Information processing device, information processing method, and program for predicting fracture risk

Abstract

This invention provides an information processing device, an information processing method, and a program that provide users with useful information on fracture risk. [Solution] The program is a program that causes a computer to function as an information processing device that supports fracture prevention, the information processing device comprising: prediction means for predicting the fracture risk of each of the multiple subjects from the health information of the multiple subjects; and selection means for selecting subjects for fracture prevention based on the fracture risk of the multiple subjects.

Description

Technical Field

[0001] The present invention relates to an information processing device, an information processing method, and a program for predicting the risk of fracture.

Background Art

[0002] As a cause of the need for care for the elderly, "fracture and fall" is the fourth most common after "dementia", "cerebrovascular disease (stroke)", and "weakness due to old age" (Ministry of Health, Labour and Welfare "National Living Conditions Survey" (Reiwa元年), URL = https: / / www.mhlw.go.jp / toukei / saikin / hw / k-tyosa / k-tyosa19 / dl / 05.pdf). The cause of fractures in the elderly is osteoporotic fractures, the number of patients with proximal femur fractures is on the increase, and the 5-year mortality rate after proximal femur fractures exceeds 50% (Osteoporosis Foundation, Osteoporosis in Numbers, URL = https: / / www.jpof.or.jp / osteoporosis / tabid265.html). Therefore, measures against osteoporotic fractures in the elderly are required (Non-Patent Document 1).

[0003] In Patent Document 1, a prediction device that predicts the health state of a user based on feature amounts including fracture information using a prediction model learned using data from the National Living Conditions Survey has been proposed.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Non-Patent Documents

[0005]

Non-Patent Document 1

[0006] The prediction device proposed in Patent Document 1 was unable to predict the risk of future fractures. Furthermore, while Non-Patent Document 1 discusses the importance of preventing fragility fractures, it does not suggest any method for predicting fracture risk.

[0007] This invention has been made in view of the above problems, and its objective is to realize a technology that can provide users with useful information on fracture risk. [Means for solving the problem]

[0008] To solve this problem, for example, the program relating to this disclosure is a program that causes a computer to function as various means of an information processing device that supports fracture prevention, wherein the information processing device comprises a prediction means that predicts the fracture risk of each of the multiple subjects from the health information of the multiple subjects, and a selection means that selects subjects for fracture prevention based on the fracture risk of the multiple subjects. [Effects of the Invention]

[0009] According to the present invention, it is possible to provide users with useful information on fracture risk. [Brief explanation of the drawing]

[0010] [Figure 1] A diagram showing an example of a fracture prevention support system according to an embodiment of the present invention. [Figure 2] A block diagram showing an example of the functional configuration of the information processing device according to the embodiment. [Figure 3] A block diagram showing an example of the functional configuration of a terminal device according to this embodiment. [Figure 4] A diagram illustrating an example of the functional configuration of a health information database according to this embodiment. [Figure 5] A flowchart illustrating a series of processes related to fracture prevention according to the embodiment. [Figure 6] A flowchart illustrating the process of generating a fracture risk prediction model according to the embodiment. [Figure 7] A diagram illustrating the process of generating a fracture risk prediction model according to the embodiment. [Figure 8] A diagram illustrating the operation of the fracture risk prediction model according to the embodiment. [Figure 9] A flowchart showing the process for selecting individuals to be targeted for fracture prevention measures according to the embodiment. [Figure 10] A diagram illustrating the behavior change probability estimation model according to this embodiment. [Figure 11] A flowchart illustrating the process of generating a behavior change probability estimation model according to the embodiment. [Figure 12] A diagram illustrating fracture prevention support using behavioral change probability according to the embodiment. [Modes for carrying out the invention]

[0011] The embodiments will be described in detail below with reference to the attached drawings. Note that the following embodiments do not limit the invention as defined in the claims, and not all combinations of features described in the embodiments are essential to the invention. Two or more features from the multiple features described in the embodiments may be combined arbitrarily. Furthermore, identical or similar configurations will be given the same reference numeral, and redundant descriptions will be omitted.

[0012] <First Embodiment> <Configuration of the fracture prevention support system> Referring to FIG. 1, the configuration of the fracture prevention support system according to an embodiment of the present invention will be described. The fracture prevention support system 100 includes, for example, an information processing device 101, a health information database 102, and a terminal device 103 used by a user. Here, the user is assumed to be an employee of an insurer, but is not limited thereto. The information processing device 101, the health information database 102, and the terminal device 103 are communicably connected by a network 104. The network may be any network such as a LAN, a WAN, or the Internet. Also, all or part of the network may be a wireless communication connection.

[0013] Examples of groups and organizations that implement fracture risk countermeasures include local governments, insurers (national health insurance associations, health insurance associations, insurers under the late-stage elderly medical system, etc., hereinafter referred to as "insurers"), pharmacies, and medical institutions. In this specification, local governments, pharmacies, medical institutions, etc. are included under the term "insurers, etc.". Insurers, etc. include local governments, health insurance associations, pharmacies, or medical institutions, but may also include other groups and organizations.

[0014] <Configuration of the information processing device> The information processing device 101 is, for example, a server managed by an insurer, etc., or a server managed by a company that provides services related to data health to an insurer, etc. In the following description, the insurer, local government, and company providing the service are not particularly distinguished, and are simply described as an insurer, etc.

[0015] The terminal device 103 is, for example, a user terminal for accessing the information processing device 101 when an insurer, etc. acquires information related to health guidance. In this embodiment, the case where the terminal device 103 is a desktop personal computer will be described as an example. However, the terminal device 103 is not limited to a desktop personal computer, and may be other devices that can access the information processing device 101, such as a notebook personal computer, a tablet terminal, or a smartphone.

[0016] The health information database 102 is a database that stores data related to various types of health information. Health information may include, for example, data on health checkups, medical claims, and long-term care certification, but it does not have to include all of these, and may also include other types of information. The health information is used for training fracture risk prediction models and for predicting fracture risk using fracture risk prediction models. There may be multiple health information databases 102 for each type of data, or there may be only one. The health information database 102 may be owned by insurers, etc., or by other organizations. It may be the National Database of Medical Claims and Specific Health Checkups (NDB) provided by the Ministry of Health, Labour and Welfare, databases owned by review and payment organizations (such as the Social Insurance Medical Fee Payment Fund, the National Health Insurance Association, and the National Health Insurance Federation) (such as KDB), or other databases owned by private companies. Furthermore, the information processing device 101 may also have a database.

[0017] <Functional Configuration of Information Processing Devices> Next, an example of the functional configuration of the information processing device 101 will be described with reference to Figure 2. Note that each of the functional blocks described may be integrated or separated, and the functions described may be implemented in other blocks. Furthermore, what is described as hardware may be implemented in software, and vice versa.

[0018] The communication unit 201 includes a communication circuit that communicates with various devices via a network. The communication unit 201 receives information processed by the control unit 204 from the communication partner device (e.g., terminal device 103) and transmits information processed by the control unit 204 to the communication partner device (e.g., terminal device 103). The power supply unit 202 is a power supply that provides the power necessary for the operation of the information processing device 101.

[0019] The storage unit 203 includes, for example, a non-volatile storage medium such as a hard disk or semiconductor memory, and includes various programs executed by the control unit 204 of the information processing device 101, various data used by the control unit 204, and DB230. The various programs include a program for performing fracture risk prediction according to this embodiment, as well as an operating system, framework, libraries, etc. The various data includes, for example, setting values ​​of the information processing device 101, data obtained from the health information database 102, data related to the prediction model, training data, and prediction result data. The storage unit 203 also stores the parameters of the prediction model. Data obtained from the health information database 102 may be stored in DB230.

[0020] The control unit 204 includes a central processing unit (CPU) 210 and RAM 211. The control unit 204 controls the operation of various parts within the control unit 204 and the operation of various parts of the information processing device 101 by loading and executing programs stored in the memory unit 203 into the RAM 211. The control unit 204 also performs fracture risk prediction and selection of individuals to be targeted for fracture risk countermeasures, which will be described later.

[0021] RAM211 includes a volatile storage medium such as DRAM, and temporarily stores parameters and processing results for the control unit 204 to execute the program.

[0022] The control unit 204 has a functional block that is implemented by the CPU 210 reading a program stored in the memory unit 203 into the RAM 211. The control unit 204 also has functional blocks including a prediction model unit 220, a selection unit 221, a learning unit 222, a data acquisition unit 223, a user interface (IF) unit 224, and a notification unit 225.

[0023] The prediction model unit 220 predicts an individual's fracture risk using a fracture risk prediction model. The fracture risk prediction model has the function of taking data included in the health information of the health information database 102 as input and outputting the fracture risk (probability). The prediction period may be for fracture risk within one year, or it may be longer or shorter than one year, not limited to one year.

[0024] The selection unit 221 selects individuals who will be targeted for guidance and intervention by insurers, etc., based on the individual fracture risk predicted by the prediction model unit 220.

[0025] The learning unit 222 trains a fracture risk prediction model to generate a trained fracture risk prediction model. The learning unit 222 trains the fracture risk prediction model using data obtained from the health information database 102. The learning unit 222 may, for example, train the fracture risk prediction model using data such as a history of fractures, a diagnosis of osteoporosis, a prescription for osteoporosis medication, or a determination of needing further examination in an osteoporosis screening. The learning unit 222 may also use data on hospitalizations due to fractures or visits to the doctor due to fractures as ground truth data.

[0026] The data acquisition unit 223 acquires data necessary for predicting fracture risk and selecting individuals to take preventative measures from the health information database 102 and other data sources. The data acquisition unit 223 may store the acquired data in the database (DB) 230 of the storage unit 203.

[0027] The user interface unit 224 functions as an interface with the terminal device 103 of the insurer, etc. The user interface unit 224 displays an input screen and a settings screen to the terminal device 103 and receives data input from the terminal device 103.

[0028] The notification unit 225 sends notifications regarding fracture prevention measures to the selected individuals. The notification unit 225 may send notifications directly to the individuals, or it may transmit the individuals' information to the terminal device 103. The insurer, etc., can send notifications regarding fracture prevention measures to the selected individuals transmitted to the terminal device 103 by mail or email.

[0029] <Terminal device configuration> An example of the functional configuration of the terminal device 103 for insurers, etc., will be explained with reference to Figure 3. Note that each of the functional blocks described may be integrated or separated, and the functions described may be implemented in other blocks. Furthermore, what is described as hardware may be implemented in software, and vice versa.

[0030] The communication unit 301 includes, for example, a communication circuit, and communicates with the information processing device 101 via mobile communication such as wired LAN or LTE, or via wireless communication such as WiFi, to send and receive necessary data.

[0031] The operation unit 303 includes buttons and a touch panel on the terminal device 103, and accepts operations from users such as insurers to display information related to health programs. The display unit 304 includes a display panel such as an LCD or OLED, and displays a GUI for various operations. For example, the display unit 304 displays presentation information generated by the notification unit 225 of the information processing device 101.

[0032] The storage unit 305 includes, for example, non-volatile memory such as an HDD or semiconductor memory, and stores programs and the like that the control unit 502 executes.

[0033] The control unit 302 includes a CPU 310 and RAM 311. For example, the CPU 310 executes a program stored in the memory unit 305 to control the operation of each functional block within the control unit 302 and each part within the terminal device 103.

[0034] <Structure of the health information database> Next, with reference to Figure 4, an example of the functional configuration of the health information database 102 will be described. Note that each of the functional blocks described may be integrated or separated, and the functions described may be implemented in other blocks. Also, what is described as hardware may be implemented in software, and vice versa. The health information database 102 is a database server that stores data of various types of health information according to this embodiment. The health information database 102 includes a communication unit 401, a power supply unit 402, a storage unit 403, and a control unit 404. The communication unit 401 includes, for example, a communication circuit, and communicates with the information processing device 101 via mobile communication such as wired LAN or LTE, or communicates with the information processing device 101 via wireless communication such as WiFi to send and receive necessary data. The power supply unit 402 is a power supply that provides the power necessary for the operation of the health information database 102.

[0035] The control unit 404 includes a CPU 410 and RAM 411. For example, the CPU 410 executes a program stored in the memory unit 403 to control the operation of each functional block within the control unit 404 and each part within the health information database 102.

[0036] The storage unit 403 includes, for example, non-volatile memory such as an HDD or semiconductor memory, and stores various health information data and programs executed by the control unit 502. The storage unit 403 stores health checkup data 420, dental checkup data 430, medical treatment data 440, prescription claim data 450, long-term care certification data 460, intervention data 470, and osteoporosis checkup data 480. The health information database 102 does not have to include all of these, and may also include other medical and health-related data. Each piece of data is recorded in association with the date on which a diagnosis, etc., was received for each patient (insured person). Medical treatment data 440 and prescription claim data 450 can be obtained from medical fee claim data. Medical treatment data 440 can be obtained from medical claim, dental claim, and long-term care claim data. Medical, dental, and long-term care claims are created separately for each patient, each month of treatment, and for inpatients and outpatients. Health information data 102 may also store health checkup data 420, dental checkup data 430, medical treatment data 440, prescription claim data 450, long-term care certification data 460, intervention data 470, and osteoporosis screening data 480, linked to the ledger. Furthermore, in recent years, with the use of My Number cards as health insurance cards, claims data is recorded in association with My Number (individual number). Therefore, My Number makes it possible to obtain comprehensive data related to individual medical information, such as records of health checkup data 420, dental checkup data 430, medical treatment data 440, prescription claim data 450, long-term care certification data 460, and osteoporosis screening data 480 for each individual, as well as vital data obtained from wearable devices, electronic medical record data, etc.

[0037] Furthermore, the health information database 102 may acquire and utilize data from the Ministry of Health, Labour and Welfare's claims information and specific health checkup information database (NDB), databases held by review and payment organizations (such as the Social Insurance Medical Fee Payment Fund, the National Health Insurance Association, and the National Health Insurance Federation) (such as KDB), and other databases held by private companies. Alternatively, the health information database 102 may be NDB, KDB, or other databases held by private companies. In NDB, patient (insured person) information is anonymized, but different types of patients are linked by the same hash ID. Also, the data acquisition unit 223 of the information processing device 101 may be configured to access NDB and acquire data. The learning unit 222 and selection unit 221 of the information processing device 101 can use data from NDB, KDB, and other databases.

[0038] The following explains each of the data points. Health checkup data 420 is data from health examinations in a given population. Local governments and corporate health insurance associations conduct health examinations for residents and members. The population of health checkup data 420 may be a single local government or a single company. Alternatively, the population of health checkup data 420 may be multiple local governments or multiple companies. In the following explanation, we will use an example where the insurer is a company that provides health guidance services and uses health checkup data 420 after obtaining permission to use it from multiple local governments and multiple health insurance associations. The same applies to the populations of the following dental checkup data 430, medical treatment data 440, prescription claim data 450, long-term care certification data 460, intervention data 470, and osteoporosis checkup data 480.

[0039] Health checkup data 420 is data accumulated annually, linked to the individual, based on the health checkups received by the target group. Health checkup data 420 stores information such as whether or not a target group received a health checkup, and the measured values ​​for each item of the checkup. Health checkup data 420 includes questionnaire data that the target group answers in advance before the checkup. Health checkup data includes specific health checkup data, which is data on the results of health checkups conducted for target groups aged 40 to 74 focusing on metabolic syndrome, and late-stage elderly health checkup data, which is conducted for target groups aged 75 and over. It may also include health checkup data conducted by insurers etc. for employees based on the Industrial Safety and Health Act. Health checkup data 420 includes data such as examinee information, specific health checkup result information, questionnaire information (medication and smoking history, etc.), determination of eligibility for metabolic syndrome criteria, and determination of eligibility for specific health guidance. Of the 420 health checkup data from the subjects, specific health checkup items included health questions (medication history, smoking history), questionnaires, physical measurements (height, weight, BMI, waist circumference), physical examination, urinalysis (urinary glucose, urinary protein), blood tests (lipids (triglycerides, HDL cholesterol, LDL cholesterol), glucose metabolism (fasting blood glucose or hemoglobin A1C), and liver function (GOT, GPT, γ-GTP)). Additional items included anemia tests (red blood cell count, hemoglobin level, hematocrit), electrocardiogram, fundus examination, and renal and urinary tract examinations.

[0040] Dental checkup data 430 is dental checkup data accumulated annually and associated with the target individuals. Dental checkup data 430 may also include data from regular checkups and specific health checkups. Dental checkup data 430 stores whether the target individuals underwent a dental checkup or not, and the results for each item of the dental checkup. Dental checkup data 430 may also include data from questionnaires that the target individuals answered during the dental checkup.

[0041] Medical data 440 is data recording the patient's visits, admissions, and discharges from medical institutions. It can be obtained from medical and dental claims data created separately for inpatients and outpatients. Medical data 440 includes the name of the medical institution, visit history, date of treatment, and name of medical procedure.

[0042] The prescription data 450 includes information about the medication dispensed at the pharmacy, information about the prescribing source, and information about the patient. The prescription data 450 is linked to each patient. The prescription data 450 includes the name of the medical institution / pharmacy, the date of dispensing, the name of the drug, the name of the active ingredient, the usage, and the dosage.

[0043] The 460 Care Needs Assessment Data set contains data related to care needs assessment. Care needs assessment is determined by a care needs assessment review committee attached to a local government. Based on the application for assessment, the committee calculates the standard care needs assessment time for five areas (direct assistance with daily living, indirect assistance with daily living, BPSD-related activities, functional training-related activities, and medical-related activities), and then determines the level of support needs (levels 1 to 5) based on the sum of these time and the dementia allowance. The 460 Care Needs Assessment Data set may also include "independent" status, which means that the individual does not require assistance such as care services to carry out daily life. Furthermore, the 460 Care Needs Assessment Data set may also include support needs data.

[0044] Intervention data 470 includes data recording measures implemented by insurers, etc., with the implementer, date of implementation, and content associated with them, as well as personal medical data obtainable from electronic medical records and vital data obtainable from wearable devices, etc. Interventions include interventions for individuals and interventions for groups. An example of an intervention for individuals is specific health guidance for those with metabolic syndrome or at risk of metabolic syndrome. An example of an intervention for groups is encouraging those who have not undergone health checkups to do so, and health promotion initiatives by insurers, etc. Regarding osteoporosis, examples include encouraging participation in osteoporosis screenings and encouraging those requiring further assessment to visit medical institutions. This fracture prevention support system handles both group and individual interventions.

[0045] The osteoporosis screening data 480 records data from osteoporosis screenings. Under the Health Promotion Act, osteoporosis screenings are conducted for women aged 40-70 in 5-year increments. Some municipalities expand the age range and sex range independently. In addition to municipalities, health insurance associations also conduct osteoporosis screenings. Osteoporosis screenings may include a medical interview, bone density test, physical measurements, and blood and urine tests. The medical interview involves recording lifestyle habits and dietary content, allowing for an understanding of the subject's nutritional and exercise status. Bone density tests include DXA, MD, and ultrasound, depending on the testing site and whether or not radiation is used. Physical measurements involve continuously measuring the subject's height; a decrease in height suggests osteoporosis. Blood and urine tests measure bone metabolism markers, providing an indicator of the rate of bone metabolism.

[0046] <Machine learning model> The fracture risk prediction model for predicting fracture risk in this embodiment may also use a machine learning model. Various data from the health information database 102 are used as explanatory variables. The data used in the health information database 102 may be some types of data or all types of data. The target variable of the machine learning model is fracture risk.

[0047] <Overall flow of this system> Referring to Figure 5, the overall flow of this embodiment will be explained. The flowchart in Figure 5 is realized when the CPU 210 of the information processing device 101 reads the program stored in the memory unit 203 into the RAM 211 and executes it. Hereafter, the step numbers of each process included in the flowchart are indicated by numbers starting with "S". The same applies to subsequent flowcharts.

[0048] In S501, the learning unit 222 of the information processing device 101 generates a fracture risk prediction model. The fracture risk prediction model is a model that takes health check data and other information for an individual subject as input and outputs the fracture risk as a probability. The fracture risk prediction model may also output fracture risks including secondary fractures and domino fractures, which will be described later, as probabilities. Details of S501 will be described later.

[0049] Next, in S502, the prediction model unit 220 of the information processing device 101 uses a prediction model to obtain the probability of individual fracture risk for a given population. Then, based on the individual fracture risk predicted by the prediction model unit 220, the selection unit 221 selects individuals to whom countermeasures against fracture risk should be taken. For example, individuals to whom countermeasures should be taken are selected in order of highest fracture risk. Details of S502 will be described later.

[0050] Finally, in S503, the notification unit 225 of the information processing device 101 notifies the high-risk individuals selected by the selection unit 221. The user interface unit 224 then transmits the information of the selected individuals to the terminal device 103. The terminal device 103 can display the information of individuals with a high fracture risk on the display unit 304. By reviewing the information of individuals subject to fracture risk countermeasures, insurers can determine whether such countermeasures are necessary for those individuals. The content notified by the notification unit 225 of the information processing device 101 may include not only information on individuals with a high fracture risk, but also effective countermeasures for those individuals. For example, individuals who have discontinued treatment or hospital visits may be advised to visit a medical institution again. If the fracture risk is high, recommendations for bone density testing, initiation of osteoporosis medication and medication guidance, exercise recommendations, and nutritional improvements may be considered. Details of S503 will be described later.

[0051] <Fragility fracture> Before explaining the fracture risk prediction model, let's explain fragility fractures and their risks. Fractures include traumatic fractures, which occur when a strong external force is applied temporarily, such as in a traffic accident or sports incident, and fragility fractures, which occur when the bone strength is weakened due to osteoporosis, etc., and the bone breaks from a small external force, such as falling from standing height. The fractures targeted by the fracture prevention support system of this embodiment are fragility fractures. A small external force in the context of fragility fractures is generally considered to be a force weaker than that of falling from standing height. For example, it is a weak force such as falling and putting your hands out to break your fall, landing on your buttocks, or lifting a heavy object. In such cases, a fracture would not normally occur, but if the bone strength is reduced, it can become a factor in fracture.

[0052] Examples of fragility fractures include distal radius fractures (wrist fractures), proximal humerus fractures (shoulder joint fractures), thoracolumbar vertebral fractures (spine compression fractures), and proximal femur fractures (hip fractures). In particular, proximal femur fractures and thoracolumbar vertebral fractures can affect the prognosis and lead to immobility, so prevention and countermeasures are important.

[0053] Fragility fractures are caused by weakness in the bones throughout the body, and therefore carry the risk of further fractures and secondary fractures. Fragility fractures can occur repeatedly, sometimes referred to as domino fractures. Secondary fractures and domino fractures are thought to be caused by a vicious cycle in which repeated fractures occur because, once a fracture occurs, the body cannot be moved during treatment, leading to decreased muscle strength and reduced stimulation to the bone, which further weakens the bone.

[0054] Fragility fractures can be identified from clinical data 440 based on a history of distal radius fractures, proximal humerus fractures, thoracolumbar vertebral fractures, and proximal femoral fractures. Furthermore, osteoporosis, a contributing factor to fragility fractures, can be identified from the diagnosis in clinical data 440 and the medications prescribed for osteoporosis in prescription claim data 450. Additionally, individuals can be identified from osteoporosis screening data 480 based on whether they received a recommendation for further examination or were designated as priority recipients of osteoporosis screening, such as those with diabetes.

[0055] Osteoporosis often goes unnoticed because it has no noticeable symptoms, and treatment is frequently discontinued. Fragility fractures can lead to being bedridden or requiring long-term care, which contributes to the soaring costs of medical care and long-term care benefits.

[0056] <Generating a fracture risk prediction model> Details of S501 in Figure 5 will be explained with reference to Figures 6 to 8. First, the generation of the fracture risk prediction model will be explained with reference to the flowchart in Figure 6 and the relationship between data and the model in Figure 7. The flowchart in Figure 6 is realized when the CPU 210 of the information processing device 101 reads the program stored in the memory unit 203 into the RAM 211 and executes it. Here, health checkup data 420, dental checkup data 430, medical treatment data 440, prescription claim data 450, and osteoporosis screening data 480 from the health information database 102 are used as training data, but only a part of these may be used, or other data may be included. Data can be appropriately selected from the health information database 102 and used as training data.

[0057] In S601, the data acquisition unit 223 of the information processing device 101 acquires health checkup data 420, dental checkup data 430, medical treatment data 440, prescription claim data 450, and osteoporosis screening data 480 for fiscal year 2021 from the health information database 102 and stores them in the storage unit 203.

[0058] Next, in S602, the data acquisition unit 223 of the information processing device 101 acquires the medical treatment data 440 for fiscal year 2022 from the health information database 102 and stores it in the storage unit 203. In S602, the learning unit 222 may acquire the health checkup values, questionnaires, medical history, prescribed drugs, and osteoporosis screening patterns for fiscal year 2022.

[0059] Next, in S603, the learning unit 222 identifies individuals who were hospitalized or received medical treatment for fractures in fiscal year 2022. The learning unit 222 may also identify individuals who were hospitalized or received medical treatment for fractures in fiscal year 2021, not just fiscal year 2022.

[0060] Next, in S604, the learning unit 222 trains a fracture risk prediction model with the patterns of health check values, questionnaires, medical history, prescribed drugs, and osteoporosis screenings from fiscal year 2021 for subjects who were hospitalized or received medical treatment for fractures in fiscal year 2022. The learning unit 222 may also extract subjects from the medical data 440 based on a history of distal radius fracture, proximal humerus fracture, thoracolumbar vertebral fracture, or proximal femur fracture as subjects who were hospitalized or received medical treatment for fractures. The learning unit 222 may also extract subjects from the medical data 440 based on the diagnosis and the therapeutic drugs for osteoporosis from the prescription data 450 as subjects who were hospitalized or received medical treatment for fractures. Furthermore, the learning unit 222 may also extract subjects from the osteoporosis screening data 480 who were judged to require further examination as subjects who were hospitalized or received medical treatment for fractures.

[0061] This allows the fracture risk prediction model to learn which characteristics indicate a high risk of fracture in individuals. In S604, the learning unit 222 may train the fracture risk prediction model with the patterns of health checkup values, questionnaires, medical history, prescribed medications, and osteoporosis screenings for individuals who were hospitalized or received medical treatment for fractures in fiscal year 2021. The learning algorithm uses known gradient boosting, but is not limited to this.

[0062] The learning unit 222 may, in S602, acquire the patterns of health checkup values, questionnaires, medical history, prescribed medications, and osteoporosis screenings for fiscal year 2022. Then, the learning unit 222 may omit S601 and, in S604, train the fracture risk prediction model using the patterns of health checkup values, questionnaires, medical history, prescribed medications, and osteoporosis screenings for fiscal year 2022. Alternatively, the learning unit 222 may, in S604, train the fracture risk prediction model using the patterns of health checkup values, questionnaires, medical history, prescribed medications, and osteoporosis screenings for both fiscal years 2021 and 2022.

[0063] As a result of the learning unit 222's training in S604, a fracture risk prediction model is generated in S605.

[0064] The fracture risk prediction model described above was trained using health data from fiscal years 2021 and 2022, but any year's health data can be used. Furthermore, the time period is not limited to fiscal years, and the periods do not need to be consecutive. The period can be a calendar year, or a period of two years or more. It can also be in units of months or several months.

[0065] The questionnaire data included in the health checkup data 420 contains information on the subjects' responses to items such as whether or not they use medication, including blood pressure lowering drugs, insulin injections or blood sugar lowering drugs, cholesterol lowering drugs, medical history such as stroke, heart disease, and renal failure, smoking habits, weight gain since age 20, exercise habits, eating habits, drinking habits, and sleep status. Since this questionnaire data is information that the subjects themselves have identified, it can serve as an indicator for risk prediction by fracture risk prediction models. For example, people who do not exercise may have low bone density and tend to have a higher risk of fracture. Also, drinking habits tend to increase the risk of falls due to alcohol consumption, and therefore tend to increase the risk of fracture.

[0066] Figure 8 shows the operation of the fracture risk prediction model generated by the learning unit 222. The fracture risk prediction model generated by the flow in Figure 6 takes health check data, dental check data, medical treatment data, prescription claim data, and osteoporosis screening data of a given subject as input and outputs the fracture risk of that subject as a probability. The input and output of the fracture risk prediction model do not necessarily have to match the training data used to train the fracture risk prediction model. For example, the fracture risk prediction model may be trained using 5 years of data, and the fracture risk may be output when a subject's health information is input. The prediction period may be for fracture risk within one year, or it may be shorter or longer than one year. Furthermore, fracture risk may be predicted by inputting health information for multiple years. Fracture risk may also be predicted by inputting multiple non-consecutive health information records.

[0067] <Persons eligible for fracture prevention measures> Referring to Figure 9, the detailed processing of S502 in Figure 5 will be explained. The flowchart in Figure 9 is realized when the CPU 210 of the information processing device 101 reads the program stored in the memory unit 203 into the RAM 211 and executes it.

[0068] The fracture risk prediction model outputs an individual's risk of fracture as a probability. Using the fracture risk prediction model, it is possible to predict the fracture risk (probability) of a given subject by inputting health check data, dental check data, medical treatment data, prescription claim data, and osteoporosis screening data. The prediction period can be for fracture risk within one year, or it can be shorter or longer than one year. In this embodiment, the fracture risk prediction model can predict not only the fracture risk of subjects with no prior history of fractures, but also the fracture risk of recurrent fractures, secondary fractures, and domino fractures.

[0069] Here, the group for which fracture risk is predicted does not necessarily have to be the same as the group of subjects used to generate the fracture risk prediction model. Generally, accuracy is required when generating a prediction model, so a larger amount of data is better. On the other hand, risk prediction using a prediction model requires the selection of target individuals for countermeasures, taking into account the size of the insurer, etc., and the composition of the target individuals, so the target group is the group that the insurer, etc., will target for fracture countermeasures.

[0070] In S901, the prediction model unit 220 of the information processing device 101 first predicts the individual fracture risk (probability) for all individuals covered by fracture prevention measures, such as those covered by insurers, using a fracture risk prediction model.

[0071] Next, in S902, the selection unit 221 of the information processing device 101 ranks the subjects in descending order of fracture risk. Then, in S903, the selection unit 221 selects subjects with high risk. The selection unit 221 may select subjects with a risk level above a predetermined level, or until a predetermined number is reached, or at a predetermined ratio to the total number of subjects.

[0072] In S902, the ranking may include not only the risk of fracture (probability of fracture), but also the following rankings from the perspective of encouraging medical consultation and leading to treatment. In this case, the risk prediction using the fracture risk prediction model in S901 may be omitted. (1) From the 440 medical data entries, select those who have not received treatment among those with a history of conditions such as distal radius fracture, proximal humerus fracture, thoracolumbar vertebral fracture, or proximal femur fracture. (2) Select individuals who have discontinued treatment for osteoporosis. (3) Select individuals who have been determined to require further examination during an osteoporosis screening and who have not yet visited a medical institution. (4) Select individuals who are at high risk of fracture but have not undergone osteoporosis screening.

[0073] The learning unit 222 can also train the fracture risk prediction model to increase the risk for the individuals described in (1) to (4) above. The selection unit 221 may prioritize the individuals described in (1) to (4) above as targets for fracture prevention measures and output the factors that led to their selection as reference information for the selected individuals. By referring to the reference information for the selected individuals, insurers and others can implement effective fracture prevention measures for those individuals.

[0074] <Second Embodiment> In the first embodiment, individuals with a high risk of fracture were selected as subjects for fracture prevention measures. In the second embodiment, subjects for prevention measures are selected based on both the risk of fracture and the probability of behavioral change.

[0075] Behavioral change refers to a change in a person's behavior as a result of intervention by an insurer or other relevant party. For example, a person who has not visited a medical institution may start visiting one after being encouraged to do so by their insurer or other party. Another example is a person receiving guidance from their insurer or other party on improving their lifestyle and then working to improve it. The probability of behavioral change is the probability that a person's behavior will change as a result of intervention by an insurer or other relevant party. For example, the probability that a person will visit a medical institution after being encouraged to do so by their insurer or other party.

[0076] When insurers or other organizations encourage eligible individuals to visit medical institutions, they may respond differently, such as completely ignoring the recommendation or failing to receive treatment or improvement even after visiting. The content of the recommendation also makes a difference. The way individuals perceive and act upon the recommendation differs depending on whether the insurer sends direct mail or makes individual phone calls or visits to encourage them to visit medical institutions. Furthermore, individuals who have never had a health checkup and those who receive health checkups annually have different health awareness levels and therefore respond differently to recommendations.

[0077] <Generation of a behavior change probability estimation model> In this embodiment, in response to intervention by insurers, etc., a behavioral change probability estimation model is trained using health check data 420 and medical treatment data 440 of subjects who visited medical institutions, etc. as training data.

[0078] The behavioral change probability estimation models of this embodiment include a consultation probability model, a response model, and an uplift model. Each will be explained with reference to Figure 10.

[0079] Figure 10(a) is a conceptual diagram illustrating the generation of a visit probability model. The visit probability model uses health check data 420, dental check data 430, medical treatment data 440, prescription claim data 450, and osteoporosis screening data 480 as training data. The data acquisition unit 223 of the information processing device 101 acquires the health check data 420, dental check data 430, medical treatment data 440, prescription claim data 450, and osteoporosis screening data 480 from the health information database 102 and stores them in the storage unit 203. The period for the health check data 420, dental check data 430, medical treatment data 440, prescription claim data 450, and osteoporosis screening data 480 acquired by the data acquisition unit 223 is arbitrary. Furthermore, only a portion of this data may be used, or other data may be included.

[0080] Next, the learning unit 222 identifies the subjects who actually took action. For example, it identifies subjects who underwent a health checkup. From the health checkup data 420, dental checkup data 430, medical treatment data 440, and prescription claim data 450 of the subjects who underwent a health checkup, the learning unit 222 identifies attributes that indicate a tendency to undergo a health checkup and generates a probability model of participation. Subjects who underwent a dental checkup instead of, or in addition to, a health checkup may also be identified.

[0081] This allows the probability model to output the probability of a person receiving a health checkup when health checkup data 420, dental checkup data 430, medical treatment data 440, and prescription claim data 450 are input. Here, assuming that individuals who receive health checkups also have a high tendency to visit medical institutions, the output of the probability model is considered to be the probability that a given individual will visit a medical institution. The probability model is an estimation model that outputs the probability that an individual will visit a medical institution even without intervention by the insurer or other relevant party encouraging the individual to do so.

[0082] Next, the response model will be explained. Figure 10(b) is a conceptual diagram illustrating the generation of the response model. In the response model, health checkup data 420, dental checkup data 430, medical treatment data 440, prescription claim data 450, osteoporosis screening data 480, and intervention data 470 are used as training data. The data acquisition unit 223 of the information processing device 101 acquires the health checkup data 420, dental checkup data 430, medical treatment data 440, prescription claim data 450, osteoporosis screening data 480, and intervention data 470 from the health information database 102 and stores them in the storage unit 203. The period for the health checkup data 420, medical treatment data 440, prescription claim data 450, and intervention data 470 acquired by the data acquisition unit 223 is arbitrary. Furthermore, only a portion of this data may be used, or other data may be included.

[0083] Intervention data 470 includes, for example, data from notifications issued by insurers, etc., encouraging subjects to visit medical institutions, etc., based on their health checkup data 420. The learning unit 222 identifies subjects whose behavior has changed as a result of the intervention, based on the intervention data 470. Here, for example, it identifies subjects who visited medical institutions, etc., in accordance with recommendations from insurers, etc., and who did so. The learning unit 222 identifies attributes that indicate a tendency to visit medical institutions, etc., from the health checkup data 420, dental checkup data 430, medical treatment data 440, prescription claim data 450, and osteoporosis checkup data 480 of subjects who visited medical institutions, etc., in accordance with recommendations, and generates a response model. In this case, the learning unit 222 may, in addition to attributes that tend to visit medical institutions, use the health checkup data 420, dental checkup data 430, medical treatment data 440, prescription claim data 450, and osteoporosis screening data 480 of individuals who were encouraged by insurers, etc., to visit medical institutions but did not, to identify attributes that make it difficult to visit medical institutions, etc., and use them as learning data to generate a response model.

[0084] When health checkup data 420, dental checkup data 430, medical treatment data 440, prescription claim data 450, and osteoporosis screening data 480 for a given individual are input into a response model, the model outputs the probability that the individual will actually visit a medical institution if they are encouraged to do so. The response model has the advantage of being able to generate an estimation model even when insurers or other organizations have little experience in encouraging visits.

[0085] Next, the uplift model will be explained with reference to Figures 10(c) and 11. Figure 10(c) is a conceptual diagram illustrating the generation of the uplift model. The flowchart in Figure 11 is realized when the CPU 210 of the information processing device 101 reads the program stored in the memory unit 203 into the RAM 211 and executes it.

[0086] In the uplift model, health checkup data 420, dental checkup data 430, medical treatment data 440, prescription claim data 450, osteoporosis screening data 480, and intervention data 470 are used as training data. In S1101, the data acquisition unit 223 of the information processing device 101 acquires health checkup data 420, dental checkup data 430, medical treatment data 440, prescription claim data 450, and osteoporosis screening data 480 from the health information database 102 and stores them in the storage unit 203. The period for the health checkup data 420, dental checkup data 430, medical treatment data 440, prescription claim data 450, and osteoporosis screening data 480 acquired by the data acquisition unit 223 is arbitrary.

[0087] In S1102, the data acquisition unit 223 of the information processing device 101 acquires intervention data 470 and stores it in the storage unit 203. The intervention data 470 includes, for example, data of notifications issued by insurers, etc., to individuals recommending that they visit a medical institution, etc., based on the individual's health checkup data 420. The intervention data 470 here includes, for example, data of individuals who were not recommended to visit a medical institution, etc., data of individuals who were recommended to visit a medical institution, etc., by regular paper notice, and data of individuals who were recommended to visit a medical institution, etc., by detailed paper notice. The recommendation notice is not limited to these and may also include, for example, telephone, email, or visit, or it may be a further modification of the content of the paper notice. The intervention data 470 prepared as training data may be notifications indicating whether or not a visit to a medical institution, etc. was recommended, or notifications indicating recommendations at multiple different levels.

[0088] Next, in S1103, the learning unit 222 of the information processing device 101 trains the uplift model on the attributes of individuals who actually visited a medical institution, etc., for those who were not notified to visit a medical institution, etc. (non-intervention). As a result, the uplift model can estimate the probability of a visit according to the attributes of the individual in the non-intervention case.

[0089] Next, in S1104, the learning unit 222 of the information processing device 101 trains the uplift model on the attributes of individuals who actually visited a medical institution or other facility after being encouraged to do so through paper-based notification (intervention). This allows the uplift model to estimate the probability of a person visiting a medical institution or other facility based on their attributes in the event of intervention.

[0090] Next, in S1105, the learning unit 222 of the information processing device 101 compares the probability of seeking medical attention in the case of no intervention as determined in S1103 with the probability of seeking medical attention in the case of notification by paper (intervention) as determined in S1104. For example, if the probability of seeking medical attention estimated in S1104 exceeds the probability of seeking medical attention estimated in S1103, the learning unit 222 can determine that this is the effect of notification by paper (intervention), excluding the effect of no intervention. Since individuals with a high level of health consciousness will voluntarily seek medical attention even without intervention from insurers, etc., S1105 can determine the effect of the intervention, excluding the effect of not intervening, and the increase in the probability of seeking medical attention as a result of the intervention.

[0091] In S1106, the learning unit 222 trains the uplift model on the difference between the probability of seeking medical attention in the non-intervention case and the probability of seeking medical attention in the intervention case, according to the attributes of the subject. In the example above, the learning unit 222 compared the probability of seeking medical attention in the non-intervention case and the probability of seeking medical attention in the intervention case, but it may also compare the probability of seeking medical attention when a normal notification is given with the probability of seeking medical attention when a detailed notification is given. In this case, it is possible to determine the increase in the probability that a person will not seek medical attention with a normal notification, but will seek medical attention when a detailed notification is given.

[0092] When health check data 420, dental check data 430, medical treatment data 440, prescription claim data 450, and osteoporosis screening data 480 of a given subject are input into the generated uplift model, the increase in the probability that the subject will actually visit a medical institution if they are encouraged to do so is output.

[0093] The learning unit 222 may further validate and retrain the generated uplift model. In S1107, the learning unit 222 inputs health checkup data 420, medical treatment data 440, and prescription claim data 450, and outputs the increase in the probability that the subject will visit a medical institution. The learning unit 222 may then retrain the uplift model depending on whether the intervened subject actually visited a medical institution (took action). If the subject actually visited a medical institution (took action), the uplift model may be modified to increase the output probability, and if the subject does not visit a medical institution (take action) within a certain period, the uplift model may be modified to decrease the output probability.

[0094] According to the behavior change probability estimation model, by inputting health information, the probability of behavior change for a target individual can be output. For example, the behavior change probability estimation model can estimate the following: Target individuals who are older or who underwent a health checkup in the previous year are more likely to undergo a health checkup when encouraged. Target individuals who have not had a health checkup for 1-2 years, or who only undergo a health checkup once every few years, are more likely to undergo a health checkup when encouraged. On the other hand, target individuals who have never undergone a health checkup in the past are less likely to undergo a health checkup even when encouraged. Target individuals who visit medical institutions with a certain frequency are more likely to respond to encouragement. Encouraging visits by telephone elicits a higher response rate from the elderly. In this way, the behavior change probability estimation model outputs the probability of behavior change by inputting health information data.

[0095] <Selection of participants using a behavior change probability estimation model> Next, with reference to Figure 12, the selection of individuals to be targeted for countermeasures using the fracture risk prediction model and the behavior change probability estimation model will be explained. The behavior change probability estimation model may be any of the following: a consultation probability model, a response model, or an uplift model. In the second embodiment, the behavior change probability estimation model is combined with the fracture risk prediction model of the first embodiment to select individuals.

[0096] First, the prediction model unit 220 of the information processing device 101 inputs the subject's individual health checkup data, dental checkup data, medical treatment data, prescription claim data, and osteoporosis health checkup data (1201) into the fracture risk prediction model 1202. Then, the fracture risk (probability) 1203 for a given subject is output. The prediction period can be for fracture risk within one year, or it can be longer or shorter than one year.

[0097] Furthermore, the prediction model unit 220 of the information processing device 101 inputs the individual's health checkup data, dental checkup data, medical treatment data, prescription claim data, and osteoporosis screening data (1204) into the behavior change probability estimation model 1205. This allows the system to obtain the probability 1206 that a particular individual will visit a medical institution if an insurer intervenes.

[0098] Next, the prediction model unit 220 calculates the probability of a subject visiting a medical institution as a result of an intervention, 1207, relative to the subject's fracture risk (probability) 1203 and the probability of a subject visiting a medical institution as a result of an intervention, 1206. The calculation performed by the prediction model unit 220 may also involve weighting either the fracture risk 1203 or the behavior change probability 1206 before taking the product.

[0099] By incorporating the probability of behavioral change resulting from intervention into the fracture risk predicted by the fracture risk prediction model, it is possible to identify individuals with a high fracture risk who require intervention and who are more likely to seek medical attention if insurers or other organizations intervene. After calculating the probability (1207) of an individual seeking medical attention through intervention based on their fracture risk, individuals to be targeted for intervention are selected according to the flow chart in Figure 9. According to the second embodiment, it is possible to select individuals for whom intervention is more effective.

[0100] The invention is not limited to the embodiments described above, and various modifications and changes are possible within the scope of the gist of the invention. [Explanation of symbols]

[0101] 101: Information processing device, 102: Health information database, 103: Terminal device

Claims

1. A program that causes a computer to function as one of the means of an information processing device that supports fracture prevention, wherein the information processing device is A prediction means for predicting the fracture risk of each of the multiple subjects from the health information of the multiple subjects, A selection means for selecting individuals to be targeted for fracture prevention measures based on the fracture risk of the aforementioned multiple individuals, A program that includes the following features.

2. A program that causes a computer to function as one of the means of an information processing device that supports fracture prevention, wherein the information processing device is A prediction means for predicting the fracture risk of each of the multiple subjects from the health information of the multiple subjects, An estimation means for estimating the probability of behavioral change due to intervention for each of the aforementioned multiple subjects, based on the health information of the aforementioned multiple subjects, A selection means for selecting individuals to be targeted for fracture prevention measures based on the fracture risk and behavioral change probability of the aforementioned multiple individuals, A program that includes the following features.

3. The aforementioned prediction means includes a prediction model that learns from the health information of the person who suffered the fracture. The program according to claim 1 or 2.

4. The selection means selects individuals to be targeted for countermeasures until the number exceeds a predetermined number or a predetermined percentage. The program according to claim 1 or 2.

5. The selection means outputs the selected subject along with the factors that led to the selection of the subject. The program according to claim 1 or 2.

6. The estimation means includes an estimation model that learns from information on interventions and information on behavioral changes. The program according to claim 2.

7. The program according to claim 6, wherein the estimation model divides the subjects into two or more groups according to whether or not an intervention is performed or the level of the intervention, and performs learning based on whether or not there is a change in the behavior of the subjects in each group.

8. An information processing device that supports fracture prevention measures, A prediction means for predicting the fracture risk of each of the multiple subjects from the health information of the multiple subjects, The system includes a selection means for selecting individuals to be targeted for fracture prevention measures based on the fracture risk of the aforementioned multiple individuals. Information processing device.

9. An information processing device that supports fracture prevention measures, A prediction means for predicting the fracture risk of each of the multiple subjects from the health information of the multiple subjects, An estimation means for estimating the probability of behavioral change due to intervention for each of the aforementioned multiple subjects, based on the health information of the aforementioned multiple subjects, The system includes a selection means for selecting individuals to be targeted for fracture prevention measures based on the fracture risk and behavioral change probability of the aforementioned multiple individuals. Information processing device.

10. An information processing method performed by an information processing device that supports fracture prevention, A prediction process that predicts the fracture risk of each of the multiple subjects based on the health information of the multiple subjects, The system includes a selection step for selecting individuals to be targeted for fracture prevention measures based on the fracture risk of the aforementioned multiple individuals. Information processing methods.

11. An information processing method performed by an information processing device that supports fracture prevention, A prediction process that predicts the fracture risk of each of the multiple subjects based on the health information of the multiple subjects, An estimation process for estimating the probability of behavioral change due to intervention for each of the aforementioned multiple subjects, based on their health information, The system includes a selection step for selecting individuals to be targeted for fracture prevention measures based on the fracture risk and behavioral change probability of the aforementioned multiple individuals. Information processing methods.

Images