Health information management device and health information management method

Abstract

From the various items included in the user's health information, appropriately identify the items that are suitable for understanding the user's health status. [Solution] The health information management device receives health information provided by multiple users, each containing explanatory variables pre-selected to describe the health status of each user, and multiple variables different from the explanatory variables. It stores this information in association with each user. From the multiple variables, it extracts candidate explanatory variables that can describe the health status of multiple first users associated with a first attribute among the multiple users. From the group of candidate important factors consisting of the explanatory variables and candidate explanatory variables, it identifies candidate important factors that have an influence on other variables included in the health information as important factors.

Description

Technical Field

[0001] The present disclosure relates to a health information management device and a health information management method for managing health information submitted by a plurality of users.

Background Art

[0002] By estimating the health status of a user from the health information provided by the user, extracting users at risk of health risks, and appropriately dealing with them, it becomes possible to maintain the health status of the user well. Among the plurality of items included in the health information, by referring to the items that have a great influence on the health status, users at risk of health risks can be appropriately extracted.

[0003] Patent Document 1 describes a system that can quantify the time data and smoking burden of smoking activities and identify a smoking trigger unique to an individual.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] In the system described in Patent Document 1, it is predetermined to refer to the time data and smoking burden among the plurality of items included in the health information. That is, the system administrator needs to consider a large number of health information and the health status of each user in advance and select items suitable for extracting users at risk of health risks.

[0006] Therefore, an object of the present disclosure is to provide a health information management device and a health information management method that can appropriately identify items suitable for grasping the health status of a user from a plurality of items included in the health information of the user.

Means for Solving the Problems

[0007] The gist of this disclosure is as follows:

[0008] (1) A health information receiving unit that receives health information provided by multiple users, each containing explanatory variables pre-selected to describe the health status of each user, and multiple variables different from the explanatory variables, and stores it in association with each user. A candidate extraction unit extracts candidate explanatory variables from the aforementioned plurality of variables that can explain the health status of a plurality of first users to which a first attribute is associated, among the plurality of users. A factor identification unit identifies candidate key factors that have an influence on other variables included in the health information from a group of candidate key factors consisting of the explanatory variables and candidate explanatory variables, A health information management device equipped with the following features.

[0009] (2) The health information management device according to (1) above, wherein the factor identification unit identifies a candidate important factor from the group of candidate important factors that has an influence on one variable predetermined from the explanatory variable and the plurality of variables, as the important factor.

[0010] (3) The factor identification unit is, Using each of the health information samples, each separately extracted from the aforementioned health information and consisting of the aforementioned candidate important factors, the regression coefficients are calculated by penalized regression analysis, which adds a penalty term to the loss function, based on a regression model for each of the aforementioned candidate important factors. Among the candidate group of key factors, those key factors that are frequently found to have non-zero regression coefficients in the penalty-based regression analysis are given priority and identified as key factors. A health information management device as described in (1) or (2) above.

[0011] (4) A health information management device according to any one of (1) to (3) above, further comprising a user extraction unit that extracts a second user from among the plurality of users whose value of the important factor in the corresponding health information is within a predetermined range as a user whose health status should be noted.

[0012] (5) The health information management device according to any one of (1) to (4) above, wherein the candidate extraction unit extracts the candidate explanatory variable using a machine learning model that has been pre-trained to extract the candidate explanatory variable from the plurality of variables contained in the health information.

[0013] (6) The machine learning model is a large-scale language model, as described in (5) above, for the health information management device.

[0014] (7) A health information receiving step that receives health information provided by multiple users, each including explanatory variables that have been pre-selected to describe the health status of each user, and multiple variables different from the explanatory variables, and stores the information associated with each user. A candidate extraction step of extracting candidate explanatory variables from the aforementioned plurality of variables that can explain the health status of a plurality of first users to which a first attribute is associated, among the plurality of users, A factor identification step of identifying key factors from a group of key factor candidates consisting of the explanatory variables and the explanatory variable candidates, which have an influence on other variables included in the health information, as key factors, A method for managing health information, including the following.

[0015] (8) The health information management method according to (7) above, wherein in the factor identification step, a candidate important factor that has an effect on the explanatory variable and one variable predetermined from among the multiple variables is identified as the important factor from the group of candidate important factors.

[0016] (9) The health information management method according to (7) or (8) above, wherein in the candidate extraction step, candidate explanatory variables are extracted using a machine learning model that has been pre-trained to extract candidate explanatory variables from the explanatory variables and the plurality of variables.

Advantages of the Invention

[0017] According to the health information management device and the health information management method according to the present disclosure, it is possible to appropriately identify items suitable for grasping the health status of the user from a plurality of items included in the health information of the user.

Brief Description of the Drawings

[0018] [Figure 1] FIG. 1 is a schematic diagram showing a schematic configuration of a health information management system. [Figure 2] FIG. 2 is a schematic diagram showing a schematic configuration of a health information management device. [Figure 3] FIG. 3 is a diagram for explaining data stored in a user attribute database. [Figure 4] FIG. 4 is a diagram for explaining data stored in a health information database. [Figure 5] FIG. 5 is a functional block diagram of a processor included in a health information management device. [Figure 6] FIG. 6 is a diagram for explaining the calculation of the frequency of candidate important factors for which the regression coefficient becomes non-zero. [Figure 7] FIG. 7 is a flowchart of a health information management process.

Modes for Carrying Out the Invention

[0019] Hereinafter, referring to the drawings, a health information management device and a health information management method capable of appropriately identifying items suitable for grasping the health status of a user from a plurality of items included in health information indicating the health status of the user will be described in detail. However, it should be understood that the present invention is not limited to the drawings or the embodiments described below.

[0020] FIG. 1 is a schematic diagram showing a schematic configuration of a health information management system.

[0021] The health information management system 100 includes a health information management device 1, a first input terminal 2, and a second input terminal 3.

[0022] Each of the first input terminal 2 and the second input terminal 3 is connected to the health information management device 1 via a wireless base station WBS and a communication network NW. Multiple wireless base stations WBS may be connected to the communication network.

[0023] Each of the first input terminal 2 and the second input terminal 3 may be directly connected to the communication network NW. In this case, each of the first input terminal 2 and the multiple second input terminals 3 is connected to the health information management device 1 in a communicative manner without going through the wireless base station WBS.

[0024] The first input terminal 2 is used by user U1 (hereinafter, a user corresponding to user U1 is also referred to as a "user associated with the first attribute," and such a user associated with the first attribute is also referred to as the "first user"). User U1 may be a user arbitrarily selected from a group of users. Preferably, the first attribute refers to a predetermined attribute of an evaluation item among the health status items. For example, if the evaluation item is an item indicating whether or not the person is on leave from work, and the first attribute is an attribute indicating that the person is on leave from work, then user U1 may be an employee on leave from work. The health information management system 100 may have multiple first input terminals 2 depending on the number of first users.

[0025] The second input terminal 3 is used by users U2-1 and U2-2 (hereinafter collectively referred to as "User U2," and users corresponding to User U2 are also referred to as "Users associated with the second attribute," and such users associated with the second attribute are also referred to as "Second Users") who are different from the first user. User U2 may be a user arbitrarily selected from multiple users, separate from User U1. Preferably, the second attribute refers to a predetermined attribute different from the first attribute regarding the evaluation item (User U2 is a user among multiple users who does not correspond to User U1). For example, if the evaluation item is an item indicating whether or not the employee is on leave from work, and the second attribute is an attribute indicating that the employee is not on leave from work, then User U2 may be an employee who is not on leave from work.

[0026] The items to be evaluated refer to the items that indicate a user's health status and are subject to evaluation. As explained here, each user may have attributes associated with the items to be evaluated. Through such attribute associations, each user is classified based on the items to be evaluated. In the example above, there are two attributes for the items to be evaluated, but the number of attributes is not limited to this and may be three or more.

[0027] In the health information management system 100, one user may use multiple terminals. Furthermore, multiple users may each use multiple accounts on a single terminal. When the first user and the second user use the same terminal with different accounts, that terminal becomes the first input terminal 2 when used with the account corresponding to the first user, and the second input terminal when used with the account corresponding to the second user.

[0028] The health information management device 1, the first input terminal 2, and the second input terminal 3 are each configured by executing a predetermined computer program on an information processing device such as a server computer, personal computer, tablet, or smartphone.

[0029] Figure 2 is a schematic diagram showing the general configuration of the health information management device 1.

[0030] The health information management device 1 comprises an input / output interface 11, a memory 12, and a processor 13.

[0031] The input / output interface 11 has an interface circuit for receiving data to be processed by the health information management device 1, or for outputting data processed by the health information management device 1. The input / output interface 11 includes, for example, a communication interface circuit for connecting the health information management device 1 to a communication network NW, or a peripheral device interface circuit for connecting the health information management device 1 to various peripheral devices such as a keyboard, mouse, and display.

[0032] The input / output interface 11 receives health information input from the first input terminal 2 and the second input terminal 3 via the communication interface circuit. The input / output interface 11 also outputs output data to output devices such as displays to show identified important factors via the peripheral device interface circuit.

[0033] The memory 12 includes, for example, at least one of a semiconductor memory, a magnetic disk device, and an optical disk device. The memory 12 stores various programs, various data, etc., used for processing by the processor 13.

[0034] The various programs include driver programs, operating system programs, and computer programs for health information management. These programs may be provided in the form of computer-readable portable recording media, such as semiconductor memory, magnetic recording media, or optical recording media.

[0035] The various data include a user attribute database that associates user attributes with each user, and a health information database that associates health information with each user.

[0036] Figure 3 illustrates the data stored in the user attribute database.

[0037] In the user attribute database 121 stored in memory 12, user attributes and categories are associated with each user. For example, the user with user identifier "0001" is associated with user attribute "B" and category "P", while the user with user identifier "0002" is associated with user attribute "A" and category "Q".

[0038] User attribute "A" corresponds to the first attribute described above, and user attribute "B" corresponds to the second attribute described above. Therefore, in the example in Figure 3, the user with user identifier "0001" is the second user, and the user with user identifier "0002" is the first user.

[0039] The categories correspond to classifications based on predetermined criteria for users, and may correspond to either the industry or place of employment for users who are employees. In categories corresponding to a user's industry, categories "P," "Q," and "R" may correspond to, for example, "Sales," "Technical," and "Administrative," respectively. Any one of categories "P," "Q," and "R" can also be called the first category. In the user attribute database 121, being associated with a category can be said to mean being classified into that category.

[0040] Figure 4 illustrates the data stored in the health information database.

[0041] In the health information database 122 stored in memory 12, health information is associated for each user and for each reporting date. As shown in Figure 4, health information may be divided into multiple categories such as health status, work performance, and daily activities.

[0042] The strings A, B, C, etc., stored for each item in the health information represent the score for that item. In other words, each string may represent the numerical range of the corresponding item's value within that score.

[0043] If the score is stored as a string corresponding to a numerical range, mathematical processing on the score may be performed using representative values ​​of the numerical range (e.g., median, maximum, minimum, etc.) as the score value.

[0044] In Figure 4, for the sake of simplicity, the scores are represented by strings corresponding to numerical ranges, but the scores may also be stored as numerical values ​​in the health information database 122. It is preferable that the scores are normalized for each item. Furthermore, any of the multiple items (e.g., daily activities) may be aggregate values ​​based on the scores of the corresponding sub-items. Each of these items representing a score can also be called a variable.

[0045] Health information includes explanatory variables and multiple variables distinct from the explanatory variables. Explanatory variables are variables that describe the health status of each user. Explanatory variables may be pre-selected, for example, by the administrator of the health information management device 1. In addition, important factors identified by the processing described later may be used as explanatory variables in subsequent processing. In the health information database 122 shown in Figure 4, the variable "health status" is pre-selected as an explanatory variable.

[0046] Explanatory variables may also be variables that represent the items being evaluated. In other words, health information may include variables corresponding to user attributes included in the user attribute database 121, and these may be selected as explanatory variables.

[0047] In this case, the user attribute database 121 and the health information database 122 may each store the evaluation items independently. If the data representing the evaluation items differs between the user attribute database 121 and the health information database 122, the data will be processed according to predetermined processing rules (for example, prioritizing the data in the health information database 122).

[0048] Alternatively, the evaluation items for users whose health information does not include evaluation items may be read from the user attribute database 121, and together with the received health information, they may be associated with the user as health information and stored in the health information database 122. Storing the evaluation items (explanatory variables) corresponding to the received health information and the received health information in association with the user in the health information database 122 is equivalent to receiving health information that includes explanatory variables and multiple variables different from the explanatory variables, and storing it in association with the user.

[0049] In the health information database 122, the health information of the first user (a user with user attribute "A") is associated with action information indicating whether or not action is required for that first user. This action information may be set by a healthcare professional, such as an industrial physician at the user's workplace, based on the health information provided by each user. Such action information can be considered information that represents the health status of the first user. Note that the health information of the second user (a user with user attribute "B") does not need to have action information set.

[0050] Returning to Figure 2, the processor 13 comprises one or more processors and their peripheral circuits. The processor 13 is a processing circuit that comprehensively controls the overall operation of the health information management device 1, and is, for example, a CPU (Central Processing Unit). The processor 13 controls the operation of the input / output interface 11, etc., so that various processes of the health information management device 1 are executed by appropriate means based on programs stored in the memory 12. The processor 13 executes processes based on various programs stored in the memory 12. In addition, the processor 13 can execute multiple programs in parallel.

[0051] Figure 5 is a functional block diagram of the processor 13 of the health information management device 1.

[0052] The processor 13 includes a health information receiving unit 131, a candidate extraction unit 132, a factor identification unit 133, and a user extraction unit 134. Each of these units in the processor 13 is a functional module implemented by a program executed on the processor 13. Alternatively, each of these units in the processor 13 may be implemented in the health information management device 1 as an independent integrated circuit, microprocessor, or firmware.

[0053] The health information reception unit 131 receives health information provided by multiple users and stores it in memory 12, associating it with each user. The health information may also be provided according to the health information response form provided by the health information reception unit 131.

[0054] For example, the health information receiving unit 131 provides a health information response form to the first input terminal 2 via the input / output interface 11 and the communication network. The first input terminal 2 displays a screen representing the response form on its display, accepts health information input from user U1, and transmits the accepted health information to the health information management device 1. The health information receiving unit 131 stores the health information received from the first input terminal 2, associating it with user U1. The receipt of health information from the first input terminal 2 can also be described as the acceptance of health information.

[0055] The candidate extraction unit 132 extracts candidate explanatory variables from multiple variables included in the health information. The candidate explanatory variables are variables that can explain the health status of user U1.

[0056] The candidate extraction unit 132 may, for example, refer to the health information database 122 and compare the values ​​of each variable in the health information of the first user that is associated with "required" action information and health information that is associated with "not required" action information, and extract variables with statistically significant differences as candidate explanatory variables. Since such variables can be said to fluctuate depending on whether or not action based on the health status of the first user is required, they can explain the health status of the first user.

[0057] The factor identification unit 133 identifies key factors from a group of key factor candidates consisting of explanatory variables and candidate explanatory variables. At this time, the factor identification unit 133 identifies key factor candidates that have an influence on other variables included in health information as key factors.

[0058] The factor identification unit 133 may prioritize identifying important factors that have a greater impact on other variables. For example, the factor identification unit 133 may identify candidate important factors if the value indicating the impact on other variables is greater than the statistical representative values ​​such as the mean, median, and mode of the other candidate important factors. Alternatively, the factor identification unit 133 may identify candidate important factors if the value indicating the impact on other variables is greater than a predetermined impact threshold.

[0059] The factor identification unit 133 may calculate the influence on other variables based on a linear regression model (linear model) such as linear regression, logistic regression, or Cox regression. Alternatively, the factor identification unit 133 may calculate the influence on other variables based on a nonlinear regression model (nonlinear model) such as a decision tree, random forest, gradient boosting, or neural network.

[0060] However, when calculating the effect using a nonlinear model, a larger number of samples are required than when calculating using a linear model, and the obtained results may be difficult to explain intuitively. Therefore, in this embodiment, it is more preferable for the factor identification unit 133 to calculate the effect using a linear model.

[0061] The factor identification unit 133 may calculate the impact by performing a penalized regression analysis based on the regression model, adding a penalty term to the loss function. In other words, the factor identification unit 133 applies the regression model with one candidate important factor as the dependent variable and other candidate important factors as independent variables. In this case, the independent variables among the candidate important factors may also be used as the dependent variable in the regression model. The factor identification unit 133 performs a penalized regression analysis by adding a penalty term to the loss function based on the regression model and optimizing the loss function.

[0062] In penalized regression analysis, the regression coefficients representing the influence of explanatory variables (other candidate important factors) on the objective function (one candidate important factor) in the regression model may be zero. Explanatory variables for which the regression coefficient is zero can be said to have no influence on the objective function in the regression model. Therefore, it is preferable for the factor identification unit 133 to identify important factors from candidate important factors for which the regression coefficient is non-zero.

[0063] Figure 6 illustrates the calculation of the frequency of candidate important factors for which the regression coefficient is non-zero.

[0064] First, the factor identification unit 133 creates a health information sample 123 containing health information samples 123A and 123B, which are obtained by separately extracting the health information of the first user from the health information database 122. Here, the health information sample 123 consists of a group of candidate important factors. The creation of the health information sample may be performed by methods such as the Bootstrap method or the Jackknife method, similar to the resampling described in Japanese Patent Application Publication No. 2020-036579. Alternatively, the creation of the health information sample may be performed according to other random or regular methods (for example, in ascending order of identifiers, in predetermined groups). The number of each health information item included in the health information sample 123 may be the same or different. In Figure 6, only two health information samples are shown for the sake of simplicity, but more health information samples may be created.

[0065] The factor identification unit 133 performs regression analysis on the health information sample 123 based on a linear model using the linear regression equation shown in (1) below.

[0066]

number

[0067] Here, x o x is a candidate important factor selected as the dependent variable. i These are candidate key factors selected as explanatory variables (but not as dependent variables), β iis the regression coefficient for the explanatory variable, p is the number of explanatory variables, and β is the regression coefficient for the explanatory variable. o This is the intercept (constant term).

[0068] The factor identification unit 133 determines that one of the candidate important factors of each health information contained in the health information sample 123 is the target variable x o Let the other be the explanatory variable x. i Extract multiple datasets in such a way.

[0069] The factor identification unit 133 performs a penalty-based regression analysis on each sample using the extracted dataset, which reduces the regression coefficients to zero. The factor identification unit 133 then identifies candidate key factors corresponding to explanatory variables whose regression coefficients do not become zero. The factor identification unit 133 performs this identification for each sample, changing the dependent variable within the candidate key factors, thereby selecting candidate key factors that should be identified as key factors.

[0070] Furthermore, if it is clear which of the candidate important factors should be the dependent variable, there is no need to change the dependent variable to another candidate important factor. For example, the factor identification unit 133 identifies important factors by using the explanatory variables from the candidate important factors as the dependent variable in the regression model and the other candidate important factors as explanatory variables. If the explanatory variables from the candidate important factors are variables that represent the items to be evaluated, the regression model can be said to be a model that explains the items to be evaluated. The factor identification unit 133 identifies candidate important factors that have a large impact on the items to be evaluated as important factors. The factor identification unit 133 may also identify candidate important factors that have a large impact on a particular item for users classified under one attribute of the items to be evaluated. In other words, the factor identification unit 133 may identify candidate important factors from the candidate important factors that have an impact on an explanatory variable and a predetermined variable from among multiple variables as important factors.

[0071] Specifically, if the item to be evaluated is an item indicating whether or not the employee is on leave from work and has two attributes, the factor identification unit 133 performs a regression analysis using a binary dependent variable corresponding to the item to be evaluated and explanatory variables corresponding to the group of candidate important factors in the regression model. From the group of candidate important factors, it identifies the candidate important factors that have a large influence on the binary dependent variable as important factors. Here, the number of values ​​that the dependent variable can take is not limited to two, but can vary depending on the number of values ​​that the variable used as the dependent variable in the regression model can take (the number of attributes for the item to be evaluated).

[0072] Furthermore, the factor identification unit 133 may perform a penalty-based regression analysis using the treatment information contained in the health information database 122 as the dependent variable (for example, setting "required" treatment information as 1 and "not required" as 0).

[0073] Penalized regression analysis methods that allow regression coefficients to be reduced to zero include, for example, Lasso (Least Absolute Shrinkage and Selection Operator), Elastic Net, and SCAD (Smoothly Clipped Absolute Deviation).

[0074] The factor identification unit 133 determines the frequency with which the regression coefficient is calculated to be non-zero for each candidate important factor using a penalty-based regression analysis. The frequency with which the regression coefficient is calculated to be non-zero can be calculated by multiplying the number of penalty-based regression analyses performed for each health information sample (the number of candidate important factors if the dependent variable is changed to each candidate important factor, or 1 if it is not changed) by the number of health information samples, and using the number of times the regression coefficient corresponding to the candidate important factor was calculated to be non-zero in a series of penalty-based regression analyses as the numerator. The frequency may also be expressed in other forms, including as a percentage.

[0075] The factor identification unit 133 prioritizes identifying important factors as such, prioritizing candidates for important factors that have a greater impact on other variables, based on the size of the statistical representative values ​​(e.g., mean, median, mode, maximum, or minimum) of the regression coefficients obtained from multiple penalty regression analyses.

[0076] Furthermore, the factor identification unit 133 may identify important factors from among the candidate important factors for which the regression coefficient is frequently calculated to be non-zero by the penalty-based regression analysis. In other words, among the candidate important factors for which the regression coefficient is frequently calculated to be non-zero by the penalty-based regression analysis, the factor identification unit 133 may prioritize identifying important factors that have a large statistical representative value of the regression coefficient obtained from multiple penalty-based regression analyses.

[0077] By operating in this manner, the health information management device 1 can appropriately identify items from among the multiple items included in the user's health information that are suitable for understanding the user's health status.

[0078] Returning to Figure 5, the user extraction unit 134 extracts users of interest from among multiple users whose values ​​for important factors in the corresponding health information fall within a predetermined range. It can also be said that the user extraction unit 134 extracts users of interest in terms of their health status (specifically, the items to be evaluated).

[0079] The user extraction unit 134 may set a predetermined range based on the values ​​of important factors in the health information provided by the first user. For example, the user extraction unit 134 sets a predetermined range based on the statistically representative values ​​of the important factors in the health information provided by the first user from the health information database 122. For example, the user extraction unit 134 may set the predetermined range as the range between the maximum and minimum values ​​of important factors in the health information provided by the first user for which treatment information is required.

[0080] The user extraction unit 134 may set a predetermined range based on the statistical representative values ​​of important factors in health information for which treatment information is "necessary" and the statistical representative values ​​of important factors in health information for which treatment information is "not necessary," among the health information provided by the first user. For example, the user extraction unit 134 may set the predetermined range to be closer to the statistical representative value of important factors in health information for which treatment information is "necessary" than to the value between the statistical representative value of important factors in health information for which treatment information is "necessary" and the statistical representative value of important factors in health information for which treatment information is "not necessary."

[0081] By operating in this manner, the health information management device 1 can extract users whose health status (specifically, the items to be evaluated) deserves attention, using items identified based on the health information of the first user.

[0082] The factor identification unit 133 may identify important factors for a first user classified into the first category among multiple categories in the user attribute database 121. In this case, the user extraction unit 134 may extract a second user of interest from the second user classified into the first category based on the values ​​of important factors in the health information provided by the second user classified into the first category.

[0083] By operating in this manner, the health information management device 1 can identify appropriate important factors for each category and use them to appropriately extract a second user for each category.

[0084] Figure 7 is a flowchart of the health information management process. The health information management device 1 can execute the health information management method according to the flowchart in Figure 7 by having the processor 13 execute a computer program for health information management.

[0085] The health information receiving unit 131 of the processor 13 of the health information management device 1 receives health information provided by multiple users from the first input terminal 2 and the second input terminal 3, respectively, via the communication network NW and input / output interface 11. The health information consists of multiple items indicating each user's health status, and includes explanatory variables pre-selected to describe each user's health status, as well as multiple variables different from the explanatory variables. The health information receiving unit 131 stores the received health information in memory 12, associating it with each user (health information receiving process, step S11).

[0086] Next, the candidate extraction unit 132 of the processor 13 of the health information management device 1 extracts candidate explanatory variables from a plurality of variables that can explain the health status of the first user (candidate extraction step, step S12). Step S12 may be performed each time a predetermined number of health information is received, may be performed at predetermined time intervals, or may be performed irregularly (for example, in accordance with the instructions of the administrator of the health information management device 1).

[0087] Then, the factor identification unit 133 of the processor 13 of the health information management device 1 identifies candidate key factors that have a significant impact on other variables included in the health information from a group of candidate key factors consisting of explanatory variables and candidate explanatory variables (factor identification step, step S13), and terminates the health information management process. Step S13 may be performed each time a predetermined number of candidate explanatory variables are extracted, at predetermined time intervals, or irregularly (for example, according to the instructions of the administrator of the health information management device 1).

[0088] By processing the data in this manner, the health information management device 1 can appropriately identify items from among the multiple items included in the user's health information that are suitable for understanding the user's health status.

[0089] Following step S13, the health information management device 1 may use the user extraction unit 134 of the processor 13 to extract users from among multiple users whose values ​​of important factors in the corresponding health information are within a predetermined range, as users whose health status (specifically, the items to be evaluated) should be noted.

[0090] By processing the data in this way, the health information management device 1 can extract users whose health status (specifically, the items to be evaluated) deserves attention, using the items identified based on the health information of the first user.

[0091] In the modified health information management device, the candidate extraction unit 132 extracts candidate explanatory variables using a machine learning model. The general configuration of the modified health information management device is the same as that of health information management device 1, so the explanation is omitted.

[0092] The processor of the modified health information management device includes a health information receiving unit 131, a candidate extraction unit 132, and a user extraction unit 134. The health information receiving unit 131 and the user extraction unit 134 in the modified health information management device are the same as those in health information management device 1, so their description is omitted.

[0093] The candidate extraction unit in the modified health information management device extracts explanatory variable candidates using a machine learning model that has been pre-trained to identify explanatory variable candidates from multiple variables contained in health information. Specifically, the candidate extraction unit in the modified health information management device inputs at least a portion of the health information stored in the health information database 122 into the machine learning model and extracts the variables output from the machine learning model as explanatory variable candidates.

[0094] The machine learning model is pre-trained to output variables that show common trends among users (e.g., User 1) associated with specific attributes, from among multiple variables.

[0095] The machine learning model may be, for example, a large-scale language model (LLM). The LLM may be trained using LoRA (Low-Rank Adaptation), which is used to fine-tune the LLM.

[0096] The modified health information management device may retrain its machine learning model each time a predetermined number of health information entries are received from first users, at predetermined intervals, or irregularly (for example, as instructed by the administrator of the modified health information management device). By processing in this way, the modified health information management device can extract second users that reflect the health information from the most recent first user.

[0097] Those skilled in the art should understand that various changes, substitutions, and modifications can be made to this disclosure without departing from its spirit and scope. [Explanation of Symbols]

[0098] 1 Health information management device 131 Health Information Reception Department 132 Candidate Extraction Unit 133 Factor identification part 134 User Extraction Unit

Claims

1. A health information receiving unit receives health information from multiple users, each containing pre-selected explanatory variables that describe the health status of each user, and multiple variables different from the aforementioned explanatory variables, and stores this information in association with each user. A candidate extraction unit extracts candidate explanatory variables from the aforementioned plurality of variables that can explain the health status of a plurality of first users to which a first attribute is associated, among the plurality of users. A factor identification unit identifies, from a group of candidate key factors consisting of the explanatory variables and candidate explanatory variables, candidate key factors that have an influence on other variables included in the health information, as key factors. A health information management device equipped with the following features.

2. The health information management device according to claim 1, wherein the factor identification unit identifies from the group of candidate important factors a candidate important factor that has an influence on one variable predetermined from the explanatory variable and the plurality of variables, and identifies that as the important factor.

3. The aforementioned factor identification unit is Using each of the health information samples, each separately extracted from the aforementioned health information and consisting of the aforementioned candidate important factors, the regression coefficients are calculated by penalized regression analysis, which adds a penalty term to the loss function, based on a regression model for each of the aforementioned candidate important factors. Among the candidate group of key factors, those key factors that are frequently found to have non-zero regression coefficients in the penalty-based regression analysis are given priority and identified as key factors. A health information management device according to claim 1 or 2.

4. The health information management device according to claim 1 or 2, further comprising a user extraction unit that extracts a second user from among the plurality of users whose values ​​of the important factors in the corresponding health information are within a predetermined range, as a user whose health status should be noted.

5. The health information management device according to claim 1 or 2, wherein the candidate extraction unit extracts candidate explanatory variables using a machine learning model that has been pre-trained to extract candidate explanatory variables from the plurality of variables included in the health information.

6. The health information management device according to claim 5, wherein the machine learning model is a large-scale language model.

7. A health information receiving process that receives health information provided by multiple users, each containing pre-selected explanatory variables that describe the health status of each user, and multiple variables different from the aforementioned explanatory variables, and stores this information in association with each user. A candidate extraction step of extracting candidate explanatory variables from the aforementioned plurality of variables that can explain the health status of a plurality of first users to which a first attribute is associated, among the plurality of users, A factor identification step of identifying key factors from a group of key factor candidates consisting of the explanatory variables and the explanatory variable candidates, which have an influence on other variables included in the health information, as key factors, A method for managing health information, including the following.

8. The health information management method according to claim 7, wherein in the factor identification step, a candidate important factor having an influence on the explanatory variable and one predetermined variable among the plurality of variables is identified as the important factor from the group of candidate important factors.

9. The health information management method according to claim 7 or 8, wherein in the candidate extraction step, candidate explanatory variables are extracted using a machine learning model that has been pre-trained to extract candidate explanatory variables from the explanatory variables and the plurality of variables.

Images