Treatment selection support device, treatment selection support method, and treatment selection support program

Abstract

To provide a treatment selection support device which improves the reliability of treatment choices.SOLUTION: A treatment selection support device includes a processor for executing a program and a storage unit for storing the program, and implements verification processing for verifying validity of a treatment choice which includes a treatment method related to an injury and disease type of a patient on the basis of the treatment choice and an analysis result based on treatment results for the injury and disease type in other patients and estimated treatment effects, and output processing for outputting a verification result by the verification processing.SELECTED DRAWING: Figure 8

Description

[Technical Field]

[0001] The present invention relates to a treatment selection support device, a treatment selection support method, and a treatment selection support program that support the selection of a treatment. [Background technology]

[0002] The medical application of large-scale language models has been attracting attention. However, large-scale language models can present false information, a phenomenon known as hallucination. To prevent hallucination, methods such as training only reliable information, establishing guardrails, and using reinforcement learning from human feedback (RLHF) under the supervision of medical experts are being implemented, but these methods require a lot of effort and cost.

[0003] Patent Literature 1 discloses a decision support system for medical diagnosis and treatment. The decision support system includes software modules embodied on a computer-readable medium, the software modules including an input / output module and a question-answering module. The decision support system receives patient case information using the input / output module, generates a medical diagnosis or treatment query based on the patient case information, and generates multiple medical diagnosis or treatment answers to the query using the question-answering module. The decision support system also calculates multiple medical evidence dimension values ​​from medical evidence sources for each answer using the question-answering module, and calculates a confidence value corresponding to each answer based on the numerical values ​​of each evidence dimension using the question-answering module. The decision support system further outputs the medical diagnosis or treatment answers, corresponding confidence values, and the numerical values ​​of each medical evidence dimension for one or more selected medical diagnosis or treatment answers using the input / output module. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] US Patent Publication No. 2012-0078062 Summary of the Invention [Problem to be solved by the invention]

[0005] In medical applications of large-scale language models, particularly in support of treatment selection, there is a problem in ensuring the reliability of the output from the large-scale language model. For example, in Patent Document 1, if the information about diagnosis or treatment generated from a query is incorrect, it is not possible to determine whether the information is correct. On the other hand, it is difficult to set the reference range of past data based on medical records alone, making it impossible to respond to revisions to guidelines.

[0006] The present invention aims to improve the reliability of treatment options. [Means for solving the problem]

[0007] A treatment selection support device according to one aspect of the invention disclosed in the present application is a treatment selection support device having a processor that executes a program and a storage device that stores the program, wherein the processor executes a verification process that verifies the validity of treatment options based on treatment options including treatment methods for a patient's illness or injury and analysis results based on treatment records and estimated treatment effects for the illness or injury of other patients, and an output process that outputs the verification results obtained by the verification process. [Effects of the Invention]

[0008] According to a representative embodiment of the present invention, it is possible to improve the reliability of treatment options. Objects, configurations, and effects other than those described above will become clear from the following description of the examples. [Brief explanation of the drawings]

[0009] [Figure 1] FIG. 1 is a block diagram illustrating an example of a hardware configuration of a treatment selection support device according to a first embodiment. [Figure 2]FIG. 2 is an explanatory diagram showing an example of the patient information DB. [Figure 3] FIG. 3 is an explanatory diagram illustrating an example of the guideline DB. [Figure 4] FIG. 4 is an explanatory diagram showing an example of the formulary DB. [Figure 5] FIG. 5 is an explanatory diagram showing an example of the medical care DB. [Figure 6] FIG. 6 is an explanatory diagram illustrating an example of the analysis viewpoint DB. [Figure 7] FIG. 7 is an explanatory diagram showing an example of a medical dictionary. [Figure 8] FIG. 8 is a block diagram of an example of a functional configuration of the treatment selection supporting device according to the first embodiment. [Figure 9] FIG. 9 is a flowchart illustrating an example of an analysis control process performed by the analysis control unit according to the first embodiment. [Figure 10] FIG. 10 is an explanatory diagram illustrating an example of a search query. [Figure 11] FIG. 11 is an explanatory diagram illustrating an example of an analysis request. [Figure 12] FIG. 12 is a flowchart illustrating an example of an extraction processing procedure performed by the extraction unit. [Figure 13] FIG. 13 is a flowchart illustrating an example of a verification process procedure performed by the verification unit. [Figure 14] FIG. 14 is an explanatory diagram illustrating an example of the reliability score table. [Figure 15] FIG. 15 is an explanatory diagram illustrating an example of the case number score table. [Figure 16] FIG. 16 is an explanatory diagram showing an example of the estimated therapeutic effect score table. [Figure 17] FIG. 17 is an explanatory diagram illustrating an example of the output information determination table. [Figure 18] FIG. 18 is an explanatory diagram showing an example of the input screen. [Figure 19] FIG. 19 is an explanatory diagram showing an example of the output screen. [Figure 20] FIG. 20 is a block diagram of an example of a functional configuration of a treatment selection supporting device according to the second embodiment. [Figure 21] FIG. 21 is a flowchart illustrating an example of analysis control processing by the analysis control unit according to the second embodiment. DETAILED DESCRIPTION OF THE INVENTION [Example]

[0010] <Figure 1. Example of hardware configuration for a treatment selection support device> FIG. 1 is a block diagram showing an example of a hardware configuration of a treatment selection assisting device according to a first embodiment. The treatment selection assisting device 100 includes a processor 101, a storage device 102, an input device 103, an output device 104, and a communication interface (communication IF) 105. The processor 101, the storage device 102, the input device 103, the output device 104, and the communication IF 105 are connected via a bus 106. The processor 101 controls the treatment selection assisting device 100. The storage device 102 serves as a working area for the processor 101. The storage device 102 is a non-transitory or temporary recording medium that stores various programs and data. Examples of the storage device 102 include a read-only memory (ROM), a random access memory (RAM), a hard disk drive (HDD), and a flash memory. The input device 103 inputs data. Examples of the input device 103 include a keyboard, a mouse, a touch panel, a numeric keypad, a scanner, a microphone, and a sensor. The output device 104 outputs data. Examples of the output device 104 include a display, a printer, and a speaker. The communication IF 105 connects to a network and transmits and receives data.

[0011] <Figure 2 Patient information DB> FIG. 2 is an explanatory diagram showing an example of a patient information DB. The patient information DB 200 has the following fields: patient ID 201, basic information 202, and pathological condition information 203. The patient ID 201 is identification information that uniquely identifies a patient. A combination of values ​​in each field in the same row forms an entry that indicates the patient information of one patient. In FIG. 2, as an example, P1, ..., Pi, ... Pn (n is an integer equal to or greater than 1, and i is an integer that satisfies 1≦i≦n) are used as the patient ID 201. Hereinafter, a patient whose patient ID 201 value is Pi will be referred to as patient Pi.

[0012] The basic information 202 is basic information about the patient Pi. In Fig. 2, B1, ..., Bi, ..., Bn are taken as the basic information 202 of the patients P1 to Pn, for example. The basic information 202 includes a name 221, an address 222, an age 223, and a sex 224, for example.

[0013] The pathological condition information 203 is information indicating the pathological condition of patient Pi. In FIG. 2, B1, ..., Ci, ..., Cn are assumed to be the pathological condition information 203 of patients P1 to Pn, for example. The pathological condition information 203 includes, for example, a consultation date 231, a chief complaint 232, a medical history 233, test results 234, a medical impression (findings) 235, and a prescription 236. The chief complaint 232 is the main symptom that patient Pi complains to the doctor. The medical history 233 includes, for example, a type of injury or illness 241, a progression 242, a treatment history 243, and a name of a concomitant injury or illness 244. The test results 234 include, for example, a heart rate 251, a blood pressure 252, and a body temperature 253.

[0014] <Figure 3 Guideline DB> 3 is an explanatory diagram showing an example of a guideline DB. The guideline DB 300 is a database that stores a group of text data describing treatment methods for illnesses and injuries, and has the following fields: guideline ID 301, injury / illness type 302, and guideline 303. The guideline ID 301 is identification information that uniquely identifies the guideline 303. The guideline 303 is text data that presents a treatment method that is considered optimal for the injury / illness type 302 to support the decision-making of a doctor and patient Pi.

[0015] <Figure 4 Formulary DB> FIG. 4 is an explanatory diagram showing an example of a formulary DB 400. Like the guideline DB 300, the formulary DB 400 is a database that stores a group of text data describing treatment methods for illnesses and injuries, and has the following fields: formulary ID 401, treatment method 402, recommendation level 403, indication 404, and contraindication 405. A combination of values ​​in each field on the same line forms an entry that defines one piece of formulary data. A formulary is text data of operational guidelines that indicates recommended treatment methods (mainly drug treatments) based on efficacy, safety, cost-effectiveness, etc.

[0016] <Figure 5 Medical DB> 5 is an explanatory diagram showing an example of a medical DB. The medical DB 500 has fields such as a patient ID 211, an injury / illness type 241, a treatment history 243, a test value 501, an adverse event 502, a treatment effect 503, a treatment date 504, a most recent medical treatment date 505, a gene mutation 506, a performance status 507, an age 508, a gender 509, and a test type 511. A combination of values ​​in each field in the same row forms an entry that indicates the medical information of one patient Pi.

[0017] The gene mutation 506 indicates the mutated gene. The performance status 507 is an index of the overall condition of the patient Pi, indicating the degree of limitation in the daily life of the patient Pi. For example, the index is classified into 0 to 4 as follows.

[0018] 0: Able to function without any problems. Can carry out daily activities without any restrictions, just like before the onset of symptoms. 1: Physically strenuous activity is limited, but walking is possible and light work and seated work can be performed. Examples: light housework, office work 2: Able to walk and care for themselves, but unable to work. Spend more than 50% of the day out of bed. 3: Limited self-care capabilities. Spend more than 50% of the day in bed or a chair. 4: Totally immobile. Cannot care for themselves at all. Confined entirely to bed or a chair.

[0019] In addition, each field item related to medical treatment, such as type of injury or illness 241, treatment history 243, test values ​​501, adverse events 502, treatment effects 503, treatment dates 504, latest medical treatment dates 505, ..., genetic mutations 506, and performance status 507, age 508, gender 509, and test type 511, is referred to as a data schema 510.

[0020] <Figure 6 Analysis perspective DB> 6 is an explanatory diagram showing an example of an analytical viewpoint DB. The analytical viewpoint DB 600 is a database that stores analytical viewpoints that serve as indicators of the perspective from which an analysis will be performed, and has the following fields: analytical viewpoint ID 601, injury / illness type 302, objective variable 603, and background adjustment factor 604. A combination of values ​​in each field on the same line becomes an entry that defines one analytical viewpoint.

[0021] The analysis viewpoint ID 601 is identification information that uniquely identifies the analysis viewpoint. The objective variable 603 is information that is desired to be obtained about the injury / illness type 302. The background adjustment factor 604 is an adjustment variable for the objective variable 603.

[0022] For example, the analysis perspective 610, which is an entry with the analysis perspective ID 601 of "V1," shows what the objective variables 603, "survival period" and "recurrence rate," will be when the analysis of the disease type 302, "gastric cancer," is performed using "age (generation)," "sex," "stage of progression," "performance status," and "previous treatment method" as background adjustment factors 604.

[0023] <Figure 7 Medical Dictionary> 7 is an explanatory diagram showing an example of a medical dictionary. A medical dictionary 700 has fields including a medical term 701, a category 702, and a synonym 703. The category 702 is a division to which the medical term belongs. The synonym 703 is another medical term 701 that has a similar meaning to the medical term 701.

[0024] The databases shown in Figures 2 to 7 may be stored in the storage device 102 within the treatment selection support device 100, or may be stored in the storage device of another computer that the treatment selection support device 100 can access via a network such as the Internet, a LAN (Local Area Network), or a WAN (Wide Area Network) via the communication IF 105.

[0025] <Figure 8: Example of functional configuration of treatment selection support device 100> 8 is a block diagram illustrating an example of a functional configuration of the treatment selection support device 100 according to the first embodiment. The treatment selection support device 100 includes an input unit 801, an analysis control unit 802, an extraction unit 803, an analysis unit 804, a verification unit 805, and an output unit 806. Specifically, the input unit 801, the analysis control unit 802, the extraction unit 803, the analysis unit 804, the verification unit 805, and the output unit 806 are realized by, for example, causing the processor 101 to execute a program stored in the storage device 102 illustrated in FIG.

[0026] (input unit 801) The input unit 801 inputs patient information (basic information 202, pathological condition information 203) of a patient Pi from the patient information DB 200. For example, when the user operates the input device 103 to input the patient ID 201 of the patient Pi, the patient information of the patient Pi is read from the patient information DB 200.

[0027] (Analysis control unit 802) The analysis control unit 802 generates a search query 821 and an analysis request 822. The analysis request 822 is generated using the guideline DB 300, the data schema 510, and the medical dictionary 700. Details of the generation of the search query 821 and the analysis request 822 will be described later with reference to FIG. 9. The analysis control unit 802 may be implemented using a language model. The analysis control unit 802 may also be implemented in a computer outside the treatment selection support device 100.

[0028] A language model is an interactive language generation AI that is trained to generate answers to queries. When a query called a prompt is input, the language model generates an answer and sends it to the analysis control unit 802. The language model includes, for example, a large-scale language model such as ChatGPT.

[0029] (Extraction part 803) The extraction unit 803 extracts treatment options 830 from the formulary DB 400 based on the search query 821. Details of the extraction process by the extraction unit 803 will be described later with reference to FIG.

[0030] (Analysis Department 804) The analysis unit 804 refers to the medical DB 500 and outputs an analysis result 840 in response to the analysis request 822. Specifically, for example, the analysis unit 804 extracts medical information of other patients Pj (i≠j) that satisfy the analysis request 822 from the medical DB 500. The analysis unit 804 counts the number of extracted pieces of medical information of other patients Pj as the number of cases.

[0031] Furthermore, the analysis unit 804 calculates the estimated treatment effect (for example, PSF (progression-free survival)) requested in the analysis request 822 based on the extracted medical information of the other patient Pj. The calculation of the estimated treatment effect is performed, for example, by inputting the values ​​of the data schema 510 of the patient Pi into a machine learning model in which the treatment effect of the other patient Pj is used as the objective variable and the values ​​of the other patient Pj indicated by the data schema 510 are used as explanatory variables. Note that in calculating the estimated treatment effect, subgroup analysis may be applied using the depth of invasion listed in the guideline DB 300 and the background adjustment factor 604 listed in the analysis perspective DB 600.

[0032] The analysis unit 804 outputs the number of cases and the estimated treatment effect as the analysis result 840 to the verification unit 805. The analysis unit 804 may be implemented using a language model. The analysis unit 804 may also be implemented in a computer outside the treatment selection supporting device 100.

[0033] (Verification unit 805) The verification unit 805 calculates a case number score indicating the value of the case number included in the analysis result 840, calculates an estimated treatment effect score indicating the value of the estimated treatment effect included in the analysis result 840, calculates a reliability score indicating the reliability of the treatment option 830, and generates selection support information 850. The selection support information 850 includes, for example, the treatment option 830 (treatment method) recommended for the patient Pi, a match score (output information), a guideline recommendation level, the case number, and the estimated treatment effect. Details of the verification unit 805 will be described later with reference to FIG. 13.

[0034] (output unit 806) The output unit 806 outputs the selection support information 850. Specifically, for example, the output unit 806 may display the selection support information 850 on a display, which is an example of the output device 104, or may transmit the selection support information 850 to another computer via the communication IF 105.

[0035] <Figure 9 Analysis control process by analysis control unit 802> 9 is a flowchart illustrating an example of analysis control processing by the analysis control unit 802 according to Example 1. Generation of a search query 821 will be described in steps S901 and S902, and generation of an analysis request 822 will be described in steps S911 to S917.

[0036] (Step S901) The analysis control unit 802 executes a pathological condition information extraction process. Specifically, for example, the analysis control unit 802 extracts pathological condition information Ci from the patient information (Bi, Ci) of the patient Pi.

[0037] (Step S902) The analysis control unit 802 executes a search query generation process based on the pathological condition information Ci. Specifically, for example, the analysis control unit 802 generates a search query 821 by calculating the logical product of the disease type 241, the progression level 242, the treatment history 243, and the name of the coexisting disease 244 in the pathological condition information Ci. In this example, the disease type 241 is a required condition for the search query 821.

[0038] The progression level 242, treatment history 243, and comorbid illness name 244 do not necessarily have to be included in the search query 821. Furthermore, the treatment history 243 and the comorbid illness name 244 may be connected by logical product (AND) or logical sum (OR). The type of connection to be made can be set by the user in advance. Furthermore, in this example, the type of illness 241 is set as a required condition for the search query 821, but the user can also set in advance which items are required as required conditions.

[0039] [Figure 10 Search query 821] Fig. 10 is an explanatory diagram showing an example of a search query 821. In Fig. 10, the query is composed of ANDs 1001 to 1003 indicating the logical product of "gastric cancer" (disease type 241), "IIA" (stage of progression 242), "endoscopic" (treatment history 243), and "hypertension" (name of coexisting disease 244).

[0040] (Step S911) 9, the analysis control unit 802 extracts guidelines 303 for the disease type 302 that matches the disease type 241 of the pathological condition information Ci extracted in step S901 from the guideline DB 303. In the case of the search query 821 in FIG. 10, text data 330 is extracted from the guideline DB 303, where the guideline 303 has the disease type 302 of "stomach cancer."

[0041] (Step S912) The analysis control unit 802 extracts named entities related to treatment methods and sentence data containing the named entities from the one or more guidelines 303 extracted in step S911. Specifically, for example, "SP therapy" and "CapeOX therapy" are extracted from sentence data 330 as named entities related to treatment methods, and sentence data 330 is extracted as a sentence containing the named entities.

[0042] In addition, if the analysis control unit 802 is implemented using a language model, in step S912, the analysis control unit 802 extracts sentence data from the sentence data 330, including the progression level 242, treatment history 243, and name of concomitant illness 244 that constitute the medical history 233 of the pathological information Ci.

[0043] (Step S913) The analysis control unit 802 extracts disease condition indicators or biomarkers contained in the extracted text data as items to be considered when selecting a treatment, using relationship analysis techniques such as dependency relationships or named entity extraction using the medical dictionary 700. Specifically, for example, "depth of invasion" contained in the text data 330 is extracted as a disease condition indicator.

[0044] (Step S914) The analysis control unit 802 extracts an analysis viewpoint of the injury / illness type 302 that matches the injury / illness type 241 of the pathological condition information Ci extracted in step S901 from the analysis viewpoint DB 600. In this example, the entry 610 in Fig. 6 is extracted as the analysis viewpoint.

[0045] (Step S915) The analysis control unit 802 adds the extracted item in step S913 to the extracted analysis viewpoint in step S914. In this example, "depth of invasion" is added as an extracted item to the background adjustment factor 604 in the entry 610 in FIG.

[0046] (Step S916) The analysis control unit 802 narrows down the extracted analysis perspective to which the extracted item has been added to items present in the data schema 510. For example, if the items present in both the background adjustment factor 604 of the entry 610 and the extracted analysis perspective are "age (generation)," "gender," "stage of progression," "performance status," and "previous treatment," these are output as the analysis perspective 900. Note that the extracted item "depth of invasion" added in step S915 may be excluded from the narrowing down targets. In this case, the analysis perspective 900 also includes "depth of invasion."

[0047] (Step S917) The analysis control unit 802 generates an analysis request 822 based on the analysis perspective 900. Specifically, for example, the analysis control unit 802 transmits the analysis perspective 900 to the language model as a prompt, and then obtains the analysis request from the language model.

[0048] [Figure 11 Analysis request 822] 11 is an explanatory diagram showing an example of an analysis request 822. When the analysis viewpoint 900 is used as a prompt, an analysis request 822 as shown in FIG.

[0049] <Fig. 12 Extraction process by the extraction unit 803> 12 is a flowchart showing an example of the extraction processing procedure by the extraction unit 803. When the extraction unit 803 acquires the search query 821 from the analysis control unit 802, it executes at least one of steps S1211 to S1213 and steps S1221 to S1223.

[0050] (Step S1211) The extraction unit 803 searches the guideline DB 300 for guideline data having an injury / illness type 302 that matches the injury / illness type 241 of the search query 821. In the above example, the injury / illness type 241 of the search query 821 is "stomach cancer." The guideline data whose injury / illness type 302 is "stomach cancer" is the entry whose guideline ID 301 is "1."

[0051] (Step S1212) The extraction unit 803 extracts a guideline 303 that corresponds to the search query 821 from the guidelines 303 of the one or more guideline data extracted in step S1211.

[0052] In this case, the extraction unit 803 may gradually relax the conditions of the search query 821. For example, the extraction unit 803 converts at least one of AND 1002 and 1003, which indicate the logical product of the search query 821, into OR, which indicates the logical sum, and searches the text data 330. The extraction unit 803 may perform this gradually relaxing until a predetermined number of sentences matching the search query 821 are found, or may perform it until all of AND 1002 and 1003 are converted into OR. However, in this example, because "stomach cancer" in the injury / illness type 241 is a required condition, AND 1001 is not converted into OR.

[0053] In the above example, the guideline 303 of the extracted guideline data is the text data 330 , and since the text data 330 includes “endoscopic resection” in the treatment history 243 of the search query 821 , the text data 330 is extracted as the guideline 303 .

[0054] (Step S1213) The extraction unit 803 extracts treatment methods and recommendation levels as treatment options 830 from the extraction guidelines of step S1212 by natural language processing such as named entity extraction technology. In the above example, a combination of "SP therapy" as a named entity for the treatment method and "evidence level A" as a named entity for the recommendation level is extracted from the text data 330 as the treatment option 830. Also, a combination of "CapeOX therapy" as a named entity for the treatment method and "evidence level B" as a named entity for the recommendation level is extracted from the text data 330 as the treatment option 830.

[0055] (Step S1221) The extraction unit 803 extracts formulary data in which the disease type 241 of the search query 821 is included in the indication 404 from the formulary DB 400. In this example, the disease type 241, "gastric cancer", is also included in the indication 404 of the entry 410 whose formulary ID 401 is "1", so the entry 410 is extracted as formulary data (hereinafter, formulary data 410).

[0056] (Step S1222) The extraction unit 803 searches for extracted formulary data from the one or more extracted formulary data in step S1221, whose indications 404 correspond to the search query 821 (hereinafter referred to as the first search) and whose contraindications 405 do not correspond to the comorbid illness name 244 of the search query 821 (hereinafter referred to as the second search).

[0057] In this case, the extraction unit 803 may gradually relax the conditions of the search query 821 in the first search. For example, the extraction unit 803 converts at least one of AND1002 and 1003, which indicate the logical product of the search query 821, into OR, which indicates the logical sum, to search for one or more extracted formulary data. Note that the extraction unit 803 may perform this gradually relaxing until a predetermined number of extracted formulary data corresponding to the search query 821 are searched for, or may perform it until all of AND1002 and 1003 are converted into OR. However, in this example, since "stomach cancer" in the disease type 241 is a required condition, AND1001 is not converted into OR.

[0058] In this example, it is assumed that step S1222 has searched for formulary data 410 in which indications 404 include "stomach cancer," which is the illness type 241 of the search query 821, and contraindications 405 do not include "high blood pressure," which is the comorbid illness name 244 of the search query 821. Note that the extraction unit 803 may execute at least one of steps S1221 and S1222.

[0059] (Step S1223) The extraction unit 803 outputs a combination of the treatment method 402 and recommendation level 403 of the formulary data retrieved in step S1221 as treatment option 830. In the above example, since the retrieved formulary data is formulary data 410, the combination of “Trastuzumab (Herceptin)” as the treatment method 402 and “First choice” as the recommendation level 403 of formulary data 410 is output as treatment option 830.

[0060] <Fig. 13 Verification process by the verification unit 805> FIG. 13 is a flowchart showing an example of the verification process procedure performed by the verification unit 805.

[0061] (Step S1301) The verification unit 805 executes a reliability score calculation process. The reliability score calculation process is a process of calculating a reliability score for each treatment option 830. In the reliability score calculation process, a reliability score is calculated based on whether or not the treatment option 830 is described in the formulary data and the recommendation level described in the guideline 303.

[0062] [Figure 14 Reliability score table] 14 is an explanatory diagram showing an example of a reliability score table. The reliability score table 1400 is referenced for each treatment option 830 when the verification unit 805 executes the reliability score calculation process. The reliability score table 1400 has fields of description conditions 1401 and reliability scores 1402.

[0063] The description conditions 1401 are conditions that specify items in a stepwise manner for determining whether the treatment option 830 is listed in the formulary data or whether the treatment option 830 corresponds to one of the recommendation levels listed in the guideline 303. When one treatment option 830 corresponds to multiple items in the description conditions 1401, the verification unit 805 acquires the maximum value of the reliability scores 1402 corresponding to each of the multiple items.

[0064] The reliability score 1402 is information that quantifies the reliability of the treatment option 830, and the larger the value, the higher the reliability. The reliability score 1402 is obtained for each treatment option 830.

[0065] For example, if treatment option 830 is a combination of "Trastuzumab (Herceptin)" as treatment method 402 in formulary data 410 and "First choice" as recommendation level 403, since the source of extraction is formulary data 410, it satisfies the judgment item "listed in formulary data" of description condition 1401, and therefore "10" is extracted as the reliability score 1402 for treatment option 830.

[0066] For example, if treatment option 830 is a combination of the treatment method of the named entity "SP therapy" in text data 330 and the recommendation level of the named entity "Evidence level A," then it falls under the judgment item "Guideline recommendation (evidence) level A" of description condition 1401, and therefore "9" is extracted as the reliability score 1402 for treatment option 830.

[0067] For example, if treatment option 830 is a combination of the named entity treatment method "CapeOX therapy" in text data 330 and the named entity recommendation level "Evidence level B," then this falls under the judgment item "Guideline recommendation (evidence) level B" of description condition 1401, and therefore "7" is extracted as the reliability score 1402 for treatment option 830.

[0068] (Step S1302) 13, the verification unit 805 executes a performance score calculation process. The performance score calculation process is a process for calculating a performance score using the analysis result 840. The performance score includes a case number score and an estimated therapeutic effect score.

[0069] [Figure 15 Case number score table] 15 is an explanatory diagram showing an example of a case number score table. The case number score table 1500 is referenced when the verification unit 805 executes the performance score calculation process. The case number score table 1500 has a case number condition 1501 and a case number score 1502 as fields.

[0070] The case number condition 1501 is a condition applied to the number of cases in the analysis result 840. The case number score 1502 is information that quantifies the evaluation of the number of cases in the analysis result 840, with a larger value indicating a higher evaluation. The verification unit 805 extracts the case number score 1502 that satisfies the case number condition 1501 from the case number score table 1500.

[0071] [Figure 16 Estimated Treatment Effect Score Table] 16 is an explanatory diagram showing an example of an estimated therapeutic effect score table. The estimated therapeutic effect score table 1600 is referred to when the verification unit 805 executes the performance score calculation process. The estimated therapeutic effect score table 1600 has fields of an estimated therapeutic effect condition 1601 and an estimated therapeutic effect score 1602.

[0072] The estimated treatment effect condition 1601 is a condition applied to the estimated treatment effect in the analysis result 840. The mean and SD (standard deviation) may be updated in accordance with updates to the medical DB 500. In addition, while FIG. 16 shows an example of PSF as the estimated treatment effect, PSF is just one example, and the estimated treatment effect may also be OS (overall survival), a test value 501 (e.g., HbA1c), or the occurrence or non-occurrence of an adverse event 502 with a specific progression level of 242 or higher.

[0073] The estimated therapeutic effect score 1602 is information obtained by quantifying the evaluation of the estimated therapeutic effect in the analysis result 840, and a larger value indicates a higher evaluation. The verification unit 805 extracts an estimated therapeutic effect score 15602 that corresponds to the estimated therapeutic effect condition 1601 from the estimated therapeutic effect score table 1600.

[0074] (Step S1303) 13, the verification unit 805 executes a match score calculation process. The match score calculation process is a process for calculating a match score using the reliability score, the number of cases score, and the estimated treatment effect score. The match score is calculated for each treatment option 830 using the following formula (1).

[0075] Agreement score = Confidence score × Case number score + W × Estimated treatment effect score (1)

[0076] W is a weight for the estimated treatment effect score, and its value varies depending on the type of estimated treatment effect (PSF, OS, etc.) and the type of disease 241 identified in the pathological information 203 of the patient Pi. When the confidence score and the case number score are both large, the agreement score indicates an established and commonly used treatment option 830. When the confidence score is large but the case number score is small, the agreement score indicates a treatment option 830 with high evidence but infrequently performed, requiring careful adaptation judgment by the physician. When the confidence score is small but the case number score is large, the agreement score indicates a treatment option 830 that is not yet established but is commonly performed, or a treatment option 830 that is performed with a low recommendation level. When the confidence score and the case number score are both small, the agreement score indicates a treatment option 830 that is not recommended and is infrequently performed. Note that the agreement score may be calculated using a machine learning model trained using the explanatory variables (clinical DB 500) and previously calculated agreement scores as a training dataset.

[0077] [Figure 17 Output information determination table] 17 is an explanatory diagram showing an example of an output information determination table. The output information determination table 1700 has a match score condition 1701 and output information 1702. The output information determination table 1700 has the match score condition 1701 and the output information 1702 as fields.

[0078] The match score condition 1701 is a condition applied to the match score. The output information 1702 is information that is selected as an output target when the match score condition 1701 is satisfied.

[0079] The verification unit 805 acquires, for each treatment option 830, output information 1702 whose match score satisfies the match score condition 1701 as selection support information 850. Note that the verification unit 805 may use the match score of the treatment option 830 as selection support information 850 without using the output information determination table 1700.

[0080] The selection support information 850 includes, for example, treatment options 830 (treatment methods) recommended for the patient Pi, a match score (output information), a guideline recommendation level, the number of cases, and an estimated treatment effect.

[0081] <Figure 18 Input screen> 18 is an explanatory diagram showing an example of an input screen. The input screen 1800 is displayed on a display, which is an example of the output device 104. The input screen 1800 has a patient name input field 1801, a patient ID input field 1802, a pathological condition information display area 1803, and an execute button 1804.

[0082] The patient name input field 1801 is an input field that accepts and displays the input of the name 221 of the patient Pi. The patient ID input field 1802 is an input field that accepts and displays the input of the patient ID 201. The pathological condition information display area 1803 is a display field that displays the name 221 of the patient Pi input in the patient name input field 1801 by the input unit 801 and the pathological condition information Ci of the patient Pi identified by the patient ID 201 input in the patient ID input field 1802 by the input unit 801. The execute button 1804 is a user interface that, when pressed, starts the execution of treatment selection support (FIGS. 9 to 17).

[0083] When the name 221 and patient ID 201 of patient Pi are entered into the input screen 1800 by the input unit 801, the treatment selection support device 100 extracts the pathological information Ci of patient Pi that matches the name 221 and patient ID 201 from the patient information DB 200 and displays it in the pathological information display area 1803.

[0084] <Figure 19 Output screen> 19 is an explanatory diagram showing an example of an output screen. The output screen 1900 is displayed on a display, which is an example of the output device 104. The output screen 1900 has a patient name display field 1901, a patient ID display field 1902, a treatment option ranking table 1903, and a setting button 1904. The patient name display field 1901 displays the name of the patient Pi input by the input unit 801. The patient ID display field 1902 displays the patient ID 201 of the patient Pi input by the input unit 801.

[0085] The treatment option ranking table 1903 is a list of information that ranks the treatment options 830 by their matching scores (output information 1702). The treatment options, matching scores (output information 1702), guideline recommendation levels, number of cases, and estimated treatment effects that make up the treatment option ranking table 1903 are information included in the selection support information 850.

[0086] The setting button 1904 is a user interface for recording the selected treatment method when pressed. [Example]

[0087] Next, a description will be given of Example 2. Example 2 is an example in which a treatment option 830 is input to the analysis control unit 802 and an analysis request 822 is generated.

[0088] <Figure 20 Example of functional configuration of treatment selection support device 100> 20 is a block diagram showing an example of the functional configuration of the treatment selection supporting device 100 according to Example 2. By using the treatment options 830 to generate the analysis request 822, the accuracy of the analysis result 840 obtained from the analysis request 822 is improved. More specifically, a comparative analysis between one or more treatment options included in the treatment options 830 is generated as the analysis request 822.

[0089] <Figure 21 Analysis control process by analysis control unit 802> 21 is a flowchart illustrating an example of analysis control processing by the analysis control unit 802 according to the second embodiment. The difference from FIG. 9 is that step S912 is changed to step S2112.

[0090] (Step S2112) The analysis control unit 802 extracts text data that includes at least a treatment method from the treatment options 830, from the one or more guidelines 303 extracted in step S911. Specifically, for example, if the treatment options 830 include "SP therapy" as a treatment method and "evidence level A" as a recommendation level, the text data 330 of the guideline 303 is extracted.

[0091] As described above, according to the above-described Examples 1 and 2, it is possible to improve the reliability of treatment options, i.e., it is possible to present realistic treatment options that are less likely to deviate from past medical data.

[0092] The present invention is not limited to the above-described embodiments, and includes various modifications and equivalent configurations within the spirit and scope of the appended claims. For example, the above-described embodiments have been described in detail to clearly explain the present invention, and the present invention is not necessarily limited to configurations including all of the described configurations. Furthermore, part of the configuration of one embodiment may be replaced with the configuration of another embodiment. Furthermore, the configuration of another embodiment may be added to the configuration of one embodiment. Furthermore, part of the configuration of each embodiment may be added to, deleted from, or replaced with other configurations.

[0093] Furthermore, the aforementioned configurations, functions, processing units, processing means, etc. may be realized in part or in whole in hardware, for example by designing them as integrated circuits, or may be realized in software by having a processor interpret and execute a program that realizes each function.

[0094] Information such as programs, tables, files, etc. that realize each function can be stored in storage devices such as memory, hard disks, SSDs (Solid State Drives), or recording media such as IC (Integrated Circuit) cards, SD cards, and DVDs (Digital Versatile Discs).

[0095] In addition, the control lines and information lines shown are those that are considered necessary for explanation, and do not necessarily represent all the control lines and information lines that are necessary for implementation. In reality, it can be assumed that almost all components are interconnected. [Explanation of symbols]

[0096] 100 Treatment selection support device 101 processors 102 Storage Devices 200 Patient information DB 202 Basic information 203 Pathological information 241 Disease species 300 Guideline DB 303 Guidelines 330 Text data 400 Formulary DB 402 Treatment 403 Recommended Level 404 Adaptation 405 Contraindications 410 Formulary Data 500 Medical DB 510 Data Schema 600 Analysis Perspective DB 700 Medical Dictionary 801 Input section 802 Analysis control section 803 Extraction part 804 Analysis Department 805 Verification Department 806 Output section 821 search queries 822 Analysis Requests 830 Treatment Options 840 Analysis results 850 Selection Support Information 1400 Confidence Score Table 1500 Case Score Table 1600 Estimated Treatment Effect Score Table 1700 Output Information Decision Table 1800 Input Screen 1900 output screen

Claims

1. A treatment selection support device having a processor that executes a program and a storage device that stores the program, The processor: a verification process for verifying the validity of the treatment options based on treatment options including treatment methods for the patient's illness or injury and analysis results based on treatment records and estimated treatment effects for the illness or injury of other patients; an output process for outputting a verification result obtained by the verification process; A treatment selection support device characterized by executing the above.

2. The treatment selection support device according to claim 1, A first database is accessible, which stores a set of text data describing treatment methods for injuries and illnesses; In the verification process, the processor calculates a reliability score indicating the reliability of the treatment option based on the presence or absence of the treatment option in the text data group of the first database, calculates a case number score evaluating the number of treatment results, calculates an estimated treatment effect score evaluating the estimated treatment effect, and verifies the validity of the treatment option based on the reliability score, the case number score, and the estimated treatment effect score. A treatment selection support device characterized by:

3. The treatment selection support device according to claim 1, A first database is accessible, which stores a set of text data describing treatment methods for injuries and illnesses; The processor: an extraction process for extracting, from the first database, the treatment options corresponding to a search query regarding a pathological condition of the patient, including the type of injury or disease; A treatment selection support device characterized by executing the above.

4. The treatment selection support device according to claim 3, the first database is accessible to a first database storing a set of sentence data describing the treatment and a recommendation level of the treatment; In the extraction process, the processor extracts, from the first database, treatment options that match the search query and include the treatment and a recommended level of the treatment. A treatment selection support device characterized by:

5. The treatment selection support device according to claim 3, the first database is accessible to a database storing a group of sentence data describing the treatment method, a recommended level of the treatment method, and adaptation information indicating an injury or illness type to which the treatment method can be adapted; In the extraction process, the processor extracts, from the first database, specific text data that corresponds to the search query and includes the adaptation information describing a treatment method applicable to the type of illness or injury of the patient, and extracts the treatment method and the recommendation level of the specific text data as the treatment option. A treatment selection support device characterized by:

6. The treatment selection support device according to claim 3, the first database is accessible to a first database storing a group of sentence data describing the treatment method, a recommended level of the treatment method, and contraindication information indicating types of injuries or illnesses to which the treatment method should not be applied; In the extraction process, the processor extracts, from the first database, specific text data that corresponds to the search query and does not include contraindication information describing a treatment that should not be applied to the type of illness or injury of the patient, and extracts the treatment method and the recommendation level of the specific text data as the treatment option. A treatment selection support device characterized by:

7. The treatment selection support device according to claim 6, a second database accessible to the patient that stores the condition; a generation process of generating the search query from the second database based on the pathological condition of the patient; A treatment selection support device characterized by executing

8. The treatment selection support device according to claim 1, A third database is accessible that stores analysis points for each type of injury or illness; an acquisition process in which the processor extracts a specific analysis aspect related to the type of illness or injury of the patient from the third database and acquires an analysis request requesting the analysis result based on the analysis aspect; A treatment selection support device characterized by executing

9. The treatment selection support device according to claim 8, A first database is accessible, which stores a set of text data describing treatment methods for each type of injury or illness; In the acquisition process, the processor extracts specific text data describing treatment methods for the type of illness or injury of the patient from the first database, extracts disease state indicators or biomarkers from the specific text data as specific items to be considered in treatment selection, and acquires the analysis request based on the specific analysis perspective including the specific items. A treatment selection support device characterized by:

10. The treatment selection support device according to claim 9, In the acquisition process, the processor inputs the specific analysis perspective including the specific item into a language model, and acquires the analysis request from the language model. A treatment selection support device characterized by:

11. The treatment selection support device according to claim 10, capable of communicating with a computer having the language model; In the acquisition process, the processor acquires the analysis request from the computer as a result of transmitting the specific analysis perspective including the specific item to the computer. A treatment selection support device characterized by:

12. The treatment selection support device according to claim 9, In the acquisition process, the processor extracts, from the first database, specific text data describing the treatments included in the treatment options, extracts disease state indicators or biomarkers from the specific text data as specific items to be considered in treatment selection, and acquires the analysis request based on the specific analysis perspective including the specific items. A treatment selection support device characterized by:

13. A treatment selection support method executed by a treatment selection support device having a processor that executes a program and a storage device that stores the program, The processor: a verification process for verifying the validity of the treatment options based on treatment options including treatment methods for the patient's illness or injury and analysis results based on treatment records and estimated treatment effects for the illness or injury of other patients; an output process for outputting a verification result obtained by the verification process; A treatment selection support method characterized by executing the above.

14. The processor a verification process for verifying the validity of the treatment options based on treatment options including treatment methods for the patient's illness or injury and analysis results based on treatment records and estimated treatment effects for the illness or injury of other patients; an output process for outputting a verification result obtained by the verification process; A treatment selection support program characterized by executing the above.

Images