JP2020071562A - Health check reception probability calculation method and health check recommendation notification support system - Google Patents

Abstract

To improve efficiency of a recommendation notification work by calculating a probability that an insured person receives a health check.SOLUTION: Parent population to be basic data of probability calculation is specified as health check data of a health check that an insured person receives during past n (n is an integer of 3 or more) years owned by a single or a plurality of autonomous bodies or a single or a plurality of health insurance organizations. A processing program extracts data of past m (m is an integer of 3 or more, m≤n) years from all the health check data to process data of (m-1) years excluding the last one year into data having x explanatory variables as a first step, performs learning using the health check data of the (m-1) years and presence or absence of health check reception of the last one year as teacher data as a second step, and calculates reception probability of an individual insured person who subscribes to a specific single autonomous body or a specific single health insurance organization by a learned model as a third step.SELECTED DRAWING: Figure 2

Description

  The present invention relates to a health checkup probability calculation method and a health checkup recommendation notification support system for supporting efficiency improvement of recommendation notification to an insured person who encourages a health insurance organization to receive a health checkup.

  The medical expenses in Japan continue to increase every year, already exceeding 40 trillion yen, and it is expected to continue to increase with the progress of aging. Among them, national health insurance, which is enrolled by the majority of the people, is public insurance that is mainly operated by local governments (prefectures and municipalities), and the increase in medical costs is a heavy burden on local governments and national finances. Has become.

  In order to reduce the ever-increasing medical expenses, each municipality implements health maintenance of insured residents, specific medical examinations for the purpose of early detection and early treatment of illness, and specific health guidance every year. There are many residents who do not receive medical examination even when notified.

  In light of this situation, each municipality has sent a recommendation notice to encourage consultation, but due to budget and personnel restrictions, we cannot afford to notify all insured persons. Is required to provide the most effective recommendation notifications.

  Conventionally, a system (patent document 1) that supports the evaluation of health services carried out based on the health status of the insured person, a new target person based on the measured values of medical examinations or medical examination history of a plurality of target persons in a predetermined period. A system for calculating an evaluation value of risk using a value based on various health checkup information and a model (Patent Document 2), a past checkup rate of a checkup of each group included in a plurality of groups that can receive a certain health checkup A system (Patent Document 3) and the like is known in which a combination of groups is determined according to information, and a group included in the determined combination is extracted as a group to be subjected to the certain medical examination, but it is effective. There was no system to support notification of medical examination recommendations.

JP, 2016-177644, A JP, 2017-117469, A JP, 2014-102797, A

  In order to solve the above problems, the present invention calculates the consultation probability of an insured who joins a specific national health insurance or a specific health insurance association, etc., based on past medical examination data, and gives a recommendation notice. Identify and exclude those who are not likely to receive the medical examination even if they do not have the medical examination, or those who are unlikely to receive the medical examination even if the notification of recommendation is made, and the remaining insured, that is, the notification of recommendation The purpose of this is to select those who may receive medical examination with high accuracy and to make the recommendation notification work more efficient.

In order to achieve the above-mentioned object, the first invention of the present application is a method for calculating a health examination consultation probability of an insured person, wherein a population serving as basic data for probability calculation is a plurality or a single municipality, or a plurality of or The health checkup data of the health checkup that the insured has received in the past n years (n is an integer of 3 or more) held by a single health insurance association, etc., and predicts the checkup behavior of the insured based on the health checkup data. As a first step, the processing program includes a processing program for extracting the data for the past m years (m is an integer of 3 or more, m ≦ n) from all the medical examination data, and for the most recent one year. The data for m-1 year excluding is processed into data having x explanatory variables, and as the second step, the medical examination data for the m-1 year and the medical examination for the most recent year are taken. Learning by using presence / absence as teacher data, the third step And, wherein the learned model, to calculate a consultation probabilities of individual insured to subscribe to a particular single municipality or specific single health insurance, it is characterized.

  A second invention of the present application is a method for calculating a probability of receiving a health checkup of an insured person, wherein a population serving as basic data for probability calculation is owned by a plurality of local governments or a plurality of health insurance associations in the past n years. (N is an integer equal to or greater than 3), including the medical examination data of the medical examination that the insured has taken, and including a processing program that predicts the medical examination behavior of the insured based on the medical examination data. As a first step, data for the past m years (m is an integer of 3 or more, m ≦ n) is extracted from all the medical examination data, and data for m-1 years excluding the most recent one year is extracted. The data is processed to have x explanatory variables from the inside, the algorithm is deep learning, cross entropy is used for the error function, and Adam is used for optimization learning. Teachers with or without medical examination By using as data, a plurality of weights Wi corresponding to a plurality of layers are learned, and as a second step, only the medical examination data of a specific single municipality or a specific single health insurance association is extracted, and transfer learning is performed by an algorithm. Is used to create a second feature amount in which a part of the weight Wi created in the first step is adjusted, and as a third step, the specific single municipality or the It is characterized by calculating the consultation probability of each insured who joins a specific independent health insurance association.

  A third invention of the present application is a health checkup recommendation notification support system for supporting the efficiency of the recommendation notification to the insured for promoting the health checkup conducted by the local government or the health insurance association, and is provided with a plurality of or A health checkup database in which the results of health checkups for the past n years (n is an integer of 3 or more) received by an insured person owned by a single municipality or a plurality of or single health insurance associations, and the health checkup database And a server in which a processing program for predicting a medical examination behavior of the insured person based on the data accumulated in the above is recorded, and the processing program includes, as a first step, a plurality of or a single processing from the medical examination database. Data for the past m years (m is an integer of 3 or more, m ≤ n) extracted by the local government, or a single or multiple health insurance associations, and data for m-1 years excluding the most recent one year, x explanatory variables It is processed into data that has, and as a second step, learning is performed using the data for the m-1 year and the presence or absence of a medical examination in the most recent year as teacher data, and as a third step, by the learned model, Calculating the probability of individual insured who join a specific municipality or a specific health insurance association, etc., and as a fourth step, a predetermined high probability value or more determined by the specific municipality or the specific health insurance association, etc. The recommendation notification target list is created only for insured persons with intermediate probability values other than a predetermined low probability value or for insured persons with a predetermined number of intermediate probability values. And

  Further, a fourth invention of the present application is a health checkup recommendation notification support system for supporting the efficiency improvement of a recommendation notification to an insured person who encourages a health checkup conducted by a local government or a health insurance association. The health checkup database in which the results of health checkups for the past n years (n is an integer of 3 or more) received by the insured held by the local government or a plurality of health insurance associations and the like are stored in the health checkup database. And a server in which a processing program for predicting a medical examination behavior of the insured person based on the data is recorded, the processing program, as a first step, from the medical examination database for the past m years (m is 3 or more). An integer, m ≦ n) of data is extracted and processed into data having x explanatory variables from the data of m−1 years excluding the most recent one year, and the algorithm is deep learning, Error function By using cross entropy, using Adam for the optimization learning, and using the data for the m-1 year and the presence or absence of the medical examination in the most recent year as the teacher data, a plurality of weights Wi corresponding to a plurality of layers To create a first feature amount, and as a second step, only data of a specific municipality or a specific health insurance association is extracted, and the weight Wi created in the first step using transfer learning as an algorithm. Create a second feature amount that adjusts a part of the above, and as a third step, by a model based on the second feature amount, of the individual insured persons who join the specific municipality or the specific health insurance association, etc. Calculating the consultation probability, and as the fourth step, an intermediate probability excluding a value higher than a predetermined high probability value and a value lower than a predetermined low probability value set by the specific municipality or the specific health insurance association, etc. Only the insured, or only for a predetermined insured of a predetermined number of intermediate probability values, creating a encouraged notified list, characterized in that.

  According to the first invention and the third invention of the present application, the population serving as the basic data for the probability calculation that the insured undergoes the medical examination is owned by a plurality or a single municipality, or a plurality or a single health insurance association. A large city with sufficient number of health checkup data to create a high-performance health checkup probability calculation model as health checkup data of health checkups received by the insured during the past n years (n is an integer of 3 or more) (Or a large health insurance association, etc.) creates a probabilistic model using the data of only that large city (or a large health insurance association, etc.) so that the data of a local small city (or a small health insurance association, etc.) does not become noise. For small local cities (or small health insurance associations, etc.) that do not have enough medical examination data to create a performance consultation probability calculation model, a probability calculation model is created using data from multiple local governments (or health insurance associations) across the country. Create Rukoto, it is possible to identify a single municipality or specific single health insurance such visits probability of individual insured that subscribe to accurately calculate (guess). As a result, the insured group of those who are likely not to undergo the medical examination even if the recommendation notice is sent, and the insured group of those who are likely to receive the medical examination even if the recommendation notice is not sent. Can be identified with high accuracy, and the insured group that belongs to a group other than these two groups, that is, the insured group that is more likely to be examined by sending out the recommended notice, can receive the recommended notice. It is possible to improve the efficiency of office work related to and to optimize the budget and personnel. Further, as a result, the number of insured persons who take a medical examination increases, which can lead to a reduction in medical expenses.

  According to the second invention and the fourth invention of the present application, a population serving as basic data for calculating the probability that the insured receives a medical examination is owned by a plurality of local governments or a plurality of health insurance associations in the past n years. (N is an integer of 3 or more), the health checkup data of the health checkup received by the insured person is used, and as the first step, the first feature amount is obtained by using all data held by a plurality of local governments or a plurality of health insurance associations. As a second step, a second feature amount is created by adjusting a part of the feature amount created in the first step using only the medical examination data of a specific single municipality or a specific single health insurance association. As the third step, the probability of the individual insured who joins a specific independent municipality or a specific independent health insurance association is calculated by a probability calculation model based on the second feature quantity, and thus the multiple municipalities are examined. Or multiple While creating the first feature amount, which is the basic form, based on a large amount of data from the health insurance association, etc., consider the behavior pattern peculiar to the insured person such as the specific single municipality to be calculated or the specific single health insurance association. You can As a result, the insured group of those who are likely not to undergo the medical examination even if the recommendation notice is sent, and the insured group of those who are likely to receive the medical examination even if the recommendation notice is not sent. Can be identified with high accuracy, and the insured group that belongs to a group other than these two groups, that is, the insured group that is more likely to be examined by sending out the recommended notice, can receive the recommended notice. It is possible to improve the efficiency of office work related to and to optimize the budget and personnel.

  As described above, each invention of the present application identifies an insured person who is likely to receive a medical examination without giving a recommendation notice, or an insured person who is unlikely to receive a medical examination even if a notice is given. The remaining insured persons, that is, those who may be examined by giving a recommendation notice are extracted with high accuracy, but when analyzed using actual data, the characteristics of each municipality Alternatively, since the difference from the actual measurement value differs depending on the characteristics of each health insurance association, it is possible to appropriately select which method is used while making the best use of the characteristics of each local government or health insurance association.

  The “health insurance association, etc.” referred to in the present invention means various health insurance associations other than national health insurance operated by local governments (health insurance associations of companies, health insurance managed by the National Health Insurance Association (so-called association Kenpo), mutual aid associations). Etc.) is meant.

Basic system configuration diagram of the present invention Algorithm of processing program according to first embodiment Diagram showing the data format that has been cleaned Explanatory variable list Diagram showing the teacher dataset Diagram showing the structure of a random forest Algorithm of processing program according to second embodiment Diagram showing the structure of deep learning Diagram showing the structure of transfer learning Figure showing the experimental results in a small municipality Diagram showing experimental results in a large-scale municipality Diagram showing the relationship between the predicted value of consultation and the effect of recommendation for consultation

  Embodiments of the present invention will be described in detail with reference to FIGS. It is needless to say that each embodiment is not intended to limit the scope of the invention of the present application, and may be appropriately implemented based on the contents described in the claims and the spirit of the invention. The present embodiment described below will be described as a system assuming that a service is provided to a specific local government as an example.

  FIG. 1 is a system configuration diagram that is the basis of the present invention. The system according to the present invention is common to both the first embodiment and the second embodiment, and is stored in the medical examination database 1 in which past medical examination consultation results of a plurality of local governments are accumulated and in the medical examination database 1. And a server 3 in which a processing program 2 for predicting the medical examination behavior of the insured person based on the stored data is recorded. The medical examination database 1 may be built in the server 3. In each of the embodiments described below, the processing method of the processing program 2 is different, and thus the method of calculating the examination probability of the insured is different.

First, the processing program 2 according to the first embodiment of the present invention will be described.
The processing program 2 is used as a basis for probability calculation between a municipality in a large city with a large amount of data (may be a prefecture unit) and a municipality in a small city with a small amount of data (may be a prefecture unit). The method of extracting data is different. First, a method of obtaining the examination probability of the insured in a small city will be described.

  In the case of a small city, since the number of data is small and it is difficult to construct a high-performance consultation probability calculation model, the data of other local governments including the large city accumulated in the medical examination database 1 are used together.

FIG. 2 is an explanatory diagram showing an outline of a processing flow by the processing program.
As a first step, the processing program 2 targets the data of a plurality of local governments accumulated in the medical examination database 1 and extracts the data for the past 6 years from the data for the purpose of creating a teacher data set, and the latest data. The data for 1 year is processed into the presence / absence of medical checkup (correct answer label data), and the data for 5 years other than the correct answer label data is processed into data with 44 explanatory variables as shown in Fig. 4. To do.

  FIG. 4 is a list of explanatory variables, and 44 items such as age, sex, height, weight, BMI, abdominal girth, systolic blood pressure, diastolic blood pressure, and neutral fat of the insured individual who has previously undergone a medical examination. Composed of. FIG. 5 is a diagram showing the structure of teacher data.

  As a second step, the processing program 2 creates a model in which the 44 items are used as explanatory variables and the consultation probability is used as an objective variable. In this embodiment, a model is constructed using a random forest as shown in FIG. From the medical examination data of a plurality of local governments, 1500 samples are randomly extracted, 6 explanatory variables are randomly extracted from 44 explanatory variables, and 100 decision trees are created. The decision tree to be a model is determined by learning by using the health checkup data for the five years and the presence or absence of the health checkup in the most recent year as teacher data.

  As a third step, the processing program 2 calculates the examination probabilities of all insured persons who are residents of the specific small city, using the decided decision tree. All insured persons do not mean all the residents actually, but all the persons whose records of medical examinations in the past exist in the medical examination database 1.

  As the fourth step, the processing program 2 lists the calculated consultation probability of each insured person as an output. If the calculated consultation rate of individual insured persons exceeds, for example, 80%, there is a possibility that the person will undergo a medical examination with a high degree of accuracy, and conversely, a person with a rate lower than 20% will have a high degree of accuracy. There is a possibility that you will not receive a medical examination at. That is, it can be seen that 20% or more and 80% or less of the persons are more likely to receive the health examination by making the recommendation notice.

  Based on the above results, the specific small city, for example, identifies only those who have a probability of 20% or more and 80% or less out of all insured persons, and gives a recommendation notice to the identified persons, thus limiting the budget. , With a limited number of personnel, it is possible to efficiently carry out the recommendation notification work.

  In addition, when the number of shipments of recommendation notifications is budgeted in advance, the probability threshold for shipment may be adjusted and changed as appropriate so as to match the number of shipments. Alternatively, when the number of shipments is set to 1,000 in advance, it is sufficient to select so that the number of intermediate persons, which excludes the persons with high and low probability values, is 1,000.

  The fourth step is only for the insured with an intermediate probability value that is higher than a predetermined high probability value and lower than a predetermined low probability value finally determined by a specific municipality, or a predetermined number of intermediate probability values. Only for the insured of the value, the recommendation notification target list will be created.

  Next, a method for obtaining the examination probability of the insured in the metropolitan government will be described. Contrary to small cities, large cities have a large amount of data, have enough samples to build a high-performance model, and the behavior patterns of rural small cities in particular can be noise. There is. Therefore, the basic data for obtaining the medical examination probability of the insured of a large city municipality should be the data of that municipality only.

  As a first step, the processing program 2 targets only the data of the large city municipality accumulated in the medical examination database 1 and, as in the case of the small city, the data for the past 6 years, Extracted for the purpose of creating a teacher data set, the data for the most recent year is processed into the presence / absence of health checkup (correct answer label data), and the data for 5 years other than the correct answer label data is shown in FIG. Such data is processed into data having 44 explanatory variables. Subsequent processing is the same as in the case of the small city, so description will be omitted.

  In the above-described first embodiment, the medical checkup data for the past 6 years is used for calculation, but it is sufficiently predictable if there is data for 3 years or more. Further, various parameters such as the number of extracted data of the random forest and the number of decision trees may be appropriately determined within the range of the purpose of increasing the accuracy.

  Furthermore, in the above-described first embodiment, an example in which a random forest is used as an algorithm for creating a model is shown, but the present invention is not limited to this, and deep learning may be used. In that case, it is preferable to use the cross entropy as the error function and use Adam as the optimization learning to calculate the consultation probability of each insured person. However, loss functions other than the cross entropy (for example, square error and hinge loss function) are used. Etc.) and other optimization algorithms such as gradient descent method other than Adam (for example, Nestov accelerated gradient, Adagrad, Adadelta, etc.) may be used without any particular limitation.

Next, the processing program 2 according to the second embodiment of the present invention will be described.
The processing program 2 obtains the examination probability of the insured by the same method for the municipalities in large cities with a large amount of data and the municipalities in small cities with a small amount of data.

  FIG. 7 is an explanatory diagram showing an outline of the processing flow. The overall schematic structure is as follows: first create a model using data from multiple local governments, modify a part of the model using data from specific local governments that require probability calculation, and The final model used in the calculation of

  As a first step, the processing program 2 targets the data of a plurality of local governments accumulated in the medical examination database 1 and extracts the data for the past 6 years from the data for the purpose of creating a teacher data set, and the latest data. The data for 1 year is processed into the presence / absence of medical checkup (correct answer label data), and the data for 5 years other than the correct answer label data is processed into data with 44 explanatory variables as shown in Fig. 4. To do.

  As shown in FIG. 8, the algorithm in the first step is deep learning, and the cross entropy is used for the error function and the Adam is used for the optimization learning, and a plurality of weights W1 corresponding to the plurality of layers are generated based on the teacher data set. , W2, W3 are learned to create the first feature amount.

  As shown in FIG. 9, as a second step, the processing program 2 extracts only the data of a specific municipality whose probability is calculated, and uses the weights W1 and W2 created in the first step by using transfer learning as an algorithm. Is left as it is, and the second feature amount in which only W3 is adjusted is created. In transfer learning, cross entropy is used for the error function and Adagrad is used for optimization learning. In deep learning and transfer learning, optimization functions such as loss functions other than cross entropy (for example, squared error and hinge loss function), and gradient descent methods other than Adam and Adagrad (for example, Nestov accelerated gradient and Adelatta). Etc.) may be used and is not particularly limited.

  As the third step, the processing program 2 calculates the consultation probability of all insured persons of the specific municipality whose probability is calculated based on the model based on the above-mentioned second characteristic amount.

  As a fourth step, the processing program 2 uses a method similar to that of the first embodiment to provide the insured with an intermediate probability value excluding a predetermined high probability value and a predetermined low probability value set by a specific municipality. The recommended notification target list only for the insured with a predetermined number of intermediate probability values.

In the second embodiment described above, the medical checkup data for the past 6 years is used for calculation, but it is possible to sufficiently predict if there is data for 3 years or more. Further, various parameters such as the number of layers in deep learning may be appropriately determined within the range of the purpose of increasing the accuracy.

  Although the respective embodiments of the invention of the present application have been described above, according to the proof experiment conducted by the applicant for many local governments and health insurance associations, the difference between the predicted value and the actual measured value is a large city with a large number of data or In a large-scale health insurance association, etc., the system of the second embodiment ≒ the system of the first embodiment (using data of a single local government, a single health insurance association, etc.) ≧ the system of the first embodiment (all local governments or all health insurance (The data of unions, etc.) has a strong tendency, and in a small city with a small amount of data or a small health insurance union, the system of the second embodiment ≧ the system of the first embodiment (all municipalities or all health insurance unions) Etc.)> There was a strong tendency to use the system of the first embodiment (using data of independent municipalities, independent health insurance associations, etc.).

  The above demonstration results are only overall trends, and based on the peculiarities of each local government and each health insurance union, it is possible to appropriately select and decide which system to use by accumulating demonstration experiments. good.

  FIG. 10 is a diagram showing the results of the verification experiment carried out on a prefecture basis, and shows the results of the tests carried out for insurers (local government A, local government B, local government C, local government D, local government E) having data of several hundred levels. is there.

  The consultation rates shown in three bar graphs for each prefecture are the actual consultation rates of those who gave the recommendation notice in the left bar graph, and the bar graph in the center is the probability calculation using the model using only the data of the prefecture. It is a predicted value, and the bar graph on the right is a predicted value calculated by probability using a model in which data of a plurality of insurers is used and transfer learning is performed using data of the insurer.

  As is clear from FIG. 10, in all five insurers with a small number of data, the predicted value calculated by the probability with the transfer learning model is closer to the actual measured value, and thus the insured selected by the method is selected. It is sufficient to give a recommendation notice only to the person who contributes to the budget reduction and personnel reduction of the local government.

  FIG. 11 shows the results of implementation for three insurers (local government X, local government Y, local government Z) having tens of thousands of levels of data. The consultation rate shown by three bar graphs for each insurer is the same as in FIG.

  As is clear from FIG. 11, even in the three insurers with a large amount of data, the predicted value calculated by the probability with the transfer learning model is closer to the actual measured value, but in the X insurer and the Y insurer, It can be understood that there is no great difference from the case of using the independent data.

  FIG. 12 is a diagram showing the difference between the predicted value of the consultation rate of a person to be predicted by an insurer and the actual consultation rate after the recommendation for consultation. As is clear from the figure, the recommended effect for the subjects having the predicted consultation rate of 30 to 70% was obviously high, and the average recommended effect for the predicted subjects was 12.6%.

  Although each of the above-described embodiments has been described as being performed as a service for a local government that is the operator of national health insurance, it can be similarly performed as a service for a health insurance association and the like. In that case, the data of a plurality of health insurance associations may be accumulated in the health checkup database 1, and processing may be performed according to the large-scale health insurance association with a large amount of data and the small-scale health insurance association with a small amount of data.

  As described above, according to the present invention, the insured group of persons who are not likely to receive the medical examination even if the recommendation notification is sent, and the possibility of receiving the medical examination even if the recommendation notification is not sent. It is possible to identify a high insured group with high accuracy, and to identify a person who belongs to a group other than these two groups, that is, an insured group that is likely to be examined by sending a recommendation notice. Therefore, the clerical work related to the recommendation notice can be streamlined and the budget and personnel can be optimized. Further, as a result, the number of insured persons who take a medical examination increases, which can lead to a reduction in medical expenses.

1 Medical examination database 2 Processing program 3 Server

In order to achieve the above-mentioned object, the first invention of the present application is a method for calculating a health examination consultation probability of an insured person, wherein a population serving as basic data for probability calculation is a plurality or a single municipality, or a plurality of or The health checkup data of the health checkup conducted by the insured in the past n years (n is an integer of 3 or more) held by a single health insurance association, etc., and the computer examines the insured based on the health checkup data. According to the processing program, the computer includes a processing program for predicting behavior, and as a first step, extracts data for the past m years (m is an integer of 3 or more, m ≦ n) from all the medical examination data. Then, the data for m-1 years excluding the latest one year is processed into data having x explanatory variables, and as a second step, the medical examination data for the m-1 year and the latest 1 Existence of health checkup for a year Perform learning using as teacher data, wherein the third step, the by model constructed by learning, and calculates the consultation probabilities of individual insured to subscribe to a particular single municipality or specific single health insurance, that And

A second invention of the present application is a method of calculating a probability of receiving a health examination of an insured person, wherein a population serving as basic data for probability calculation is owned by a plurality of local governments or a plurality of health insurance associations in the past n years. (n is an integer of 3 or more) and medical examination of medical examination data insured was admitted between, on the basis of the medical examination data comprises a processing program computer to predict the visit behavior of the insured, the computer According to the processing program, as a first step, data for the past m years (m is an integer of 3 or more, m ≦ n) is extracted from the medical examination data of a plurality of insurers, and the latest one year is extracted. The data for m-1 years that has been removed is processed into data having x explanatory variables, and the algorithm is deep learning, and the data for the m-1 years and the presence / absence of a medical examination in the most recent year are taught. Use as data With, a plurality of weights Wi corresponding to a plurality of layers are learned to create a first feature amount, and as a second step, only the medical examination data of a specific single local government or a specific single health insurance association is extracted. , A second feature amount in which a part of the weight Wi created in the first step is adjusted by using transfer learning as an algorithm, and as a third step, the identification is performed by a model based on the second feature amount. The medical examination probability of each insured person who joins the independent municipality or the specific independent health insurance association is calculated.

A third invention of the present application is a health checkup recommendation notification support system for efficiently giving a recommendation notice to an insured to encourage a health checkup conducted by a local government or a health insurance association based on the calculation of the examination probability. A health checkup database in which the results of health examinations for the past n years (n is an integer of 3 or more) that have been examined by insured persons held by multiple or independent municipalities or multiple or independent health insurance associations; And a server in which a processing program for predicting the medical examination behavior of the insured based on the data accumulated in the medical examination database is recorded, wherein the server follows the processing program and, as a first step, Data for the past m years (m is an integer greater than or equal to 3, m ≦ n) of multiple or independent local governments or multiple or independent health insurance associations was extracted from the medical examination database, and the last one year was excluded. The data for one year, and processed into data having x number of explanatory variables, using as a second step, a medical examination whether visits in the m-1_Nenbun'nodetaoyobichokkin 1 year as teacher data learning As a third step, the model constructed by the learning is used to calculate the consultation probability of each insured who joins a specific municipality or a specific health insurance association, and as a fourth step, the specific municipality or Only for those insured with an intermediate probability value excluding a predetermined high probability value or more and a predetermined low probability value or less specified by the specific health insurance association, etc., or an insured person with a predetermined number of intermediate probability values. It is characterized in that a list of recommended notification targets is created only for.

A fourth invention of the present application is a health checkup recommendation notification support system for efficiently giving a recommendation notice to an insured to encourage a health checkup conducted by a local government or a health insurance association, etc. based on a consultation probability calculation. The medical checkup database in which the results of the medical examinations of the past n years (n is an integer of 3 or more) received by the insured held by a plurality of local governments or a plurality of health insurance associations and the medical examination database are accumulated. And a server in which a processing program for predicting a medical examination behavior of the insured person based on the accumulated data is recorded, the server according to the processing program, and as a first step , the past from the medical examination database. Data for m years (m is an integer greater than or equal to 3, m ≦ n) is extracted and processed into data having x explanatory variables from the data for m-1 years excluding the most recent one year. , The algorithm is This is the sweep learning, and by using the data for the m-1 year and the presence / absence of the medical examination consultation in the most recent year as the teacher data, the plurality of weights Wi corresponding to the plurality of layers are learned to determine the first feature amount. A second feature in which, as a second step, only data of a specific municipality or a specific health insurance association is extracted and a part of the weight Wi created in the first step is adjusted by using transfer learning as an algorithm. Create a quantity, as a third step, by the model based on the second characteristic amount, calculate the consultation probability of each insured to join the specific municipality or the specific health insurance association, the fourth step As for the insured with an intermediate probability value excluding a predetermined high probability value or more and a predetermined low probability value or less defined by the specific municipality or the specific health insurance association, etc., or Only for a predetermined insured of a predetermined number of intermediate probability values, creating a encouraged notified list, characterized in that.

Claims (14)

A method of calculating the probability of receiving a health checkup of the insured,
The population that is the basic data for probability calculation is held by multiple or independent municipalities, or multiple or independent health insurance associations, etc. for the past n years (n is an integer of 3 or more) As medical examination data,
Including a processing program for predicting the examination behavior of the insured based on the medical examination data,
The processing program is
As the first step,
Data for the past m years (m is an integer of 3 or more, m ≦ n) is extracted from all the medical examination data,
The data for m-1 years excluding the latest one year is processed into data with x explanatory variables,
As a second step,
Learning using the medical examination data for the m-1 year and the presence or absence of the medical examination for the most recent year as teacher data,
As the third step,
By the learned model, calculate the consultation probability of individual insured persons who join a specific single municipality or a specific single health insurance association,
A method for calculating a probability of receiving a medical examination of an insured person, which is characterized by the following. The algorithm of the processing program is a random forest, and a samples are randomly extracted from the medical examination data of a plurality of or a single municipality, or a plurality of or a single health insurance association, and from x explanatory variables. b extracted variables are randomly extracted to create y decision trees, and the ensemble of all y decision trees is used to calculate the consultation probability of each individual insured person,
The method for calculating the health checkup probability of an insured person according to claim 1, wherein. The algorithm of the processing program is deep learning, and based on the medical examination data of a plurality of or a single municipality, or a plurality of or a single health insurance association, cross entropy is used as an error function, and Adam is used for optimization learning. , Calculating the consultation probability of the individual insured,
The method for calculating the health checkup probability of an insured person according to claim 1, wherein. The x explanatory variables are 43 or more items created by processing the date of birth, sex, height, weight, various inquiry results, various test values, and the like.
The method for calculating a health checkup probability of an insured person according to any one of claims 1 to 3, wherein: A method of calculating the probability of receiving a health checkup of the insured,
The population, which is the basic data for the probability calculation, is used as the health checkup data of the health checkup conducted by the insured in the past n years (n is an integer of 3 or more) held by a plurality of local governments or health insurance associations,
Including a processing program for predicting the examination behavior of the insured based on the medical examination data,
The processing program is
As the first step,
Data for the past m years (m is an integer of 3 or more, m ≦ n) is extracted from the medical examination data of a plurality of insurers,
The data for m-1 years excluding the latest one year is processed into data with x explanatory variables,
The algorithm is deep learning,
By using the data for the m-1 year and the presence / absence of a medical examination in the most recent year as teacher data, a plurality of weights Wi corresponding to a plurality of layers are learned to create a first feature amount,
As a second step,
Extracting only the medical examination data of a specific single municipality or a specific single health insurance association, and using transfer learning as an algorithm to create a second feature amount by adjusting a part of the weight Wi created in the first step. Then
As the third step,
By the model based on the second characteristic amount, calculate the examination probability of each insured person who joins the specific single municipality or the specific single health insurance association, etc.
A method for calculating a probability of receiving a medical examination for an insured person, which is characterized by the following. The deep learning uses cross entropy for the error function, and Adam for the optimization learning,
In the transfer learning, cross entropy is used for the error function, and Adagrad is used for the optimization learning.
The method for calculating a health checkup probability of an insured person according to claim 5, wherein. The x explanatory variables are 43 or more items created by processing the date of birth, sex, height, weight, various inquiry results, various test values, and the like.
7. The method for calculating the health checkup probability of an insured person according to claim 5 or 6. A medical checkup recommendation notification support system for supporting the efficiency improvement of the recommendation notification to the insured who encourages the health examination conducted by the local government or the health insurance association,
A medical examination database in which the results of medical examinations for the past n years (n is an integer of 3 or more), which have been examined by insured persons held by multiple or independent municipalities or multiple or independent health insurance associations,
A server in which a processing program for predicting the medical examination behavior of the insured person based on the data accumulated in the medical examination database is recorded;
The processing program, as a first step,
Data of the past m years (m is an integer of 3 or more, m ≦ n) of a plurality or a single municipality, or a plurality or a single health insurance association is extracted from the medical examination database,
The data for m-1 years excluding the latest one year is processed into data with x explanatory variables,
As a second step,
Learned by using the data for the m-1 year and the presence or absence of medical examination in the most recent year as teacher data,
As the third step,
By the learned model, calculate the consultation probability of each insured who joins a specific municipality or a specific health insurance association,
As the fourth step,
Only for the insured with an intermediate probability value excluding a predetermined high probability value or more and a predetermined low probability value or less set by the specific municipality or the specific health insurance association, or a predetermined number of intermediate probabilities Create a list of recommended notifications only for insured with value,
A medical examination recommendation notification support system characterized by the following. The algorithm of the processing program is a random forest, a samples are randomly extracted from a plurality of or a single municipality, or a plurality or a single health insurance association, and b explanatory variables are selected from x explanatory variables. To randomly generate y decision trees, and calculate an examination probability of each individual insured by an ensemble of all y decision trees.
The medical examination recommendation notification support system according to claim 8. The algorithm of the processing program is deep learning, based on the medical examination data such as a plurality of or a single municipality, or a plurality of or a single health insurance association accumulated in the medical examination database, and a cross entropy in an error function. And using Adam for optimization learning, calculate the consultation probability of the individual insured,
The medical examination recommendation notification support system according to claim 8. The x explanatory variables are 43 or more items created by processing the date of birth, sex, height, weight, various inquiry results, various test values, and the like.
The medical examination recommendation notification support system according to any one of claims 8 to 10, wherein: A medical checkup recommendation notification support system for supporting the efficiency improvement of the recommendation notification to the insured who encourages the health examination conducted by the local government or the health insurance association,
A medical examination database in which the results of the medical examinations of the past n years (n is an integer of 3 or more) received by the insured held by a plurality of local governments or a plurality of health insurance associations are accumulated,
A server in which a processing program for predicting the medical examination behavior of the insured person based on the data accumulated in the medical examination database is recorded;
The processing program is
As the first step,
Data for the past m years (m is an integer of 3 or more, m ≦ n) is extracted from the medical examination database,
From the data for m-1 years excluding the latest one year, process into data with x explanatory variables,
The algorithm is deep learning,
By using the data for the m-1 year and the presence / absence of a medical examination in the most recent year as teacher data, a plurality of weights Wi corresponding to a plurality of layers are learned to create a first feature amount,
As a second step,
Only data of a specific municipality or a specific health insurance association is extracted, and a second feature amount is created by adjusting a part of the weight Wi created in the first step by using transfer learning as an algorithm,
As the third step,
By the model based on the second feature amount, calculate the consultation probability of each insured person who joins the specific municipality or the specific health insurance association,
As the fourth step,
Only for the insured with an intermediate probability value excluding a predetermined high probability value or more and a predetermined low probability value or less set by the specific municipality or the specific health insurance association, or a predetermined number of intermediate probabilities Create a list of recommended notifications only for insured with value,
A medical examination recommendation notification support system characterized by the following. The deep learning uses cross entropy for the error function, and Adam for the optimization learning,
In the transfer learning, cross entropy is used for the error function, and Adagrad is used for the optimization learning.
The medical examination recommendation notification support system according to claim 12. The x explanatory variables are 43 or more items created by processing the date of birth, sex, height, weight, various inquiry results, various test values, and the like.
14. The medical examination recommendation notification support system according to claim 12 or 13.