JP7634464B2 - Intervention effect analysis system and intervention effect analysis method - Google Patents

Description

The present invention relates to an intervention effect analysis system that analyzes the effects of intervention measures.

To extend the healthy life expectancy of residents and optimize social security costs, local governments are required to analyze the effectiveness of intervention measures such as health checkups, health programs, and nursing care prevention programs. To implement effective and efficient intervention measures, it is important to utilize data accumulated by local governments to determine whether intervention measures are effective or ineffective, prioritize budget allocation to effective programs, and realize sustainable social security services.

The following prior art is included as background technology in this technical field. Patent Document 1 (JP 2019-192065 A) describes a nursing care intervention effect verification device that is an apparatus for verifying the effect of nursing care intervention on a subject, and includes a means for generating a group consisting of a similar portion of a plurality of subjects based on past data on the plurality of subjects, a means for dividing the portion of subjects constituting the group into a first subgroup in which nursing care intervention is performed and a second subgroup in which nursing care intervention is not performed, a means for receiving verification data including values of predetermined evaluation indexes for the first subgroup and the second subgroup after a specific period of time has elapsed, and a means for determining the effect of the nursing care intervention based on a comparison result of the values of the predetermined evaluation indexes for the first subgroup and the second subgroup.

Patent Document 2 (JP Patent Publication 2014-225177A) describes an analysis system having a processor that executes a program and a memory that stores the program, and that analyzes the effects of a health project by executing the program, characterized in that the analysis system has an input unit for inputting subscribers' medical expenses, intervention information indicating whether or not an intervention service is provided, and a start date of the intervention service, a propensity score calculation unit that analyzes the relationship between the medical expenses before the provision of the intervention service and the intervention information, and calculates a propensity score of the intervention group indicating the probability that the intervention service is provided and a propensity score of the non-intervention group indicating the probability that the intervention service is not provided from the medical expenses before the provision of the intervention service, and an adjusted medical expense calculation unit that calculates the adjusted medical expenses of the intervention group before and after the provision of the intervention service by multiplying the medical expenses of the intervention group before and after the provision of the intervention service by the reciprocal of the propensity score of the intervention group, and calculates the adjusted medical expenses of the non-intervention group before and after the provision of the intervention service by multiplying the medical expenses of the non-intervention group before and after the provision of the intervention service by the reciprocal of the propensity score of the non-intervention group.

JP 2019-192065 A JP 2014-225177 A

Even when comparing groups that differ not only in the presence or absence of an intervention but also in background information such as age and medical visit status, it is difficult to properly evaluate the effectiveness of an intervention measure. Therefore, it is necessary to use propensity scores to equalize the distribution of the groups, create groups that differ only in the presence or absence of the intervention, and track these groups to properly evaluate the effectiveness of the intervention.

In the technology described in Patent Document 1, subjects are divided into a care intervention group and a non-care intervention group so that the variance in background information of subjects before care intervention is equalized, and the effects on care level and care costs after the intervention are examined. However, the technology described in Patent Document 1 separates the intervention group and the non-intervention group in the future, and it is difficult to apply it to past data in which intervention has already been carried out. Furthermore, Patent Document 1 does not specifically describe a method for equalizing the variance in background information.

In addition, the technology described in Patent Document 2 uses a propensity score to adjust, compare, and analyze background information (covariates) between the intervention group and the non-intervention group as a method of analyzing the effects of the intervention. Here, a propensity score is a score that consolidates background information into one, and is the probability of intervention calculated from the background information. However, the technology described in Patent Document 2 uses a single propensity score to adjust background information and analyze the effects of the intervention, but does not take into consideration stable adjustment of the background information.

Furthermore, the prior art mentioned above takes into consideration the alignment of background information between two groups, the intervention group and the non-intervention group, but does not take into consideration the alignment of background information between three or more groups, including multiple intervention groups.

The purpose of this invention is to accurately analyze the effects of interventions such as health care projects and nursing care prevention projects using nursing care, medical care, health care projects, and nursing care prevention data.

A representative example of the invention disclosed in the present application is as follows: That is, an intervention effect analysis system for analyzing the effects of a plurality of intervention measures includes a calculation device that executes a predetermined process, an input unit to which data is input, and a storage device accessible to the calculation device, the input unit receives input of intervention data of a plurality of intervention groups in which one of the intervention measures has been implemented, and non-intervention data of a non-intervention group in which the intervention measure is not implemented, the calculation device calculates a propensity score for each of the plurality of intervention data and the non-intervention data from background information contained in each of the plurality of intervention data and the non-intervention data using a plurality of machine learning methods, sets one of the plurality of intervention data or the non-intervention data as a reference intervention group , and calculates a propensity score for each of the plurality of intervention data and the non-intervention data from background information contained in each of the plurality of intervention data and the non- intervention data ... using a plurality of machine learning methods. the method comprises: generating paired data of a plurality of individuals in which the difference in the calculated propensity scores is small from a number of intervention data and non-intervention data; calculating a standardized difference indicating a difference in background information of the plurality of paired data; calculating the propensity scores of the data of the reference intervention group, the other intervention groups, and the non-intervention group using a plurality of methods; selecting data having a propensity score similar to that of the data of the reference intervention group from the other intervention groups and the non-intervention groups; creating paired data for each calculation method of the propensity score; selecting paired data in which a representative value of the standardized difference calculated for the background information of the paired data is small; calculating an intervention effect from the selected paired data; and outputting a comparative analysis result between the plurality of intervention data or the non-intervention data and the reference intervention group .

According to one aspect of the present invention, the intervention effects of health care projects and nursing care prevention projects can be properly analyzed from nursing care, medical care, health care project, and nursing care prevention data. In particular, since it is possible to compare three or more groups, it is possible to understand the effectiveness of an intervention measure and the relative merits of multiple intervention measures. Issues, configurations, and effects other than those mentioned above will be made clear through the explanation of the following examples.

FIG. 1 is a diagram showing the concept of processing by an intervention effect analysis system to which the present invention is applied. FIG. 1 is a block diagram showing the configuration of an intervention effect analysis system according to a first embodiment. FIG. 4 is a diagram illustrating a configuration example of an intervention data storage unit according to the first embodiment. FIG. 4 is a diagram illustrating a configuration example of an intervention data storage unit according to the first embodiment. FIG. 2 is a diagram illustrating an example of the configuration of a non-intervention data storage unit according to the first embodiment. FIG. 2 is a diagram illustrating an example of the configuration of a propensity score storage unit according to the first embodiment. FIG. 2 is a diagram illustrating an example of the configuration of a propensity score storage unit according to the first embodiment. FIG. 2 is a diagram illustrating an example of the configuration of a propensity score storage unit according to the first embodiment. FIG. 2 is a diagram illustrating an example of the configuration of a matching data storage unit according to the first embodiment. FIG. 2 is a diagram illustrating an example of the configuration of a matching data storage unit according to the first embodiment. FIG. 13 is a diagram showing an example of the format of the standardized difference in the first embodiment. FIG. 11 is a diagram showing a process performed by a matching data extraction unit in the first embodiment. 1 is a flowchart of an intervention effect analysis process according to the first embodiment. FIG. 13 is a diagram illustrating an example of a machine learning method setting screen in the first embodiment. FIG. 13 is a diagram showing an example of a background information setting screen according to the first embodiment; FIG. 13 is a diagram showing an example of a reference intervention setting screen in the first embodiment. FIG. 13 is a diagram showing another example of the reference intervention setting screen in the first embodiment. FIG. 13 is a diagram showing an example of an intervention effect index setting screen in Example 1. FIG. 13 is a diagram showing an example of an intervention effect display screen in Example 1. FIG. 11 is a block diagram showing the configuration of an intervention effect analysis system according to a second embodiment. 13 is a flowchart of an intervention effect analysis process according to the second embodiment. FIG. 11 is a diagram illustrating a process performed by a pair data creation unit according to the second embodiment. FIG. 13 is a diagram illustrating a process performed by a pair data extraction unit according to the second embodiment. FIG. 11 is a block diagram showing the configuration of an intervention effect analysis system according to a third embodiment. FIG. 13 is a diagram showing an example of a background information group setting screen according to the third embodiment. FIG. 13 is a diagram showing an example of an intervention effect display screen in Example 3.

Figure 1 shows the concept of processing by an intervention effect analysis system to which the present invention is applied.

To analyze the effects of interventions from data on people (e.g. 1,000 people) who have received nursing care, medical care, or health care services, a group that has received two different types of intervention, intervention 1 and intervention 2, and a non-intervention group that has received neither intervention, are extracted. The extracted intervention 1 group, intervention 2 group, and non-intervention group have different distributions of background information (gender, age, medical care and nursing care status, etc.). Each group may have a different number of people, but if the background information is different, it is not possible to know whether subsequent changes are the effect of the intervention or are due to different background information.

For this reason, propensity score matching data is used to extract people with complete background information from multiple groups (multiple intervention groups and non-intervention groups) from intervention 1, intervention 2, and non-intervention groups, creating groups in which the effects of the interventions can be compared. Then, by tracking the health checkup results (test values), diagnosis results (presence or absence of disease), medical expenses, level of care, care expenses, etc. of people in each group and comparing changes in health status after the intervention is implemented, the effects of interventions such as multiple health programs and care prevention programs can be accurately analyzed.

Example 1
FIG. 2 is a block diagram showing the configuration of the intervention effect analysis system according to the first embodiment.

The intervention effect analysis system of Example 1 has an analysis terminal 100 and a database 130.

The analysis terminal 100 is a computer having an input unit 102, an output unit 103, a processor (CPU) 104, a memory 105, and a storage medium 106.

The input unit 102 is a human interface such as a mouse or keyboard, and accepts input to the analysis terminal 100. The output unit 103 is a display or printer that outputs the results of calculations performed by the analysis terminal 100. The input unit 102 and the output unit 103 may include an interface that inputs and outputs data via a storage medium or a network. The storage medium 106 is a storage device that stores various programs that execute the intervention effect analysis process performed by the analysis terminal 100, and the execution results of the intervention effect analysis process, and is composed of, for example, a non-volatile storage medium (magnetic disk drive, non-volatile memory, etc.).

The memory 105 includes a ROM, which is a non-volatile storage element, and a RAM, which is a volatile storage element. The ROM stores unchanging programs (e.g., BIOS) and the like. The RAM is a high-speed, volatile storage element such as a DRAM (Dynamic Random Access Memory), and temporarily stores programs executed by the processor 104 and data used when executing the programs. In other words, the programs stored in the storage medium 106 are deployed in the memory 105.

The processor 104 is a computing device that executes a program loaded into the memory 105, and is, for example, a CPU or a GPU. The processes and calculations described below are executed by the processor 104. Note that some of the processes that the processor 104 executes by executing a program may be executed by another computing device (for example, hardware such as an FPGA (Field Programmable Gate Array) or an ASIC (Application Specific Integrated Circuit)).

The program executed by the processor 104 is provided to the analysis terminal 100 via removable media (CD-ROM, flash memory, etc.) or a network, and is stored in a non-volatile storage device, which is a non-transitory storage medium. For this reason, the analysis terminal 100 may have an interface for reading removable media.

The intervention effect analysis system of this embodiment is a computer system configured on one computer or on multiple logically or physically configured computers, and may operate in separate threads on the same computer, or may operate on a virtual computer constructed on multiple physical computer resources.

The storage medium 106 stores programs that realize the machine learning method setting unit 111, the background information setting unit 112, the propensity score calculation unit 113, the matching data creation unit 114, the standardized difference calculation unit 115, the matching data extraction unit 116, the intervention effect index setting unit 117, the intervention effect calculation unit 118, and the reference intervention setting unit 119.

The machine learning method setting unit 111 sets the machine learning method to be used in the intervention effect analysis and stores the selection result in the machine learning method storage unit 133.

The background information setting unit 112 sets the background information to be used in the intervention effect analysis, and stores the selected background information in the background information storage unit 134. The background information is information that can be a covariate for the presence or absence of an intervention, and is shown as options on the background information setting screen (Figure 13), including the subject's basic information (gender, age, address, etc. from the resident basic register), nursing care information (level of nursing care required, nursing care costs, etc.), medical information (presence or absence of disease, medical costs by disease, etc.), and health check information (test values, interview results, etc.).

The propensity score calculation unit 113 calculates the propensity score using a machine learning model based on the machine learning method selected by the machine learning method setting unit 111, and stores the calculation result in the propensity score storage unit 135. The propensity score is the probability of intervention (probability of intervention) obtained from background information, and is calculated using a machine learning model with the background information as the explanatory variable and the probability of intervention as the objective variable.

The matching data creation unit 114 creates matching data that combines data with similar background information using the propensity scores, and stores the matching data in the matching data storage unit 136.

The standardized difference calculation unit 115 evaluates the matching results and calculates the standardized difference. The standardized difference is a standardized value that indicates the difference in background information between a certain intervention group and other intervention groups (including a non-intervention group) in each matching data. The standardized difference is also called the standardized effect size.

The matching data extraction unit 116 selects the best machine learning method based on the evaluation results of the matching data using standardized differences.

The intervention effect index setting unit 117 sets the intervention effect index selected by the user as the intervention effect index to be displayed on the intervention effect display screen (Figure 17). The intervention effect index is a numerical value for quantitatively evaluating the effect of the intervention, and may be the presence or absence of disease, disease incidence rate, medical expenses by disease, nursing care expenses, nursing care level, test values, interview results, etc.

The intervention effect calculation unit 118 calculates the intervention effect index from the matching data in the machine learning method selected based on the evaluation results of the matching data, and generates display data for the intervention effect display screen (Figure 17).

The standard intervention setting unit 119 displays the basic intervention setting screen (Figures 14 and 15) and sets the standard intervention selected by the user.

The database 130 is composed of an intervention data storage unit 131, a non-intervention data storage unit 132, a machine learning method storage unit 133, a background information storage unit 134, a propensity score storage unit 135, and a matching data storage unit 136.

The intervention data storage unit 131 stores data on background information of people in multiple intervention groups. As shown in Figures 3 and 4, the intervention data storage unit 131 includes a personal ID 1311, basic information 1312 of the subject, nursing care information 1313, medical information 1314, health check information 1315, and intervention information 1316. The personal ID 1311 is a unique identifier indicating a single resident. The basic information 1312 of the subject is the gender, age, address, etc. of the subject, and can be obtained from the resident basic register. The nursing care information 1313 is information related to nursing care, such as the level of nursing care required and nursing care costs. The medical information 1314 is information related to medical care, such as the presence or absence of a disease and medical costs by disease. The health check information 1315 is information on the results of a health check, such as test values and interview results. Intervention information 1316 is information on whether or not an intervention has been received, i.e., whether an intervention measure has been received to improve health or quality of life (QOL) related to medical care, health checkups, or nursing care, such as health programs (health checkups, health guidance, etc.), nursing care prevention activities, or medical procedures. Of the background information, nursing care information 1313, medical information 1314, and health checkup information 1315 change over time, and are therefore recorded annually (year T, year T+1, etc.).

The non-intervention data storage unit 132 stores background information data for people in the non-intervention group. As shown in FIG. 5, the non-intervention data storage unit 132 includes the same items as the intervention data storage unit 131, such as a personal ID 1321, basic information 1322 of the subject (gender, age, address, etc. from the resident basic register), nursing care information 1323 (level of nursing care required, nursing care costs, etc.), medical information 1324 (presence or absence of disease, medical costs by disease, etc.), health check information 1325 (test values, interview results, etc.), and intervention information 1326 (presence or absence of intervention). Of the background information, the nursing care information 1323, medical information 1324, and health check information 1325 change over time, and are therefore recorded annually (year T, year T+1, etc.).

The propensity score storage unit 135 stores the propensity scores of the intervention group and the non-intervention group calculated by multiple machine learning methods. As shown in FIG. 6A, FIG. 6B, and FIG. 6C, the propensity score storage unit 135 includes the same items as the intervention data storage unit 131 and the non-intervention data storage unit 132, such as an individual ID 1351, basic information 1352 of the subject (gender, age, address, etc. in the resident basic register), nursing care information 1353 (level of nursing care required, nursing care costs, etc.), medical information 1354 (presence or absence of disease, medical costs by disease, etc.), medical checkup information 1355 (test values, interview results, etc.), and intervention information 1356 (presence or absence of intervention). Furthermore, the propensity score storage unit 135 includes propensity scores 1357 of the intervention group and the non-intervention group calculated using machine learning models using multiple machine learning methods. In this embodiment, as shown in FIG. 6A, FIG. 6B, and FIG. 6C, the propensity score is calculated using methods such as logistic regression analysis and neural network.

The matching data storage unit 136 stores the results of matching based on the propensity score calculated by a certain machine learning method. As shown in FIG. 7 and FIG. 8, the matching data storage unit 136 includes the same items as the propensity score storage unit 135, such as an individual ID 1361, basic information 1362 of the subject (gender, age, address, etc. in the resident basic register), nursing care information 1363 (level of nursing care required, nursing care costs, etc.), medical information 1364 (presence or absence of disease, medical costs by disease, etc.), medical checkup information 1365 (test values, interview results, etc.), intervention information 1366 (presence or absence of intervention), and propensity score 1367. Furthermore, the matching data storage unit 136 includes matching information 1368 between data of the intervention group and data of the non-intervention group that have similar propensity scores. In the figure, the matching information 1368 is represented by a line that associates rows of a data table, but a column for recording the individual ID of the other party may be provided, or the information may be recorded in a table that associates data of the intervention group with data of the non-intervention group.

Even if the background information is the same, the propensity scores differ for each machine learning method, and the matching pairs differ. For this reason, it is advisable to create a matching data storage unit 136 for each propensity score calculation method (Figures 7 and 8).

9 is a diagram showing an example of the format of the standardized difference calculated by the standardized difference calculation unit 115. The standardized difference indicates the standardized difference in background information between the reference intervention group (intervention 1 in this embodiment) set by the reference intervention setting unit 119 and other intervention groups (intervention 2, intervention 3, ..., non-intervention group), and is stored in the memory 105 or storage medium 106 as an internal variable of the intervention effect analysis system in the table format shown in the figure. The standardized difference is calculated for each index of the same items as those in the matching data storage unit 136, such as the subject's basic information 1402 (gender, age, address, etc. in the resident basic register), nursing care information 1403 (level of nursing care required, nursing care costs, etc.), medical information 1404 (presence or absence of disease, medical costs by disease, etc.), and health check information 1405 (test values, interview results, etc.). Furthermore, the standardized difference includes a propensity score calculation method 1401 and a representative value 1406 of the standardized difference. The representative value 1406 of the standardized difference is a value obtained by statistically processing the standardized difference of the data pair in each propensity score calculation method. Various statistical values such as the mean, maximum, minimum, and median can be used for statistical processing of standardized differences.

The standardized difference for each index shown in FIG. 9 may be output from the output unit 103 and displayed on the screen. For example, for each index, the standardized difference for a simple comparison without using a propensity score and the standardized difference using a specific propensity score may be displayed in a graph format. Such a screen display allows the balance of background information to be confirmed.

Figure 11 is a flowchart of the intervention effect analysis process in Example 1.

First, the input unit 102 receives input of intervention data and non-intervention data, and stores the data in the intervention data storage unit 131 and the non-intervention data storage unit 132 (S101).

Next, the machine learning method setting unit 111 displays a machine learning method setting screen (FIG. 12), sets the machine learning method selected on the machine learning method setting screen as the machine learning method to be used in the intervention effect analysis, and stores the selection result in the machine learning method storage unit 133 (S102).

Next, the background information setting unit 112 displays a background information setting screen (FIG. 13), sets the background information selected on the background information setting screen as background information (covariates) to be used in the intervention effect analysis, and stores the selection result in the background information storage unit 134 (S103).

Next, the reference intervention setting unit 119 displays a basic intervention setting screen (FIGS. 14 and 15) and sets the reference intervention selected by the user (S104). The reference intervention setting unit 119 may select the reference intervention according to a predetermined policy without displaying the basic intervention setting screen and asking the user to select the reference intervention. For example, the group with the largest overlap of the propensity score distribution with other groups (inclusion rate shown in FIG. 15) among the groups to be compared may be selected as the reference intervention, a group implementing the intervention to be evaluated may be selected as the reference intervention, or a group with a small distribution of propensity scores may be selected as the reference intervention.

Next, the propensity score calculation unit 113 calculates a propensity score indicating a baseline intervention probability from the background information using a machine learning model based on the machine learning method selected by the machine learning method setting unit 111, and stores the calculation result in the propensity score storage unit 135 (S105). If multiple machine learning methods are selected in step S102, the propensity score is calculated using a machine learning model based on one of the selected machine learning methods.

Next, the matching data creation unit 114 creates multi-group matching data for the calculated propensity scores and stores it in the matching data storage unit 136 (S106). For example, data with similar propensity scores between the intervention group and the non-intervention group are matched to create a data pair. When multiple propensity scores are calculated, matching data including different data pairs for each propensity score is created.

Next, the standardized difference calculation unit 115 evaluates the matching result, calculates the standardized difference (S107), and calculates a representative value of the calculated standardized difference (S108). For example, the standardized difference calculation unit 115 calculates the age difference between the matched data pairs of the reference intervention group and the other intervention group, and calculates the standardized difference by dividing the average value of the calculated age difference by the standard deviation of the age difference. The standardized difference makes it possible to know the magnitude of the difference in specific background information (e.g., age), and to compare multiple pieces of background information without considering the unit or the magnitude of the value. The standardized difference calculated by the standardized difference calculation unit 115 is stored in the memory 105 or the storage medium 106 as an internal variable of the intervention effect analysis system in the table format shown in FIG. 9. Then, the standardized difference calculation unit 115 statistically processes the standardized difference of the data pairs in each propensity score calculation method to calculate a representative value of the standardized difference. Various statistical values such as the average value, maximum value, minimum value, and median value can be adopted as representative values for the statistical processing of the standardized difference. When multiple propensity scores are calculated, a different standardized difference and a representative value of the standardized difference are calculated for each propensity score.

If multiple machine learning methods are selected in step S102, steps S105 to S108 are repeated for each selected machine learning method, and the standardized difference and the representative value of the standardized difference are calculated for each machine learning method.

Next, the matching data extraction unit 116 selects the best machine learning method based on the evaluation results of the matching data based on the standardized difference. For example, as shown in FIG. 10, the calculation method of the propensity score with the smallest representative value is selected, and matching data based on the propensity score calculated by that calculation method is extracted (S109).

Next, the intervention effect index setting unit 117 displays an intervention effect index setting screen (FIG. 16), and sets the intervention effect index selected on the intervention effect index setting screen as the intervention effect index to be displayed on the intervention effect display screen (FIG. 17) (S110). As shown in FIG. 16, the intervention effect indexes include the level of care, care costs, presence or absence of disease, medical costs by disease, test values, and interview results.

Next, the intervention effect calculation unit 118 calculates the intervention effect index from the matching data in the machine learning method selected based on the evaluation results of the matching data, and generates display data for the intervention effect display screen (Figure 17) (S111).

Next, the intervention effect calculation unit 118 calculates the significance probability of the intervention effect calculated in step S109 and tests the calculated value of the intervention effect (S112).

Figure 12 shows an example of a machine learning method setting screen.

The machine learning method setting screen includes a list of pre-prepared machine learning methods and a selection field for selecting each machine learning method. The machine learning method setting screen is displayed by the machine learning method setting unit 111 in step S102, and prompts the user to select the machine learning method to be used in the intervention effect analysis. After selecting one or more machine learning methods in the selection field, the user operates the "Done" button to store the selection result in the machine learning method storage unit 133.

Figure 13 shows an example of a background information setting screen.

The background information setting screen includes a list of background information, which is the items of intervention data and non-intervention data recorded in the intervention data storage unit 131 and non-intervention data storage unit 132, and a selection field for selecting each piece of background information. The background information setting screen is displayed by the background information setting unit 112 in step S103, and prompts the user to select background information to be used in the intervention effect analysis. After selecting one or more pieces of background information in the selection field, the user operates the "Done" button to store the selection results in the background information storage unit 134.

Figures 14 and 15 show examples of the standard intervention setting screen.

The reference intervention setting screen shown in FIG. 14 displays the intervention methods (including non-intervention), and the intervention methods to be compared can be selected using check boxes. For example, when the intervention methods to be compared have been decided, such as when comparing intervention 1 and intervention 2, the reference intervention setting screen shown in FIG. 14 is effective.

The reference intervention setting screen shown in FIG. 15 displays the distribution of propensity scores for intervention methods (including non-intervention) and the inclusion rate for the propensity scores of other intervention methods, and allows the user to select the intervention method to be compared using a check box. The inclusion rate indicates the proportion of this intervention method that is included in other intervention methods; 100% indicates complete inclusion, and a higher inclusion rate allows for more accurate comparison. Note that instead of the inclusion rate, the average or minimum value of the propensity score may be used. For example, when comparing intervention methods with similar propensity scores by looking at the propensity score distribution, the reference intervention setting screen shown in FIG. 15 is effective.

Figure 16 shows an example of the intervention effect indicator setting screen.

The intervention effect index setting screen includes a list of intervention effect indexes calculated as a result of the intervention effect analysis, a selection field for selecting each intervention effect index, and an input field for the period for calculating the intervention effect. The intervention effect index setting screen is displayed by the intervention effect index setting unit 117 in step S109, and prompts the user to select the intervention effect index to be displayed as a result of the intervention effect analysis. The user selects one or more intervention effect indexes in the selection field, inputs the effect calculation period, and then operates the "Complete" button to transmit the selection result to the intervention effect index setting unit 117. In addition, the intervention effect calculation unit 118 calculates the intervention effect according to the effect calculation period inputted in the intervention effect index setting screen.

Figure 17 shows an example of the intervention effect display screen for Example 1.

The intervention effect display screen includes an intervention effect indicator selection area in which the user selects the intervention effect indicator they wish to display, a number display area that displays the difference in numbers between multiple intervention and non-intervention groups in simple comparisons and comparisons using matching data, a graph display area that displays the intervention effects in graphs in simple comparisons and comparisons using matching data, and a numerical display area that displays the intervention effects numerically.

The user selects an intervention effect index selected on the intervention effect index setting screen (Figure 16) in the intervention effect index selection area to switch between the intervention effect index displayed in the graph display area and the numerical value display area. The graph display area displays the intervention effect for the period entered in the effect calculation period on the intervention effect index setting screen so that multiple intervention groups and non-intervention groups can be compared. The numerical value display area displays the difference in the intervention effect between multiple intervention groups and non-intervention groups for the period entered in the effect calculation period on the intervention effect index setting screen, and the significance probability of the numerical value of the intervention effect. The significance probability allows the user to check the statistical certainty of the intervention effect value.

In addition, when comparing multiple intervention groups with a non-intervention group, the intervention effect indexes before the intervention are different in a simple comparison, but in the matching data, the intervention effect indexes before the intervention are equal, so the difference in the intervention effect indexes between multiple intervention groups and the non-intervention group after the intervention can be seen, allowing the correct intervention effect to be analyzed.

In the above-described Example 1, an example was described in which the intervention effects were compared with other intervention groups and a non-intervention group using one intervention group as a benchmark, but the intervention effects of three or more intervention groups may be compared without including the non-intervention group in the comparison.

As described above, in Example 1, matching data is created for multiple intervention groups, and matching data with similar background effects is selected. This makes it possible to know the true effects of the intervention measures.

Example 2
In the second embodiment of the present invention, the standardized difference of the background information of each paired data extracted by multiple propensity scores is evaluated sequentially, and multi-group matching data is created from the paired data that minimizes the standardized difference. Note that in the second embodiment, differences from the first embodiment are mainly described, and the description of the same configuration and effects as the first embodiment is omitted.

Figure 18 is a block diagram showing the configuration of the intervention effect analysis system of Example 2.

The intervention effect analysis system of Example 2 has an analysis terminal 100 and a database 130.

The analysis terminal 100 is a computer having an input unit 102, an output unit 103, a processor (CPU) 104, a memory 105, and a storage medium 106.

The storage medium 106 stores programs that realize the machine learning method setting unit 111, the background information setting unit 112, the propensity score calculation unit 113, the standardized difference calculation unit 115, the paired data creation unit 120, the paired data extraction unit 121, the intervention effect index setting unit 117, the intervention effect calculation unit 118, and the reference intervention setting unit 119.

The paired data creation unit 120 creates paired data between the personal data of the reference intervention data and personal data with similar propensity scores. The paired data extraction unit 121 selects the paired data with the smallest representative value of the standardized difference in the background information of each paired data.

Figure 19 is a flowchart of the intervention effect analysis process in Example 2.

Steps S101 to S104 and steps S110 to S112 are the same as the intervention effect analysis process of Example 1 described above (Figure 11).

After the baseline intervention is set in step S104, the propensity score calculation unit 113 uses multiple machine learning methods to calculate a propensity score indicating the baseline intervention probability from the background information, and stores the calculation result in the propensity score storage unit 135 (S115).

Next, the paired data creation unit 120 extracts one personal data item from the reference intervention data (S116), selects personal data items with propensity scores close to the extracted personal data items from other data groups (intervention data and non-intervention data to be compared), and creates paired data with the selected personal data items (S117).

For example, as shown in FIG. 20, the propensity scores of the personal data of intervention data 1, which is the reference intervention group, are calculated using different calculation methods. Furthermore, the propensity scores of the personal data of the other intervention groups and non-intervention groups for which paired data is to be created in comparison with the reference intervention group are calculated using different calculation methods. Then, one piece of personal data is extracted from intervention data 1, which is the reference intervention group, and data with similar propensity scores is selected from the other groups, and paired data is created for each propensity score calculation method. For this reason, as shown in the figure, the paired data created differs depending on the propensity score calculation method.

Next, the paired data extraction unit 121 calculates a representative value of the standardized difference of the background information of each paired data. If there is already selected paired data, the standardized difference is calculated including the already selected paired data, and the representative value is calculated. In the statistical processing of the standardized difference, various statistical values such as the average, maximum, minimum, and median can be used as the representative value (S118). Furthermore, the paired data extraction unit 121 selects the paired data with the smallest representative value of the standardized difference (S119).

For example, as shown in FIG. 21, the standardized difference in background information and its representative value (e.g., average value) are calculated for each paired data, and the paired data with the smallest representative value of the standardized difference is selected. If the representative value of the standardized difference in background information is small, the background information is similar, so for propensity score 1, paired data is selected between intervention 1 and intervention 2, and for propensity score 2, paired data is selected between intervention 1 and intervention 3 or intervention 1 and no intervention. For this reason, for a person with personal ID K0001, a pair is created with person K0021, who has a similar propensity score 1, and a pair is created with person K0013, who has a similar propensity score 2.

Then, in step S110, the intervention effect index is set, and the processing from step S110 onwards is executed.

As described above, according to the second embodiment of the present invention, since matching is performed using multiple propensity scores, it is possible to create multi-group matching data in which the distribution of background information is more accurately aligned than in the first embodiment. Therefore, it is possible to achieve both individual optimization and group optimization, and to compare data from each group with better aligned background information.

Example 3
In a third embodiment of the present invention, subgroups are analyzed. In the third embodiment, differences from the first and second embodiments will be mainly described, and descriptions of the same configurations and effects as the first and second embodiments will be omitted.

Figure 22 is a block diagram showing the configuration of the intervention effect analysis system of Example 3.

The intervention effect analysis system of Example 3 has an analysis terminal 100 and a database 130.

The analysis terminal 100 is a computer having an input unit 102, an output unit 103, a processor (CPU) 104, a memory 105, and a storage medium 106.

The storage medium 106 stores a program that realizes the intervention effect calculation unit 118 and the background information group setting unit 123. The storage medium 106 may also store a program that realizes the other functional blocks described in Example 1 (machine learning method setting unit 111, background information setting unit 112, propensity score calculation unit 113, matching data creation unit 114, standardized difference calculation unit 115, matching data extraction unit 116, intervention effect index setting unit 117, reference intervention setting unit 119, etc.).

The background information group setting unit 123 displays a background information group setting screen (Figure 23) and sets a group of background information selected on the background information group setting screen. The intervention effect is calculated for each group of background information that has been set.

The database 130 is composed of an intervention data storage unit 131, a non-intervention data storage unit 132, and a matching data storage unit 136. The database 130 may also store other data described in Example 1 (such as a machine learning method storage unit 133, a background information storage unit 134, and a propensity score storage unit 135).

Figure 23 shows an example of a background information group setting screen for Example 3.

The background information group setting screen includes a list of background information, which is an item of intervention data and non-intervention data recorded in the intervention data storage unit 131 and the non-intervention data storage unit 132, and a selection field for selecting each background information. The background information group setting screen is displayed by the background information group setting unit 123, and prompts the user to select background information for analyzing the intervention effect for each group. After selecting one or more background information in the selection field, the user operates the "Complete" button to transmit the selection result to the background information group setting unit 123. The background information group may be set in advance for each background information (for example, ages from 60 to 79 years old may be divided into three groups of 10 years each, such as the 60s, 70s, and 80s), or after operating the "Complete" button, a sub-screen for setting the details of the group may be displayed to prompt the user to input the group range (60 to 79 years old) and interval (every 10 years).

Figure 24 shows an example of the intervention effect display screen for Example 3.

The intervention effect display screen of Example 3 includes an intervention effect index selection area in which the user selects the intervention effect index they wish to display, an effect calculation method selection area in which they select between displaying a simple comparison and displaying matching data, a number display area in which the difference in number of people for each group of background information between multiple intervention groups and non-intervention groups in a simple comparison and a comparison using matching data, a graph display area in which the intervention effect is displayed in a graph for each group of background information in a simple comparison and a comparison using matching data, and a numerical display area in which the intervention effect for each group of background information is displayed numerically.

The user selects the intervention effect index selected on the intervention effect index setting screen (Figure 16) in the intervention effect index selection area to switch the intervention effect index displayed in the graph display area and numerical value display area. In addition, the user selects either simple comparison or matching data in the effect calculation method selection area to switch the data displayed in the graph display area and numerical value display area. The graph display area displays the intervention effect for the period entered in the effect calculation period on the intervention effect index setting screen for each group of background information so that multiple intervention groups and non-intervention groups can be compared. The numerical value display area displays the difference between the multiple intervention groups and non-intervention groups in the intervention effect for the period entered in the effect calculation period on the intervention effect index setting screen and the significance probability of the numerical value of the intervention effect for each group of background information. The user can check the statistical certainty of the intervention effect value from the significance probability.

As described above, in the third embodiment, the effectiveness of an intervention can be known for each group of background information (e.g., age) by grouping the data that combines background information as a whole, without having to combine background information for each group. This makes it possible to know the demographic for which an intervention is suitable.

The present invention is not limited to the above-described embodiments, but includes various modified examples and equivalent configurations within the spirit of the appended claims. For example, the above-described embodiments have been described in detail to clearly explain the present invention, and the present invention is not necessarily limited to having all of the configurations described. Furthermore, part of the configuration of one embodiment may be replaced with the configuration of another embodiment. Furthermore, the configuration of another embodiment may be added to the configuration of one embodiment. Furthermore, part of the configuration of each embodiment may be added, deleted, or replaced with other configurations.

Furthermore, each of the configurations, functions, processing units, processing means, etc. described above may be realized in part or in whole in hardware, for example by designing them as integrated circuits, or may be realized in software by a processor interpreting and executing a program that realizes each function.

Information such as programs, tables, and files that realize each function can be stored in a storage device such as a memory, hard disk, or SSD (Solid State Drive), or in a recording medium such as an IC card, SD card, or DVD.

In addition, the control lines and information lines shown are those considered necessary for explanation, and do not necessarily represent all control lines and information lines necessary for implementation. In reality, it is safe to assume that almost all components are interconnected.

100 Analysis terminal 102 Input unit 103 Output unit 104 Processor 105 Memory 106 Storage medium 111 Machine learning method setting unit 112 Background information setting unit 113 Propensity score calculation unit 114 Matching data creation unit 115 Standardized difference calculation unit 116 Matching data extraction unit 117 Intervention effect index setting unit 118 Intervention effect calculation unit 119 Reference intervention setting unit 120 Background information group setting unit 130 Database 131 Intervention data storage unit 132 Non-intervention data storage unit 133 Machine learning method storage unit 134 Background information storage unit 135 Propensity score storage unit 136 Matching data storage unit

Claims (9)

An intervention effect analysis system for analyzing the effects of multiple intervention measures,
The present invention comprises a computing device that executes a predetermined process, an input unit to which data is input, and a storage device that can be accessed by the computing device,
The input unit receives input of intervention data of a plurality of intervention groups in which one of the intervention measures is implemented, and non-intervention data of a non-intervention group in which the intervention measure is not implemented,
The computing device includes:
Calculating a propensity score for each of the plurality of intervention data and the non-intervention data from background information included in each of the plurality of intervention data and the non-intervention data using a plurality of machine learning methods;
Set one of the plurality of intervention data or the non-intervention data to a reference intervention group ;
generating paired data of a plurality of individuals having a small difference in the calculated propensity scores from the data of the individuals in the set reference intervention group and the plurality of intervention data and non-intervention data;
Calculating a standardized difference indicating a difference in background information of the plurality of paired data;
Calculating propensity scores for the data of the baseline intervention group, other intervention groups, and non-intervention groups in multiple ways;
Select data having a propensity score similar to the data of the reference intervention group from other intervention groups and non-intervention groups, and create paired data for each calculation method of the propensity score;
An intervention effect analysis system characterized by selecting paired data with the smallest representative value of the standardized difference calculated for the background information of the paired data, calculating the intervention effect from the selected paired data, and outputting the results of a comparative analysis between the multiple intervention data or the non-intervention data and the reference intervention group . The intervention effect analysis system according to claim 1,
The intervention effect analysis system, wherein the calculation device sets the reference intervention group in accordance with a predetermined policy. The intervention effect analysis system according to claim 2,
The intervention effect analysis system is characterized in that the calculation device sets the group that has the largest overlap in distribution of propensity scores with other groups among the groups to be compared as the reference intervention group . The intervention effect analysis system according to claim 1,
The intervention effect analysis system is characterized in that the calculation device outputs propensity scores of an intervention group and a non-intervention group that can be used as comparison targets in a comparable manner, and prompts a user to select a reference intervention group . The intervention effect analysis system according to claim 1,
The calculation device is an intervention effect analysis system characterized in that it outputs, for each indicator for evaluating the intervention effect, a standardized difference from a simple comparison without using a propensity score and a standardized difference using a specific propensity score in a comparable manner. The intervention effect analysis system according to claim 1,
The calculation device is an intervention effect analysis system characterized in that it generates data for displaying an intervention effect display screen including an intervention effect index selection area for selecting the intervention effect index to be displayed, a number display area for displaying the difference in numbers between the intervention group and the non-intervention group, a graph display area for displaying the intervention effect in a graph, and a numerical display area for displaying the intervention effect as a numerical value. The intervention effect analysis system according to claim 1,
The intervention effect analysis system is characterized in that the calculation device divides the background information into groups and calculates the intervention effect for each group. The intervention effect analysis system according to claim 1,
The intervention effect analysis system is characterized in that the intervention measure is an action for improving health or quality of life related to medical care, health checkups, or nursing care. An intervention effect analysis method in which a computer analyzes the effects of multiple intervention measures, comprising:
The computer includes an arithmetic unit that executes a predetermined process, an input unit to which data is input, and a storage device that is accessible by the arithmetic unit;
The method comprises:
The input unit receives input of intervention data of a plurality of intervention groups in which one of the intervention measures is implemented, and non-intervention data of a non-intervention group in which the intervention measure is not implemented,
The computing device calculates a propensity score for each of the plurality of intervention data and the non-intervention data from background information included in each of the plurality of intervention data and the non-intervention data using a plurality of machine learning methods;
The arithmetic device sets one of the plurality of intervention data or the non-intervention data to a reference intervention group;
The computing device generates paired data of a plurality of individuals having a small difference in the calculated propensity scores from the data of the individuals in the set reference intervention group, the plurality of intervention data, and the plurality of non-intervention data;
The computing device calculates a standardized difference indicating a difference in background information of the plurality of paired data,
The intervention effect analysis method is characterized in that the calculation device calculates an intervention effect from the plurality of paired data based on the calculated standardized difference, and outputs the results of a comparative analysis between the plurality of intervention data or the non-intervention data and the reference intervention group.

Images