drugs prescribed in the year. , the number of medicines prescribed annually, and the number of types of high-frequency medicines in the annual prescription records. Since the iForest anomaly detection model is not suitable for anomaly detection of high-dimensional data, a two-stage anomaly detection model is designed in this method. According to the data of six dimensions related to the number of times and the total medical insurance amount, the preliminary abnormal detection is completed according to the total annual drug purchase information of the insured.
  • the abnormal value detection in the described step S3 includes utilizing some modules in the iForest model to calculate the abnormal degree values of all the insured persons to be detected and sort them. Considering that the number of insured persons to be screened is large (60W), in ensuring In the case of sorting speed and stability, the quick sorting algorithm is selected, and the threshold processing criteria are the number of annual visits and the total amount of annual prescriptions. Such an evaluation method will treat the two extreme cases of "overdose drug purchase" and "small amount of drug purchase” as abnormal. Therefore, after sorting out the abnormal values, it is necessary to take a threshold value according to the number of annual visits and the total amount of drugs. The first certain number of insured persons with abnormal degree values are regarded as abnormal results predicted by the one-stage model.
  • the analysis data in the step S5 is the analysis data of the two-stage abnormality detection model, and the analysis data includes the number of months of prescription, the average number of prescriptions per month, the number of prescriptions per month, and the number of prescriptions per month, and one year is used as the analysis.
  • Data cycle, the two-stage anomaly detection model analyzes data including twenty-six dimensions.
  • the anomaly detection of the second stage in the step S6 adopts the OneClassSVM anomaly detection method.
  • Steps S4, S5, and S6 are the data processing and abnormality detection process of the two-stage model. Since the data analysis dimension in the one-stage abnormality detection process is only the overall data, the specific high-frequency drug prescription of the insured person is not added to the model. Anomaly detection is performed, so in the two-stage model, the analysis and anomaly detection are focused on specific high-frequency prescription records. In data processing, firstly, the medicines in the top 1/4 of the prescribing frequency of abnormal personnel are regarded as high-frequency drugs, and then for each high-frequency drug, the prescribing records of abnormal personnel are extracted and analyzed by the OneClassSVM method. Anomaly detection, while giving outliers of insured persons predicted by the two-stage model as anomalies.
  • a detection device for abnormal drug purchase by medical insurance based on a two-stage screening process comprising:
  • memory for storing the computer's executable instructions and data used or generated in the execution of the executable instructions
  • a processor for executing computer-executable instructions stored in a memory, in communication with the memory;
  • the detection device is used to implement the above-mentioned method.
  • the present invention has the following beneficial effects:
  • the anomaly detection method in machine learning data mining is used to detect the abnormality of medical insurance over-purchasing drugs, which greatly improves the efficiency of medical insurance anomaly detection.
  • a two-stage anomaly detection process is adopted, which improves the detection accuracy while maintaining a good algorithm efficiency.
  • FIG. 1 is an algorithm flow chart of a method and device for detecting abnormal drug purchases by medical insurance based on a two-stage screening process of the present invention.
  • An analytical dimension model is established, and according to the needs of the model, the data records of drug purchases by medical insurance insured persons are extracted from the medical insurance database.
  • the dimensions of the insured's drug purchase record extracted from the medical insurance database include but are not limited to: transaction serial number, insured individual number, transaction item amount, transaction item medical insurance reimbursement amount, transaction item code (drug code), transaction Item name (drug name), item unit price, item quantity, item price, item self-pay amount, item self-care amount and transaction time.
  • the Groupby method is used to group statistics according to the individual number of the insured person, and a total of 370,000 insured persons to be detected are obtained.
  • the one-stage anomaly detection model obtained by data preprocessing The analysis data includes: the number of visits per year, the annual medical insurance reimbursement amount, the number of types of drugs prescribed per year, the total amount of drugs prescribed per year, and the number of types of high-frequency drugs in the annual prescription records.
  • the one-stage anomaly detection model uses some modules in the iForest model to calculate the anomaly degree values of all insured persons to be detected, sort the anomaly values, and perform threshold processing according to the number of annual visits and the total amount of drugs, and treat the abnormality after threshold processing.
  • the first 5,000 insured persons of the degree value are considered as the preliminary abnormal results predicted by the one-stage model.
  • the prescription records of the abnormal people are obtained and the frequency is counted.
  • the upper quartile is taken as the high-frequency drugs, a total of 300 kinds, and each high-frequency drug corresponds to a second-stage drug.
  • the data analysis dimensions of each model include: the number of months of prescription, the average number of prescriptions per month, the number of prescriptions and the number of prescriptions per month (two dimensions per month, a total of 24 dimensions), two-stage
  • the anomaly detection method uses OneClassSVM to detect the anomaly of all the abnormal insured persons in the first stage, and gives the outliers of the insured persons predicted to be abnormal.
  • the information of all the high-frequency drugs corresponding to the two-stage model is predicted to be abnormal insured persons by union and deduplication, and the final abnormal insured persons with over-purchasing drugs are obtained.

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Strategic Management (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Economics (AREA)
  • General Physics & Mathematics (AREA)
  • Human Resources & Organizations (AREA)
  • Marketing (AREA)
  • Development Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Finance (AREA)
  • General Business, Economics & Management (AREA)
  • Accounting & Taxation (AREA)
  • Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Primary Health Care (AREA)
  • Data Mining & Analysis (AREA)
  • Chemical & Material Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Medicinal Chemistry (AREA)
  • Epidemiology (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Databases & Information Systems (AREA)
  • Public Health (AREA)
  • Computational Linguistics (AREA)
  • Educational Administration (AREA)
  • Technology Law (AREA)
  • Game Theory and Decision Science (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • Medical Treatment And Welfare Office Work (AREA)

Abstract

本发明属于医疗数据处理技术领域，尤其是涉及一种基于两阶段筛查流程的医保异常购药的检测方法及装置。一种基于两阶段筛查流程的医保异常购药的检测方法，包括以下步骤：S1、建立分析维度模型；S2、对步骤S1模型中的数据依据参保人的个人编号进行分组统计，将统计得出的每个参保人的分析数据输入一阶段模型；S3、对步骤S2所得的一阶段模型中的数据进行异常值检测，并将检测出的异常参保人进行阈值处理。本发明提供了一种解决实际医疗保险中参保人超量购药异常检测中的准确率不高和速度慢的问题、基于两阶段筛查流程的医保异常购药的检测方法及装置。

Description

一种基于两阶段筛查流程的医保异常购药的检测方法及装置 技术领域
本发明属于医疗数据处理技术领域，尤其是涉及一种基于两阶段筛查流程的医保异常购药的检测方法及装置。
背景技术
随着人民生活水平的不断提高，广大公民自身的社会保障意识得到不断提升，同时，医疗保险制度以其不可或缺的重要性在我国社会保障体系中占据着重要的地位。医疗保险制度自实施以来，医保基金在解决人民大众就医看病方面起到了重要作用，而医保也被认为是介于参保人，医疗机构和医保机构三方之间的博弈的过程，三方利益的驱动往往容易诱发欺诈行为的发生。在这种情况下，愈演愈烈的医保欺诈行为使医保基金遭受巨大的损失，威胁着医保基金的正常运转，因此，为了实现医保基金长期的正常运转，加大对医保欺诈行为的检测至关重要。
国家医保局成立后，在2018年9月开始打击欺诈骗保，一直延续到2019年年底。在这期间要求各省市医保局做省内飞行检查和专项稽核，各级医疗机构做自查整改。各地医保局在媒体和官网上公布欺诈骗保典型案例，主要集中在医疗机构重复收费、超标准收费、串换项目收费、超范围治疗、低标准入院、挂床住院等行为；定点药店存在串换药品、诱导消费等，参保人存在冒名就医、超量取药、因药而病等。
目前，大部分对于医保违规欺诈行为的审核，主要还是靠人工手段去解决，但是医保数据数量庞大，人工解决工作量极大，效率低下，最重要的是，由于人工检查过程中的疏漏，导致很多的医保欺诈行为不能被成功检测出来。随着数据挖掘和机器学习方法在医疗领域的广泛应用，利用数据挖掘等大数据技术来解决医疗保险中欺诈行为的需求应运而生，相当一部分的学者以及一些医保机构也开始逐渐把机器学习数据挖掘技术用于医疗保险欺诈行为的检测，目前大部分的研究和检测都是使用单一的异常检测模型对医保欺诈异常进行检测，单一检测模型对于其他的医保欺诈异常能较好得检测出异常，但是对于医保超量购药异常，需要把具体的高频开药数据也进纳入分析和检测的范围，如果只用一阶段的模型的话，数据分析维度过高，检测算法效率低下，并且高维数据过于复杂，也在很大程度上影响了模型的性能。
发明内容
本发明所要解决的技术问题是提供一种解决实际医疗保险中参保人超量购药异常检测中的准确率不高和速度慢的问题、基于两阶段筛查流程的医保异常购药的检测方法及装置。为此，本发明采用以下技术方案：
一种基于两阶段筛查流程的医保异常购药的检测方法，包括以下步骤：
S1、建立分析维度模型，并根据该模型需要将医保参保人购药数据记录从医保数据库中进行提取；
S2、对步骤S1模型中的数据依据参保人的个人编号进行分组统 计，将统计得出的每个参保人的分析数据输入一阶段模型；
S3、对步骤S2所得的一阶段模型中的数据进行异常值检测，并将检测出的异常参保人进行阈值处理；
S4、将步骤S3中检测出的异常参保人的开药记录从所有开药记录中筛选出，统计药物的开药频次，将开药频次上四分位点的药物视为高频药物；
S5、将异常参保人对步骤S4中获得的高频药物中的每一种药物的购药记录进行筛选，并计算出异常参保人的分析数据输入二阶段模型；
S6、将步骤S5中的二阶段模型输入数据结合步骤S4中获得的高频药物中的每一种药物，进行二阶段的异常检测，并得出预测为异常的参保人的异常值；
S7、将步骤S6所得的预测为异常的参保人进行求并集和去重，以得到检测结果。
其中，四分位点即四分位数(Quartile)，是指在统计学中把所有数值由小到大排列并分成四等份，处于三个分割点位置的数值，它是一组数据排序后处于25％和75％位置上的值。四分位数是通过3个点将全部数据等分为4部分，其中每部分包含25％的数据。很显然，中间的四分位数就是中位数，因此通常所说的四分位数是指处在25％位置上的数值(称为下四分位数)和处在75％位置上的数值(称为上四分位数)。
在采用上述技术方案的基础上，本发明还可采用以下进一步的技 术方案：
所述步骤S1中提取的购药数据记录至少包括交易流水号、参保人个人编号、交易项目金额、交易项目医保报销金额、交易项目编码、交易项目名称、项目单价、项目数量、项目价格、项目自费金额、项目自理金额和交易时间。
在提取数据的过程中，为了方便处理以及提升医保异常检测模型的效率，我们在医保数据库中提取参保人购药数据记录中剔除了处理起来较为繁杂且对模型训练不力的部分中药以及部分非常常见的药物如氯化钠注射液及葡萄糖注射液。
所述步骤S2中的分组统计方法采用Groupby方法，所述分析数据为一阶段异常检测模型分析数据，该分析数据包括年就诊次数、年医保报销金额、年开药种类数、年开药总金额、年开药数量、年开药记录中高频药的种类数。由于iForest异常检测模型不适合使用对高维的数据进行异常检测，故在此方法中设计了两个阶段的异常检测模型，在第一阶段模型中，对所有参保人只提取与总的就诊次数以及总的医保金额等相关的六个维度的数据，根据参保人年度总的购药信息完成初步的异常检测。
所述步骤S3中的异常值检测包括利用iForest模型中的部分模块计算所有待检测参保人的异常程度值并进行排序，考虑到待筛查的参保人人数较多(60W)，在保证排序速度和稳定性的情况下选择快速排序算法，所述阈值处理的标准为年就诊次数和年开药总金额。这样的评估方法会将两个极端的情况“过量购药”和“少量购药”视为异 常，所以在异常值排序之后需要根据年就诊次数和药物总金额进行取阈值处理，将阈值处理后异常程度值的前一定数目的参保人视为一阶段模型预测的异常结果。
所述步骤S5中的分析数据为二阶段异常检测模型分析数据，该分析数据包括开药月份数、月均开药数、每月开药次数、每月开药种数，以一年为分析数据周期，二阶段异常检测模型分析数据包括二十六个维度。
所述步骤S6中二阶段的异常检测采用OneClassSVM异常检测方法。其中，参数：kernel：核函数(一般用高斯核)；nu：设定训练误差(0，1]，方法：fit(x)：训练，根据训练样本和上面两个参数探测边界。predict(x)：返回预测值，+1就是正常样本，-1为异常样本。decision_function(X)：返回各样本点到超平面的函数距离(signed distance),正的为正常样本，负的为异常样本。通过该异常检测方法，可以比较有效地对二阶段高维数据的特征进行特诊提取和聚类分析并检测出其中的异常值。
步骤S4、S5、S6为二阶段的模型的数据处理和异常检测过程，由于在一阶段异常检测过程中的数据分析维度只是整体数据，并未将参保人具体的高频药物开药加入模型进行异常检测，所以在二阶段的模型中重点针对具体的高频开药记录进行分析和异常检测。在数据处理中，首先将异常人员的开药频次前1/4的药物视为高频药物，然后针对对于每一种高频药，都对其异常人员的开药记录进行提取并用OneClassSVM方法进行异常检测，同时给出二阶段模型预测为异常的 参保人的异常值。
进一步地，本发明同时还提供以下技术方案：
一种基于两阶段筛查流程的医保异常购药的检测装置，包括：
存储器，用于存储计算机的可执行指令、和在执行所述可执行指令时使用或产生的数据；
处理器，用于执行存储器存储的计算机可执行指令，与所述存储器通信连接；
所述检测装置用于实现上述的方法。
与现有技术相比，本发明具有以下有益效果：
1)将机器学习数据挖掘中的异常检测方法用于医保超量购药异常的检测，很大程度上提高了医保异常检测的效率。
2)采用两个阶段的异常检测流程，在提升检测准确率的同时维持了不错的算法效率。
3)能有效提升医保工作人员工作效率，为国家节省大量的医保基金。
附图说明
图1为本发明一种基于两阶段筛查流程的医保异常购药的检测方法及装置的算法流程图。
具体实施方式
为了进一步理解本发明，下面结合具体实施方式对本发明提供的一种基于两阶段筛查流程的医保异常购药的检测方法及装置进行具体描述，但本发明并不限于此，该领域技术人员在本发明核心指导思 想下做出的非本质改进和调整，仍然属于本发明的保护范围。
S1、提取数据
建立分析维度模型，并根据该模型需要将医保参保人购药数据记录从医保数据库中进行提取。
具体地，从医保数据库中提取的参保人购药记录维度包括但不限于：交易流水号，参保人个人编号，交易项目金额，交易项目医保报销金额，交易项目编码(药物编码)，交易项目名称(药物名称)，项目单价，项目数量，项目价格，项目自费金额，项目自理金额和交易时间。在剔除部分中药以及诊疗项目之后，共有记录2500多万条。
S2、数据预处理
对提取的所有购药记录利用Groupby方法对根据参保人个人编号进行分组统计，合计得待检测参保人共37万人次，对于每个参保人，数据预处理得到的一阶段异常检测模型分析数据包括：年就诊次数，年医保报销金额，年开药种类数，年开药总金额和年开药记录中高频药的的种类数。
S3、一阶段异常检测
一阶段异常检测模型利用iForest模型中的部分模块计算所有待检测参保人的异常程度值并对该异常值进行排序，并根据年就诊次数和药物总金额进行取阈值处理，将阈值处理后异常程度值的前5000个参保人视为一阶段模型预测的初步异常结果。
S4、二阶段异常检测
首先根据一阶段异常人群的开药记录统计得出异常人群的开药 记录并统计频次，取上四分位点为高频药，共计300种，每一种高频药都对应一个二阶段的异常检测模型，每个模型的数据分析维度包括：开药月份数，月均开药数，每个月的开药次数及种数(每个月两个维度共24个维度)，二阶段的异常检测方法采用OneClassSVM对所有一阶段的异常参保人进行异常检测，并给出预测为异常的参保人的异常值。
S5、后处理及结果汇总
对所有的高频药对应的二阶段模型预测为异常参保人信息进行求并集并去重，得出了最终的超量购药异常参保人。
虽然本发明已通过参考优选的实施例进行了图示和描述，但是，本专业普通技术人员应当了解，在权利要求书的范围内，可作形式和细节上的各种各样变化。

Claims (7)

  1. 一种基于两阶段筛查流程的医保异常购药的检测方法，其特征在于包括以下步骤：
    S1、建立分析维度模型，并根据该模型需要将医保参保人购药数据记录从医保数据库中进行提取；
    S2、对步骤S1模型中的数据依据参保人的个人编号进行分组统计，将统计得出的每个参保人的分析数据输入一阶段模型；
    S3、对步骤S2所得的一阶段模型中的数据进行异常值检测，并将检测出的异常参保人进行阈值处理；
    S4、将步骤S3中检测出的异常参保人的开药记录从所有开药记录中筛选出，统计药物的开药频次，将开药频次上四分位点的药物视为高频药物；
    S5、将异常参保人对步骤S4中获得的高频药物中的每一种药物的购药记录进行筛选，并计算出异常参保人的分析数据二阶段模型；
    S6、将步骤S5中的二阶段模型输入数据结合步骤S4中获得的高频药物中的每一种药物，进行二阶段的异常检测，并得出预测为异常的参保人的异常值；
    S7、将步骤S6所得的预测为异常的参保人进行求并集和去重，以得到检测结果。
  2. 根据权利要求1所述的一种基于两阶段筛查流程的医保异常购药的检测方法，其特征在于所述步骤S1中提取的购药数据记录至少包括交易流水号、参保人个人编号、交易项目金额、交易项目医保报销 金额、交易项目编码、交易项目名称、项目单价、项目数量、项目价格、项目自费金额、项目自理金额和交易时间。
  3. 根据权利要求1所述的一种基于两阶段筛查流程的医保异常购药的检测方法，其特征在于所述步骤S2中的分组统计方法采用Groupby方法，所述分析数据为一阶段异常检测模型分析数据，该分析数据包括年就诊次数、年医保报销金额、年开药种类、年开药数量、年开药总金额、年开药记录中高频药的种类数。
  4. 根据权利要求3所述的一种基于两阶段筛查流程的医保异常购药的检测方法，其特征在于所述步骤S3中的异常值检测包括利用iForest模型中的部分模块计算所有待检测参保人的异常程度值并进行排序，所述阈值处理的标准为年就诊次数和年开药总金额。
  5. 根据权利要求1所述的一种基于两阶段筛查流程的医保异常购药的检测方法，其特征在于所述步骤S5中的分析数据为二阶段异常检测模型分析数据，该分析数据包括开药月份数、月均开药数、每月开药次数、每月开药种数，以一年为分析数据周期，二阶段异常检测模型分析数据包括二十六个维度。
  6. 根据权利要求1所述的一种基于两阶段筛查流程的医保异常购药的检测方法，其特征在于所述步骤S6中二阶段的异常检测采用OneClassSVM异常检测方法。
  7. 一种基于两阶段筛查流程的医保异常购药的检测装置，包括：存储器，用于存储计算机的可执行指令、和在执行所述可执行指令时使用或产生的数据，处理器，用于执行存储器存储的计算机可执行指令， 与所述存储器通信连接，其特征在于所述检测装置用于实现如权利要求1-6中任意一项所述的方法。
PCT/CN2020/129363 2020-10-19 2020-11-17 一种基于两阶段筛查流程的医保异常购药的检测方法及装置 WO2022082908A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202011118465.9A CN112289403A (zh) 2020-10-19 2020-10-19 一种基于两阶段筛查流程的医保异常购药的检测方法及装置
CN202011118465.9 2020-10-19

Publications (1)

Publication Number Publication Date
WO2022082908A1 true WO2022082908A1 (zh) 2022-04-28

Family

ID=74496481

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/129363 WO2022082908A1 (zh) 2020-10-19 2020-11-17 一种基于两阶段筛查流程的医保异常购药的检测方法及装置

Country Status (2)

Country Link
CN (1) CN112289403A (zh)
WO (1) WO2022082908A1 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113626488B (zh) * 2021-08-04 2024-06-21 挂号网（杭州）科技有限公司 数据处理方法、装置、电子设备及存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109119137A (zh) * 2018-08-24 2019-01-01 腾讯科技（深圳）有限公司 一种异常检测方法、装置、服务器及存储介质
CN109636623A (zh) * 2018-10-19 2019-04-16 平安医疗健康管理股份有限公司 医疗数据异常检测方法、装置、设备及存储介质
CN109657059A (zh) * 2018-12-13 2019-04-19 平安医疗健康管理股份有限公司 异常参保人购药数据检测方法、装置、设备及存储介质
CN109873812A (zh) * 2019-01-28 2019-06-11 腾讯科技（深圳）有限公司 异常检测方法、装置及计算机设备
CN110213724A (zh) * 2019-05-17 2019-09-06 国家计算机网络与信息安全管理中心 一种伪基站运动轨迹的识别方法

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190087737A1 (en) * 2017-09-18 2019-03-21 Callidus Software, Inc. Anomaly detection and automated analysis in systems based on fully masked weighted directed
CN108921710A (zh) * 2018-06-08 2018-11-30 东莞迪赛软件技术有限公司 医保异常检测的方法及系统
CN109635113A (zh) * 2018-12-13 2019-04-16 平安医疗健康管理股份有限公司 异常参保人购药数据检测方法、装置、设备及存储介质
CN109636652A (zh) * 2018-12-13 2019-04-16 平安医疗健康管理股份有限公司 购药行为异常的监控方法、监控服务端及存储介质
CN109785162A (zh) * 2018-12-13 2019-05-21 平安科技（深圳）有限公司 医保异常检测方法、装置、设备和计算机存储介质
CN109829733B (zh) * 2019-01-31 2023-02-03 重庆大学 一种基于购物行为序列数据的虚假评论检测系统和方法
CN110009516A (zh) * 2019-04-01 2019-07-12 太平洋医疗健康管理有限公司 医保智能审核方法及系统
CN110781222A (zh) * 2019-10-14 2020-02-11 平安医疗健康管理股份有限公司 异常医保申请检测方法、装置、计算机设备和存储介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109119137A (zh) * 2018-08-24 2019-01-01 腾讯科技（深圳）有限公司 一种异常检测方法、装置、服务器及存储介质
CN109636623A (zh) * 2018-10-19 2019-04-16 平安医疗健康管理股份有限公司 医疗数据异常检测方法、装置、设备及存储介质
CN109657059A (zh) * 2018-12-13 2019-04-19 平安医疗健康管理股份有限公司 异常参保人购药数据检测方法、装置、设备及存储介质
CN109873812A (zh) * 2019-01-28 2019-06-11 腾讯科技（深圳）有限公司 异常检测方法、装置及计算机设备
CN110213724A (zh) * 2019-05-17 2019-09-06 国家计算机网络与信息安全管理中心 一种伪基站运动轨迹的识别方法

Also Published As

Publication number Publication date
CN112289403A (zh) 2021-01-29

Similar Documents

Publication Publication Date Title
Chandola et al. Knowledge discovery from massive healthcare claims data
Kirlidog et al. A fraud detection approach with data mining in health insurance
CN108921710A (zh) 医保异常检测的方法及系统
Megdad et al. Fraudulent financial transactions detection using machine learning
Verma et al. Fraud detection and frequent pattern matching in insurance claims using data mining techniques
Bologa et al. Big data and specific analysis methods for insurance fraud detection.
CN112991079B (zh) 多卡共现就医欺诈行为检测方法、系统、云端及介质
CN113657548A (zh) 医保异常检测方法、装置、计算机设备及存储介质
CN114781805B (zh) 基于大数据的护理人员护理技能评估方法、系统及装置
CN113642672B (zh) 医保数据的特征加工方法、装置、计算机设备及存储介质
Hassan et al. An apriori algorithm-based association rule analysis to detect human suicidal behaviour
Duman et al. Heath care fraud detection methods and new approaches
WO2022082908A1 (zh) 一种基于两阶段筛查流程的医保异常购药的检测方法及装置
CN113674867A (zh) 临床数据挖掘方法、装置、电子设备和存储介质
Sun et al. Heterogeneous network-based chronic disease progression mining
US11023820B2 (en) System and methods for trajectory pattern recognition
Cao et al. Using PCA to improve the detection of medical insurance fraud in SOFM neural networks
CN112231420A (zh) 数据分析方法、数据分析装置、电子设备及存储介质
Nikolaou et al. Fast decliner phenotype of chronic obstructive pulmonary disease (COPD): Applying machine learning for predicting lung function loss
CN111582879A (zh) 一种基于遗传算法的医疗保险反欺诈识别方法
Zhang [Retracted] A Novel Deep Learning Model for Analyzing Psychological Stress in College Students
Jeong et al. Effectiveness of psychotherapy on prevention of suicidal re-attempts in psychiatric emergencies: a systematic review and network Meta-analysis of randomized controlled trials
CN112884593A (zh) 一种基于图聚类分析的医保骗保行为检测方法及预警装置
Yan et al. A CNN Model with Discretized Mobile Features for Depression Detection
Mehraby et al. Abnormal behavior detection in health insurance assessment process

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20958486

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20958486

Country of ref document: EP

Kind code of ref document: A1