JP6506201B2 - System and method for determining explanatory variable group corresponding to objective variable - Google Patents

Description

  The present invention relates to a system and method for determining an explanatory variable group corresponding to a target variable.

  In recent years, in the medical field, an analysis using machine learning or statistical analysis is performed on a data set consisting of measured values of a plurality of explanatory variables (for example, disease names and prescription drugs) from accumulated data. The prediction of the event indicated by the objective variable is performed.

  As background art which selects the explanatory variable included in a data set in order to perform the said prediction etc., there exists Unexamined-Japanese-Patent No. 2000-20504 (patent document 1). In Patent Document 1, “the explanatory variable combining unit S1 to Sn that generates explanatory variables based on candidate explanatory variables, and the relationship that explains the change of the target variable based on the explanatory variables that the explanatory variable combining units S1 to Sn generate A regression analysis execution unit 1 for generating an equation, an appropriateness determination unit 2 for quantitatively evaluating the appropriateness of a relational expression generated by the regression analysis execution unit 1, an explanatory variable and a regression equation having the highest appropriateness are searched The best regression equation determination unit 3 is provided, and the candidate explanatory variables input to the explanatory variable combining units S1 to Sn are set by the explanatory variable generation parameter output from the best regression equation determination unit 3. The explanatory variable combination unit S1 It is described that “̃Sn can output not only the candidate explanatory variable itself but also a combination of candidate explanatory variables and the result of performing some arithmetic processing on the candidate explanatory variables as an explanatory variable.” .

JP 2000-20504 A

  Consider an example in which the technology described in Patent Document 1 selects explanatory variables to be included in the above-described data set from disease names included in the international disease classification. The international disease classification consists of a plurality of disease names, and each disease name belongs to any of a large classification, a middle classification, or a small classification. In addition, the international disease classification is defined in a tree structure from the superordinate concept to the subordinate concept. That is, each disease name included in the minor classification is a subordinate concept of any disease name included in the middle classification, and each disease name included in the middle classification is a subordinate concept of any disease name included in the large classification.

  When selecting explanatory variables from explanatory variable candidates defined in a tree structure such as international disease classification, it is necessary to consider the tree structural connection between the explanatory variable candidates. However, the technology described in Patent Document 1 does not consider the connection.

  For example, consider an example where there is a middle class "sick name group α" that includes the minor classes "sick name A", "sick name B" and "sick name C" as subordinate concepts. For example, when the influence on the objective variable by "disease name B" is strong, even if the influence to the objective variable by "disease name A" and "disease name C" is weak, the "disease name" which is a superordinate concept of "disease name B" There is a high possibility that the influence of the group α on the objective variable is strong.

  At this time, the technique described in Patent Document 1 is highly likely to select “disease name group α” as an explanatory variable in addition to “disease name B”. However, when “disease name B” and “disease name group α” are selected as explanatory variables, the user determines whether there is an influence on the objective variable by “disease name A” and “disease name C”. I can not

  In addition, when “disease name B” and “disease name group α” are selected as explanatory variables, the explanatory variable “disease name B” has double effects in predicting the value of the objective variable, so the prediction accuracy of the value of the objective variable May decrease. Therefore, in such a case, a new explanatory variable candidate obtained by excluding the “disease name B” from the lower concept of the “disease name group α” and the “disease name group α” which is the explanatory variable candidate in consideration of the tree structure connection It is desirable to regenerate the same as the disease name group, that is, a disease name group that is a new explanatory variable candidate consisting of "disease name A" and "disease name C".

  Also, in the medical field, for example, there are several thousand to several tens of thousands of candidate explanatory variables such as disease names, prescription drugs, and examination items. Therefore, in the technique described in Patent Document 1, it is difficult for the user to manually specify the selection of explanatory variables, the determination of explanatory variable candidates to be regenerated, the method of regenerating explanatory variables, and the like appropriately.

  In order to solve the above-mentioned subject, one mode of the present invention adopts the following composition. A system for determining an explanatory variable group corresponding to an objective variable, comprising: a processor and a storage device, wherein the storage device comprises one or more tree structures having one or more layers, and a plurality of explanatory variable candidates Conceptual structure information indicating a correspondence relation indicating any one of the nodes of the one or more tree structures, and measurement value information indicating measurement values of the plurality of explanatory variable candidates and measurement values of the target variable , Each node of the one or more tree structures is a concept, and in the one or more tree structures, a parent node is a superordinate concept of a child node, and the processor determines from the explanatory variable candidate group the The determination process of determining an explanatory variable to be included in an explanatory variable group is repeated, and the explanatory variable candidate group in the first determination process is the plurality of explanatory variable candidates, and the processor determines the determination process. Calculating the first association degree between each of the explanatory variable candidates of the explanatory variable candidate group and the objective variable based on the measurement value information, and based on each of the first association degree, the explanatory variable group Selected explanatory variable candidates to be included in the explanatory variable candidate group, the selected explanatory variable candidate is included in the explanatory variable group, and the selected explanatory variable candidate is excluded from the explanatory variable candidate group; With reference to the information, the explanatory variable candidate which is a superordinate concept of the selected explanatory variable candidate is specified from the explanatory variable candidate group, and the conceptual structure information of the explanatory variable candidate corresponds to each of the specified explanatory variable candidates. A change process is performed to change the selected explanatory variable candidate from the lower concept to the explanatory variable candidate excluding the selected explanatory variable candidate, and the sibling node indicated by the conceptual structure information of the selected explanatory variable candidate is measured. A value is acquired from the measurement value information, a measurement value of each explanatory variable candidate after the change in the change process is calculated based on the acquired measurement value, and the calculated measurement value is included in the measurement value information. ,system.

  According to one aspect of the present invention, an explanatory variable with an explanatory power to a target variable can be selected, and overlapping of the meaning of the selected explanatory variable can be avoided.

  Problems, configurations, and effects other than the above are clarified by the description of the embodiments below.

FIG. 1 is a block diagram showing a configuration example of a variable optimization system in a first embodiment. 5 is an example of a conceptual structure table in Embodiment 1. 7 is an example of a patient information table in Embodiment 1. 7 is an example of an inspection information table in the first embodiment. It is an example of a prescription information table in the first embodiment. 7 is an example of an explanatory variable candidate table in the first embodiment. 7 is an example of a data set storage table according to the first embodiment. It is an example of the data set storage table after the explanatory variable extraction process and the regeneration process are executed in the first embodiment. 7 is an example of a degree-of-association table in the first embodiment. It is an example of the degree-of-association table after the explanatory variable extraction process and the regeneration process are executed in the first embodiment. 5 is a flowchart illustrating an example of a variable optimization process according to the first embodiment. 5 is a flowchart illustrating an example of association degree calculation processing according to the first embodiment. 5 is a flowchart illustrating an example of an explanatory variable extraction process and an explanatory variable regeneration process according to the first embodiment. 7 is an example of a various analysis option reception screen according to the first embodiment. 7 is an example of an analysis processing result display screen in the first embodiment.

  Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. It should be noted that the present embodiment is merely an example for realizing the present invention, and does not limit the technical scope of the present invention. The same reference numerals are given to the same configuration in each drawing.

  The present embodiment describes a variable optimization system. The variable optimization system selects one or more explanatory variables corresponding to the target variable from the plurality of explanatory variable candidates. Each of the plurality of explanatory variable candidates is defined as any one of the nodes of the tree structure having one or more layers.

<System configuration>
FIG. 1 is a block diagram showing an example of the configuration of the information system of this embodiment. The present embodiment describes an example where the variable optimization system is applied to hospital information. The information system of the present embodiment includes, for example, a variable optimization system 100, a hospital information system 120, and an input / output terminal 130 connected to each other via a network 140.

  The network 140 uses wired communication with a LAN (Local Area Network) cable or wireless communication with a wireless LAN. The network 140 can also use other wide area networks such as the Internet, VPN, a cellular phone communication network, a PHS communication network, and the like.

  The variable optimization system 100 calculates the degree of association between each explanatory variable candidate and the objective variable using the accumulated measured values of the objective variable and explanatory variable candidate, and extracts an explanatory variable candidate from the explanatory variable candidate based on the degree of association. Determine the explanatory variables. Further, the explanatory variable candidate selected based on the explanatory variable to be extracted is regenerated.

  The hospital information system 120 is, for example, a computer having a storage device that stores a patient information database 121, an examination information database 122, and a prescription information database 123. Each database included in the hospital information system 120 stores variable measurement values.

  The patient information database 121 stores basic information and the like for each patient in the hospital. The examination information database 122 stores examination information for each patient. The prescription information database 123 stores prescription information for each patient. The information stored in the database included in the hospital information system 120 is provided to the variable optimization system 100 via the network 140.

  The input / output terminal 130 communicates with, for example, an input device (not shown) such as a keyboard, a mouse or a touch panel, an output device (not shown) such as a display, and a communication unit (not shown) And one or more personal computers. In addition, the input / output terminal 130 is, for example, a PDA, a PHS, a mobile phone, a smartphone, or an input device such as a button or a touch panel, an output device such as a display, and a communication unit that communicates with the variable optimization system 100 or the like. It may be a portable terminal such as a tablet terminal.

  The input / output terminal 130 is installed in, for example, a medical institution (health care provider) such as a hospital or a clinic. The variable optimization system 100 is installed, for example, in a data center. By installing the variable optimization system 100 in the data center, it is possible to centrally manage privacy information such as patient's personal information and data collected from the patient, so that security management such as information leakage prevention can be simplified. The variable optimization system 100 may be installed in a healthcare provider depending on the mode of operation.

  A doctor at a medical institution, a person in charge of analysis, a pharmacist, a manager, and a manager are examples of a user of the input / output terminal 130 (hereinafter referred to as a user). The variable optimization system 100 extracts an explanatory variable corresponding to the target variable according to an instruction from the input / output terminal 130 from the user, and outputs a data set including the extracted explanatory variable and the measured value of the target variable.

  The variable optimization system 100 is configured by, for example, a computer including a control unit 101, an output unit 102, a memory 103, a communication unit 104, and an auxiliary storage device 116 connected to one another. The control unit 101 is, for example, a processor that executes a program stored in the memory 103, and controls the variable optimization system 100.

  The memory 103 includes a ROM, which is a non-volatile storage element, and a RAM, which is a volatile storage element. The ROM stores an immutable program (for example, BIOS). The RAM is a high-speed and volatile storage element such as a dynamic random access memory (DRAM), and temporarily stores a program executed by the control unit 101 and data used when the program is executed.

  The auxiliary storage device 116 is, for example, a large-capacity non-volatile storage device such as a magnetic storage device (HDD) or a flash memory (SSD), and the program executed by the control unit 101 and the data used when executing the program Store. That is, the program is read from the auxiliary storage device 116, loaded into the memory 103, and executed by the control unit 101. Note that part or all of the data stored in the auxiliary storage device 116 may be stored in the memory 103. Also, part or all of the data stored in the memory 103 may be stored in the auxiliary storage device 116.

  The auxiliary storage device 116 stores, for example, an integrated database 113, a conceptual structure database 114, and an analysis data storage database 115. The integrated database 113 stores information acquired from a database included in the hospital information system 120.

  Specifically, the integrated database 113 stores, for example, a variable that can be a candidate for an explanatory variable to be extracted, and a measured value of the variable. Hereinafter, candidates for the explanatory variable to be extracted are referred to as explanatory variable candidates. In the present embodiment, the measured value is a concept including an actual measurement value, a predicted value, and an estimated value. The conceptual structure database 114 stores data indicating the hierarchical structure to which the variable belongs. The analysis data storage database 115 stores, for example, data indicating the processing result of variable optimization processing.

  The control unit 101 is, for example, a processing unit that executes processing for realizing the function of the variable optimization system 100, for example, the display screen generation unit 105, the data extraction unit 106, the objective variable generation unit 107, and the explanatory variable generation unit 108 includes a data set creation unit 109, a degree of association calculation unit 110, an explanatory variable re-generation unit 111, a conceptual structure editing unit 112, and an accumulated data acquisition unit 119.

  For example, the control unit 101 operates as the display screen generation unit 105 by operating according to the display screen generation program loaded into the memory 103, and operates as the data extraction program loaded into the memory 103 to operate as the data extraction unit. It functions as 106. The same applies to the other units included in the control unit 101. Note that each unit included in the control unit 101 may be realized by dedicated hardware.

  The display screen generation unit 105 generates information for displaying an analysis option and an analysis processing result to be described later on the output unit 102, and the generated information is transmitted to the display device or the like via the output unit 102 or the communication unit 104. The data is output to the input / output terminal 130 etc.

  The data extraction unit 106 extracts data from the database stored in the auxiliary storage device 116. The target variable generation unit 107 generates a target variable based on, for example, an instruction from the user and the information stored in the conceptual structure database 114. The explanatory variable generation unit 108 generates candidates for explanatory variables to be extracted based on, for example, an instruction from the user and the information stored in the conceptual structure database 114.

  The data set creation unit 109 creates a data set storage table stored in the analysis data storage base. Details of the data set storage table will be described later. The degree of association calculation unit 110 calculates the degree of association between variables. The explanatory variable regeneration unit 111 regenerates explanatory variable candidates, which are high-level concepts of the extracted explanatory variables. The conceptual structure editor 112 edits the conceptual structure table stored in the conceptual structure database 114. Details of the conceptual structure table will be described later.

  The accumulated data acquisition unit 119 acquires data stored in the patient information database 121, the examination information database 122, and the prescription information database 123, and stores the acquired data in each table described later included in the integrated database 113. The accumulated data acquisition unit 119 acquires the data, for example, automatically according to an instruction from the input / output terminal 130 or at a time designated in advance by the user. Also, when the hospital information system 120 updates the data of the patient information database 121, the examination information database 122, or the prescription information database 123, the accumulated data acquisition unit 119 acquires the data according to a notification from the hospital database, for example. May be

  The output unit 102 is an interface to which a display device, a printer, and the like are connected, and which outputs the execution result of the program in a format that can be viewed by the operator. The variable optimization system 100 may further include an input unit. The input unit is an interface to which a keyboard, a mouse and the like are connected and which receives an input from an operator.

  The communication unit 104 is a network interface device connected to the network 140 and controlling communication with another device according to a predetermined protocol. The communication unit 104 also includes, for example, a serial interface such as USB.

  The program executed by the control unit 101 is provided to the variable optimization system 100 via removable media (CD-ROM, flash memory, etc.) or via a network, and is stored in the non-volatile auxiliary storage device 116 which is a non-temporary storage medium. Be done. To this end, the variable optimization system 100 may have an interface for reading data from removable media.

  Note that the variable optimization system 100, the input / output terminal 130, and the hospital information system 120 are computer systems configured physically on one computer or on a plurality of logically or physically configured computers. , And may operate on separate threads on the same computer, or may operate on a virtual computer built on multiple physical computer resources.

  Hereinafter, an example of the table which comprises the integrated database 113 is demonstrated. In the present embodiment, the information used by the variable optimization system 100 may be expressed by any data structure without depending on the data structure. Although the following shows an example in which a table stores information, for example, a data structure appropriately selected from a list, a database or a queue can store information.

  The integrated database 113 stores, for example, a patient information table managing basic information and the like for each patient, an examination information table storing examination information for each patient, and a prescription information table storing prescription information for each patient. The information included in the plurality of tables in the present embodiment may be included in one table. For example, the integrated database 113 may manage information using one table integrated with a patient information table, an examination information table, and a prescription information table.

  FIG. 3 is an example of a patient information table. The patient information table 300 indicates, for example, a field 301 storing a patient ID for identifying a patient, a field 302 storing a gender of the patient, and whether the corresponding record is hospitalization data or outpatient data. A field 303 for storing an admission flag, a field 304 for storing an outpatient diagnosis date to a medical institution, a field 305 for storing an admission date to a medical institution, and a discharge date from a medical institution It includes a field 306 and fields 307 to 309 storing diseased disease names.

  The record in which the admission flag is "1" is the data of the inpatient, and "0" is stored in the field 304 indicating the diagnosis date. In each of the field 305 indicating the hospitalization date and the field 306 indicating the hospital discharge date, the hospitalization date and hospital discharge date for the hospitalized patient are stored. In addition, "0" is stored in the field 306 which shows the discharge date of the patient of the undischarged hospital. On the other hand, a record in which the admission flag is "0" is data of an outpatient, and a field 304 indicating a diagnosis date indicates a value corresponding to the outpatient, a field 305 indicating an admission date, and a discharge date. The field 306 stores “0”.

  In each record (patient information record) of the patient information table 300, information on one hospitalization or one diagnosis of one patient is stored. That is, one hospitalization or one outpatient diagnosis of one patient corresponds to one record.

  For example, in the example of FIG. 3, in the patient information record 310, the patient of the patient ID "# 1" has sex "male", hospitalization flag "1", diagnosis date "0", hospitalization date "2014/04 / Discharge date "2014/04/26", disease name 1 is "disease name A", disease name 2 is "disease name B", and disease name 3 is "NULL" indicating that no information is registered. To indicate that That is, the patient information record 310 indicates that the male patient identified by the patient ID “# 1” was hospitalized from April 1, 2014 to April 26, 2014 for the disease name A and the disease name B. . Further, the patient information table 300 may include four or more fields for storing disease names.

  The fields included in the patient information table 300 are not limited to those illustrated in the drawings. Also, the patient information table 300 may not include part of the fields illustrated in the drawings. The same applies to the other tables.

  FIG. 4 is an example of the examination information table 400. The examination information table 400 includes, for example, a field 401 for storing a patient ID, a field 402 for storing information for identifying examination items, a field 403 for storing examination dates, and a field for storing examination results (for example, examination values). And 404.

  In the example of FIG. 4, the value of “test item X” of the patient with the patient ID “# 1” is “41” in “2014/04/1” and “62” in “2014/04/7”. It was "180" on April 13th, "220" on "2014/04/15", "196" on "2014/04/18", and "120" on "2014/04/25" Also, it indicates that the value of the inspection item “inspection item Y” is “38/2014/1”.

  Further, the field 404 for storing the inspection value may store, for example, image information which is an inspection result by an image inspection (for example, a CT image inspection). Also, the field 404 may store, for example, a character string such as information on the patient's symptoms such as "nausea" or "vomiting".

  FIG. 5 is an example of the prescription information table 500. The prescription information table 500 includes, for example, a field 501 for storing a patient ID, a field 502 for storing information (for example, drug name) for identifying a drug, a field 503 for storing a prescription start date of each drug, and And a field 504 for storing a prescription end date.

  In the example shown in FIG. 5, the drug names "Drug A" ranged from "2014/04/01" to "2014/04/07" and "2014/04/08" to "2014 for patients with patient ID" # 1 ". The drug name "Drug B" is prescribed from "2014 04/03" to "2014 Apr 8" and the drug name "Drug D" is from "2014 04/16". It indicates that the drug was prescribed up to "2014/04/20", and the drug name "drug T" was prescribed from "2014 Apr. 14" to "2014 Apr. 18".

  In addition, the prescription information table 500 may further include, for example, a field for storing the prescription usage of each medicine and a field for storing the dosage of each medicine prescription. Hereinafter, an example of the table which comprises the conceptual structure database 114 is demonstrated. The conceptual structure database 114 includes, for example, a conceptual structure table that manages the names of the conceptual structures and the conceptual levels corresponding to the conceptual structures.

  FIG. 2 is an example of a conceptual structure table. The conceptual structure table 200 stores information specifying one or more tree structures and a concept that is a node of each of the one or more tree structures. Each concept stored in the conceptual structure table 200 can be an explanatory variable candidate.

  The conceptual structure table 200 includes, for example, a field 201 for storing information on the name of a conceptual structure, and a field 202 for storing information on a concept belonging to a conceptual level 1 which is the highest hierarchy of the conceptual structure corresponding to the conceptual structure name. , A field 203 for storing information on concepts belonging to concept level 2 which is a hierarchy one level below concept level 1 and a field for storing information on concepts belonging to concept level 3 one layer lower than concept level 2 And 204.

  The conceptual structure table 200 may be determined in advance, or may be created by processing to be described later by the conceptual structure editing unit 112. A concept corresponding to an ancestor node (including a parent node) of a certain concept is a superordinate concept of the certain concept. Also, a concept corresponding to a descendant node (including a child node) of a certain concept is a subordinate concept of the certain concept.

  In the conceptual structure record 200A, the conceptual structure record 200B, and the conceptual structure record 200C, which are the records of the conceptual structure table 200 in FIG. 2, “pathname group 1” belongs to the concept level 1 of the concept structure “pathname concept 1” The “disease name group α” belonging to 2 is a subordinate concept of “disease name group 1”, and “disease name A”, “disease name B” and “disease name C” belonging to concept level 3 are subordinate concepts of “disease name group α” Indicates that there is.

  The concept structure table 200 may further include a field for storing information of concepts belonging to a concept level further lower than concept level 3, ie, concept level 4 or less. That is, the conceptual structure may include four or more levels.

  Also, one concept may be included in a plurality of conceptual structures. In the example of FIG. 2, the conceptual structure record 200D and the conceptual structure record 200E indicate that “drug group θ” is included in both the conceptual structure “drug concept 1” and the conceptual structure “knowledge concept 1”. In addition, in the case where one concept belongs to a plurality of conceptual structures as in the example of FIG. 2, the conceptual level to which the one concept belongs in the conceptual structures included in the plurality of conceptual structures, and the plurality of conceptual structures The concept level to which the one concept belongs in the other concept structures included may be different from each other.

  Hereinafter, an example of the table which comprises the analysis data storage database 115 is demonstrated. The analysis data storage database 115 includes, for example, an explanatory variable candidate table, a data set storage table, and an association degree table.

  FIG. 6 is an example of an explanatory variable candidate table. The explanatory variable candidate table 600 includes, for example, a field 601 for storing the name of an explanatory variable candidate, a field 602 for storing an extraction flag indicating whether the explanatory variable candidate has become an explanatory variable to be extracted, and an explanatory variable candidate. A field 603 for storing a regeneration flag indicating whether or not to be regenerated, and a field for storing variable information of lower concepts to be excluded from the explanatory variable candidate when the explanatory variable candidate becomes a target for regeneration And 604, and a field 605 for storing a non-target flag indicating whether the explanatory variable candidate has become out of target for extraction.

  The initial value stored in the field 602, the field 603, and the field 605 is "0", and the initial value stored in the field 604 is "NULL". When the explanatory variable candidate is determined as the explanatory variable to be extracted, the value of the cell corresponding to the explanatory variable candidate in the field 602 is changed to “1”.

  When the explanatory variable candidate becomes a target of regeneration, the value of the cell corresponding to the explanatory variable candidate in the field 603 is changed to “1”, and the value of the cell corresponding to the explanatory variable candidate in the field 604 is regenerated. Is changed to a value that indicates the explanatory variable that caused the When the explanatory variable candidate is not a target of extraction, the value corresponding to the explanatory variable candidate in the field 605 is changed to “1”.

  In the example of FIG. 6, the explanatory variable candidates "disease name B" and "treatment drug of disease name Z" are determined as the explanatory variables of the extraction target, and in the conceptual structure, the explanatory variable "disease name B" which is the extraction target is subordinate It means that the explanatory variable candidate "disease name group α" to be included is to be regenerated. Further, the example of FIG. 6 means that the explanatory variable “disease name group α” is regenerated as an explanatory variable not including the disease name B.

  Further, in the example of FIG. 6, the explanatory variable candidate “drug group θ” which is a subordinate concept of the explanatory variable “treatment drug with disease name Z” determined to be the extraction target is not the extraction target and is not the extraction target The explanatory variable candidate “drug A”, which is a subordinate concept of the candidate “drug group θ”, is also excluded from the extraction target.

  FIG. 7 is an example of a data set storage table 700 for storing the data set generated by the data set generation unit 109. The data set storage table 700 includes, for example, a field 701 for storing a record identifier, a field 702 for storing values of one or more explanatory variables generated by the explanatory variable generation unit 108, and a target variable generation unit 107. And a field 703 for storing the value of the objective variable. The record identifier in the example of FIG. 7 is the patient ID, and the explanatory variables include “disease name A”, “disease name B”, “disease name C”, “disease name group α”, and “treatment drug for disease name Z”.

  FIG. 8 is an example of a data set storage table 700 for storing the data set after the explanatory variable candidate regeneration process and the explanatory variable extraction process are performed. In FIG. 8, a field 702 stores the values of one or more explanatory variables determined as extraction targets. The explanatory variable “disease name group α (except for the disease name B)” in FIG. 8 indicates a disease name group obtained by removing the disease name B from the disease name group α indicated by the conceptual structure table 200, and the explanatory variable regeneration unit 111 "Is an explanatory variable regenerated from".

  FIG. 9 is an example of a degree-of-association table 900 that stores the degree of association between the explanatory variables calculated by the degree-of-relevance calculation unit 110. The association degree table 900 includes, for example, a field 901 for storing names of unextracted explanatory variable candidates, a field 902 for storing association degrees between unextracted explanatory variable candidates and target variables, and unextracted explanatory variable candidates. And a field 903 for storing the degree of association with the already extracted explanatory variable. When there is no already extracted explanatory variable, the field 903 stores “NULL” as in the example of FIG.

  In FIG. 10, the explanatory variable "disease name B" is extracted by the explanatory variable regenerating unit 111, and the explanatory variable candidate which is a superordinate concept of "disease name B" is "disease name group α" is "disease name group α (except for disease name B) The “relevancy degree” is an example of the relevancy degree table 900 that stores the relevancy degree calculated by the relevancy degree calculation unit 110 after “regenerating”.

  In the example of FIG. 10, the association degree table 900 does not include a record corresponding to the already extracted explanatory variable “disease name B”. Further, the example of the association degree table 900 of FIG. 10 is that the explanatory variable candidate “disease name group α” is changed to the regenerated explanatory variable candidate “disease name group α (except for the disease name B)”. This is different from the example of the 9 relevance table 900. In the example of FIG. 10, the field 903 stores the degree of association between each explanatory variable candidate and the already extracted explanatory variable “disease name B”, which is calculated by the degree of association calculation unit 110.

  Hereinafter, an operation example of the variable optimization system 100 according to the present embodiment will be described. FIG. 11 is a flow chart showing an operation example of the variable optimization system 100 of the present embodiment. First, the data extraction unit 106 reads out the patient data accumulated in the integrated database 113 into the memory 103 (S1101). The patient data is, for example, a generic name of data related to a patient, such as information identifying a patient, information on a prescribed medicine corresponding to the patient, and information on test results corresponding to the patient.

  Hereinafter, an example will be described in which the data extraction unit 106 uses the patient information table 300, the examination information table 400, and the prescription information table 500 as patient data. In step S1101, the data extraction unit 106 may read all patient data included in the integrated database 113, or may read only patient data of a patient designated by the user.

  Next, the data extraction unit 106 reads out, for example, the conceptual structure table 200 from the conceptual structure database 114 as conceptual structure data, and stores the conceptual structure table 200 in the memory 103 (S1102). In step S1102, the data extraction unit 106 may read all data included in the conceptual structure database 114 or may read a part of the conceptual structure. Next, the display screen generation unit 105 displays, for example, an analysis option reception screen on the input / output terminal 130 via the communication unit 104, and receives input of various analysis options (S1103).

  FIG. 14 is an example of the analysis option reception screen. The analysis option reception screen 1400 includes, for example, an area 1401 for selecting a conceptual structure to be used, an area 1402 for the user to specify an explanatory variable included in the extraction target, and a calculation method of the degree of association. An area 1403; an area 1404 for setting an extraction / regeneration method of an explanatory variable; an area 1405 for setting a generation method of an objective variable; an area 1406 for setting the number of explanatory variables to be extracted; A button 1407 for creating a conceptual structure using knowledge data, a button 1408 for editing a conceptual structure, and an analysis execution button 1409 are included.

  The check boxes in the area 1401 are used to select a conceptual structure to be used from a plurality of conceptual structures displayed in the area 1401. The radio button in the area 1402 is used to designate the presence or absence of an explanatory variable which is forcibly included in the regenerated data set regardless of the value of the degree of association. The pull-down list 14021 in the area 1401 is a pull-down list for selecting a conceptual structure including an explanatory variable which the user desires to designate when the radio button of “Yes” is selected.

  The explanatory variable search box 14022 in the area 1401 is a search box for receiving the input of the explanatory variable name of the search target and designating the searched explanatory variable. The check box in the explanatory variable list display area 14023 in the area 1402 is used to determine the explanatory variable to be specified. The example shown in FIG. 14 indicates that “disease name A” and “disease name group β” are designated as explanatory variables for extraction. For example, when an explanatory variable is designated in the area 1402, the area 1402 is configured such that an explanatory variable which is a subordinate concept of the explanatory variable can not be selected.

  The pull-down list 14031 in the area 1403 is a pull-down list for selecting a method of calculating the degree of association. The example of FIG. 14 shows that the degree of association is calculated using the correlation coefficient. The radio button in the area 1403 is used to specify whether to calculate the degree of association with the already-extracted explanatory variable and exclude the explanatory variable whose degree of association with the already-extracted explanatory variable is equal to or higher than the threshold value. Used. The up-down button in the area 14032 included in the area 1403 is used to set the threshold value of the degree of association when performing the exclusion process when the radio button “with” is selected. Also, the area 14032 may directly receive the input of the threshold.

  Each check box in the area 1404 determines whether or not the “subordinate concept of the extracted explanatory variable is not subject” process, and “the upper concept of the extracted explanatory variable is only the regenerated explanatory variable. It is used to specify whether or not to execute "targeted" processing.

  When the check box “Exclude the lower concept of the extracted explanatory variable” is valid, the explanatory variable candidates belonging to the lower concept of the extracted explanatory variable are not extracted. For example, when “treatment drug of disease name group Z” is extracted, “drug group θ” and “drug group ψ”, which are subordinate concepts of “treatment drug of disease name group Z”, and subordinate concepts of “drug group θ” Sub-concepts of “drug A” and “drug B” and “drug group ψ” are excluded from extraction.

  If the check box "The upper concept of the extracted explanatory variable is to be analyzed is only the regenerated explanatory variable" is not valid, the explanatory variable of the upper concept including the extracted explanatory variable and the extracted explanatory variable Both of the explanatory variable candidates regenerated as the explanatory variable candidates not including H are included in the explanatory variable candidates.

  For example, when the explanatory variable "disease name B" is extracted, both of the regenerated explanatory variable "disease name group α (except for disease name B)" and the original explanatory variable "disease name group α" are explanatory variable candidates. It becomes. On the other hand, if the explanatory variable “disease name B” is extracted when the check box is valid, the explanatory variable candidate group does not include the “disease name group α”.

  Radio buttons in the area 1404 are used to select the extraction priority of the explanatory variable. When the radio button corresponding to "the degree of association with the objective variable" is selected, the explanatory variable having a high degree of association with the objective variable is preferentially extracted. When the radio button corresponding to “the degree of association with the already extracted explanatory variable” is selected, the explanatory variable having a low degree of association with the already extracted explanatory variable is preferentially extracted.

  The up-down button in the area 1406 is used to set the number of explanatory variables to be extracted. The area 1406 may directly receive the input of the number of explanatory variables to be extracted. Note that the number of extraction target explanatory variables specified in area 1406 may or may not include the number of explanatory variables specified in area 1402.

  Radio buttons in the area 1405 are used to specify the type of objective variable to be created. Types of objective variables are, for example, binary variables, categorical variables, and quantitative variables. The binary variable is, for example, a variable that takes one of two values, such as a variable representing “0” or “1” whether or not an event for analysis purpose has occurred.

  The categorical variable is, for example, a plurality of categorical values, such as a variable represented by “low”, “medium”, “high” or “1”, “2”, “3”, for the importance of the event for analysis purpose. It is a possible variable. A quantitative variable is a variable that represents quantitative information on an event of analysis purpose by a value such as "1," "10," "100," and the like. For example, a categorical variable is an ordinal scale or a nominal scale, and when the categorical variable is an objective variable, the range of the categorical variable includes three or more values. Also, for example, the quantitative variable is an interval scale or a proportional scale, and when the quantitative variable is a target variable, the range of the quantitative variable includes three or more values.

  When a categorical variable or a quantitative variable is designated as an objective variable, the association degree calculation unit 110 uses, for example, a predetermined function having a value range of 0 or more and 1 or less when calculating the association degree of the explanatory variable and the objective variable. Then, the value of the objective variable may be standardized in the range of “0” to “1”.

  The plurality of pull-downs in the area 1405 are used to perform detailed setting of the target variable to be created after the type of the target variable to be created is specified. The example of FIG. 14 shows an example of creating a target variable indicating "presence or absence of the event XX within 30 days after discharge from hospital" by specifying three pull-downs, and if there is the event XX, the target variable It is shown that “1” is stored as “0”, and “0” is stored as an objective variable in other cases.

  An item selectable in the pull-down 14051 in the area 1405 corresponds to the type of the target variable designated by the radio button in the area 1405. That is, for example, when the radio button corresponding to the binary variable is selected, only the item indicating the binary variable can be selected in the pull-down 14051. Further, only when the radio button corresponding to the binary variable is selected, the input area 14052 in the area 1405 receives the input of the value indicating the state of “when there is an event”.

  A button 1407 is a button for causing the conceptual structure editing unit 112 to create the conceptual structure table 200 using the knowledge data. Medical papers, drug inserts, master information such as drugs and disease names, and various guidelines are all examples of knowledge data. The knowledge data is stored, for example, in the auxiliary storage device 116, the hospital information system 120, or an external database connected to the variable optimization system 100.

  Hereinafter, an example of a conceptual structure editing process will be described using medical literature including a description “the drug Z is used as a treatment for the disease name Z” as knowledge data. The conceptual structure editing unit 112, for example, executes syntactic analysis based on a predetermined rule on the description to specify a noun phrase in the description. The conceptual structure editing unit 112 determines, for example, based on a predetermined rule such as whether or not an expression included in a predetermined dictionary is included in the description, the plurality of noun phrases in the description have a relationship between the upper concept and the lower concept. Determine if it is. The predetermined dictionary and the predetermined rule are stored in advance in, for example, the memory 103 or the auxiliary storage device 116. In the example of the description, the conceptual structure editing unit 112 determines that “treatment of disease name Z” is a superordinate concept of “drug Z”.

  The conceptual structure editing unit 112 determines, for example, a conceptual structural name according to an instruction from the user, the determined conceptual structural name in the field 201, the concept “treatment of the disease name Z” in the field 202, and the concept “disease name The “treatment action of Z” is stored in the field 203, and the concept “drug Z” is stored in the field 204. As a result of the process, in the conceptual structure table 200, "treatment drug for disease name Z" and "drug Z" are generated as subordinate concepts of "treatment for disease name Z" and "treatment drug for disease name Z".

  In addition, for example, when it is determined that “drug Z” belongs to a specific drug group with reference to the concept structure table 200, the conceptual structure editing unit 112 selects between “the therapeutic drug of disease name Z” and “drug Z”. For example, a conceptual structure including the concept “drug group Z” may be created. When an attached document is used as the knowledge data, for example, the conceptual structure editing unit 112 extracts a drug efficacy class name and a drug name from the package insert, and as a subordinate concept of “drug effect class XX”, “drug name XX” Generate

  For example, in the conceptual structure table 200, for example, as in the case where "treatment of disease name Z" is already registered as a subordinate concept of "drug Z", the upper-lower relation between the concept indicated by the knowledge data and the concept If the upper-lower relation of the concept indicated by the structure table 200 is different, the conceptual structure editing unit 112 may edit the conceptual structure table 200 according to the instruction of the user, for example. The conceptual structure table 200 may be edited to be a subordinate concept of “treatment action”.

  It returns to the explanation of FIG. A button 1408 is, for example, a button for causing the display screen generation unit 105 to create a conceptual structure editing screen including the contents of the conceptual structure table 200 and displaying the screen on the input / output terminal 130 or the like. For example, the user performs editing of a concept included in each concept level in the conceptual structure, addition and deletion of a conceptual structure, and the like through the conceptual structure editing screen.

  It returns to the explanation of FIG. When receiving the notification that the analysis execution button 1409 has been selected from the input / output terminal 130, the objective variable generation unit 107, based on the various analysis options accepted in step S1103 and the patient data read out in step S1101, An objective variable is generated (S1104).

  In the area 1405, creating a target variable which is a binary variable indicating "presence or absence of the event XX within 30 days after discharge from the hospital", and the value of the binary variable when there is the event XX is "1". An example of the process of step S1104 when it is specified will be described.

  First, the objective variable generation unit 107 acquires, from the patient information table 300, information on the discharge date of a patient who has a discharge experience. Next, the objective variable generation unit 107 verifies the presence or absence of occurrence of the event XX within 30 days from the discharge date of each patient with reference to the patient information table 300, the examination information table 400, the prescription information table 500, and the like. Thus, the value of the objective variable of each patient is determined.

  Information of patient data referred to by the objective variable generation unit 107 for determining the value of the objective variable in each patient for each objective variable that can be designated in the area 1405 and the method of determining the value of the objective variable are as follows: For example, it is predetermined.

  The objective variable generation unit 107 stores the patient ID of each patient in the field 701. In addition, if the event XX occurs within 30 days from the discharge date of the patient for each patient, the objective variable generation unit 107 sets “1” to the cell corresponding to the patient of the field 703, If the event XX has not occurred within 30 days from the discharge date of the patient, "0" is stored in the cell corresponding to the patient of the field 703. In the creation of the objective variable, information other than the above-described table may be used.

  Next, the explanatory variable generation unit 108 generates explanatory variable candidates (step S1105). Specifically, for example, the explanatory variable generation unit 108 acquires, from the conceptual structure table 200, the concepts included in the conceptual structure selected in the area 1401. The explanatory variable generation unit 108 determines, as explanatory variable candidates, each concept included in the item of the patient data read out in step S1101 among the acquired concepts. The explanatory variable generation unit 108 stores the names of the explanatory variable candidates in the field 601 of the explanatory variable candidate table 600 and the field 901 of the association degree table 900, respectively.

  Next, the data set creation unit 109 creates a data set, and registers the created data set in the data set storage table 700 (S1107). Specifically, for example, the data set creation unit 109 stores the names of the explanatory variable candidate variables in the item name column of the field 702 of the data set storage table. The data set creation unit 109 acquires the value of each patient corresponding to each of the variables from patient data, and stores the value in the corresponding cell of the field 702.

  Next, the degree-of-association calculation unit 110 calculates the degree of association between the objective variable created in step S1104 and each of the explanatory variable candidates created in step S1105 (S1107).

  FIG. 12 is a flowchart showing an example of details of the degree-of-association calculation process by the degree-of-association calculation unit 110, that is, the process of step S1107. The degree-of-association calculation unit 110 reads out the data set storage table 700 and the explanatory variable candidate table 600 onto the memory 103 (S1201). Next, the degree-of-association calculation unit 110 determines whether the degree of association with the target variable has been calculated for all the explanatory variable candidates (S1202).

  When it is determined that there is an explanatory variable candidate for which the degree of association with the objective variable is not calculated (S1202: unprocessed), the association degree calculation unit 110 selects one explanatory variable candidate from the explanatory variable candidates and selects it. The degree of association between the candidate explanatory variables and the objective variable is calculated (S1203).

  The degree-of-association calculation unit 110 corresponds to, for example, a column vector generated by extracting as it is the value of the column corresponding to the selected explanatory variable candidate from the data set storage table 700 and the target variable from the data set storage table 700. The correlation coefficient between the column vector and the column vector generated by extracting the column value as it is is calculated as the degree of association. Further, the degree-of-association calculation unit 110 may calculate, for example, the non-linear correlation coefficient (Maximum information coefficient) or the like of the two column vectors as the degree of association.

  Next, the degree of association calculation unit 110 associates the explanatory variable candidate selected in step S1203 with the degree of association calculated in step S1203 and stores the result in the field 901 and the field 902 of the degree of association table 900 (S1204). ). When the degree of association calculating unit 110 determines that the degree of association with the target variable is calculated for all the explanatory variable candidates (S1202: all ends), the process of FIG. 12 ends, and the process transitions to step S1108 of FIG.

  It returns to the explanation of FIG. The explanatory variable regenerating unit 111 generates the data set, the objective variable generated in step S1104, the explanatory variable candidate generated in step S1105, the degree of association calculated in step S1107, and the conceptual structure read out in step S1102. , And extracts and reproduces explanatory variable candidates (S1108).

  FIG. 13 is a flowchart showing an example of the explanatory variable extraction and regeneration processing by the explanatory variable regeneration unit 111, that is, the details of the processing of step S1108. The explanatory variable regenerating unit 111 reads out the conceptual structure table 200, the data set storage table 700, and the association degree table 900 into the memory 103 (S1301).

  Next, the explanatory variable regenerating unit 111 determines whether or not extraction specification of the explanatory variable of the area 1402 is specified in the reception of various analysis options in step S1103 (S1302). If the explanatory variable regeneration unit 111 determines that there is an extraction variable extraction specification (S1302: Yes), it sets an explanatory variable specified in the area 1402 as an extraction target (S1303), and transitions to step S1308.

  When the explanatory variable regenerating unit 111 determines that the extraction specification of the explanatory variable is not specified (S1302: No), whether or not there is the already extracted explanatory variable, that is, whether the process of step S1308 has been performed at least once. Is determined (S1304). Hereinafter, in the explanatory variable candidate table 600, an explanatory variable candidate in which both the extraction flag and the non-target flag are 0 is referred to as an extracted explanatory variable candidate.

  If the explanatory variable regeneration unit 111 determines that there is no already-extracted explanatory variable (S1304: No), the objective variable calculated in step S1107, for example, is referred to with reference to the explanatory variable candidate table 600 and the association degree table 900. The extraction explanatory variable candidate having the highest degree of association with is determined as the explanatory variable to be extracted (S1305), and the process transitions to step S1308.

  If there are a plurality of extracted explanatory variable candidates having the highest degree of association in step S1305, the explanatory variable regenerating unit 111 extracts, for example, one extracted explanatory variable candidate randomly selected from the plurality of extracted explanatory variable candidates. Decide on. Further, the display screen generation unit 105 may output the plurality of extracted explanatory variable candidates to the input / output terminal 130, and allow the user to select an extraction target.

  If the explanatory variable regeneration unit 111 determines that there is an already extracted explanatory variable (S1304: Yes), it calculates the degree of association between the already extracted explanatory variable and each of the extracted explanatory variable candidates, and stores it in the field 903 ( S1306). When a plurality of already-extracted explanatory variables exist, the explanatory variable regenerating unit 111 may, for example, calculate the sum of the degree of association between each already extracted explanatory variable and the extracted explanatory variable candidate. It is calculated as the degree of association between the variable and the extracted explanatory variable candidate. Also, instead of the sum, statistical values such as mean, median, etc. may be used.

  Next, the explanatory variable regenerating unit 111 determines, as the explanatory variable to be extracted, the extraction explanatory variable candidate whose degree of association calculated in step S1306 is equal to or less than a predetermined threshold and whose degree of association with the objective variable is the largest. (S1307). If it is determined in step S1307 that there is no extracted explanatory variable candidate whose degree of association calculated in step S1306 is equal to or less than the predetermined threshold value, the explanatory variable regenerating unit 111 ends the process, for example.

  In step S1307, if there are a plurality of extracted explanatory variable candidates for which the degree of association calculated in step S1306 is equal to or less than a predetermined threshold and the degree of association is maximum, the explanatory variable regenerating unit 111 One extraction explanatory variable candidate randomly selected from the extraction explanatory variable candidates of is determined as an extraction target. Further, the display screen generation unit 105 may output the plurality of extracted explanatory variable candidates to the input / output terminal 130, and allow the user to select an extraction target.

  When the radio button corresponding to “none” in the area 1403 is selected, the explanatory variable regenerating unit 111 does not calculate the degree of association in step S1306, and the degree of association with the objective variable in step S1307. The extraction explanatory variable candidate which is the largest is determined as the explanatory variable to be extracted.

  Next, the explanatory variable regeneration unit 111 changes the value of the extraction flag of the explanatory variable determined as the extraction target in the explanatory variable candidate table 600 to “1”, that is, extracts the explanatory variable determined as the extraction target It excludes from a candidate (S1308). At this time, the explanatory variable regenerating unit 111 may exclude the information of the explanatory variable determined as the extraction target from the association degree table 900. Further, the explanatory variable regenerating unit 111 may store information indicating the extracted order of the explanatory variables determined as the extraction target in the explanatory variable candidate table 600.

  Next, the explanatory variable regenerating unit 111 executes the process of regenerating explanatory variable candidates (S1309). Specifically, the explanatory variable regenerating unit 111 refers to, for example, the conceptual structure table 200, and determines all explanatory variable candidates that are high-level concepts of the explanatory variable to be extracted as the explanatory variable candidate to be regenerated. The value of the regeneration flag corresponding to the explanatory variable candidate to be regenerated in the explanatory variable candidate table 600 is changed to 1.

  The explanatory variable regeneration unit 111 regenerates each explanatory variable candidate to be regenerated as an explanatory variable which does not include the explanatory variable to be extracted. Specifically, the explanatory variable regenerating unit 111 changes the name of the explanatory variable candidate to be regenerated in the field 901 of the association degree table 900 to the name of the regenerated explanatory variable. Further, the explanatory variable regenerating unit 111 stores the name of the explanatory variable to be extracted in the cell corresponding to the explanatory variable candidate to be regenerated in the field 604 of the explanatory variable candidate table 600.

  Further, the explanatory variable regenerating unit 111 adds a column of explanatory variable candidates after regeneration to the field 702 of the data set storage table 700. The explanatory variable regenerating unit 111 is, for example, a subordinate concept of the explanatory variable to be regenerated with reference to the conceptual structure indicated by the conceptual structure table 200, and is a sibling node of the explanatory variable candidate determined to be the extraction target. , Identify candidate explanatory variables. The explanatory variable regenerating unit 111 determines the value of the added column based on the value of the specified explanatory variable candidate stored in the field 702, for example.

  When the check box “The upper concept of the extracted explanatory variable is to be analyzed is only the explanatory variable that is regenerated” in the area 1404 is not valid, the explanatory variable regenerating unit 111 generates the explanatory variable candidate table 600. In the field 604, the values of the regeneration flag and the exclusion variable corresponding to the explanatory variable candidate to be regenerated are not changed. At this time, the explanatory variable regenerating unit 111 adds a record corresponding to the explanatory variable candidate to be regenerated to the explanatory variable candidate table 600, and stores the initial value in the fields 602 to 605 of the record.

  Hereinafter, an example of the process of regenerating explanatory variable candidates in the case where “disease name B” is determined as the extraction target will be described. The explanatory variable regeneration unit 111 refers to the conceptual structure table 200, and determines the explanatory variable candidates “disease name group α” and “disease name group 1”, which are high-level concepts of “disease name B”, as explanatory variable candidates to be regenerated. . The explanatory variable regenerating unit 111 excludes the explanatory variable candidate "disease name group a (except for the disease name B)" obtained by removing "disease name B" from the "disease name group a", and removes "disease name B" from "disease name group 1". The explanatory variable candidate "disease name group 1 (except for disease name B)" is regenerated.

  The explanatory variable regeneration unit 111 sets “disease name group α” to “disease name group α (except for disease name B)” in the field 901 of the degree of association table 900 and “disease name group 1” to “disease name group α (disease name B Change to "Excluding." In addition, the explanatory variable regenerating unit 111 stores “disease name B”, which is an explanatory variable to be extracted, in each cell of the field 604 corresponding to “disease name group α” and “disease name group 1” of the explanatory variable candidate table 600. Do.

  Further, the explanatory variable regenerating unit 111 adds columns of “disease name group α (except for disease name B)” and “disease name group 1 (except for disease name B)” to the field 702 of the data set storage table 700. The explanatory variable regeneration unit 111 refers to the conceptual structure table 200, and identifies “disease name A” and “disease name C” that are subordinate concepts of “disease name group α” and are sibling nodes of “disease name B”. Do.

  The explanatory variable regenerating unit 111 acquires the values of “disease name A” and “disease name C” of each patient from the data set storage table 700. The explanatory variable regeneration unit 111 determines that the value of the “patient name group α (excluding the patient name B)” of the patient is 0 when the values of the “patient name A” and the “patient name B” of the patient are all 0. Otherwise, decide to 1. The explanatory variable regeneration unit 111 determines the value of “disease name group 1 (except for disease name B)” of each patient in the same manner.

  Next, the explanatory variable regenerating unit 111 calculates the degree of association between each of the explanatory variable candidates regenerated in step S1309 and the target variable, and the field corresponding to each of the regenerated explanatory variable candidates of the relevance degree table 900. It registers in the cell of 902 (S1310).

  Next, the explanatory variable regenerating unit 111 refers to the conceptual structure table 200 to identify all explanatory variable candidates of the lower concept of the explanatory variable to be extracted, and excludes the identified explanatory variable candidates from the extraction target ( S1311). Specifically, for example, the explanatory variable regenerating unit 111 changes the non-target flag of the explanatory variable candidate specified in the explanatory variable candidate table 600 to “1”. If the check box of “do not target lower concepts of extracted explanatory variables” in the area 1404 is not valid, the explanatory variable regenerating unit 111 does not execute the process of step S1311.

  Next, the explanatory variable regenerating unit 111 determines whether there is an explanatory variable candidate whose non-target flag is “0” in the explanatory variable candidate table 600 (S1312). If the explanatory variable regenerating unit 111 determines that the explanatory variable candidate having the non-target flag of “0” does not exist (S1312: No), the process of FIG. 13 ends, and the process transitions to step S1109. If the explanatory variable regenerating unit 111 determines that there is an explanatory variable candidate whose non-target flag is “0” (Yes in 1312), the process transitions to step S1313.

  The explanatory variable regenerating unit 111 determines whether the number of already extracted explanatory variables is less than the set value of the area 1406 (S1313). If the explanatory variable regeneration unit 111 determines that the number of already-extracted explanatory variables is equal to or larger than the set value (S1313: No), the process of FIG. 13 ends, and the process transitions to step S1109. If the explanatory variable regeneration unit 111 determines that the number of already-extracted explanatory variables is less than the set value (S1313: Yes), the process returns to step S1304.

  When a plurality of explanatory variables are specified as extraction targets in the area 1402, for example, the explanatory variable regenerating unit 111 sequentially performs the processing of step S1303 and steps S1308 to S1313 for each of the plurality of explanatory variables. Run.

  It returns to the explanation of FIG. The data set creation unit 109 re-creates the data set based on the explanatory variable candidate table 600, and registers the re-created data set in the data set storage table 700 (S1109). Specifically, for example, the data set creation unit 109 creates a data set storage table 700 in which the field 702 is information of an explanatory variable whose extraction flag is 1 in the explanatory variable candidate table 600.

  Next, the display screen generation unit 105 generates a display screen displaying the analysis processing result based on, for example, the information of the conceptual structure table 200 and the explanatory variable candidate table 600 (S1110). Details of the display screen will be described later. The display screen generation unit 105 displays the screen generated in step S1110 on the input / output terminal 130 via the output unit 102 (S1111).

  The data set creation unit 109 is a variable including the objective variable and the extracted explanatory variables included in the data set re-created in step S1109, and the degree of association between each of the extracted explanatory variables and the objective variable. The set may be stored, for example, in the memory 103 or the auxiliary storage device 116.

  When the same objective variable as the objective variable included in the variable set is designated in step S1104 in the process of FIG. 11 from the next time onwards, the explanatory variable regenerating unit 111 determines that the association degree in the variable set is a predetermined threshold, for example. The explanatory variable which is the above may be included in the explanatory variable of extraction specification in step S1303.

  That is, the explanatory variable regenerating unit 111 determines the explanatory variable whose degree of association in the variable set is equal to or more than the predetermined threshold as the explanatory variable of the extraction target without newly calculating the degree of association. If the explanatory variable whose association degree in the variable set is equal to or higher than the predetermined threshold includes the regenerated explanatory variable, the extraction flag of the original explanatory variable candidate corresponding to the regenerated explanatory variable is generated in step S1308. Change to 1

  Thus, in the process of extracting again the explanatory variable corresponding to the same objective variable, the explanatory variable regeneration unit 111 can extract an explanatory variable with high explanatory power for the objective variable, and execute the process at high speed. can do.

  FIG. 15 shows an example of the analysis processing result display screen displayed on the input / output terminal 130 in step S1111. The display screen 1500 includes, for example, an area 1501 displaying a list of extracted explanatory variables, and an area 1502 displaying a relationship diagram of the extracted explanatory variables and the conceptual structure.

  The area 1501 displays, for example, the extracted explanatory variables (that is, the explanatory variables whose extraction flag is 1), the extraction order of each of the explanatory variables, and the degree of association between each of the explanatory variables and the objective variable. In the example of FIG. 15, the area 1501 displays “disease name B”, “treatment drug with disease name Z”, “disease name group α (except for disease name B)” in the extracted order as the explanatory variable extracted ing.

  The area 1502 displays, for example, a conceptual structure including the extracted explanatory variables, the explanatory variables regenerated because the explanatory variables are extracted, and the explanatory variables to be excluded. In the area 1502, dotted lines indicate the explanatory variables before regeneration corresponding to the regenerated explanatory variables, and the explanatory variables belonging to the same conceptual structure as the explanatory variables before the regeneration and belong to the adjacent hierarchy. tie. Further, in the area 1502, solid lines connect the explanatory variables belonging to the same conceptual structure and belonging to adjacent layers.

  The user refers to the display screen 1500 to find out at which position of the conceptual structure the extracted explanatory variable being displayed is an explanatory variable, which explanatory variable has been excluded, and which explanatory variable is reproduced. It can be easily grasped that it was made. As a result, the user can easily verify whether extraction and regeneration of explanatory variables are appropriate.

  As described above, the variable optimization system 100 can extract an explanatory variable with high explanatory power for the objective variable by determining an explanatory variable having a high degree of association with the objective variable as an extraction target. Further, the variable optimization system 100 regenerates the explanatory variable candidate of the upper concept to be extracted as an explanatory variable candidate not including the explanatory variable to be extracted. By the process, the variable optimization system 100 can automatically generate a new concept, and can use the generated new concept as an explanatory variable candidate.

  In addition, since the regenerated explanatory variable candidate is a concept that does not include the explanatory variable to be extracted, even when the variable optimization system 100 selects the regenerated explanatory variable candidate as the extraction target, It is possible to suppress the duplication of the meaning of the extracted explanatory variables. Also, the regenerated explanatory variable candidate is a variable for which the user can easily understand the medical meaning, such as a disease name group not including a specific disease name.

  In addition, the variable optimization system 100 can exclude the explanatory variables having the same meaning as well as the efficiency of the extraction processing can be performed by excluding the lower concepts of the explanatory variables to be extracted. it can. Consequently, the variable optimization system 100 can create a data set in which different explanatory variables having the same meaning can be excluded from the data set, that is, the medical meaning can be easily grasped.

  In addition, the variable optimization system 100 can extract explanatory variables that are less likely to cause multicollinearity by excluding explanatory variables that have a high degree of association with already-extracted explanatory variables, so that the target event can be obtained. Data sets suitable for prediction and verification can be generated. That is, the variable optimization system 100 can select explanatory variables to improve the prediction accuracy and verification accuracy of the target event.

  The present invention is not limited to the embodiments described above, but includes various modifications. For example, the embodiments described above are described in detail in order to explain the present invention in an easy-to-understand manner, and are not necessarily limited to those having all the configurations described. In addition, with respect to a part of the configuration of each embodiment, it is possible to add, delete, and replace other configurations.

  Each of the above configurations, functions, processing units, processing means, etc. may be realized by hardware, for example, by designing part or all of them with an integrated circuit. Further, each configuration, function, and the like described above may be realized by software by the processor interpreting and executing a program that realizes each function. Information such as programs, tables, and files that realize each function is memory, hard disk drive, storage device such as SSD (Solid State Drive), or computer readable non-temporary data such as IC card, SD card, DVD, etc. It can be stored in a storage medium.

  Also, the drawings show control lines and information lines that are considered to be necessary to explain the embodiment, and not necessarily all control lines and information lines included in an actual product to which the present invention is applied. Not necessarily. In practice, almost all configurations may be considered to be mutually connected.

  Reference Signs List 100 variable optimization system, 101 control unit, 102 output unit, 103 memory, 104 communication unit, 105 display screen generation unit, 106 data extraction unit, 107 accumulated data acquisition unit, 108 explanatory variable generation unit, 109 data set generation unit, 110 Relevance degree calculation unit, 111 explanatory variable regeneration unit, 112 conceptual structure editing unit, 113 integrated database, 114 conceptual structure database, 115 analysis database, 116 auxiliary storage unit, 119 accumulated data acquisition unit

Claims (15)

A system for determining an explanatory variable group corresponding to a target variable,
Including processor and storage,
The storage device is
Conceptual structure information indicating one or more tree structures having one or more layers, and a correspondence indicating that each of a plurality of explanatory variable candidates is any one of the nodes of the one or more tree structures;
Holding measurement values of the plurality of explanatory variable candidates and measurement value information indicating the measurement values of the target variable;
Each of the one or more tree-structured nodes is a concept,
In the one or more tree structures, a parent node is a superordinate concept of a child node,
The processor repeats a determination process of determining explanatory variables to be included in the explanatory variable group from the explanatory variable candidate group,
The explanatory variable candidate group in the first determination process is the plurality of explanatory variable candidates,
In the determination process, the processor
Based on the measurement value information, a first degree of association between each explanatory variable candidate of the explanatory variable candidate group and the objective variable is calculated;
An explanatory variable candidate to be included in the explanatory variable group is selected from the explanatory variable candidate group based on each of the first degree of association,
The selected explanatory variable candidate is included in the explanatory variable group,
Excluding the selected explanatory variable candidate from the explanatory variable candidate group;
Identifying explanatory variable candidates, which are high-level concepts of the selected explanatory variable candidates, from the explanatory variable candidate group with reference to the conceptual structure information;
A change process is performed to change each of the identified explanatory variable candidates into explanatory variable candidates excluding the selected explanatory variable candidate from the lower concept indicated by the conceptual structure information of the explanatory variable candidate.
The measurement value of the sibling node indicated by the conceptual structure information of the selected explanatory variable candidate is acquired from the measurement value information,
Based on the acquired measurement value, the measurement value of each explanatory variable candidate after the change in the change process is calculated,
The system which includes the calculated measured value in the measured value information. The system according to claim 1, wherein
The processor determines whether or not the second and subsequent determination processes include:
Based on the measurement value information, a second degree of association between the explanatory variable group and each explanatory variable candidate of the explanatory variable candidate group is calculated;
The system which selects the explanatory variable candidate included in the said explanatory variable group from the explanatory variable candidate of the said explanatory variable candidate group whose said 2nd related degree is below a threshold value based on each said 1st degree of association. The system according to claim 1, wherein
In the determination process, the processor
Identifying the explanatory variable candidate which is a subordinate concept of the selected explanatory variable candidate from the explanatory variable candidate group with reference to the conceptual structure information;
The system which excludes the explanatory variable candidate which is the specified lower concept from the explanatory variable candidate group. The system according to claim 1, wherein
The system according to claim 1, wherein the first explanatory variable candidate included in the plurality of explanatory variable candidates belongs to a plurality of tree structures. The system according to claim 1, wherein
The storage device holds knowledge information indicating a first concept and a second concept which is a superordinate concept of the first concept,
The processor is
When all the tree structures included in the conceptual structure information do not include the first concept and the second concept, a tree structure in which the second concept is a parent node of the first concept is used as the conceptual structure information. Including
When the first tree structure included in the conceptual structure information includes the first concept and a third concept which is a superordinate concept of the first concept, one of the second concept or the third concept is the other parent. A system including a tree structure that is a node and the other is a parent node of the first concept in the conceptual structure information. The system according to claim 1, wherein
The processor is
Storing, in the storage device, variable information indicating an explanatory variable group determined by repeating the determination process, and the first relevance of each of the explanatory variables of the determined explanatory variable group and the objective variable;
Implement a redetermination process to redetermine the explanatory variable group corresponding to the objective variable,
The re-determination process consists of repeating the determination process,
In the first determination process in the redetermination process,
The explanatory variable indicated by the variable information, which is the explanatory variable indicated by the variable information, is selected as an explanatory variable candidate to be included in the explanatory variable group, based on each of the first degrees of association indicated by the variable information.
Identifying explanatory variable candidates before change by the change process of the selected explanatory variable with reference to the conceptual structure information;
The system which excludes the specified explanatory variable candidate before change from the explanatory variable candidate group. The system according to claim 1, wherein
The processor, in the determination process, selects an explanatory variable candidate of the explanatory variable candidate group having the largest first degree of association as an explanatory variable candidate to be included in the explanatory variable group. The system according to claim 1, wherein
Connected to the display,
The processor is
The system which outputs the said explanatory variable group and the explanatory variable candidate after the change in the said change process to the said display apparatus. A method in which a system determines an explanatory variable group corresponding to a target variable,
The system
Conceptual structure information indicating one or more tree structures having one or more layers, and a correspondence indicating that each of a plurality of explanatory variable candidates is any one of the nodes of the one or more tree structures;
Holding measurement values of the plurality of explanatory variable candidates and measurement value information indicating the measurement values of the target variable;
Each of the one or more tree-structured nodes is a concept,
In the one or more tree structures, a parent node is a superordinate concept of a child node,
The method repeats the determination process in which the system determines an explanatory variable to be included in the explanatory variable group from the explanatory variable candidate group,
The explanatory variable candidate group in the first determination process is the plurality of explanatory variable candidates,
In the method, the system determines whether
Based on the measurement value information, a first degree of association between each explanatory variable candidate of the explanatory variable candidate group and the objective variable is calculated;
An explanatory variable candidate to be included in the explanatory variable group is selected from the explanatory variable candidate group based on each of the first degree of association,
The selected explanatory variable candidate is included in the explanatory variable group,
Excluding the selected explanatory variable candidate from the explanatory variable candidate group;
Identifying explanatory variable candidates, which are high-level concepts of the selected explanatory variable candidates, from the explanatory variable candidate group with reference to the conceptual structure information;
A change process is performed to change each of the identified explanatory variable candidates into explanatory variable candidates excluding the selected explanatory variable candidate from the lower concept indicated by the conceptual structure information of the explanatory variable candidate.
The measurement value of the sibling node indicated by the conceptual structure information of the selected explanatory variable candidate is acquired from the measurement value information,
Based on the acquired measurement value, the measurement value of each explanatory variable candidate after the change in the change process is calculated,
The method of including the calculated measured value in the measured value information. The method according to claim 9, wherein
In the second and subsequent determination processes, the system
Based on the measurement value information, a second degree of association between the explanatory variable group and each explanatory variable candidate of the explanatory variable candidate group is calculated;
A method of selecting an explanatory variable candidate to be included in the explanatory variable group based on each of the first degree of association, from explanatory variable candidates of the explanatory variable candidate group whose second degree of association is equal to or less than a threshold. The method according to claim 9, wherein
In the determination process, the system
Identifying the explanatory variable candidate which is a subordinate concept of the selected explanatory variable candidate from the explanatory variable candidate group with reference to the conceptual structure information;
The method of excluding the explanatory variable candidate which is the specified lower concept from the explanatory variable candidate group. The method according to claim 9, wherein
The first explanatory variable candidate included in the plurality of explanatory variable candidates belongs to a plurality of tree structures. The method according to claim 9, wherein
The system holds knowledge information indicating a first concept and a second concept which is a superordinate concept of the first concept,
The method comprises:
When all the tree structures included in the conceptual structure information do not include the first concept and the second concept, a tree structure in which the second concept is a parent node of the first concept is used as the conceptual structure information. Including
When the first tree structure included in the conceptual structure information includes the first concept and a third concept which is a superordinate concept of the first concept, one of the second concept or the third concept is the other parent. A method, comprising: a tree structure which is a node and the other is a parent node of the first concept in the conceptual structure information. The method according to claim 9, wherein
The system
Storing, in the system, variable information indicating an explanatory variable group determined by repeating the determination process and the first relevance of each of the explanatory variables of the determined explanatory variable group and the objective variable;
Implement a redetermination process to redetermine the explanatory variable group corresponding to the objective variable,
The re-determination process consists of repeating the determination process,
In the first determination process in the redetermination process,
The explanatory variable indicated by the variable information, which is the explanatory variable indicated by the variable information, is selected as an explanatory variable candidate to be included in the explanatory variable group, based on each of the first degrees of association indicated by the variable information.
Identifying explanatory variable candidates before change by the change process of the selected explanatory variable with reference to the conceptual structure information;
The method for excluding the specified explanatory variable candidate before change from the explanatory variable candidate group. The method according to claim 9, wherein
The method is such that, in the determination process, the system selects an explanatory variable candidate of the explanatory variable candidate group having the largest first degree of association as an explanatory variable candidate to be included in the explanatory variable group.

Images