JP7645213B2 - Preventive medical support system and preventive medical support method - Google Patents

Description

The present invention relates to a preventive medical support system and a preventive medical support method.

In recent years, it is expected that health checkup results will be effectively used as preventive medicine. The health management support device of Patent Document 1 is disclosed to include a storage medium that stores information representing the health condition of a user, including information representing the results of the health checkup, and information representing lifestyle habits, associated with each of a plurality of users; a first selection unit that selects a first user to be supported in health management from among the plurality of users based on the stored information representing the health condition; a second selection unit that selects a second user having a target health condition for the selected first user from among the plurality of users based on the stored information representing the health condition and the stored information representing the lifestyle habits; a support information generation unit that reads information representing the lifestyle habits of the selected second user from the storage medium and generates support information for supporting the health management of the first user based on the read information representing the lifestyle habits; and an information output unit that outputs the generated support information.

Patent No. 6957324

Patent Document 1 aims to provide health management support and support information to individuals, but the current situation is that preventive medical care and health management measures at companies and other organizations are uniform measures for all targets. This has led to the issue that sufficient analysis has not been done from the perspective of utilizing health check result data and user attribute information to apply measures in line with the target's attributes and health condition.

The present invention has been made to solve the above-mentioned problems, and aims to provide a preventive medical support system and a preventive medical support method that can analyze the characteristics of health check results for user attributes (groups) and use the results to make decisions on whether to continue preventive medical care or select measures.

To achieve the above object, the preventive medical support system of the present invention is characterized by having a classification unit that classifies health check result data into groups based on user attribute information, a machine learning unit that extracts health check result items related to the classification of the groups, a health check result extraction unit that extracts characteristic health check results from among the health check results for the groups for the health check result items extracted by the machine learning unit, an output processing unit that outputs the extracted characteristic health check results, and a significant difference verification unit that verifies whether there is a significant difference between the groups for the extracted characteristic health check results using statistical methods. Other aspects of the present invention will be described in the embodiments described below.

According to the present invention, the characteristics of health check results for user attributes (groups) can be analyzed, and the results can be used to make decisions about continuing preventive medical care and selecting policies.

1 is a block diagram showing a preventive medical care support system according to an embodiment. FIG. 11 is a diagram showing an example of medical checkup result data. FIG. 11 is a diagram illustrating an example of user attribute information. FIG. 11 is a diagram illustrating an example of a user attribute master. 13 is a diagram showing an example of medical checkup result data and user attribute information after anonymization processing by an anonymization unit. FIG. FIG. 13 is a diagram illustrating an example of anonymization processing by an anonymization unit. FIG. 13 is a diagram illustrating an example of analysis target data. 13 is a flowchart showing a preventive medical care support process. FIG. 13 is a diagram showing examples of health check result items (explanatory variables). FIG. 13 is a diagram showing examples of health check result items (explanatory variables). FIG. 13 is a diagram showing examples of health check result items (explanatory variables). FIG. 13 is a diagram showing an example of the importance of health check result items (explanatory variables) determined by the machine learning unit. FIG. 13 is a violin diagram showing a list of medical checkup result data by the medical checkup result extraction unit. FIG. 13 is a violin diagram showing the test values of neutral fats for each cluster. FIG. 13 is a diagram showing characteristics based on the test reference values of neutral fat for each cluster. FIG. 11 is a diagram showing an example of an output by an output processing unit. 13 is a diagram showing an example of a verification result by a significant difference verification unit. FIG.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment for carrying out the present invention will be described in detail with reference to the accompanying drawings.
Fig. 1 is a block diagram showing a preventive medical support system PMSS according to an embodiment. The preventive medical support system PMSS (Preventive Medical Support System) aims to analyze the characteristics of medical checkup results for user attributes (groups) and to utilize the results in decision-making for the continuation of preventive medical care and the selection of measures. In particular, the present invention aims to "look at the entire group and find current problems."

The preventive medical support system PMSS is composed of a preventive medical support device 100 and a business establishment server 200. The preventive medical support device 100 and the business establishment server 200 are communicatively connected via a network NW. Business establishments include shops, factories, offices, sales offices, banks, branches, schools, temples, hospitals, inns, smelters, mines, power plants, etc.

The business server 200 is a server that supports processing on the client side, and performs preprocessing for creating the analysis target data 21. The business server 200 has a processing unit 210 and a memory unit 220. The memory unit 220 stores health check result data 221, which is the results of the subject's health check, user attribute information 222 indicating the subject's user attributes, a user attribute master 223 used to classify subjects by attribute, and the like.

Here, user attributes are attributes used for analyzing business establishments. For example, if a business establishment has multiple factories, the attribute would be each factory. Specifically, Factory A, Factory B, Factory C, Factory D, Factory E, etc. If the factories are spread across Japan, the attribute would be each region. Specifically, Kanto region, Chubu region, Kansai region, Chugoku region, Kyushu region, etc. If there is a wide variety of work content, the attribute may be each job type. Specifically, sales, development, distribution, management, general affairs, etc.

The processing unit 210 has a classification unit 211 that classifies medical checkup result data into groups based on user attribute information, and an anonymization unit 212 that anonymizes customer medical checkup result data 221 and user attribute information 222. The anonymization process by the processing unit 210 enables the preventive medical support device 100 to perform analysis without handling personal information. The preventive medical support device 100 provides an anonymization tool to the business server 200. The anonymization unit 212 may be included in the processing of the classification unit 211.

Anonymization tools use a hash algorithm to anonymize data, making it possible to match data after anonymization. Hashing is the process of using a hash function to find a non-reversible, fixed-length value and replace the data. The same hash value will always be obtained from the same data, but even slight differences in the data will generate a completely dissimilar hash value.

Figure 2 is a diagram showing an example of medical checkup result data 221. Medical checkup result data 221 includes medical checkup result data for 150 items such as name, sex, date of birth, systolic blood pressure, fasting blood glucose, and triglyceride. The name, sex, and date of birth are key information for the name matching process described below.

When blood is being pumped from the heart to the aorta, the heart contracts to push the blood out, and the pushed-out blood exerts pressure on the aortic blood vessel wall. Systolic blood pressure refers to the blood pressure in this state.

Figure 3 is a diagram showing an example of user attribute information 222. User attribute information 222 includes name, gender, date of birth, user attribute code, etc. Name, gender, and date of birth are key information for the name matching process.

Figure 4 is a diagram showing an example of the user attribute master 223. The user attribute master 223 includes attribute codes and attribute names. In Figure 4, the attribute names are X group, Y group, Z group, W group, etc. The attribute names correspond to the aforementioned Factory A, Factory B, Factory C, Factory D, Factory E, etc.

Figure 5 shows an example of medical checkup result data 221A and user attribute information 222A after anonymization processing by the anonymization unit 212. In the medical checkup result data 221A and the user attribute information 222A, the name among the key information is hashed and the date of birth is also date-substituted. The date is uniformly converted to the 1st so that the date of birth cannot be identified. The conversion method is not limited to the 1st, and it may be converted to other symbols (for example, **).

Figure 6 is a diagram showing an example of anonymization processing S212 by the anonymization unit 212. The anonymization unit 212 converts the medical checkup result data 221 into medical checkup result data 221A and converts the user attribute information 222 into user attribute information 222A through the anonymization processing S212.

Returning to FIG. 1, the preventive medical support device 100 will be described.
The preventive medical support device 100 includes a processing unit 10, a storage unit 20, an input unit 30, an output unit 40, a communication unit 50, etc. The processing unit 10 includes a data creation unit 11 that creates analysis target data 21 from data sent from the establishment server 200, a machine learning unit 12 that extracts health check result items related to group classification, a health check result extraction unit 13 that extracts characteristic health check results from among the health check results for the groups for the health check result items extracted by the machine learning unit 12, an output processing unit 14 that outputs the extracted characteristic health check results, and a significant difference verification unit 15 that verifies whether there is a significant difference between groups for the extracted characteristic health check results using a statistical method, etc.

The memory unit 20 stores data to be analyzed 21 (see FIG. 7), health check result item importance information 22 (see FIG. 10), health check result information for groups 23 (see FIG. 11), output information 24 (see FIG. 14), health check result items 25 (see FIG. 9A to FIG. 9C), proportions per cluster based on standard values 26 (see, for example, FIG. 13), number of subjects per cluster based on standard values 27 (see, for example, FIG. 13), etc. The proportions per cluster based on standard values 26 and number of subjects per cluster based on standard values 27 will be described later.

In FIG. 1, the processing unit 10 is a central processing unit (CPU) that executes various programs stored in a RAM, HDD, etc. The memory unit 20 is a HDD that stores various data for the preventive medical support device 100 to execute processing. The input unit 30 is a device for inputting instructions to a computer, such as a keyboard or mouse, and inputs instructions such as starting a program. The output unit 40 is a display or the like that displays the execution status and execution results of processing by the preventive medical support device 100. The communication unit 50 exchanges various data and commands with other devices via the network NW.

The data creation unit 11 performs a name matching process on the health check result data 221A and user attribute information 222A sent from the business server 200 to create the analysis target data 21.

Figure 7 is a diagram showing an example of the analysis target data 21. The analysis target data 21 includes 150 items of health check result data such as hashed name, sex, date-substituted date of birth, systolic blood pressure, fasting blood glucose, and triglyceride, as well as user attribute code (attribute code), attribute name, etc.

Figure 8 is a flowchart showing the preventive medical support process S100. As described above, the data creation unit 11 creates the analysis target data 21 (process S110). The machine learning unit 12 extracts health check result items that are involved in group classification through machine learning (process S120) and outputs the importance of the health check result items (process S130). The health check result extraction unit 13 extracts health check results for each group for the health check result items extracted by the machine learning unit 12, and outputs a list of health check results for the groups (process S140).

The health check result extraction unit 13 extracts the characteristics of the cluster corresponding to the group (process S150). In particular, the health check result extraction unit 13 calculates the proportion 26 of each group based on the standard value of the health check result for the health check result items extracted by the machine learning unit 12, and the number of subjects 27 of each group based on the standard value. The output processing unit 14 outputs the characteristics of the cluster that is the extracted characteristic health check result (process S160). The significant difference verification unit 15 uses statistical methods to verify whether there is a significant difference between the two selected clusters, and outputs the significant difference verification result (process S170).

Each part will be described in detail.
9A, 9B, and 9C are diagrams showing examples of health check result items (explanatory variables). 99 health check result items 25 are shown in Fig. 9A to Fig. 9C. Among these, it is necessary to narrow down the health check result items that are important for preventive medical support.

Since it is practically difficult to analyze all of the large number of health check result items, the machine learning unit 12 aims to narrow down the analysis to important items, thereby narrowing down the health check result items that are important for determining user attributes.

One of the feature selection methods or variable selection methods is Boruta. The machine learning unit 12 uses the feature selection method (Boruta) to narrow down important items based on their importance. At that time, tuning is performed so that the quantile of importance becomes the number of items that can be analyzed practically.

Figure 10 is a diagram showing an example of the importance of health check result items (explanatory variables) determined by the machine learning unit 12. Figure 10 shows the results of narrowing down by machine learning. Referring to Figure 10, it can be seen that age at time of health check, BMI test value, waist circumference test value, general blood and red blood cell count test value, ..., triglyceride test value, fasting blood glucose test value, ..., etc. are important.

FIG. 11 is a violin diagram showing the list information 23a of the medical checkup result data by the medical checkup result extraction unit 13. In each diagram, the horizontal axis indicates the cluster (CL) corresponding to the group, and the vertical axis indicates each medical checkup result data. The clusters in each diagram are CL1, CL2, CL3, CL4, and CL5 from the left, showing five clusters. The violin diagram is a chart that combines a box-and-whisker plot and a density curve, and in addition to expressing the distribution of numerical data like the box-and-whisker plot, the average value is also visualized. The chart also shows the distribution as a density curve. The average (or median) and the interquartile range are drawn inside the violin diagram. The medical checkup result extraction unit 13 examines the change in the average value of each cluster, the degree of distribution, etc., and extracts important medical checkup result data in detail.
11 shows an example of five clusters, but the number of clusters is not limited to this example and may be changed as appropriate according to the number of groups classified by the classification unit 211 (FIG. 1).

Figure 12 is a violin diagram showing the triglyceride test values for each cluster. The horizontal axis shows the five clusters, CL1, CL2, CL3, CL4, and CL5, while the vertical axis shows the triglyceride test values. Health check result data 23c is obtained by excluding the outlying values of health check result data 23b. Looking at health check result data 23b (1), there is almost no difference in the median values for each cluster, but CL2 and CL3 vary widely. Looking at health check result data 23c (2), there is a tendency for CL4 and CL5 to tend to have lower values than the other clusters.

Figure 13 is a diagram showing the characteristics of each cluster based on the standard values for triglyceride testing. Figure 13 shows the proportion of each cluster based on the standard values 26 and the number of subjects in each cluster based on the standard values 27. Looking at (3) of the proportion of each cluster based on the standard values 26, the proportion of subjects exceeding the standard values for triglycerides (30-149 mg/dl: normal) is highest in CL2 and lowest in CL5.

Figure 14 is a diagram showing an example of output by the output processing unit 14. The display screen of the output unit 40 has a display area 24W for the health check result data extracted by the health check result extraction unit 13 in the center, and a task bar at the bottom. The task bar has a name matching icon 24A, a machine learning icon 24B, an overall display icon 24C, a selection display icon 24D, a feature extraction icon 24E, a significant difference verification icon 24F, a back button 24M, a forward button 24N, etc.

When the matching icon 24A is clicked, the data creation unit 11 starts processing, executes the matching process, and creates analysis target data 21 from the data sent from the business server 200. When the machine learning icon 24B is clicked, the machine learning unit 12 starts processing, and extracts the health check result items involved in the classification of groups (clusters).

When the full display icon 24C is clicked, the medical check result extraction unit 13 starts processing, and among the medical check result items extracted by the machine learning unit 12, medical check result items with a high degree of importance are displayed in full view (see FIG. 11). When the user clicks on a medical check result item on the full display list screen and then clicks the selection display icon 24D, the medical check result is enlarged and displayed. When multiple medical check result items are selected, multiple medical check results are displayed.

When the user clicks on a medical checkup result item on the screen in full view or enlarged view, and then clicks the feature extraction icon 24E, the medical checkup result extraction unit 13 extracts characteristic medical checkup results from the medical checkup results for the group, and displays them as shown in Figure 14.

In FIG. 14, when the user clicks two clusters on the screen and then clicks the significant difference verification icon 24F, the significant difference verification unit 15 uses statistical methods to verify whether there are significant differences between groups (clusters) for the characteristic health check results extracted. In this embodiment, the F-test and t-test are shown as examples.

Figure 15 is a diagram showing an example of the verification results by the significant difference verification unit 15. The significant difference verification unit 15 uses F-tests and t-tests to verify whether there is a significant difference in the test values of the selected clusters. In the example of Figure 15, (1) F-test: a test of variance using two samples is performed. Here, a case is shown in which CL2 and CL5 are selected based on Figure 13. Since the P-value for the variance result using the two samples of CL2 and CL5 is less than 0.05, the null hypothesis (the variances are equal) is rejected. Therefore, it is determined that "the variances of the triglyceride test values of CL2 and CL5 are not equal."

Next, the significant difference verification unit 15 performs a t-test (2) assuming that the variances are not equal. Since the P value is less than 0.05 (significance level), the null hypothesis is rejected. In other words, it is determined that "there is a significant difference in the test values of triglycerides between CL2 and CL5."
From the verification result of the significant difference verification unit 15, the user can confirm whether or not there is a significant difference in the test value of the selected cluster.

The preventive medical support system of the present embodiment described above has the following features.
(1) The preventive medical support system PMSS of this embodiment is characterized by having a classification unit 211 that classifies medical checkup result data into groups based on user attribute information, a machine learning unit 12 that extracts medical checkup result items related to group classification, a medical checkup result extraction unit 13 that extracts characteristic medical checkup results from the medical checkup results for the groups for the medical checkup result items extracted by the machine learning unit 12, an output processing unit 14 that outputs the extracted characteristic medical checkup results, and a significant difference verification unit 15 that verifies whether there is a significant difference between the groups for the extracted characteristic medical checkup results by a statistical method. This makes it possible to analyze the characteristics of medical checkup results for user attributes (groups) and utilize the results in decision-making for determining whether to continue preventive medical care or select measures.

In this embodiment, the preventive medical support device 100 and the business site server 200 are described separately, but this is not limited to this. The preventive medical support device 100 may also have a classification unit 211.

(2) The classification unit 211 anonymizes the names included in the medical checkup result data and the user attribute information by using a hash algorithm. This allows analysis to be performed without handling personal information.

(3) The machine learning unit 12 uses user attribute information as the objective variable and the medical check result items as explanatory variables, and outputs the medical check result items in order of importance. This makes it possible to narrow down the large number of medical check result items, for example up to 150 items, to the most important items.

(4) The health check result extraction unit 13 can calculate the proportion of each group based on the standard value of the health check result and the number of subjects in each group based on the standard value for the health check result items extracted by the machine learning unit 12. This makes the differences between groups (clusters) clear.

(5) The output processing unit 14 displays on the display unit a diagram including the distribution and average values of the numerical data, with the horizontal axis being classified into groups and the vertical axis being the medical checkup values of the medical checkup result items. This allows the user to understand the differences between groups (clusters). In this embodiment, the display is described as being on the output unit 40 (display unit) of the preventive medical support device 100, but this is not limited to this. It may also be viewable from the business server 200 via a web server.

(6) The significant difference verification unit 15 tests whether the variances of the groups (clusters) are equal, tests the case where the residuals are not equal, and verifies whether there is a significant difference between the groups. This makes it possible to grasp whether there are differences in the distributions of the two selected groups.

(7) User attribute information is information that indicates the physical and social characteristics of a user. In this embodiment, preventive medical care within a company is described by factory, region, job type, etc., but is not limited to this. It can be used for group analysis by hospital, etc.

(8) A preventive medical support method for a preventive medical support device that analyzes characteristics of health checkup results, the preventive medical support device classifies health checkup result data into groups based on user attribute information, extracts health checkup result items that contribute to the group classification, extracts characteristic health checkup results from the health checkup results for the extracted health checkup result items for the group, outputs the extracted characteristic health checkup results, and uses statistical methods to verify whether there are significant differences between groups for the extracted characteristic health checkup results. This allows the characteristics of the health checkup results for user attributes (groups) to be analyzed, and the results to be used in decision-making for continuing preventive medical care and selecting measures.

REFERENCE SIGNS LIST 10 Processing section 11 Data creation section 12 Machine learning section 13 Health check result extraction section 14 Output processing section 15 Significant difference verification section 20 Memory section 21 Data to be analyzed 22 Health check result item importance information 23 Health check result information for group 24 Output information 30 Input section 40 Output section (display section)
50 Communication unit 100 Preventive medical support device 200 Business server 210 Processing unit 211 Classification unit 220 Storage unit 221 Medical checkup result data 222 User attribute information 223 User attribute master NW Network PMSS Preventive medical support system

Claims (8)

A classification unit that classifies the medical examination result data into groups based on user attribute information;
A machine learning unit that extracts health check result items related to the classification of the groups;
A health check result extraction unit that extracts characteristic health check results from the health check results for the group for the health check result items extracted by the machine learning unit;
an output processing unit that outputs the extracted characteristic medical examination results;
and a significant difference verification unit that uses a statistical method to verify whether there is a significant difference between the groups for the extracted characteristic medical examination results. The preventive medical support system according to claim 1 , wherein the classification unit anonymizes the names included in the medical checkup result data and the user attribute information by using a hash algorithm. The preventive medical support system according to claim 1 , wherein the machine learning unit uses the user attribute information as a target variable, the medical checkup result items as explanatory variables, and outputs the medical checkup result items in order of importance. The preventive medical support system of claim 1, characterized in that the health check result extraction unit calculates, for the health check result items extracted by the machine learning unit, the proportion for each group based on a standard value of the health check result, and the number of subjects for each group based on the standard value. The preventive medical support system according to claim 1, characterized in that the output processing unit displays on a display unit a graph including the distribution and average values of the numerical data, with the horizontal axis representing the classified groups and the vertical axis representing the medical checkup values of the medical checkup result items. The preventive medical support system according to claim 1, characterized in that the significant difference verification unit tests whether the variances of the groups are equal, tests the case where the variances are not equal, and verifies whether there is a significant difference for each of the groups. The preventive medical support system according to claim 1 , wherein the user attribute information is information indicating physical and social characteristics of the user. A preventive medical support method for a preventive medical support device that analyzes characteristics of a medical examination result, comprising:
The preventive medical support device includes:
Classify the medical checkup result data into groups based on user attribute information,
Extracting health check result items related to the classification of the groups;
Extracting characteristic health check results from among the health check results for the group for the extracted health check result items;
Outputting the extracted characteristic medical examination results;
A preventive medical support method comprising: verifying, by a statistical method, whether there is a significant difference between the groups with respect to the extracted characteristic medical examination results.

Images