Information processing system and information processing method

Abstract

To extract a cross-connection factor with a small amount of calculation.SOLUTION: Information processing holds, for each of a plurality of targets, a value of an intervention variable, a value of a result variable candidate as a candidate for a result variable, and a value of a cross-connection factor candidate as a candidate for a cross-connection factor between the intervention variable and the result variable; sequentially selects combinations of three variables which are one intervention variable, one result variable candidate and one cross-connection factor candidate; searches for a cause and effect among the three variables on the basis of the values of the three variables included in the selected combinations for the plurality of targets and a predetermined algorithm; and extracts a cross-connection factor from the cross-connection factor candidates on the basis of the cause and effect found.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to an information processing system and an information processing method. [Background technology]

[0002] Background art in this technical field is found in Japanese Patent No. 7326640 (Patent Document 1), which states that "an information processing device acquires input data related to a plurality of variables, acquires information related to causal relationships of the plurality of variables estimated based on the input data, outputs the acquired information related to the causal relationships, accepts selection of a target variable from the plurality of variables, acquires an estimation result of values of the plurality of variables that optimize the target variable, and outputs the acquired estimation result" (see abstract). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 7326640 Summary of the Invention [Problem to be solved by the invention]

[0004] In order to accurately analyze the effect of an intervention on a subject, it is necessary to extract confounding factors. Patent Document 1 describes estimating the causal relationship between multiple variables, but does not describe extracting confounding factors. Therefore, one aspect of the present invention extracts confounding factors with a small amount of calculation. [Means for solving the problem]

[0005] In order to solve the above-mentioned problems, one embodiment of the present invention employs the following configuration: An information processing system includes a processor and a memory, wherein the memory stores, for each of a plurality of subjects, a value of an intervention variable, a value of a candidate outcome variable that is a candidate outcome variable, and a value of a candidate confounder that is a candidate confounder between the intervention variable and the outcome variable, wherein the intervention variable is a variable that indicates the status of an intervention on the subject, and the outcome variable is a variable that is affected by the intervention on the subject, and the processor sequentially selects combinations of three variables each consisting of one of the intervention variables, one of the candidate outcome variable, and one of the candidate confounder, searches for causality between the three variables based on the values of each of the three variables included in the selected combinations of the plurality of subjects and a predetermined algorithm, and extracts the confounder from the candidate confounders based on the searched causality. [Effects of the Invention]

[0006] According to one aspect of the present invention, confounding factors can be extracted with a small amount of calculation.

[0007] Problems, configurations, and effects other than those described above will become apparent from the following description of the embodiments. [Brief explanation of the drawings]

[0008] [Figure 1] 1 is a block diagram showing an example of the configuration of a causal exploration system according to a first embodiment. [Figure 2] FIG. 3 is a diagram showing an example of the data configuration of intervention data in the first embodiment. [Figure 3] FIG. 2 is a diagram showing an example of the data configuration of observation data in the first embodiment. [Figure 4] FIG. 1 is a diagram showing an example of a screen configuration of a confounding factor candidate setting screen in Example 1. [Figure 5] FIG. 10 is a diagram showing an example of a screen configuration of a result variable candidate setting screen in the first embodiment. [Figure 6] 1 is a flowchart showing an example of a confounding factor extraction process in the first embodiment. [Figure 7]3 is an example of causal search data in the first embodiment. [Figure 8] FIG. 2 is an explanatory diagram showing an example of a causal search space of three variables in the first embodiment. [Figure 9] FIG. 1 is an explanatory diagram illustrating an example of a confounding factor extraction process in Example 1. [Figure 10] FIG. 2 is a diagram showing an example of the data configuration of confounding factor data in Example 1. [Figure 11] 10 is a flowchart illustrating an example of a result variable extraction process according to the first embodiment. [Figure 12] FIG. 2 is an explanatory diagram showing an example of a two-variable causal search space in the first embodiment. [Figure 13] FIG. 10 is an explanatory diagram illustrating an example of a result variable extraction process in the first embodiment. [Figure 14] 10 is a flowchart illustrating an example of a causal graph creation process according to the first embodiment. [Figure 15] FIG. 10 is a diagram illustrating an example of a screen configuration of a causal graph display screen in the first embodiment. [Figure 16] FIG. 10 is a diagram illustrating an example of a screen configuration of a causal graph display screen in the first embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0009] Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings. In this embodiment, the same components are generally designated by the same reference numerals, and repeated explanations will be omitted. It should be noted that this embodiment is merely an example for realizing the present invention, and does not limit the technical scope of the present invention. [Example]

[0010] 1 is a block diagram showing an example of the configuration of a causal exploration system. The causal exploration system includes, for example, a causal exploration device 100 and a database 120 connected to each other. The causal exploration device 100 is configured by a computer having, for example, a CPU (Central Processing Unit) 101, a memory 102, an auxiliary storage device 103, an input device 104, an output device 105, and a communication device 106.

[0011] The CPU 101 is an example of a processor, and executes programs stored in the memory 102. The memory 102 includes a ROM (Read Only Memory), which is a nonvolatile storage element, and a RAM (Random Access Memory), which is a volatile storage element. The ROM stores unchanging programs (e.g., a BIOS (Basic Input / Output System)). The RAM is a high-speed, volatile storage element such as a DRAM (Dynamic Random Access Memory), and temporarily stores programs executed by the CPU 101 and data used when the programs are executed.

[0012] The auxiliary storage device 103 is a large-capacity, non-volatile storage device such as a magnetic storage device (HDD (Hard Disk Drive)) or a flash memory (SSD (Solid State Drive)), and stores programs to be executed by the CPU 101 and data to be used when the programs are executed. That is, the programs are read from the auxiliary storage device 103, loaded into the memory 102, and executed by the CPU 101.

[0013] The input device 104 is a device such as a keyboard or mouse that receives input from an operator. The output device 105 is a device such as a display device or printer that outputs the results of program execution in a format that can be viewed by the operator.

[0014] The communication device 106 is a network interface device that controls communication with other devices in accordance with a predetermined protocol, and includes a serial interface such as a USB (Universal Serial Bus).

[0015] Some or all of the programs executed by CPU 101 may be provided to causal exploration device 100 from removable media (CD-ROM, flash memory, etc.) which are non-transitory storage media, or from an external computer equipped with a non-transitory storage device via a network, and may be stored in non-volatile auxiliary storage device 103 which is also non-transitory storage media. For this reason, causal exploration device 100 should have an interface for reading data from removable media.

[0016] The causal exploration device 100 is a computer system configured on one physical computer or on multiple logically or physically configured computers, and may operate in separate threads on the same computer, or on a virtual computer constructed on multiple physical computer resources.

[0017] The CPU 101 includes, for example, a confounding factor candidate setting unit 111, a result variable candidate setting unit 112, a causal search unit 113, a confounding factor extraction unit 114, a result variable extraction unit 115, and a causal graph creation unit 116, all of which are functional units.

[0018] The confounding factor candidate setting unit 111 sets confounding factor candidates. The outcome variable candidate setting unit 112 sets outcome variable candidates. The causal search unit 113 searches for causal relationships between multiple variables. The confounding factor extraction unit 114 extracts confounding factors from the confounding factor candidates using the search results of the causal relationships between the intervention variables, confounding factor candidates, and outcome variable candidates by the causal search unit 113.

[0019] The outcome variable extraction unit 115 extracts outcome variables (with an intervention effect) from the outcome variable candidates using the search results for causality between the intervention variables and outcome variable candidates after fixing confounding factors by the causal search unit 113. The causal graph creation unit 116 creates a causal graph including the intervention variables, confounding factors, and outcome variables.

[0020] The intervention variable indicates the implementation status of an intervention performed on a subject. In this embodiment, an example in which a health care intervention is performed on an individual will be described, but the subject of the intervention is not limited to humans (for example, it may be an animal or an object), and intervention other than health care (for example, sales activities in marketing) may also be performed.

[0021] The outcome variable is a variable that is the target of measuring the effect of the intervention. In this embodiment, only variables that are affected by the intervention (variables that are affected positively or negatively by the intervention) may be called outcome variables.

[0022] A confounding factor is a variable (factor) that is related to an intervention variable and that affects an outcome variable. The causal exploration device 100 extracts confounding factors and adjusts the influence of the confounding factors in order to more accurately identify outcome variables that have an intervention effect due to the intervention variable and to more accurately measure the intervention effect.

[0023] For example, CPU 101 functions as confounding factor candidate setting unit 111 by operating in accordance with a confounding factor candidate setting program loaded into memory 102, and functions as outcome variable candidate setting unit 112 by operating in accordance with an outcome variable candidate setting program loaded into memory 102. The same relationship between programs and functional units applies to other functional units included in CPU 101.

[0024] Note that some or all of the functions of the functional units included in the CPU 101 may be realized by hardware such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field-Programmable Gate Array).

[0025] The database 120 includes, for example, an input data storage unit 121, a confounding factor storage unit 122, a result variable storage unit 123, and a causal graph storage unit 124, all of which are realized by areas for storing data.

[0026] The input data storage unit 121 stores, for example, information indicating values of intervening variables, values of candidate confounding factors, and values of candidate outcome variables for multiple individuals. The confounding factor storage unit 122 stores confounding factors extracted by the confounding factor extraction unit 114. The outcome variable storage unit 123 stores outcome variables extracted by the outcome variable extraction unit 115. The causal graph storage unit 124 stores the causal graph created by the causal graph creation unit 116.

[0027] It should be noted that some or all of the information stored in the database 120 may be stored in the memory 102 or in the auxiliary storage device 103 .

[0028] In this embodiment, the information used by the causal discovery system does not depend on the data structure and may be expressed in any data structure. For example, the information can be stored in a data structure appropriately selected from a table, a list, a database, or a queue.

[0029] 2 is a diagram showing an example of the data configuration of intervention data stored in the input data storage unit 121. The intervention data 200 stores, for example, attribute information of each individual at the reference time and the values of intervention variables of each individual at the reference time, in association with an individual ID that identifies the individual. In this embodiment, an example will be described in which the reference time is year T.

[0030] The attribute information at the baseline time is information that may be a confounding factor candidate, and includes, for example, gender, age, address, etc. The intervention variable includes, for example, a flag indicating whether or not a health program intervention was performed on the individual at the baseline time (for example, whether or not health guidance was provided, whether or not a specific medical procedure was performed, etc.) (for example, "1" if an intervention was performed on the individual in year T, and "0" if an intervention was not performed on the individual in year T).

[0031] 3 is a diagram showing an example of the data configuration of observation data stored in input data storage unit 121. Observation data 300 stores, for example, attribute information for each individual at the reference time, health information for each individual at the reference time, and the value of health information for each individual at a specific timing after the reference time, all associated with an individual ID. Note that, due to space limitations in the drawing, observation data 300 is shown divided into two tables, one above the other, in FIG. 3, but observation data 300 is data in which the lower table is joined to the right of the upper table in FIG. 3.

[0032] The attribute information at baseline for each individual in the observation data 300 is the same as the attribute information stored in the intervention data 200. Note that the attribute information at baseline for each individual may be stored in only one of the intervention data 200 or the observation data 300.

[0033] Health information includes, for example, nursing care information, medical information, and health checkup information. Health information at the baseline time is information that can be a candidate confounding factor. Health information at a specific timing after the baseline time is information that can be a candidate outcome variable. In this embodiment, an example will be described in which the specific timing is year T+1, year T+2, etc. (i.e., one year, two years after the baseline time, etc.).

[0034] The nursing care information includes, for example, a value indicating the level of nursing care required for an individual and a value indicating the nursing care costs for the individual. The medical information includes, for example, a value indicating the presence or absence of each individual's disease and a value indicating the medical costs for each individual's disease. The health checkup information includes, for example, test values for each individual's test type and interview results for each individual's interview content.

[0035] 4 is a diagram showing an example of the screen configuration of the confounding factor candidate setting screen. The confounding factor candidate setting screen 400 is displayed on the output device 105, and includes, for example, a confounding factor candidate setting area 410 and a done button 420.

[0036] The confounding factor candidate setting area 410 displays, for example, each item of attribute information for year T indicated by the intervention data 200 and / or observation data 300, each item of health information for year T indicated by the observation data 300, and check boxes for selecting these items. When the complete button 420 is selected, the items whose check boxes are selected in the confounding factor candidate setting area 410 are set as confounding factor candidates.

[0037] 5 is a diagram showing an example of the screen layout of the result variable candidate setting screen. The result variable candidate setting screen 500 is displayed on the output device 105, and includes, for example, a result variable candidate setting area 510, a year setting area 520, and a done button 530.

[0038] The outcome variable candidate setting area 510 displays, for example, each item of health information from year T+1 onwards indicated by the observation data 300, and check boxes for selecting these items. The year setting area 520 is an area for specifying, using a pull-down menu, the year from year T+1 onwards indicated by the observation data 300. When the done button 530 is selected, the items for the year specified in the year setting area 520, whose check boxes are selected in the outcome variable candidate setting area 510, are set as outcome variable candidates.

[0039] The result variable candidate setting screen 500 does not have to include the year setting area 520. In this case, for example, for each year from year T+1 onwards, the items whose checkboxes are selected in the result variable candidate setting area 510 are set as result variable candidates, and the processing from step S603 onwards in Figure 6 and the processing in Figure 11 described below are executed.

[0040] Fig. 6 is a flowchart showing an example of the confounding factor extraction process. Before the process of Fig. 6 starts, it is assumed that the intervention data 200 and the observation data 300 are stored in advance in the input data storage unit 121. The intervention data 200 and the observation data 300 are set, for example, by inputting them to the input device 104 and / or by transmitting them from an external device.

[0041] The confounding factor extraction unit 114 acquires the intervention data 200 and the observation data 300 relating to multiple individuals (which may be all individuals, or multiple individuals specified by the user) from the input data storage unit 121 (S601).

[0042] The confounding factor candidate setting unit 111 displays a confounding factor candidate setting screen 400 on the output device 105 and accepts the setting of confounding factor candidates via input to the input device 104, and the result variable candidate setting unit 112 displays a result variable candidate setting screen 500 on the output device 105 and accepts the setting of result variable candidates via input to the input device 104 (S602).

[0043] In step S602, the confounding factor extraction unit 114 combines the intervention data 200 and the observation data 300 acquired in step S601 using the individual ID as a key, and acquires, from the combined data, the values of the intervention variables, the values of the set candidate confounding factors, and the values of the set candidate outcome variables as data for causal exploration. Note that the confounding factor extraction unit 114 may store the data for causal exploration in the input data storage unit 121.

[0044] 7 shows an example of the causal discovery data acquired in step S602. The causal discovery data 700 indicates values of an intervening variable, a potential confounding factor, and a potential outcome variable for each individual.

[0045] Returning to the explanation of Figure 6, the confounding factor extraction unit 114 selects one unselected combination from among combinations consisting of one outcome variable candidate and one confounding factor candidate included in the causal exploration data 700 (S603). The confounding factor extraction unit 114 searches for causality among three variables, i.e., the intervening variable, the outcome variable candidate included in the selected combination, and the confounding factor candidate included in the selected combination, by applying a predetermined algorithm to the values of the three variables indicated in the causal exploration data 700 (S604).

[0046] Hereinafter, an example will be described in which LiNGAM (Linear Non-Gaussian Acyclic Model) is used as the predetermined algorithm, but other algorithms such as a Bayesian network or a method using deep learning such as GAN (Generative Adversarial Network) or GNN (Graph Neural Network) may also be used. Note that when LiNGAM is used, the accuracy of causality estimation in this embodiment is high.

[0047] 8 is an explanatory diagram showing an example of a causal search space for three variables searched in step S604. If a candidate confounding factor is X, a candidate outcome variable is Y, and an intervention variable is Z, the causal relationship (causal graph) among the three variables X, Y, and Z is defined by the presence or absence and direction of an edge between X and Y, the presence or absence and direction of an edge between Y and Z, and the presence or absence and direction of an edge between Z and X.

[0048] As will be described later, there are 27 types of causal graphs between three variables, and FIG. 8 shows three causal graphs 801 to 803 between three variables as examples. Causal graph 801 includes an edge from X to Z, an edge from X to Y, and an edge from Z to Y. Causal graph 802 includes an edge from X to Z and an edge from X to Y (but does not include an edge between Z and Y). Causal graph 803 includes an edge from Z to X, an edge from X to Y, and an edge from Z to Y. The confounding factor extraction unit 114 identifies the causal graph showing the causal relationship between the three variables from the 27 types of causal graphs described above, thereby identifying the causal relationship between the three variables.

[0049] The causal graph between n variables is 3^( n C2) ("^" indicates a power), that is, the causal search space between n variables is 3^( n As the value of n increases, the number of elements in the causal search space between n variables increases explosively, and as a result, the amount of calculation required to identify a causal graph showing the causality between the n variables from the causal search space also increases explosively.

[0050] Therefore, in this embodiment, the causal search unit 113 searches for causality among three variables in step S604 (selecting one confounding factor candidate and one outcome variable candidate, and searching for causality among the confounding factor candidate, outcome variable, and candidate intervening variable), thereby specifying a causal graph showing the causality among the confounding factor candidate, outcome variable candidate, and intervening variable from a causal search space with a small number of elements (27). In other words, the causal search unit 113 can specify the causal relationships among the confounding factor candidate, outcome variable candidate, and intervening variable with a small amount of calculation.

[0051] Returning to the explanation of Fig. 6, the confounding factor extraction unit 114 extracts confounding factors based on the causality identified in the most recent step S604, associates the extracted confounding factors with the candidate outcome variables of the selected combination, and stores them as confounding factor data (to be described later) (S605). The confounding factor data is stored in the confounding factor storage unit 122.

[0052] Specifically, for example, if the causal graph identified in the most recent step S604 includes an edge from a confounding factor candidate X to an intervening variable Z and an edge from the confounding factor candidate X to a result variable candidate Y (regardless of the presence or direction of an edge between the intervening variable Z and the result variable candidate Y), the confounding factor extraction unit 114 extracts the confounding factor candidate X as a confounding factor, links the extracted confounding factor to the result variable candidate Y, and stores the extracted confounding factor in the confounding factor data.

[0053] The confounding factor extraction unit 114 determines whether all combinations consisting of one outcome variable candidate and one confounding factor candidate have been selected (S606). If the confounding factor extraction unit 114 determines that there is an unselected combination (S606: NO), it returns to step S603, and if it determines that all the combinations have been selected (S606: YES), it ends the confounding factor extraction process.

[0054] 6, in step S603, the confounding factor extraction unit 114 selected a combination of one outcome variable candidate and one confounding factor candidate. However, a combination of one outcome variable candidate and multiple confounding factor candidates may be selected. However, even in this case, in step S604, the confounding factor extraction unit 114 does not search for causality among the multiple confounding factor candidates. In other words, the confounding factor extraction unit 114 performs processing equivalent to sequentially selecting one confounding factor candidate from the multiple confounding factor candidates one by one and searching for causality among the three variables: the selected confounding factor candidate, the outcome variable candidate, and the intervention variable.

[0055] Although FIG. 6 illustrates an example in which there is one intervening variable, when there are multiple intervening variables, the confounding factor extraction unit 114 selects a combination of one intervening variable, one outcome variable candidate, and one confounding factor candidate in step S603, and stores the extracted confounding factor in step S605 in the confounding factor data in association with the intervening variable and the outcome variable candidate.

[0056] 9 is an explanatory diagram showing an example of the confounding factor extraction process in step S605. Causal graph set 901 is a set of causal graphs for which the candidate outcome variable Y is "weight" among the causal graphs for which the causal effects were identified in step S604. Causal graph set 902 is a set of causal graphs for which the candidate outcome variable Y is "blood pressure" among the causal graphs for which the causal effects were identified in step S604.

[0057] In the causal graph set 901, the causal graph in which the confounding factor candidate X is "age" and the causal graph in which the confounding factor candidate X is "weight" include an edge from the confounding factor candidate X to the intervening variable Z and an edge from the confounding factor candidate X to the outcome variable candidate Y, so "age" and "weight" are extracted as confounding factors between the intervening variable Z and the outcome variable candidate "weight". On the other hand, in the causal graph set 901, the causal graph in which the confounding factor candidate X is "blood pressure" does not include an edge from the confounding factor candidate X to the outcome variable candidate Y, so "blood pressure" is not extracted as a confounding factor between the intervening variable Z and the outcome variable candidate "weight".

[0058] In the causal graph set 902, the causal graph in which the confounding factor candidate X is "age" and the causal graph in which the confounding factor candidate X is "blood pressure" include an edge from the confounding factor candidate X to the intervention variable Z and an edge from the confounding factor candidate X to the outcome variable candidate Y, so "age" and "blood pressure" are extracted as confounding factors between the intervention variable Z and the outcome variable candidate "blood pressure". On the other hand, in the causal graph set 902, the causal graph in which the confounding factor candidate X is "weight" does not include an edge from the confounding factor candidate X to the outcome variable candidate Y, so "weight" is not extracted as a confounding factor between the intervention variable Z and the outcome variable candidate "blood pressure".

[0059] 10 is a diagram showing an example of the data configuration of the confounding factor data stored in the confounding factor storage unit 122. The confounding factor data 1000 holds a confounding factor group Xg consisting of one or more confounding factors X identified in step S605 in association with outcome variable candidates.

[0060] The above-described confounding factor extraction process enables automatic extraction of confounding factors to be adjusted according to the intervention variables and outcome variables with a small amount of calculation, and by using the extracted confounding factors in the process described below, accurate analysis of the intervention effect becomes possible.

[0061] 11 is a flowchart showing an example of the result variable extraction process. The result variable extraction unit 115 acquires intervention data 200 and observation data 300 related to multiple individuals from the input data storage unit 121 (S1101). The data acquired in step S1101 is the same as the data acquired in step S601, for example.

[0062] The outcome variable extraction unit 115 acquires the confounding factor group Xg for each outcome variable candidate extracted by the confounding factor extraction unit 114, that is, the confounding factor data 1000, from the confounding factor storage unit 122 (S1102).

[0063] The result variable extraction unit 115 selects one unselected result variable candidate from the result variable candidates included in the confounding factor data 1000 (S1103). The result variable extraction unit 115 fixes the confounding factor group Xg corresponding to the selected result variable candidate in the confounding factor data 1000, and then searches for causality between the two variables, the intervention variable and the selected result variable candidate, by applying the above-mentioned predetermined algorithm to the values of the two variables indicated by the causality search data 700 (S1104).

[0064] Specifically, for example, in step S605, the degree of influence of each confounding factor X included in the confounding factor group Xg on the intervening variable and the degree of influence on the selected outcome variable candidate are calculated, and the outcome variable extraction unit 115 then excludes all of these influence degrees and searches for the causality between the two variables, the intervening variable and the selected outcome variable candidate.

[0065] 12 is an explanatory diagram showing an example of a two-variable causal search space searched in step S1104. Each causal graph included in the causal search space searched in step S1104 includes an edge from the confounding factor group Xg to Z and an edge from the confounding factor group Xg to Y.

[0066] Furthermore, in step S1104, the group of confounding factors Xg is fixed, and then the causality between the two variables, the intervention variable Z and the candidate outcome variable Y, is searched for, so the causal search space to be searched in step S1104 consists of three causal graphs 1201 to 1203.

[0067] Causal graph 1201 includes an edge from Z to Y. Causal graph 1202 does not include an edge between Z and Y. Causal graph 1203 includes an edge from Y to Z. The confounding factor extraction unit 114 identifies the causal graph showing the causal relationship between two variables from the three causal graphs, thereby identifying the causal relationship between the two variables.

[0068] In this embodiment, the causality between two variables is searched for in step S1104 (one candidate outcome variable is selected, the confounding factor group Xg is fixed, and then the causality between the outcome variable and the candidate intervening variable is searched for), thereby specifying a causal graph showing the causality between the candidate outcome variable and the intervening variable from a causal search space with a small number of elements (three). In other words, the causal search unit 113 can specify the causal relationship between the candidate outcome variable and the intervening variable with a small amount of calculation after adjusting for the influence of the candidate confounding factors.

[0069] Returning to the explanation of FIG. 11, the result variable extraction unit 115 extracts candidate result variables with an intervention effect as result variables and stores the extracted result variables in the result variable storage unit 123 (S1105). Specifically, for example, if the causal graph identified in the most recent step S1104 includes an edge from the intervention variable Z to the candidate result variable Y, the result variable extraction unit 115 determines the candidate result variable as a result variable with an intervention effect. Note that "having an intervention effect" may be a concept that includes both a positive and negative effect on the result variable due to the intervention variable. Furthermore, the result variable extraction unit 115 may store the extracted result variables in the result variable storage unit 123 by linking them to a confounding factor group Xg corresponding to the result variable.

[0070] The result variable extraction unit 115 determines whether all result variable candidates have been selected (S1106). If the result variable extraction unit 115 determines that there are unselected result variable candidates (S1106: NO), it returns to step S1103, and if it determines that all result variable candidates have been selected (1106: YES), it ends the result variable extraction process.

[0071] Although Figure 11 illustrates an example in which there is one intervening variable, if there are multiple intervening variables, the confounding factor extraction unit 114 selects a combination of one intervening variable and one candidate outcome variable indicated by the confounding factor data 1000 in step S1103, and links the extracted outcome variable to the intervening variable in step S1105 and stores it in the outcome variable storage unit 123.

[0072] 13 is an explanatory diagram showing an example of the outcome variable extraction process in step S1105. A confounding factor group Xg is fixed in each of the causal graphs 1301 to 1303. Each of the causal graphs 1301 to 1303 includes an edge from the confounding factor group Xg to the intervening variable Z and an edge from the confounding factor group Xg to the outcome variable candidate Y.

[0073] Causal graph 1301 is a causal graph in which the candidate outcome variable Y is "weight", and includes an edge from intervention variable Z to candidate outcome variable Y. Therefore, the candidate outcome variable "weight" is a candidate outcome variable with an intervention effect, and is extracted as the outcome variable.

[0074] Causal graph 1302 is a causal graph in which the outcome variable candidate Y is "blood pressure," and does not include an edge from intervention variable Z to outcome variable candidate Y. Therefore, the outcome variable candidate "blood pressure" is an outcome variable candidate with no intervention effect, and is not extracted as an outcome variable.

[0075] Causal graph 1301 is a causal graph in which the candidate outcome variable Y is "blood glucose," and includes an edge from intervention variable Z to candidate outcome variable Y. Therefore, the candidate outcome variable "blood glucose" is a candidate outcome variable for which the intervention has an effect, and is extracted as the outcome variable.

[0076] The above-described outcome variable extraction process allows automatic extraction of outcome variables that show an effect of an intervention after adjusting for confounding factors, with a small amount of calculation.

[0077] 14 is a flowchart showing an example of the causal graph creation process. The causal graph creation unit 116 acquires intervention data 200 and observation data 300 related to multiple individuals from the input data storage unit 121 (S1401). The data acquired in step S1401 is the same as the data acquired in step S601, for example.

[0078] The causal graph creation unit 116 acquires the confounding factor group Xg for each outcome variable candidate extracted by the confounding factor extraction unit 114, i.e., the confounding factor data 1000, from the confounding factor storage unit 122 (S1402). The causal graph creation unit 116 acquires the outcome variables extracted by the outcome variable extraction unit 115 from the outcome variable storage unit 123 (S1403).

[0079] The causal graph creation unit 116 estimates the intervention effect for each of the outcome variables acquired in step S1403 (S1404). Specifically, for example, the causal graph creation unit 116 refers to the data acquired in step S1401 for each outcome variable, and calculates a predetermined statistical value (e.g., average value) of the comparison result (e.g., difference) between the value of the outcome variable in the target year and the value of the outcome variable in year T (baseline) for each of a group of individuals whose intervention variable has a value of 1 (intervention) and a group of individuals whose intervention variable has a value of 0 (no intervention). Furthermore, for example, the causal graph creation unit 116 calculates the comparison result (e.g., difference) of the predetermined statistical value between a group of individuals whose intervention variable has a value of 1 (intervention) and a group of individuals whose intervention variable has a value of 0 (no intervention) as the intervention effect.

[0080] For example, in step S605, the influence of each of the confounding factors X included in the confounding factor group Xg on the intervention variable and the influence of each of the confounding factors X on the selected outcome variable candidate are calculated, and therefore the causal graph creation unit 116 may estimate the intervention effect after excluding all of these influences in step S1404. Also, for example, in step S1104, the influence of the intervention variable on each outcome variable is calculated, and therefore the causal graph creation unit 116 may determine the influence as the intervention effect.

[0081] The causal graph creation unit 116 creates a causal graph including the intervention variables, the outcome variables with intervention effects extracted by the outcome variable extraction unit 115, and confounding factors linked to the outcome variables, and stores the created causal graph in the causal graph storage unit 124 (S1405). The causal graph creation unit 116 displays a causal graph display screen showing the causal graph created in step S1405 and the intervention effects estimated in step S1404 on the output device 105 (S1406), and ends the causal graph creation process.

[0082] 15 is a diagram showing an example of the screen configuration of the causal graph display screen displayed in step S1406. A causal graph display screen 1500 includes, for example, a causal graph display area 1510 and an intervention effect display area 1520.

[0083] The causal graph display area 1510 displays the causal graph created in step S1405. The causal graph display area 1510 includes, for example, an intervention variable and a list of outcome variables with an intervention effect, and when one outcome variable is selected from the list of outcome variables, a list of confounding factors corresponding to the selected outcome variable can be displayed using a pull-down menu. In addition, in the causal graph display area 1510, for example, the edge between the intervention variable and the selected outcome variable and the edge between the confounding factor and the selected outcome variable are displayed in thick lines, and other edges are displayed in thin lines.

[0084] In the intervention effect display area 1520, the intervention effect estimated in step S1404 for the outcome variable selected in the causal graph display area 1510 is displayed.

[0085] In the example of Figure 15, the outcome variable "Y: weight" is selected in the causal graph display area 1510, so a list of confounding factors (including "X: weight") corresponding to the outcome variable "Y: weight" can be displayed in a pull-down menu, and the edges between the intervention variable and the outcome variable "Y: weight" and the edges between the confounding factors including "X: weight" and the outcome variable "Y: weight" are displayed in thick lines, and the intervention effect related to the outcome variable "Y: weight" is displayed in the intervention effect display area 1520.

[0086] Note that, although only result variables with an intervention effect are displayed in the causal graph display area 1510, result variables without an intervention effect may also be displayed in the causal graph display area 1510 (so that they can be distinguished from result variables with an intervention effect).

[0087] Furthermore, since outcome variables with intervention effects and confounding factors corresponding to the outcome variables may change depending on the year of the outcome variables, the causal graph display area 1510 may also display the year corresponding to each outcome variable.

[0088] Fig. 16 is a diagram showing an example of the screen configuration of the causal graph display screen 1500 displayed in step S1406. The example of the screen configuration in Fig. 16 is the same as that in Fig. 15. In the example of Fig. 16, the result variable "Y: Blood glucose" is selected in the causal graph display area 1510, and therefore a list of confounding factors (including "X: Blood glucose") corresponding to the result variable "Y: Blood glucose" can be displayed in a pull-down menu, and the edges between the intervention variable and the result variable "Y: Blood glucose" and the edges between the confounding factors including "X: Blood glucose" and the result variable "Y: Blood glucose" are displayed in bold lines, and furthermore, the intervention effect related to the result variable "Y: Blood glucose" is displayed in the intervention effect display area 1520.

[0089] The causal graph display screen 1500 allows the user to easily and accurately grasp outcome variables with intervention effects, confounding factors corresponding to those outcome variables, and estimated intervention effects, thereby assisting the user in analyzing the intervention effects.

[0090] The present invention is not limited to the above-described embodiments, but includes various modifications. 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 those including all of the described configurations. Furthermore, it is possible to replace part of the configuration of one embodiment with the configuration of another embodiment, or to add the configuration of another embodiment to the configuration of one embodiment. Furthermore, it is possible to add, delete, or replace part of the configuration of each embodiment with other configurations.

[0091] Furthermore, the above-described configurations, functions, processing units, processing means, etc. may be partially or entirely implemented in hardware, for example, by designing them as integrated circuits. The above-described configurations, functions, etc. may also be implemented in software, with a processor interpreting and executing a program that implements each function. Information such as the programs, tables, and files that implement each function can be stored in a memory, a recording device such as a hard disk or SSD (Solid State Drive), or a recording medium such as an IC card, SD card, or DVD.

[0092] In addition, the control lines and information lines shown are those that are considered necessary for the explanation, and do not necessarily show all the control lines and information lines in the product. In reality, it can be assumed that almost all components are interconnected. [Explanation of symbols]

[0093] 100 Causal exploration device, 101 CPU, 102 memory, 103 auxiliary storage device, 104 input device, 105 output device, 106 communication device, 111 confounding factor candidate setting unit, 112 outcome variable candidate setting unit, 113 causal exploration unit, 114 confounding factor extraction unit, 115 outcome variable extraction unit, 116 causal graph creation unit, 120 database, 121 input data storage unit, 122 confounding factor storage unit, 123 outcome variable storage unit, 124 causal graph storage unit, 200 intervention data, 300 observation data, 700 causal exploration data, 1000 confounding factor data, 1500 causal graph display screen

Claims

1. An information processing system, a processor and a memory, The memory holds, for each of a plurality of subjects, a value of an intervention variable, a value of a candidate outcome variable that is a candidate outcome variable, and a value of a candidate confounder that is a candidate confounder between the intervention variable and the outcome variable; The intervention variable is a variable indicating the status of intervention for the subject, the outcome variable is a variable affected by the intervention in a subject; The processor: Sequentially selecting combinations of three variables consisting of one of the intervention variables, one of the candidate outcome variables, and one of the candidate confounders; Searching for causality between the three variables based on the values of each of the three variables included in the selected combination of the plurality of targets and a predetermined algorithm; An information processing system that extracts the confounding factor from the confounding factor candidates based on the searched causality.

2. 2. The information processing system according to claim 1, The processor: In the search for causality among the three variables, a causality graph showing the relationship between the three variables is identified based on the values of each of the three variables included in the selected combination of the plurality of targets and the predetermined algorithm; an information processing system that extracts, from the identified causal graph, a confounding factor candidate included in a causal graph that includes an edge from the confounding factor candidate to the intervention variable and an edge from the confounding factor candidate to the outcome variable candidate, as the confounding factor.

3. 2. The information processing system according to claim 1, The processor: sequentially selecting a combination of two variables consisting of one of the intervention variables and one of the outcome variable candidates corresponding to the extracted confounding factor; After fixing the extracted confounding factors, searching for causality between the two variables based on the values of each of the two variables included in the selected combination of the plurality of subjects and the predetermined algorithm; An information processing system that extracts the result variable from the result variable candidates based on the causality between the two variables that have been searched for.

4. 4. The information processing system according to claim 1, The predetermined algorithm includes a Linear Non-Gaussian Acyclic Model (LiNGAM).

5. 4. The information processing system according to claim 3, The processor: In the search for causality between the two variables, the extracted confounding factor is fixed, and then a causal graph showing the relationship between the extracted confounding factor and the two variables is identified based on the values of each of the two variables included in the selected combination of the plurality of subjects and the predetermined algorithm; An information processing system that extracts, from the identified causal graphs, candidate outcome variables included in a causal graph that includes an edge from the intervention variable to the candidate outcome variable as the candidate outcome variable.

6. 4. The information processing system according to claim 3, The processor: creating a causal graph showing the relationship between the extracted confounding factors, the extracted outcome variables, and the intervention variables corresponding to the extracted confounding factors and the extracted outcome variables; an information processing system that generates data for displaying the created causal graph;

7. 4. The information processing system according to claim 3, the value of the intervention variable held in the memory is a value at a reference time; the value of the result variable candidate stored in the memory is a value at a predetermined timing after the reference time, The memory further stores values of variables of the same type as the candidate outcome variables at the reference time; The processor: classifying the plurality of subjects into a plurality of groups based on values of intervention variables corresponding to the extracted outcome variables; calculating a comparison result between the value of the extracted result variable at the predetermined timing and the value of the same type of variable at the reference time for each of the plurality of groups; An information processing system that compares the comparison results between the multiple groups and estimates the effect of an intervention indicated by an intervention variable corresponding to the extracted outcome variable on the outcome variable.

8. An information processing method by an information processing system, the information processing system includes a processor and a memory; The memory holds, for each of a plurality of subjects, a value of an intervention variable, a value of a candidate outcome variable that is a candidate outcome variable, and a value of a candidate confounder that is a candidate confounder between the intervention variable and the outcome variable; The intervention variable is a variable indicating the status of intervention for the subject, the outcome variable is a variable affected by the intervention in a subject; The information processing method includes: The processor sequentially selects a combination of three variables consisting of one of the intervention variables, one of the candidate outcome variables, and one of the candidate confounders; The processor searches for causality between the three variables based on the values of each of the three variables included in the selected combination of the plurality of targets and a predetermined algorithm; An information processing method, wherein the processor extracts the confounding factor from the confounding factor candidates based on the searched causality.

Images