Program, information processing method, and information processing device

Abstract

A program according to one embodiment of the present invention causes a computer to execute processing for acquiring trial information related to a clinical trial, acquiring a base document related to the acquired trial information, and generating a clinical trial-related document related to the clinical trial, by using a language model which uses the trial information and the base document. Due to this configuration, it is possible to generate a clinical trial-related document by using a language model.

Description

Program, information processing method and information processing device

[0001] The present invention relates to a program, an information processing method, and an information processing device.

[0002] In recent years, there has been active development of technologies relating to clinical trial-related documents used in clinical trials at medical institutions. For example, Patent Literature 1 discloses a test planning support device that searches for multiple sentences related to input clinical trial-related information from multiple document data related to clinical trials that have already been conducted, classifies the retrieved multiple sentences into multiple clusters based on similarity, and outputs information about the sentences classified into the clusters.

[0003] Japanese Patent Application Laid-Open No. 2020-035036

[0004] However, the invention of Patent Document 1 has a problem in that it is not possible to generate clinical trial-related documents using a language model.

[0005] One aspect of the present invention is to provide a program or the like that is capable of generating clinical trial-related documents using a language model.

[0006] A program according to one aspect causes a computer to perform a process of acquiring test information related to a clinical trial, acquiring base documents related to the acquired test information, and generating clinical trial-related documents related to the clinical trial using a language model that uses the test information and the base documents.

[0007] In one aspect, the language model allows for the generation of clinical trial related documents.

[0008] 1 is an explanatory diagram showing an overview of a clinical trial-related document generation system. FIG. 1 is a block diagram showing an example of the configuration of a server. FIG. 2 is an explanatory diagram showing an example of the record layout of a trial information DB and a role information DB. FIG. 3 is an explanatory diagram showing an example of the record layout of a base document DB and a generation history DB. FIG. 4 is an explanatory diagram showing an example of the record layout of a format DB. FIG. 5 is an explanatory diagram showing an example of the record layout of a task information DB. FIG. 6 is a block diagram showing an example of the configuration of a terminal. FIG. 7 is an explanatory diagram explaining a process for generating clinical trial-related documents. FIG. 8 is an explanatory diagram showing an example of a generation screen for clinical trial-related documents. FIG. 9 is an explanatory diagram showing an example of a generation screen for clinical trial-related documents using an existing assistant. FIG. 10 is a flowchart showing the processing steps for generating clinical trial-related documents. FIG. 11 is a flowchart showing the processing steps for outputting the similarity between a clinical trial-related document and a training document. FIG. 12 is an explanatory diagram explaining a process for generating clinical trial-related documents for each writer. FIG. 13 is an explanatory diagram showing an example of a generation screen for review documents for clinical trial-related documents. FIG. 14 is an explanatory diagram showing an example of a generation screen for review documents for clinical trial-related documents using an existing assistant. FIG. 15 is a flowchart showing the processing steps for generating review documents for clinical trial-related documents. 10 is a flowchart showing the processing steps when generating a second clinical trial-related document that is revised for a clinical trial-related document. FIG. 11 is a flowchart showing the processing steps when generating a review document for a third clinical trial-related document. FIG. 12 is a flowchart showing the processing steps when generating an evaluation document for a clinical trial-related document. FIG. 13 is a flowchart showing the processing steps when retraining a document generation model. FIG. 14 is a flowchart showing the processing steps when retraining a document evaluation model. FIG. 15 is an explanatory diagram showing an overview of a clinical trial-related document generation system in embodiment 5. FIG. 16 is a flowchart showing the processing steps when storing prompt information, test information, and a base document in a blockchain. FIG. 17 is a flowchart showing the processing steps when issuing a document NFT for a clinical trial-related document.1 is an explanatory diagram illustrating a process for identifying clinical trial-related documents after a debate. FIG. 1 is a flowchart illustrating the processing steps for identifying clinical trial-related documents after a debate. FIG. 1 is a flowchart illustrating the processing steps for identifying clinical trial-related documents after a debate. FIG. 1 is an explanatory diagram illustrating an example of a clinical trial-related document integration screen. FIG. 2 is a flowchart illustrating the processing steps for integrating clinical trial-related documents. FIG. 2 is an explanatory diagram illustrating an example of an evaluation screen for updated clinical trial-related documents. FIG. 3 is a flowchart illustrating the processing steps for generating an evaluation document for updated clinical trial-related documents. FIG. 3 is an explanatory diagram illustrating a process for generating clinical trial-related documents corresponding to a clinical trial outline. FIG. 4 is a flowchart illustrating the processing steps for generating clinical trial-related documents corresponding to a clinical trial outline. FIG. 4 is a flowchart illustrating the processing steps for generating review documents for updated clinical trial-related documents by multiple reviewers. FIG. 5 is a flowchart illustrating the processing steps for fine-tuning a document generation model. FIG. 5 is an explanatory diagram illustrating an example of a database search screen. FIG. 6 is a flowchart illustrating the processing steps for searching literature data from a database. FIG. 6 is an explanatory diagram illustrating an example of a screen for accepting debate settings. FIG. 7 is a flowchart illustrating the processing steps for generating a final clinical trial-related document. FIG. 7 is an explanatory diagram illustrating an example of a screen for displaying review results by multiple reviewers. FIG. 8 is a flowchart illustrating the processing steps for generating a clinical trial-related document based on the results of multiple reviews.

[0009] The present invention will be described in detail below with reference to the drawings showing embodiments thereof.

[0010] (Embodiment 1) Embodiment 1 relates to a form in which clinical trial-related documents relating to clinical trials are generated using a language model. Fig. 1 is an explanatory diagram showing an overview of a clinical trial-related document generation system. The system of this embodiment includes an information processing device 1 and an information processing terminal 2, and each device transmits and receives information via a network N such as the Internet.

[0011] Clinical trials (clinical development) include clinical trials (pre-market testing for pharmaceutical applications of drugs or medical devices) in accordance with ICH-GCP (Guidelines for the Conduct of Clinical Trials of Pharmaceuticals Based on the International Conference on Harmonisation of Technical Requirements for Registration of Pharmaceuticals for Human Use) and GCP (Good Clinical Practice) ministerial ordinances, post-marketing clinical trials, clinical research in accordance with the Clinical Research Act, or observational studies in accordance with the ethics review committee of each institution.

[0012] Clinical trial-related documents for clinical trials include synopses, protocols, patient consent forms, statistical analysis plans, statistical analysis reports, clinical study reports, or papers or treatment guidelines written based on the results of clinical trials. Clinical trials include not only company-sponsored clinical trials led by companies such as pharmaceutical or medical device manufacturers, or investigator-initiated clinical trials led by doctors at medical research institutions such as university hospitals, but also company-sponsored clinical research, investigator-initiated clinical research, epidemiological studies, database studies, etc. conducted as part of medical research activities.

[0013] Clinical trial-related documents are important documents for evaluating the conduct of clinical trials or the quality of data, and for ensuring the reliability and quality of clinical trials. However, creating clinical trial-related documents requires specialized knowledge and is a time-consuming and labor-intensive task. To solve this problem, in this embodiment, clinical trial-related documents are automatically generated using a language model, thereby reducing the burden of document creation and improving efficiency.

[0014] The information processing device 1 is an information processing device that processes, stores, and transmits / receives various types of information. The information processing device 1 is, for example, a server device, a personal computer, a general-purpose tablet PC (personal computer), etc. In this embodiment, the information processing device 1 is assumed to be a server device, and for the sake of brevity, will be referred to as server 1 below.

[0015] The information processing terminal 2 is a terminal device that accepts test information and receives and displays generated clinical trial-related documents. The information processing terminal 2 is, for example, an information processing device such as a personal computer terminal, a smartphone, a tablet, a mobile phone, or a wearable device such as a smart watch. For simplicity, the information processing terminal 2 will be referred to as the terminal 2 below.

[0016] The server 1 according to this embodiment acquires test information related to a clinical trial. The server 1 acquires a base document related to the acquired test information from a storage unit. The server 1 generates a clinical trial-related document related to the clinical trial using a document generation model (language model) that uses the test information and the base document. The document generation model will be described later.

[0017] 2 is a block diagram showing an example of the configuration of the server 1. The server 1 includes a control unit 11, a storage unit 12, a communication unit 13, a reading unit 14, and a large-capacity storage unit 15. Each component is connected by a bus B.

[0018] The control unit 11 includes an arithmetic processing device such as a CPU (Central Processing Unit), an MPU (Micro-Processing Unit), a GPU (Graphics Processing Unit), an FPGA (Field Programmable Gate Array), a DSP (Digital Signal Processor), or a quantum processor. The control unit 11 reads and executes a control program 1P (program product) stored in the storage unit 12, thereby performing various information processing, control processing, and the like related to the server 1.

[0019] The control program 1P can be deployed to run on a single computer, or on multiple computers located at one site, or distributed across multiple sites and interconnected by a communications network. While the control unit 11 is described in Figure 2 as a single processor, it may also be a multiprocessor.

[0020] The storage unit 12 includes memory elements such as RAM (Random Access Memory) and ROM (Read Only Memory), and stores the control program 1P or data required for the control unit 11 to execute processing. The storage unit 12 also temporarily stores data required for the control unit 11 to execute arithmetic processing. The communication unit 13 is a communication module for performing communication-related processing, and transmits and receives information to and from the terminal 2, etc. via the network N.

[0021] The reading unit 14 reads a portable storage medium 1a including a CD (Compact Disc)-ROM or a DVD (Digital Versatile Disc)-ROM. The control unit 11 may read the control program 1P from the portable storage medium 1a via the reading unit 14 and store it in the mass storage unit 15. Alternatively, the control unit 11 may download the control program 1P from another computer via a network N or the like and store it in the mass storage unit 15. Alternatively, the control unit 11 may read the control program 1P from the semiconductor memory 1b.

[0022] The mass storage unit 15 includes a recording medium such as a hard disk drive (HDD) or a solid state drive (SSD), etc. The mass storage unit 15 includes a document generation model 151, a test information database (DB) 152, a role information DB 153, a base document DB 154, a generation history DB 155, a format DB 156, and a task information DB 157.

[0023] The document generation model 151 is a generator that generates clinical trial-related documents related to a clinical trial based on test information related to the clinical trial and base documents related to the test information, and is a trained model generated by machine learning. The test information DB 152 stores test information related to the clinical trial.

[0024] The role information DB 153 stores role information of generators such as writers or reviewers. The base document DB 154 stores base documents related to study information. The generation history DB 155 stores the history of clinical trial-related documents generated by the document generation model 151. The format DB 156 stores the formats of clinical trial-related documents.

[0025] The task information DB 157 stores instruction scripts for preset tasks. Tasks include, for example, generating clinical trial-related documents, generating review documents for the clinical trial-related documents, etc. Note that instead of storing the task information DB 157, the instruction scripts may be stored in a text file.

[0026] In this embodiment, the storage unit 12 and the large-capacity storage unit 15 may be configured as an integrated storage device. Furthermore, the large-capacity storage unit 15 may be configured with a plurality of storage devices. Furthermore, the large-capacity storage unit 15 may be an external storage device connected to the server 1.

[0027] The server 1 may execute various information processing and control processing on a single computer, or may execute the processing in a distributed manner on multiple computers. The server 1 may also be realized by multiple virtual machines provided in a single server, or may be realized by using a cloud server.

[0028] 3 is an explanatory diagram showing an example of the record layout of the trial information DB 152 and the role information DB 153. The trial information DB 152 includes a trial ID column, a title column, a phase column, a trial treatment column, and a control treatment column. The trial ID column stores a trial ID of the trial information that is uniquely specified in order to identify the trial information related to each clinical trial.

[0029] The title column stores the title of the clinical trial. The title is a short description or name of the clinical trial. For example, the title may be "NSCLC treatment drug trial" or "Breast Cancer treatment trial."

[0030] The phase column stores the phase of the clinical trial. The phase indicates the stage of the clinical trial. For example, four phases may be set according to the stage of the clinical trial. Phase I is the stage of initial safety testing, Phase II is the stage of evaluating efficacy and side effects, Phase III is the stage of comparing the new drug with existing treatments, and Phase IV is the stage of post-marketing follow-up surveys.

[0031] The Test Treatment column stores the test treatment for the clinical trial. The test treatment indicates the treatment or procedure used in the trial (such as administration of a new drug, a specific surgery, or lifestyle changes). The Control Treatment column stores the control treatment for the clinical trial. The control treatment indicates the standard for comparison. The standard can be a "comparison group" for evaluating the test treatment, such as a placebo or an existing treatment.

[0032] In addition to the title, phase, test treatment, and control treatment, the trial information DB 152 may also store the objective, trial design, indication, eligibility criteria, endpoint, sample size, or citation. The objective is the goal that the trial aims to achieve. The trial design indicates how the trial will be conducted. The indication indicates the disease or condition that will be the target of the trial. The eligibility criteria indicate the conditions of patients who will be the target of the trial (age, sex, type or severity of disease, etc.).

[0033] Endpoints indicate specific indicators for evaluating the results of the study. Indicators are determined based on the objectives of the study, such as survival time, improvement of symptoms, or occurrence of side effects. Sample size indicates the number of patients included in the study to ensure the reliability or statistical significance of the results. Citations indicate reference sources, including literature or materials in which the results of the study were published.

[0034] The role information DB 153 includes a role ID column, a type column, a personality column, and a role column. The role ID column stores a unique ID of role information to identify the role information of each writer or reviewer. The type column stores the type of role. The role types include a writer who generates clinical trial-related documents, a reviewer who reviews the generated clinical trial-related documents, or a researcher who searches for base documents that are highly similar to clinical trial-related documents.

[0035] The personality column stores the personality of a writer, reviewer, researcher, or the like. The personality of a writer includes, for example, "analytical" or "creative." "Analytical" is a personality that generates more scientifically based drafts. "Creative" is a personality that generates more creative drafts. Note that the personality of a writer is not limited to "analytical" or "creative," but may also include, for example, "balanced." "Balance" is a personality that generates drafts by appropriately combining analytical thinking and creative thinking.

[0036] The reviewer's personality is set according to the degree or viewpoint of criticism. The reviewer's personality includes, for example, "strict" or "easy." "Strict" is a personality that leads to severe criticism. "Long-hearted" is a personality that leads to light criticism. The reviewer's personality is used in embodiment 2. The reviewer's personality is not limited to "strict" or "lenient," but may also include, for example, "economy" or "innovation." "Economy" is a personality that leads to criticism from the perspective of the cost and effort of the test. "Innovation" is a personality that leads to criticism from the perspective of the novelty of the test.

[0037] The researcher's characteristics include, for example, "precise" or "broad." "Precise" refers to the characteristic of searching for base documents with high similarity. "Broad" refers to the characteristic of searching for base documents with a wide search range. For example, if similarity levels are set from 1 to 5 in descending order of similarity, "broad" may refer to the characteristic of searching for base documents with a wide search range corresponding to similarity levels from 1 to 5. Alternatively, "broad" may refer to the characteristic of searching for base documents using both similarity and keywords. Note that the researcher's characteristics are not limited to "precise" or "broad," and may also include, for example, "problem" or "solution." "Problem" refers to the characteristic of searching for base documents with similar problems. "Solution" refers to the characteristic of searching for base documents with similar solutions.

[0038] The role column stores role information such as writer, reviewer, or researcher. Role information includes personality. For example, the role information of a writer may be "You are a writer who generates creative documents." The role information of a reviewer may be "You are a reviewer who gives harsh criticism at a pharmaceutical company." The role information of a researcher may be "You are a researcher who searches for base documents with high similarity."

[0039] Note that the role column may store role information that does not include personality. For example, the role information of a writer may be "You are a writer who creates documents."

[0040] FIG. 4 is an explanatory diagram showing an example of the record layout of the base document DB 154 and the generation history DB 155. The base document DB 154 stores base documents related to clinical trials. The base documents are documents such as clinical trial implementation plans or protocols, which describe detailed plans for conducting clinical trials, such as the clinical trial's objectives, design, methodology, statistical considerations, or eligibility criteria. The base documents include, for example, references obtained from external databases, documents related to the test treatment or target treatment, the test design, and document formats. The base documents provide important standards for ensuring the quality of clinical trials and protecting the rights, safety, and welfare of participants.

[0041] The base document DB 154 includes a document ID column, a document type column, a title column, a document content column, and a file column. The document ID column stores a document ID of a uniquely specified base document to identify each base document. The document type column stores the document type. The document type includes a synopsis, a protocol, a patient informed consent form, a statistical analysis plan, a statistical analysis report, a clinical study report, a paper, or a clinical practice guideline.

[0042] The title column stores the title of the clinical trial. Examples of clinical trial titles include a clinical trial for a drug to treat non-small cell lung cancer, or a clinical trial for a treatment for breast cancer. The document content column stores the content of the base document. The file column stores the path or file name of the file that stores the base document.

[0043] The generation history DB 155 includes a history ID column, an assistant ID column, a prompt information column, a test information column, a base document column, a document data column, and a generation date and time column. The history ID column stores a history ID of the generation history data that is uniquely specified in order to identify the generation history data of each clinical trial-related document.

[0044] The assistant ID column stores an assistant ID for identifying an assistant. An assistant is a robot unit, and includes, for example, a writer who generates clinical trial-related documents using the document generation model 151, a reviewer who reviews the generated clinical trial-related documents, or a researcher who searches for highly similar base documents.

[0045] The prompt information column stores prompt information given to the document generation model 151. The test information column stores test information related to clinical trials. The base document column stores base documents related to the test information. The document data column stores data of clinical trial-related documents generated by the document generation model 151. The generation date and time column stores date and time information when the clinical trial-related documents were generated.

[0046] 5 is an explanatory diagram showing an example of a record layout of the format DB 156. The format DB 156 includes a format ID column, a document type column, a name column, an item column, and a display destination column. The format ID column stores a unique format ID to identify the format of each clinical trial-related document. The document type column stores the type of clinical trial-related document (synopsis, protocol, etc.). The name column stores the name of the clinical trial-related document.

[0047] The item column stores the headers of clinical trial-related documents. The items include, for example, the clinical trial title (TITLE), phase (PHASE), objectives (OBJECTIVES), study design (STUDY_DESIGN), indication (INDICATION), intervention (INTERVENTION), control treatment (CONTROL), eligibility criteria (ELIGIBILITY), endpoints (END_POINTS), sample size (SAMPLE_SIZE), and citation (CITATION). The display destination column stores the display address of each item.

[0048] 6 is an explanatory diagram showing an example of a record layout of the task information DB 157. The task information DB 157 stores task information for each assistant (writer, reviewer, researcher, etc.) regarding a task for generating or reviewing clinical trial-related documents. The task information DB 157 may be manually created in advance by a user.

[0049] The task information DB 157 includes a task ID column, a document type column, an assistant type column, a task type column, and a task content column. The task ID column stores a unique ID of task information to identify each piece of task information. The document type column stores the type of document (synopsis, protocol, patient informed consent form, statistical analysis plan, etc.).

[0050] The assistant type column stores the type of assistant (writer, reviewer, researcher, etc.). The task type column stores the type of instruction information in the task. The instruction information in the task includes, for example, "quick," "medium," or "long."

[0051] The task types for the writer may include "Please extract and execute only the main items of the task information required for document generation," "Please extract and execute all the task information required for document generation," or "Please execute all the task information required for document generation," etc. The task information types for the reviewer may include "Please review the appropriateness of some of the description items," "Please review only the appropriateness of all the description items," or "Please review each of all the description items using multiple evaluation items," etc.

[0052] The task content column stores multiple task contents for a task, which are set according to the document type, the assistant type, and the option type. For example, if the document type is a synopsis, the assistant type is a writer, and the option type is medium, the task information may be "WRMEDIUM001: The purpose of this study is to evaluate the efficacy and safety of paclitaxel and bevacizumab combination therapy compared with paclitaxel monotherapy as the first-line treatment for metastatic breast cancer. WRMEDIUM002: Please write only the phases of the new clinical trial."

[0053] The storage format of each DB described above is an example, and other storage formats may be used as long as the relationships between the data are maintained.

[0054] 7 is a block diagram showing an example of the configuration of the terminal 2. The terminal 2 includes a control unit 21, a storage unit 22, a communication unit 23, an input unit 24, and a display unit 25.

[0055] The control unit 21 includes a processing unit such as a CPU or an MPU, and performs various information processing and control processing related to the terminal 2 by reading and executing a control program 2P (program product) stored in the storage unit 22.

[0056] 7, the control unit 21 is described as a single processor, but it may be a multi-processor. The control unit 21 may execute various information processing or control processes by the same processor within the terminal 2, or may execute various information processing or control processes by different processors within the terminal 2.

[0057] The storage unit 22 includes memory elements such as RAM or ROM, and stores the control program 2P or data required for the control unit 21 to execute processing. The storage unit 22 also temporarily stores data required for the control unit 21 to execute arithmetic processing.

[0058] The communication unit 23 is a communication module for performing communication-related processing, and transmits and receives information to and from the server 1, etc. via the network N. The input unit 24 may be a keyboard, a mouse, or a touch panel integrated with the display unit 25. The display unit 25 is a liquid crystal display, an organic EL (electroluminescence) display, or the like, and displays various information according to instructions from the control unit 21.

[0059] 8 is an explanatory diagram illustrating the process of generating clinical trial-related documents. Terminal 2 acquires trial information related to the clinical trial. For example, terminal 2 may accept input (selection) of trial information. The trial information may include, for example, the name, phase, objectives, trial design, indications, test treatment, control treatment, eligibility criteria, endpoints, sample size, or citation of the clinical trial.

[0060] Terminal 2 accepts input of writer role information including the writer's characteristics. For example, the accepted role information may be "You are a writer who creates documents based on scientific evidence." Terminal 2 may also accept selection of target writer role information from multiple preset writer role information via a clinical trial-related document creation screen ( FIG. 9 ) described below.

[0061] Terminal 2 accepts input of task information for generating clinical trial-related documents. The task information is a set of multiple pieces of task information. For example, the accepted task information may be, "Please extract only the main items from the instructions required for document generation, and generate a synopsis for the clinical trial using the following trial information and base document." Terminal 2 may also accept selection of target task information from multiple pieces of task information preset for the task of generating clinical trial-related documents via a clinical trial-related document generation screen ( FIG. 9 ) described below.

[0062] The terminal 2 transmits the acquired test information and the received writer role information and task information to the server 1. The server 1 receives the test information, writer role information and task information transmitted from the terminal 2.

[0063] Based on the received test information, the server 1 retrieves (searches) base documents related to the test information from the base document DB 154. The base documents may be single or multiple. The search process may be performed by text search, vector search, or the like. For example, the server 1 searches for base documents that are highly similar to clinical trial-related documents by text search, vector search, or the like. The server 1 may also accept single or multiple base documents manually searched by the user.

[0064] The server 1 identifies the type of clinical trial-related document to be generated (e.g., synopsis) from the received task information. Based on the identified type, the server 1 obtains format information for the corresponding document from the format DB 156. The format information includes a format ID and items required for generating the document (e.g., title, phase, objectives, or study design), or a display destination for the items. The terminal 2 may also accept selection of a target document format from multiple pre-set document formats via a clinical trial-related document generation screen ( FIG. 9 ) described below.

[0065] The server 1 identifies the content corresponding to each item from the received task information according to each item included in the acquired format information. For example, the server 1 acquires the name or display destination of each item from the format information. The server 1 identifies the display destination by mapping the format information with the display destination information attached to the task information.

[0066] The server 1 provides the document generation model 151 with the received trial information (such as the name, phase, objectives, or trial design of the clinical trial), writer role information, the searched base document (such as the name or content of the base document), the acquired format information, and prompt information including content corresponding to each identified item.

[0067] The document generation model 151 is a language model that uses test information and base documents, and is used as a program module that is part of artificial intelligence software. The document generation model 151 is a constructed language model (language generation model) that receives test information related to clinical trials, base documents related to the test information, or prompt information including writer role information, etc., as input, and outputs clinical trial-related documents.

[0068] The document generation model 151 is a language model constructed by pre-training using large-scale text data (dataset). As the document generation model 151, for example, large language models (LLMs) such as ALBERT (A Lite BERT), GPT (Generative Pre-trained Transformer)-3, GPT-4, Llama2 (Large Language Model Meta AI 2), LLaVA (Large Language and Vision Assistant), or BERT (Bidirectional Encoder Representations from Transformers) can be used.

[0069] Instead of storing the document generation model 151 in the mass storage unit 15, the server 1 may access an external language processing server or language processing platform and read it out.

[0070] The prompt information is an instruction or an input sentence that is created in a format that can be understood by the document generation model 151 and is given as an input to the document generation model 151. The document generation model 151 interprets the input prompt information and outputs an appropriate response (e.g., a clinical trial-related document).

[0071] As an example, the document generation model 151 divides the prompt information into tokens to convert it into a format that can be processed by the document generation model 151. The document generation model 151 performs a context understanding process by calculating the association between each token in the prompt information and other tokens.

[0072] The document generation model 151 performs a response generation process for the prompt information based on linguistic knowledge obtained through pre-learning, fine-tuning, etc. For example, the document generation model 151 selects optimal tokens using a generation method such as greedy decoding, beam search, or sampling. The document generation model 151 performs a decoding process on the selected tokens to return them to a text format and generate a clinical trial-related document (e.g., a synopsis or protocol) as output data.

[0073] As shown in the figure, as an example, the prompt information may be, "You are a writer who generates creative documents. For a clinical trial of a therapeutic drug for non-small cell lung cancer, please extract only the main items from the instructions required for document generation and generate a synopsis for the clinical trial using the trial information and base document below, following the format including title, phase, objectives, and citations. Title: Clinical trial of a therapeutic drug for non-small cell lung cancer Phase: Phase 1 Indication: XXX Test treatment: Docetaxel plus Bevacizumab Document ID of base document: B-001."

[0074] The server 1 inputs the above-mentioned prompt information into the document generation model 151 to generate a synopsis for the clinical trial. Note that, although an example of a synopsis has been described in Fig. 8, the present invention is not limited to this, and can be similarly applied to, for example, a protocol, a patient informed consent form, a statistical analysis plan, a statistical analysis report, a clinical study report, or a paper or clinical practice guideline written based on the results of a clinical trial.

[0075] The prompt information may include task information for estimating specific information. For example, the server 1 may estimate the number of subjects that meets a statistical significance level based on epidemiological information regarding the target number of subjects included in the study. The server 1 inputs the acquired prompt information into the document generation model 151 and generates a synopsis including an estimate of the required number of subjects.

[0076] Alternatively, the prompt information may include an instruction to generate a synopsis with a specified document length. The server 1 inputs the prompt information including the specified document length to the document generation model 151, and generates a synopsis with the specified document length while taking into consideration study information (such as the name, phase, purpose, or study design of the clinical trial), the format, content, or purpose of the clinical trial-related document, etc.

[0077] The server 1 transmits the generated synopsis to the terminal 2. The terminal 2 receives the synopsis transmitted from the server 1 and displays the received synopsis on the screen.

[0078] As shown in the figure, as an example, the synopsis output from the document generation model 151 may be "Synopsis of clinical trial of drug for treating non-small cell lung cancer Item Writer Title: XXX Phase: Phase 1 Objective: XXX Study design: XXX Indication: XXX Study treatment: Docetaxel plus Bevacizumab Control treatment: Docetaxel Eligibility criteria: XXX Endpoint: XXX Sample size: XXX Citation: XXX."

[0079] The output synopsis may be a file, such as a text file, a Portable Document Format (PDF) file, a Word® file, an Excel® file, an image file, or an HTML file that constitutes a web page.

[0080] 9 is an explanatory diagram showing an example of a clinical trial-related document generation screen. The screen includes a test information reception field 11a, a document information reception field 11f, a writer information reception field 11b, a generation status display field 11c, and a document display field 11d.

[0081] The trial information reception field 11a is a field for receiving input (selection; setting) of trial information including the clinical trial name, phase, objectives, trial design, indication, test treatment, control treatment, eligibility criteria, endpoints, sample size or citation, etc. The trial information reception field 11a includes a trial name selection field 12a, a phase display field 12b, a test treatment display field 12c, and a control treatment display field 12d.

[0082] The trial name selection field 12a is a selection field that accepts the selection of the clinical trial name. Note that the trial name selection field 12a may also accept direct input of the clinical trial name. The phase display field 12b is a display field that displays the phase of the clinical trial. The test treatment display field 12c is a display field that displays the test treatment of the clinical trial. The control treatment display field 12d is a display field that displays the control treatment of the clinical trial.

[0083] The document information reception field 11f is a field for receiving input (selection; setting) of document information (format, etc.). The document information reception field 11f includes a document format selection field 121. The document format selection field 121 is a field for receiving selection of various document formats.

[0084] The writer information reception field 11b is a field for receiving input (selection; setting) of writer information. The writer information reception field 11b includes a writer mode selection field 12e, a writer role information selection field 12f, and a writer task information type selection field 12g.

[0085] The writer mode selection field 12e is a field for accepting the selection of a writer mode. The writer mode includes, for example, "Create a new assistant" or "Use an existing assistant." "Create a new assistant" is a mode for creating a new assistant (writer). "Use an existing assistant" is a mode for loading an existing assistant file. Assistant files will be described later.

[0086] The writer role information selection field 12f is a field for accepting selection of writer role information. The writer task information type selection field 12g is a field for accepting selection of task information for clinical trial-related documents. Note that different items are displayed in the writer role information selection field 12f and the writer task information type selection field 12g depending on the mode accepted in the writer mode selection field 12e.

[0087] In FIG. 9, an example of the writer mode "create a new assistant" is described, but the writer mode "use an existing assistant" will be described later with reference to FIG.

[0088] The generation status display field 11c is a display field that displays the generation status of the clinical trial-related document. The generation status includes a progress status such as "Questioning," "In Progress," or "Completed," progress data indicating the progress (e.g., "XXX information obtained"), generation details (prompt information or parameters, etc.), error information, etc.

[0089] The document display field 11d is a display field that displays clinical trial-related documents generated by the document generation model 151. The document display field 11d includes an item display field 12h and a content display field 12i. The item display field 12h is a display field that displays each item included in the document format of the clinical trial-related document. The content display field 12i is a display field that displays the content corresponding to each item.

[0090] The server 1 acquires the name of each clinical trial from the trial information DB 152 based on the trial ID of each clinical trial. The server 1 transmits the acquired name of each clinical trial to the terminal 2. The terminal 2 adds the name of each clinical trial transmitted from the server 1 to the trial name selection field 12a as an item for selection.

[0091] When terminal 2 receives a selection operation in the trial name selection field 12a, it acquires the name of the selected clinical trial (e.g., a clinical trial of a drug for treating non-small cell lung cancer). Based on the trial ID corresponding to the acquired clinical trial name, terminal 2 acquires the phase, test treatment, and control treatment of the clinical trial from the trial information DB 152 of server 1. Terminal 2 displays the acquired phase in the phase display field 12b, the acquired test treatment in the test treatment display field 12c, and the acquired control treatment in the control treatment display field 12d.

[0092] The terminal 2 may directly accept input of the clinical trial name, phase, test treatment, and control treatment. While Fig. 9 shows an example of trial information including the name, phase, test treatment, and control treatment, this is not limiting. For example, the trial information reception field 11a may be provided with input fields (display fields) for the clinical trial purpose, study design, indications, eligibility criteria, endpoints, sample size, citations, etc.

[0093] When the terminal 2 receives a selection process in the document format selection field 12l, it acquires the format of the selected document (e.g., synopsis) from the format DB 156. The format information includes a format ID and items required for generating the document (e.g., title, phase, objectives, or study design), or the display destination of the items.

[0094] When the terminal 2 receives a selection operation in the writer mode selection field 12e, it acquires the selected writer mode. If the acquired writer mode is "Create a new assistant," the terminal 2 acquires multiple pieces of writer role information from the role information DB 153 of the server 1. The terminal 2 adds the acquired multiple pieces of role information to the writer role information selection field 12f as selection items.

[0095] Items added to the writer role information selection field 12f include, for example, "You are a writer who creates documents based on scientific evidence," or "You are a medical school writer who creates creative documents," etc. When the terminal 2 receives a selection operation in the writer role information selection field 12f, it acquires the selected role information.

[0096] The terminal 2 acquires multiple types of task information for clinical trial-related documents that have been set in advance from the task information DB 157 of the server 1. The types of task information may be stored in the storage unit 22 of the terminal 2. The task information for generating a clinical trial-related document is multiple task types including content that should be included in the target document or content that should be excluded. The multiple task types may include, for example, "Please extract and execute only the main items of the task information required for document generation," "Please extract and execute all of the task information required for document generation," or "Please execute all of the task information required for document generation."

[0097] The terminal 2 adds the acquired multiple task types to the writer task information type selection field 12g as selection items. When the terminal 2 receives a selection operation in the writer task information type selection field 12g, it acquires multiple pieces of task information that include the selected task type. Note that the terminal 2 may also directly receive input of multiple pieces of task information by the user via the writer task information type selection field 12g.

[0098] When the terminal 2 receives selections (settings; inputs) for all fields included in the test information reception field 11a and the writer information reception field 11b, it transmits the received test information, writer role type, and task type to the server 1. The server 1 receives the test information, writer role type, and task type transmitted from the terminal 2.

[0099] The server 1 acquires prompt information. The prompt information includes study information (e.g., the name of the clinical trial, the phase, the test treatment, and the control treatment), writer role information, the base document, format information, and content corresponding to each item. The server 1 inputs the acquired prompt information into the document generation model 151 and executes processing for generating a synopsis for the clinical trial.

[0100] The server 1 acquires the generation status of clinical trial-related documents in real time from the document generation model 151. The server 1 transmits the generation status acquired in real time to the terminal 2. The terminal 2 receives the generation status transmitted from the server 1 and displays the received generation status in the generation status display field 11c. As shown in the figure, the generation status includes a progress status (e.g., "Questioning the document generation model..."), progress data indicating the intermediate progress (e.g., "A short instruction set has been acquired"), etc.

[0101] The server 1 acquires the synopsis output from the document generation model 151 together with document format information. The server 1 transmits the acquired synopsis to the terminal 2. The terminal 2 receives the synopsis transmitted from the server 1 and displays the received synopsis in the document display field 11d. Specifically, the terminal 2 displays each item included in the document format in the item display field 12h (item), and displays the content corresponding to each item in the corresponding content display field 12i (writer).

[0102] FIG. 10 is an explanatory diagram showing an example of a screen for generating clinical trial-related documents using an existing assistant. The same reference numerals are used to designate content that overlaps with that in FIG. 9, and a description thereof will be omitted. FIG. 10 illustrates an example of a writer mode that is "use an existing assistant." FIG. 10 includes a writer selection display field 12k. The writer selection display field 12k is a display field that displays role information of the selected writer.

[0103] When the terminal 2 receives a selection operation in the writer mode selection field 12e, it acquires the selected writer mode. If the acquired writer mode is "use existing assistant," the terminal 2 acquires an assistant file from the storage unit 12 or the mass storage unit 15 of the server 1. The assistant file is a history file that was automatically generated when the document generation model 151 generated clinical trial-related documents, and includes a prompt ID, a document type, an assistant type, role information, etc.

[0104] The terminal 2 adds the acquired assistant file corresponding to each existing writer as an item for selection to the writer role information selection field 12f. When the terminal 2 accepts a selection in the writer role information selection field 12f, it acquires the corresponding assistant item from the selected assistant file. The terminal 2 displays the file name of the selected assistant file in the writer selection display field 12k. The terminal 2 displays the acquired item in the writer selection display field 12k. (For example, "Barbara Analytical ID: 123") may be used.

[0105] The terminal 2 adds the type of task information to the writer task information type selection field 12g as a selection item. The types of task information include, for example, "1. Quick task information," "2. Medium task information," or "3. Long task information." When the terminal 2 receives a selection operation in the writer task information type selection field 12g, it acquires the selected type of task information.

[0106] Terminal 2 transmits the received test information, the file name of the assistant file, and the type of selected task information to server 1. Server 1 receives the test information, the file name of the assistant file, and the type of task information transmitted from terminal 2. Server 1 extracts the writer's role information by reading the corresponding assistant file based on the file name of the received assistant file. Server 1 obtains task information from task information DB 157 based on the type of received task information.

[0107] The server 1 acquires prompt information (such as test information, writer role information, base document, format information, and content corresponding to each item). The server 1 inputs the acquired prompt information into the document generation model 151 and outputs a synopsis for the clinical trial. The server 1 transmits the output synopsis to the terminal 2. The terminal 2 receives the synopsis transmitted from the server 1 and displays the received synopsis in the document display field 11d.

[0108] 11 is a flowchart showing the processing steps for generating a clinical trial-related document. The control unit 11 of the server 1 acquires multiple pieces of task information for the clinical trial-related document from the storage unit 12 or the mass storage unit 15 (step S101). The control unit 11 then transmits the acquired multiple pieces of task information to the terminal 2 via the communication unit 13 (step S102).

[0109] The control unit 21 of the terminal 2 receives the plurality of pieces of task information transmitted from the server 1 via the communication unit 23 (step S201). The control unit 21 displays the received plurality of pieces of task information on the display unit 25 (step S202). The control unit 21 of the terminal 2 accepts input (selection) of trial information including the clinical trial name, phase, objectives, trial design, indications, test treatment, control treatment, eligibility criteria, endpoints, sample size, or citation via the input unit 24 (step S203).

[0110] The control unit 21 receives an input (selection) of writer role information, including the writer's personality, via the input unit 24 (step S204). The control unit 21 receives a selection of a target task type from a plurality of task types via the input unit 24 (step S205). Note that the control unit 21 may also receive a task type directly input by the user via the input unit 24.

[0111] The control unit 21 transmits the test information, writer role information, and task information to the server 1 via the communication unit 23 (step S206). The control unit 11 of the server 1 receives the test information, writer role information, and task information transmitted from the terminal 2 via the communication unit 13 (step S103). Based on the received test information, the control unit 11 searches the base document DB 154 of the mass storage unit 15 for base documents that are highly similar to the clinical trial-related documents by text search, vector search, or the like (step S104).

[0112] The control unit 11 identifies the type of clinical trial-related document to be generated (e.g., synopsis) from the received task information (step S105). Based on the identified type, the control unit 11 obtains format information of the corresponding document (format ID, items, display destination, etc.) from the format DB 156 of the mass storage unit 15 (step S106).

[0113] The control unit 11 identifies the content corresponding to each item from the received task information according to each item included in the acquired format information (step S107). The control unit 11 acquires the received study information (such as the name, phase, objectives, or study design of the clinical trial), writer role information, the searched base document (such as the link, name, or content of the base document), the acquired format information, and prompt information including the content corresponding to each of the identified items (step S108).

[0114] The control unit 11 inputs the acquired prompt information into the document generation model 151 (step S109) and outputs the clinical trial-related document (step S110). The control unit 11 stores the output clinical trial-related document in the generation history DB 155 of the mass storage unit 15 (step S111). Specifically, the control unit 11 assigns a history ID to the clinical trial-related document output from the document generation model 151. The control unit 11 associates the assigned history ID with the prompt information input to the document generation model 151, the test information, the base document related to the test information, the document data of the clinical trial-related document, and the date and time information when the clinical trial-related document was generated, and stores them as a single record in the generation history DB 155. Note that when an assistant file corresponding to an existing writer is used, the writer's assistant ID is stored in the generation history DB 155.

[0115] The control unit 11 transmits the clinical trial-related documents output from the document generation model 151 to the terminal 2 via the communication unit 13 (step S112). The control unit 21 of the terminal 2 receives the clinical trial-related documents transmitted from the server 1 via the communication unit 23 (step S207) and displays the received clinical trial-related documents on the display unit 25 (step S208). The control unit 21 then terminates the process.

[0116] Furthermore, it is possible to output the similarity between the clinical trial-related document generated by the document generation model 151 and the clinical trial training document. Fig. 12 is a flowchart showing the processing procedure for outputting the similarity between the clinical trial-related document and the training document. The control unit 11 of the server 1 acquires the clinical trial-related document generated by the document generation model 151 from the generation history DB 155 of the mass storage unit 15 based on the history ID of the clinical trial-related document (step S121).

[0117] The control unit 11 acquires training documents for the clinical trial (step S122). The training documents may be clinical trial-related documents summarizing the results or data of the clinical trial (e.g., clinical trial reports), or clinical trial-related documents selected by the user. The training documents may be stored in the storage unit 12 or the mass storage unit 15 of the server 1, or may be stored in an external information processing device. For example, if the training documents have been stored in the mass storage unit 15 in advance, the control unit 11 acquires the training documents from the mass storage unit 15.

[0118] The control unit 11 calculates the similarity between the clinical trial-related documents generated by the document generation model 151 and the acquired training documents (step S123). For example, the control unit 11 extracts feature words from the generated clinical trial-related documents to generate feature word vectors. The control unit 11 similarly generates feature word vectors for the acquired training documents. The control unit 11 calculates the similarity based on the generated feature word vectors.

[0119] The control unit 11 may calculate the similarity using a trained learning model capable of evaluating document similarity, such as Bag of Words (BoW), FastText, or Word2Vec, or may have the similarity evaluated by a document generation model that is the same as or different from the document generation model 151. Alternatively, the control unit 11 may evaluate the clinical trial-related document by comparing the merits of the clinical trial-related document generated by the document generation model 151 with the training document, using, for example, a BLEU (Bilingual Evaluation Understudy) score or the like.

[0120] Furthermore, the control unit 11 may evaluate clinical trial-related documents by utilizing a language model. For example, the control unit 11 inputs a prompt to the language model, saying, "Please compare and evaluate which document is superior. The evaluation should be based on three perspectives: whether it is based on facts (factuality), whether references are included (citations), and whether necessary items are comprehensively described (comprehensiveness).", and outputs evaluation information for the clinical trial-related document.

[0121] The control unit 11 transmits the clinical trial-related documents and the similarities to the terminal 2 via the communication unit 13 (step S124). The control unit 21 of the terminal 2 receives the clinical trial-related documents and the similarities transmitted from the server 1 via the communication unit 23 (step S221), and displays the received clinical trial-related documents and the similarities on the display unit 25 (step S222). The control unit 21 then terminates the process.

[0122] Next, we will explain the process of multiple users editing the clinical trial-related documents generated by the document generation model 151. The users include assistants who are writers, and users who manually create the clinical trial-related documents.

[0123] Specifically, each user's terminal 2 accepts edits of the clinical trial-related document by the user. Each terminal 2 transmits the accepted edited data of the clinical trial-related document to a server 1 installed in the document editing tool or an external document server via a cloud-based document editing tool that allows multiple people to edit simultaneously. This facilitates collaboration between team members and enables efficient work.

[0124] According to this embodiment, it is possible to generate clinical trial-related documents using the document generation model 151 based on the trial information related to the clinical trial and the base document related to the trial information.

[0125] According to this embodiment, by automatically generating clinical trial-related documents, the burden of document creation can be reduced and documents can be created efficiently and accurately.

[0126] According to this embodiment, it is possible to use the document generation model 151 to generate clinical trial-related documents that suit the writer's characteristics.

[0127] According to this embodiment, it is possible to output the similarity between the clinical trial-related document generated by the document generation model 151 and the clinical trial training document.

[0128] According to this embodiment, it is possible to search for base documents that are highly similar to clinical trial-related documents.

[0129] According to this embodiment, it is possible to generate clinical trial-related documents in accordance with the document format of clinical trial-related documents.

[0130] According to this embodiment, it is possible to accept the selection of the target task type from a plurality of task information types for clinical trial-related documents.

[0131] According to this embodiment, it is possible to accept edits of clinical trial-related documents by multiple users.

[0132] <Modification 1> The following describes a process for searching for a base document similar to a clinical trial-related document using the document generation model 151. In this modification 1, for simplicity, the document generation model 151 will be referred to as a base document search model 201.

[0133] The server 1 acquires prompt information including study information (such as the name, phase, objectives, or study design of the clinical trial) and role information of a researcher who searches for highly similar base documents from the task information DB 157. The role information of the researcher may be, for example, "You are a researcher who investigates study information of clinical trials and searches for highly similar base documents based on that information."

[0134] As an example, the prompt information may be, "You are a researcher who investigates clinical trial information and searches for highly similar base documents based on that information. Analyze the following trial information and search for related base documents from the XXX database. Title: Clinical trial of a drug for treating non-small cell lung cancer Phase: Phase 1 Indication: XXX Trial treatment: Docetaxel plus Bevacizumab..."

[0135] The server 1 inputs the acquired prompt information into the base document retrieval model 201 and outputs the search results of searching for base documents similar to the clinical trial-related documents. As an example, the search results output from the base document retrieval model 201 may be "Highly similar base documents: 1. Phase II trial results of XXX - published in 2023 2. Design and analysis of XXX trial - published in 2021 3. XXX of anti-cancer drug treatment - published in 2022."

[0136] The search results may include the files of each base document or links to the documents (e.g., URLs (Uniform Resource Locators)). Then, similar to the process of the first embodiment, the server 1 acquires prompt information including the test information and the base documents. The server 1 inputs the acquired prompt information into the document generation model 151 to generate clinical trial-related documents.

[0137] According to this modification, it is possible to use the base document retrieval model 201 to retrieve base documents similar to clinical trial-related documents.

[0138] <Modification 2> This section describes a process for generating clinical trial-related documents from the dialogue content obtained by a dialogue between a user and the document generation model 151. The server 1 acquires task information based on the dialogue content between the user and the document generation model 151 from the task information DB 157. The dialogue content includes, for example, questions from the assistant to the user and answers from the user to the questions.

[0139] For example, the dialogue could be: "Assistant: What is the purpose of this trial? User: The purpose of this trial is to demonstrate the efficacy and safety of a new compound called XXX, which is used to treat non-small cell lung cancer. Assistant: Please attach any similar trials you may have. User: That is the paper on XXX from XXX University that was presented at ASCO this year. I have just uploaded it. Assistant: ..."

[0140] The server 1 acquires a prompt from the acquired dialogue content by utilizing a language model (e.g., a document generation model 151). The server 1 inputs the acquired prompt into the document generation model 151 and outputs a clinical trial-related document. The server 1 transmits the output clinical trial-related document to the terminal 2. The terminal 2 receives and displays the clinical trial-related document transmitted from the server 1.

[0141] According to this modification, clinical trial-related documents can be generated from the content of the dialogue between the user and the assistant.

[0142] <Modification 3> A process will be described in which there are multiple writers with different roles for generating clinical trial-related documents, and each writer generates a clinical trial-related document.

[0143] 13 is an explanatory diagram illustrating the process of generating clinical trial-related documents for each writer. As shown in the figure, there are writer A, writer B, and writer C, and writer A's role is to "generate documents based on scientific evidence," writer B's role is to "generate creative documents," and writer C's role is to "generate XXX documents." Note that while FIG. 13 illustrates an example with three writers, the number of writers is not particularly limited.

[0144] The server 1 acquires task information including writer role information for each writer from the task information DB 157. The server 1 generates clinical trial-related documents for each writer by providing the generated prompt information for each writer to the document generation model 151. Note that the process of generating clinical trial-related documents by the document generation model 151 is the same as in the first embodiment, and therefore a description thereof will be omitted.

[0145] As shown in the figure, the server 1 inputs the prompt information of each of writers A, B, and C into the document generation model 151, and outputs a clinical trial-related document AA corresponding to writer A, a clinical trial-related document BB corresponding to writer B, and a clinical trial-related document CC corresponding to writer C.

[0146] According to this modification, clinical trial-related documents can be generated for each writer with a different role.

[0147] Second Embodiment The second embodiment relates to a form in which a review document of a clinical trial-related document generated by the document generation model 151 is generated. Note that a description of the contents overlapping with the first embodiment will be omitted.

[0148] In the first embodiment, the document generation model 151 is a language model that generates clinical trial-related documents. In the present embodiment, in order to distinguish it from the document generation model 151 in the first embodiment, the document generation model 151 is replaced with a document review model 202. The document review model 202 is a language model for generating review documents of the clinical trial-related documents generated by the document generation model 151.

[0149] 14 is an explanatory diagram illustrating the process of generating review documents for clinical trial-related documents. The terminal 2 acquires the clinical trial-related documents generated by the document generation model 151. The terminal 2 reads the acquired clinical trial-related documents to acquire the names, types (e.g., synopses), contents, etc. of the clinical trial-related documents.

[0150] The terminal 2 accepts input of reviewer role information including the characteristics of the reviewer who reviews the clinical trial-related document. For example, the accepted role information may be "You work for a pharmaceutical company and are responsible for conducting rigorous reviews." The terminal 2 may also accept selection of the role information of a target reviewer from among multiple preset reviewer role information via a review document generation screen ( FIG. 15 ) for clinical trial-related documents, which will be described later.

[0151] Terminal 2 accepts input of task information for generating a review document for clinical trial-related documents. For example, the accepted task information may be, "For the generated synopsis, please refer to the XXX paper and confirm the consistency between the title and purpose, the consistency between the purpose and the study design, and whether the indication for the experimental drug is metastatic breast cancer." Terminal 2 may also accept selection of a target task type from multiple task information types preset for the review task of clinical trial-related documents via a screen for generating a review document for clinical trial-related documents ( FIG. 15 ), which will be described later.

[0152] The terminal 2 transmits information about the acquired clinical trial-related documents, including the names, types, or contents of the clinical trial-related documents, and the role information and task information of the received review, to the server 1. The server 1 receives the information about the clinical trial-related documents, the role information and task information of the review transmitted from the terminal 2. The server 1 identifies the type of the clinical trial-related document (e.g., synopsis) from the received information about the clinical trial-related document.

[0153] Based on the identified document information, the server 1 acquires format information of the corresponding document (format ID, items or display destination of items, etc.) from the format DB 156. The server 1 acquires prompt information to be provided to the document review model 202, including information about the received clinical trial-related document (name, type, content, etc.), reviewer role information, task information, and document format. Note that the terminal 2 may accept selection of the target document format from multiple preset document formats via a clinical trial-related document review document generation screen ( FIG. 15 ) described below.

[0154] Note that Figure 14 illustrates an example of a synopsis review document, but it is not limited to this and can be similarly applied to review documents such as protocols, patient consent forms, statistical analysis plans, statistical analysis reports, clinical reports, or papers or clinical guidelines.

[0155] As shown in the figure, for example, the prompt information could be, "You work for a pharmaceutical company and are responsible for conducting a rigorous review. Please refer to the XXX paper for the generated synopsis of a clinical trial of a non-small cell lung cancer drug, check the consistency between the title and objectives, the consistency between the objectives and the study design, and confirm whether the indication of the study drug is metastatic breast cancer, and generate a review document according to a format including title, phase, objectives, and citations. The information about the synopsis can be as follows: Title: XXX Objective: XXX Study design: XXX Indications: XXX."

[0156] The server 1 inputs the above-described prompt information into the document review model 202 and generates a review document of the synopsis of the clinical trial of a non-small cell lung cancer therapeutic drug. The server 1 transmits the generated review document to the terminal 2. The terminal 2 receives the review document transmitted from the server 1 and displays the received review document on its screen.

[0157] As shown in the figure, as an example, the review document output from the document review model 202 may be "Clinical trial synopsis of a drug for the treatment of non-small cell lung cancer Item Writer Reviewer Title: XXX Match Phase: Phase 1 XXX Objective: XXX XXX Study design: XXX XXX Indication: XXX XXX Study treatment: Docetaxel plus Bevacizumab XXX Control treatment: Docetaxel XXX Eligibility criteria: XXX XXX Endpoint: XXX XXX Sample size: XXX XXX Citation: XXX XXX."

[0158] The output review document may be a file, such as a text file, a PDF file, a Word file, an Excel file, an image file, or an HTML file that constitutes a web page.

[0159] 15 is an explanatory diagram showing an example of a screen for generating review documents for clinical trial-related documents. Note that the same reference numerals are used to designate the same contents as those in FIG. 9, and the description thereof will be omitted.

[0160] The screen includes a reviewer information reception field 11e. The reviewer information reception field 11e is a field for receiving input (selection; setting) of reviewer information. The reviewer information reception field 11e includes a reviewer mode selection field 13a, a reviewer role information type selection field 13b, and a reviewer task information type selection field 13c.

[0161] The reviewer mode selection field 13a is a field for accepting the selection of a reviewer mode. The reviewer mode includes, for example, "Create a new assistant" or "Use an existing assistant." "Create a new assistant" is a mode for creating a new assistant who is a reviewer. "Use an existing assistant" is a mode for reading the task information DB 157 of an existing assistant.

[0162] The reviewer role information type selection field 13b is a field for accepting the selection of the type of reviewer role information. The reviewer task information type selection field 13c is a field for accepting the selection of the type of task information in the review document generation task of clinical trial-related documents. Note that different items are displayed in the reviewer role information type selection field 13b and the reviewer task information type selection field 13c depending on the mode accepted in the reviewer mode selection field 13a.

[0163] In addition, in FIG. 15, an example of a reviewer mode of "create a new assistant" is described, but a reviewer mode of "use an existing assistant" will be described later with reference to FIG. 16.

[0164] When the terminal 2 receives a selection operation in the reviewer mode selection field 13a, it acquires the mode of the selected reviewer. When the acquired reviewer mode is "Create a new assistant", the terminal 2 acquires role information of multiple reviewers from the role information DB 153 of the server 1. The terminal 2 adds the acquired types of role information to the reviewer role information type selection field 13b as selection items.

[0165] The items added to the reviewer role information type selection field 13b include, for example, "You work for a pharmaceutical company and are responsible for conducting rigorous reviews," or "You work for a pharmaceutical company and are responsible for conducting quick reviews," etc. When the terminal 2 receives a selection operation in the reviewer role information type selection field 13b, it acquires the type of role information that has been selected.

[0166] The terminal 2 acquires, from the task information DB 157 of the server 1, multiple types of task information for a clinical trial-related document review that have been set in advance. The types of task information may be stored in the storage unit 22 of the terminal 2. The task information for a document review is specific instruction information for a review task of a clinical trial-related document, including an evaluation of each item of the clinical trial-related document (such as the title, phase, objectives, study design, indications, test treatment, or eligibility criteria), items to be noted, or suggestions for improvement. The multiple types of task information may include, for example, "Please review the appropriateness of some of the described items," "Please review only the appropriateness of all of the described items," or "Please review each of all described items using multiple evaluation items."

[0167] The terminal 2 adds the acquired multiple task information types to the reviewer task information type selection field 13c as selection items. When the terminal 2 receives a selection operation in the reviewer task information type selection field 13c, it acquires the selected task information type. Note that the terminal 2 may also directly receive an input of the task information type by the user through the reviewer task information type selection field 13c.

[0168] When terminal 2 receives selection (setting; input) of all fields included in reviewer information reception field 11e, it transmits the document type (e.g., synopsis), assistant type (e.g., reviewer), task information type, information about the clinical trial-related document (name, type, content, etc.), and reviewer role information to server 1. The server 1 receives the document type, assistant type, task information type, information about the clinical trial-related document, and reviewer role information transmitted from terminal 2.

[0169] The server 1 identifies the type of clinical trial-related document (e.g., synopsis) from the received information about the clinical trial-related document. Based on the identified type, the server 1 obtains format information of the corresponding document (format ID, items, display destination of items, etc.) from the format DB 156.

[0170] The server 1 acquires task information from the task information DB 157 according to the type of received document, the type of assistant, and the type of task information. The server 1 acquires prompt information including information about the received clinical trial-related document, reviewer role information, task information, and document format. The server 1 inputs the acquired prompt information into the document review model 202 and executes processing to generate review documents of the clinical trial-related documents.

[0171] The server 1 acquires the generation status of the review document in real time from the document review model 202. The server 1 transmits the generation status acquired in real time to the terminal 2. The terminal 2 receives the generation status transmitted from the server 1 and displays the received generation status in the generation status display field 11c. As shown in the figure, the generation status includes a progress status (e.g., "Questioning the document review model..."), progress data indicating the progress, or result data (e.g., "Generation of the review document was successful"), etc.

[0172] The server 1 acquires review documents of clinical trial-related documents that conform to the document format included in the prompt information output from the document review model 202. The server 1 transmits the acquired review documents to the terminal 2. The terminal 2 receives the review documents transmitted from the server 1 and displays the received review documents in the document display field 11d.

[0173] The document display field 11d includes an item display field 12h, a content display field 12i, and a review display field 12j. The review display field 12j is a display field that displays the results of a review corresponding to each item included in the document format of the clinical trial-related document.

[0174] Specifically, the terminal 2 displays each item included in the document format in an item display field 12h (item), displays the content corresponding to each item in a corresponding content display field 12i (writer), and displays the review results corresponding to each item in a corresponding review display field 12j (reviewer). The review results may be, for example, "It is consistent with the purpose of the trial," or "The indication for the test drug in the new clinical trial is certainly metastatic breast cancer," etc.

[0175] Figure 16 is an explanatory diagram showing an example of a screen for generating a review document for a clinical trial-related document using an existing assistant. The same reference numerals are used to designate the same content as in Figure 15, and the description thereof will be omitted. Figure 16 illustrates an example of a reviewer mode that is "use an existing assistant." Figure 16 includes a reviewer selection display field 13d. The reviewer selection display field 13d is a display field that displays the role of the selected reviewer.

[0176] When the terminal 2 receives a selection operation in the reviewer mode selection field 13a, it acquires the mode of the selected reviewer. If the acquired reviewer mode is "Use existing assistant", the terminal 2 acquires an assistant file corresponding to each existing reviewer from the memory unit 12 or the mass storage unit 15 of the server 1.

[0177] The terminal 2 adds the assistant file corresponding to each acquired existing reviewer as a selection item to the reviewer role information type selection field 13b. When the terminal 2 accepts the selection in the reviewer role information type selection field 13b, it acquires the corresponding assistant item from the selected assistant file. The terminal 2 displays the file name of the acquired assistant file in the reviewer selection display field 13d.

[0178] The terminal 2 adds the type of task information as a selection item to the reviewer task information type selection field 13c. The task information types include, for example, "1. Easy Review," "2. Regular Review," or "3. Thorough Review." When the terminal 2 receives a selection operation in the reviewer task information type selection field 13c, it acquires the selected type of task information.

[0179] The terminal 2 transmits information about the clinical trial-related document (such as name, type, or content), the file name of the assistant file, and the type of task information to the server 1. The server 1 receives the information about the clinical trial-related document, the file name of the assistant file, and the type of task information transmitted from the terminal 2. The server 1 extracts the role information of the reviewer by reading the corresponding assistant file based on the file name of the received assistant file. The server 1 obtains the task information from the task information DB 157 based on the type of received task information.

[0180] The server 1 identifies the type of clinical trial-related document (e.g., synopsis) from the received information about the clinical trial-related document. Based on the identified type, the server 1 obtains format information of the corresponding document (format ID, items, display destination of items, etc.) from the format DB 156.

[0181] The server 1 acquires prompt information including information about the received clinical trial-related document, reviewer role information, task information, and document format. The server 1 inputs the acquired prompt information into the document review model 202 and outputs a review document of the clinical trial-related document. The server 1 transmits the output review document to the terminal 2. The terminal 2 receives the review document transmitted from the server 1 and displays the received review document in the document display field 11d.

[0182] 17 is a flowchart showing the processing steps for generating a review document for clinical trial-related documents. The control unit 11 of the server 1 acquires multiple pieces of task information for the review document for the clinical trial-related documents from the storage unit 12 or the mass storage unit 15 (step S131). The control unit 11 then transmits the acquired multiple pieces of task information to the terminal 2 via the communication unit 13 (step S132).

[0183] The control unit 21 of the terminal 2 receives the plurality of pieces of task information transmitted from the server 1 via the communication unit 23 (step S231). The control unit 21 displays the received plurality of pieces of task information on the display unit 25 (step S232). The control unit 21 reads the clinical trial-related documents generated by the document generation model 151, thereby acquiring information about the clinical trial-related documents, including the names, types, or contents of the clinical trial-related documents (step S233).

[0184] The control unit 21 receives input of reviewer role information, including the characteristics of the reviewer who will review the clinical trial-related document, via the input unit 24 (step S234). The control unit 21 receives selection of target task information from multiple pieces of task information via the input unit 24 (step S235). The control unit 21 transmits information about the acquired clinical trial-related document, and the received review role information and task information to the server 1 via the communication unit 13 (step S236).

[0185] The control unit 11 of the server 1 receives the information about the clinical trial-related document, the reviewer's role information, and the task information transmitted from the terminal 2 via the communication unit 13 (step S133). The control unit 11 identifies the type of the clinical trial-related document (e.g., synopsis) from the received information about the clinical trial-related document (step S134). Based on the identified type, the control unit 11 obtains format information for the corresponding document from the format DB 156 of the mass storage unit 15 (step S135).

[0186] The control unit 11 acquires prompt information (step S136). The prompt information includes information about the clinical trial-related document (such as name, type, or content), reviewer role information, task information, and document format. The control unit 11 inputs the acquired prompt information into the document review model 202 (step S137) and outputs a review document of the clinical trial-related document (step S138).

[0187] The control unit 11 transmits the review document of the clinical trial-related document output from the document review model 202 to the terminal 2 via the communication unit 13 (step S139). The control unit 21 of the terminal 2 receives the review document of the clinical trial-related document transmitted from the server 1 via the communication unit 23 (step S237). The control unit 21 displays the received review document of the clinical trial-related document on the display unit 25 (step S238). The control unit 21 ends the process.

[0188] According to this embodiment, the document review model 202 can be used to generate review documents of generated clinical trial-related documents.

[0189] According to this embodiment, by automatically generating review documents of clinical trial-related documents, it is possible to prevent human oversights and provide more accurate information.

[0190] According to this embodiment, the document review model 202 can be used to generate review documents of clinical trial-related documents that suit the reviewer's characteristics.

[0191] According to this embodiment, it is possible to accept the selection of the type of task information to be targeted from a plurality of types of task information for review documents of clinical trial-related documents.

[0192] <Modification 4> This section describes a process for generating review documents of clinical trial-related documents from the content of a dialogue obtained through a dialogue between a user and the document review model 202. The server 1 acquires task information based on the content of the dialogue between the user and the document review model 202 from the task information DB 157. The content of the dialogue includes, for example, questions from the assistant to the user and answers from the assistant to the questions.

[0193] As an example, the dialogue could be: "Assistant: Do you want to emphasize the scientific or economic perspective in your review? User: I would rather emphasize the economic perspective. Assistant: I understand. Please tell me your approximate budget. User: The budget is xx billion yen. Assistant: I understand. I will review the exam to make sure it fits within that budget."

[0194] The server 1 acquires a prompt from the acquired dialogue content by utilizing a language model (e.g., the document review model 202). The server 1 inputs the acquired prompt information into the document review model 202 and outputs a review document of the clinical trial-related documents. The server 1 transmits the output review document of the clinical trial-related documents to the terminal 2. The terminal 2 receives and displays the review document of the clinical trial-related documents transmitted from the server 1.

[0195] According to this modification, it is possible to generate review documents of clinical trial-related documents from the content of the dialogue between the user and the document review model 202 .

[0196] <Modification 5> A process will be described in which there are multiple reviewers with different roles for generating review documents of clinical trial-related documents, and review documents of clinical trial-related documents are generated for each reviewer.

[0197] 18 is an explanatory diagram illustrating the process of generating review documents for clinical trial-related documents for each reviewer. As shown in the figure, there are reviewers A1, B1, and C1, and the role of reviewer A1 is to "generate review documents from a scientific perspective," the role of reviewer B1 is to "generate review documents from an economic perspective," and the role of reviewer C1 is to "generate review documents from an innovative perspective." Note that while FIG. 18 illustrates an example of three reviewers, the number of reviewers is not particularly limited.

[0198] The server 1 generates, for each reviewer, prompt information including the clinical trial-related document generated by the document generation model 151 and the reviewer's role information. For example, the prompt information for reviewer A1 includes information about clinical trial-related document AA and the role information of reviewer A1. The prompt information for reviewer B1 includes information about clinical trial-related document BB and the role information of reviewer B1. The prompt information for reviewer C1 includes information about clinical trial-related document CC and the role information of reviewer C1.

[0199] The server 1 generates review documents of clinical trial-related documents for each reviewer by providing the generated prompt information for each reviewer to the document review model 202. Note that the process of generating review documents by the document review model 202 is the same as in the second embodiment, and therefore a description thereof will be omitted.

[0200] As shown in the figure, the server 1 inputs prompt information of reviewer A1 into the document review model 202 and outputs a review document AA1 for clinical trial-related document AA. The server 1 inputs prompt information of reviewer B1 into the document review model 202 and outputs a review document BB1 for clinical trial-related document BB. The server 1 inputs prompt information of reviewer C1 into the document review model 202 and outputs a review document CC1 for clinical trial-related document CC.

[0201] According to this modification, it is possible to generate review documents of clinical trial-related documents for each reviewer with a different role.

[0202] <Modification 6> A process will be described in which there are a plurality of writers and a plurality of reviewers, and review documents are generated for clinical trial-related documents corresponding to each of the reviewers.

[0203] 19 is an explanatory diagram illustrating the process of generating review documents of clinical trial-related documents corresponding to each writer for each reviewer. Note that the description of the contents overlapping with those in FIGS. 13 and 18 will be omitted.

[0204] As shown in the figure, there are writer A, writer B, reviewer A1, reviewer B1, and reviewer C1, each with a different role. The server 1 generates prompt information including the writer's role information for each writer. The server 1 inputs the prompt information for writer A and writer B into the document generation model 151, and outputs a clinical trial-related document AA corresponding to writer A and a clinical trial-related document BB corresponding to writer B.

[0205] The server 1 acquires, for each reviewer, prompt information including information about the clinical trial-related document corresponding to each writer and the reviewer's role information. For example, for a clinical trial-related document AA corresponding to writer A, the prompt information for reviewer A1 includes information about the clinical trial-related document AA and the role information of reviewer A1, the prompt information for reviewer B1 includes information about the clinical trial-related document AA and the role information of reviewer B1, and the prompt information for reviewer C1 includes information about the clinical trial-related document AA and the role information of reviewer C1.

[0206] Furthermore, for the clinical trial-related document BB corresponding to writer B, the prompt information for reviewer A1 includes information about the clinical trial-related document BB and the role information of reviewer A1, the prompt information for reviewer B1 includes information about the clinical trial-related document BB and the role information of reviewer B1, and the prompt information for reviewer C1 includes information about the clinical trial-related document BB and the role information of reviewer C1.

[0207] The server 1 provides the generated prompt information for each reviewer to the document review model 202, thereby generating review documents of clinical trial-related documents corresponding to each writer for each reviewer.

[0208] As shown in the figure, a review document A1_AA of clinical trial-related document AA corresponding to writer A and a review document A1_BB of clinical trial-related document BB corresponding to writer B are generated for reviewer A1. Also, a review document B1_AA of clinical trial-related document AA corresponding to writer A and a review document B1_BB of clinical trial-related document BB corresponding to writer B are generated for reviewer B1. Also, a review document C1_AA of clinical trial-related document AA corresponding to writer A and a review document C1_BB of clinical trial-related document BB corresponding to writer B are generated for reviewer C1.

[0209] According to this modified example, each reviewer generates a review document for each clinical trial-related document corresponding to each writer, allowing them to check each other, thereby improving the quality of clinical trial-related documents and reducing misunderstandings or errors.

[0210] <Variation 7> A process for generating a second clinical trial-related document that is a revision of a clinical trial-related document will be described. Fig. 20 is a flowchart showing the processing steps for generating a second clinical trial-related document that is a revision of a clinical trial-related document. The control unit 21 of the terminal 2 acquires the clinical trial-related document generated by the document generation model 151 from the generation history DB 155 of the mass storage unit 15 based on the history ID of the clinical trial-related document (step S241).

[0211] The control unit 21 receives input (selection) of reviewer role information via the input unit 24 (step S242). The control unit 21 receives input of revision information for the acquired clinical trial-related document via the input unit 24 (step S243).

[0212] As an example, for a clinical trial-related document that is a synopsis, the revision information may be, "For the protocol with document ID "B-001", please recreate the synopsis based on the new objectives and study design. Objective: XXX Study design: XXX."

[0213] The control unit 21 transmits the acquired second clinical trial-related document and the accepted reviewer role information and revision information to the server 1 via the communication unit 23 (step S244). The control unit 11 of the server 1 receives the second clinical trial-related document, role information, and revision information transmitted from the terminal 2 via the communication unit 13 (step S141). The control unit 11 acquires prompt information including the received second clinical trial-related document, role information, and revision information (step S142).

[0214] As an example, the prompt information obtained may be, "You work for a pharmaceutical company and are responsible for conducting rigorous reviews. Please revise the protocol with document ID "B-001". Revision version: 2.0 Revision date: April 10, 2024 Revision details: Objective: XXX Study design: XXX."

[0215] The control unit 11 inputs the acquired prompt information into the document review model 202 (step S143), and outputs a second clinical trial-related document that is a revision of the clinical trial-related document (step S144).The control unit 11 then transmits the second clinical trial-related document output from the document review model 202 to the terminal 2 via the communication unit 13 (step S145).

[0216] The control unit 21 of the terminal 2 receives the second clinical trial-related document sent from the server 1 via the communication unit 23 (step S245). The control unit 21 displays the received second clinical trial-related document on the display unit 25 (step S246). The control unit 21 then ends the process.

[0217] According to this modification, it is possible to use the document review model 202 to generate a second clinical trial-related document that is a revision of the clinical trial-related document.

[0218] <Variation 8> The process of generating a review document of a third clinical trial-related document created by a user will be described. Figure 21 is a flowchart showing the processing steps for generating a review document of a third clinical trial-related document. The control unit 21 of the terminal 2 acquires the third clinical trial-related document (e.g., a synopsis) created by the user from the storage unit 22 (step S251). The control unit 21 may also acquire a link or file path of the third clinical trial-related document.

[0219] The control unit 21 accepts input (selection) of reviewer role information via the input unit 24 (step S252). The control unit 21 transmits the acquired third clinical trial-related document and reviewer role information to the server 1 via the communication unit 23 (step S253). The control unit 11 of the server 1 receives the third clinical trial-related document and role information transmitted from the terminal 2 via the communication unit 13 (step S151). The control unit 11 acquires the received third clinical trial-related document and prompt information including the reviewer role information (step S152).

[0220] As an example, the obtained prompt information could be, "You work for a pharmaceutical company and are responsible for conducting rigorous reviews. Please refer to the XXX paper and check the consistency between the title and objectives, the consistency between the objectives and the study design, and whether the indication for the study drug is metastatic breast cancer. Then, generate a review document according to the format including title, phase, objectives, and citations. Alternatively, the link to the synopsis created by the user: XXX."

[0221] The control unit 11 inputs the acquired prompt information into the document review model 202 (step S153), and outputs the review document of the third clinical trial-related document (step S154).The control unit 11 transmits the review document of the clinical trial-related document output from the document review model 202 to the terminal 2 via the communication unit 13 (step S155).

[0222] The control unit 21 of the terminal 2 receives the review document of the clinical trial-related documents sent from the server 1 via the communication unit 23 (step S254). The control unit 21 displays the received review document of the clinical trial-related documents on the display unit 25 (step S255). The control unit 21 accepts editing of the third clinical trial-related document by the user via the input unit 24 (step S256). The control unit 21 ends the process.

[0223] According to this modification, the document review model 202 can be used to generate review documents of third clinical trial related documents created by the user.

[0224] (Third Embodiment) The third embodiment relates to a form in which an evaluation document is generated for a clinical trial-related document generated by the document generation model 151. Note that a description of the contents overlapping with the first and second embodiments will be omitted.

[0225] In the first embodiment, the document generation model 151 is a language model that generates clinical trial-related documents. In the present embodiment, in order to distinguish it from the document generation model 151 in the first embodiment, the document generation model 151 is replaced with a document evaluation model 203. The document evaluation model 203 is a language model for generating evaluation documents of clinical trial-related documents generated by the document generation model 151.

[0226] 22 is a flowchart showing the processing procedure for generating an evaluation document for a clinical trial-related document. The control unit 21 of the terminal 2 acquires the clinical trial-related document generated by the document generation model 151 from the generation history DB 155 of the mass storage unit 15 based on the history ID of the clinical trial-related document (step S261).

[0227] The control unit 21 receives, via the input unit 24, input (selection) of role information of the evaluator who will evaluate the acquired clinical trial-related document (step S242). For example, the evaluator's role information may be, "You are an evaluator who evaluates the clinical trial-related document in order to ensure the reliability of its quality." Note that the terminal 2 may receive selection of target task information from multiple pieces of task information preset for evaluation tasks of clinical trial-related documents.

[0228] The control unit 21 acquires the evaluation index for the clinical trial-related document from the storage unit 12 or the mass storage unit 15 (step S263). The evaluation index includes, for example, accuracy, completeness, consistency, or similarity of the content.

[0229] The accuracy of content is an evaluation index for evaluating whether the information provided by a clinical trial-related document is accurate. For example, the accuracy of content may be expressed as "whether the number of sample test subjects is correctly displayed." The completeness is an evaluation index for evaluating whether the clinical trial-related document is missing necessary information. For example, the completeness may be expressed as "whether the primary endpoint is missing."

[0230] Consistency is an evaluation metric for assessing whether clinical trial-related documents are consistent. Similarity is an evaluation metric for assessing how similar a clinical trial-related document is to a training document. Similarity may be calculated using algorithms such as cosine similarity or Jaccard similarity.

[0231] The control unit 21 transmits the acquired clinical trial-related document, the accepted evaluator role information, and the acquired evaluation index to the server 1 via the communication unit 23 (step S264). The control unit 11 of the server 1 receives the clinical trial-related document, the evaluator role information, and the evaluation index transmitted from the terminal 2 via the communication unit 13 (step S161). The control unit 11 acquires prompt information including the received clinical trial-related document, the evaluator role information, and the evaluation index (step S162).

[0232] As an example, the obtained prompt information may be, "You are an evaluator who evaluates clinical trial-related documents. For the protocol with document ID "B-001", please generate an evaluation document for the clinical trial-related document based on evaluation indicators including the accuracy, completeness, consistency, or similarity of the content. Also, when evaluating based on similarity, please refer to XXX for the training document."

[0233] The control unit 11 inputs the obtained prompt information into the document evaluation model 203 (step S163), and outputs an evaluation document of the clinical trial-related document (step S164).

[0234] As an example, the evaluation document output from the document evaluation model 203 may be: "Document ID: B-001 Evaluation date: XXX Accuracy of content: The cited information is accurate. Completeness: Contains all necessary items. Consistency: The use of technical terms is consistent. Similarity: The similarity to the training document is high."

[0235] The control unit 11 transmits the evaluation document of the clinical trial-related document output from the document evaluation model 203 to the terminal 2 via the communication unit 13 (step S165). The control unit 21 of the terminal 2 receives the evaluation document of the clinical trial-related document transmitted from the server 1 via the communication unit 23 (step S265). The control unit 21 displays the received evaluation document of the clinical trial-related document on the display unit 25 (step S266). The control unit 21 ends the process.

[0236] According to this embodiment, it is possible to generate evaluation documents for clinical trial-related documents using the document evaluation model 203.

[0237] According to this embodiment, by evaluating clinical trial-related documents, it is possible to ensure the reliability of the quality of clinical trial-related documents.

[0238] Fourth Embodiment The fourth embodiment relates to a form in which the document generation model 151 or the document evaluation model 203 is retrained. Note that a description of the contents that overlap with the first to third embodiments will be omitted.

[0239] First, the process of re-learning the document generation model 151 based on feedback information will be described. The feedback information includes a first clinical trial-related document, a second clinical trial-related document, etc. The first clinical trial-related document is a clinical trial-related document selected by the user from multiple clinical trial-related documents generated by the document generation model 151. The second clinical trial-related document is a clinical trial-related document that has been revised from the clinical trial-related document generated by the document generation model 151.

[0240] The server 1 acquires the first clinical trial-related document. Specifically, the terminal 2 accepts a selection of the first clinical trial-related document to be adopted from the multiple clinical trial-related documents generated by the document generation model 151. The terminal 2 transmits the accepted first clinical trial-related document to the server 1.

[0241] The server 1 acquires the second clinical trial-related document. Specifically, the terminal 2 accepts the selection of the second clinical trial-related document generated by the processing of Variation 7. The terminal 2 transmits the accepted second clinical trial-related document to the server 1.

[0242] The server 1 stores the acquired first clinical trial related document or second clinical trial related document as a base document in the base document DB 154. The server 1 re-trains (fine-tunes) the document generation model 151 based on the feedback information.

[0243] Specifically, based on the document ID of the first clinical trial related document or the second clinical trial related document, the server 1 retrieves the first clinical trial related document or the second clinical trial related document from the base document DB 154 as text data (document data) for learning processing.

[0244] The server 1 performs a conversion process to convert the acquired text data into a format (tokens) that can be processed by the document generation model 151. That is, the server 1 performs a training process for a tokenizer that divides the acquired text data into tokens. The server 1 uses the divided tokens to retrain the document generation model 151 for a specific task (for example, sentence generation or question answering). Note that reinforcement learning may be performed on the document generation model 151.

[0245] The server 1 evaluates the performance of the re-trained document generation model 151 using evaluation indices such as precision, recall, or precision. The server 1 adjusts hyperparameters (learning rate, batch size, number of epochs, etc.) and repeats the above-mentioned re-training and hyperparameter adjustment until the performance reaches a predetermined threshold.

[0246] 23 is a flowchart showing the processing procedure when relearning the document generation model 151. The control unit 21 of the terminal 2 acquires the clinical trial-related document selected by the user as the first clinical trial-related document via the input unit 24 (step S271). For example, the control unit 21 accepts, via the input unit 24, the selection of the first clinical trial-related document by the user from the multiple clinical trial-related documents generated by the document generation model 151.

[0247] The control unit 21 accepts the selection of the second clinical trial-related document generated by the processing of Modification Example 7 via the input unit 24 (step S272). The terminal 2 transmits the accepted first clinical trial-related document and second clinical trial-related document as feedback information to the server 1 via the communication unit 23 (step S273). Note that in FIG. 23, the control unit 21 transmits both the first clinical trial-related document and the second clinical trial-related document to the server 1, but this is not limited thereto. The control unit 21 may transmit only either the first clinical trial-related document or the second clinical trial-related document to the server 1.

[0248] The control unit 11 of the server 1 receives the feedback information transmitted from the terminal 2 via the communication unit 13 (step S171). The control unit 11 stores the acquired feedback information as a base document in the base document DB 154 of the mass storage unit 15 (step S172).

[0249] Specifically, the control unit 11 assigns a document ID to the first clinical trial-related document or the second clinical trial-related document included in the feedback information, and stores the document type (e.g., synopsis), title, document content, and file information (file path, file name, etc.) in association with the assigned document ID as a single record in the base document DB 154.

[0250] Based on the feedback information, the control unit 11 performs processes such as training the tokenizer and adjusting hyperparameters, and causes the document generation model 151 to retrain (step S173).The control unit 11 then ends the process.

[0251] Next, a process of re-learning the document evaluation model 203 based on evaluation documents of clinical trial-related documents and user evaluation information on the clinical trial-related documents will be described.

[0252] The server 1 acquires the evaluation documents of the clinical trial-related documents obtained in the processing of embodiment 3 from the terminal 2. The server 1 acquires user evaluation information for the clinical trial-related documents from the terminal 2. The evaluation information is information evaluated based on evaluation indicators including accuracy, completeness, consistency, or similarity of the content. The server 1 retrains (fine-tunes) the document evaluation model 203 based on the acquired evaluation documents of the clinical trial-related documents and the user evaluation information.

[0253] Specifically, the server 1 uses the evaluation documents (first evaluations) of the clinical trial-related documents and the user's evaluation information (second evaluations) on the clinical trial-related documents to fine-tune the parameters (e.g., weights) of the document evaluation model 203 and optimize the document evaluation. The server 1 evaluates the performance of the document evaluation model 203 by calculating a loss function (e.g., cross-entropy loss) so that the first evaluation approaches the second evaluation (correct value). After calculating the loss, the server 1 backpropagates the loss to update the parameters of the document evaluation model 203, thereby re-training the document evaluation model 203.

[0254] 24 is a flowchart showing the processing steps for relearning the document evaluation model 203. The control unit 21 of the terminal 2 receives, via the input unit 24, a selection of evaluation documents for clinical trial-related documents obtained in the processing of embodiment 3 (step S281). The control unit 21 receives, via the input unit 24, input of user evaluation information for the clinical trial-related documents (step S282). The control unit 21 transmits the received evaluation documents for the clinical trial-related documents and the user evaluation information to the server 1 via the communication unit 23 (step S283).

[0255] The control unit 11 of the server 1 receives the evaluation document of the clinical trial-related document and the user's evaluation information transmitted from the terminal 2 via the communication unit 13 (step S181). Based on the received evaluation document of the clinical trial-related document and the user's evaluation information, the control unit 11 performs processes such as hyperparameter optimization and tokenizer training, and retrains the document evaluation model 203 (step S182). The control unit 11 then terminates the process.

[0256] According to this embodiment, the document generation model 151 is retrained using feedback information, thereby making it possible to improve the accuracy of generating clinical trial-related documents.

[0257] According to this embodiment, the document evaluation model 203 is retrained using evaluation documents of clinical trial-related documents and user evaluation information, thereby making it possible to improve the evaluation accuracy for clinical trial-related documents.

[0258] Fifth Embodiment A fifth embodiment relates to a form in which the prompt information, test information, and base document given to the document generation model 151 are stored in a blockchain system. Note that a description of the contents overlapping with the first to fourth embodiments will be omitted.

[0259] 25 is an explanatory diagram showing an overview of a clinical trial-related document generation system according to embodiment 5. Note that the same reference numerals are used to designate the same parts as those in FIG. 1, and the description thereof will be omitted. This embodiment includes a blockchain system 3.

[0260] The blockchain system 3 is a distributed ledger technology or a distributed network. The blockchain system 3 is composed of multiple nodes 31 that execute consensus processing. Note that while FIG. 25 shows an example in which the blockchain system 3 is composed of five nodes 31, it may be composed of an appropriate number of nodes depending on the consensus algorithm or the number of network participants.

[0261] Each node 31 holds a copy of the blockchain data through the execution of the consensus process. The blockchain system 3 generates units of data called blocks at regular intervals and stores data by linking them together like a chain.

[0262] The blockchain system 3 is autonomously managed using a peer-to-peer network and a distributed timestamp server. Because data is stored in a chain, once data in a block is stored, it is difficult to retroactively change the data. The blockchain system 3 may be public, private, or consortium type. The data unit may be individual transactions rather than blocks. Furthermore, data may be stored in a format other than a chain, such as a directed acyclic graph. For simplicity, the blockchain system 3 will be referred to as blockchain 3 below.

[0263] 26 is a flowchart showing the processing procedure for storing prompt information, test information, and a base document in the blockchain 3. The control unit 11 of the server 1 acquires data to be stored from the generation history DB 155 of the mass storage unit 15 based on the history ID of the clinical trial-related document generated by the document generation model 151 (step S191). The data to be stored includes the prompt information provided to the document generation model 151, the test information corresponding to the clinical trial-related document, and the base document.

[0264] The control unit 11 uses a cryptographic hash function to calculate a hash value of the acquired data to be stored (step S192). Specifically, the control unit 11 employs a hash function using, for example, the SHA2-256 algorithm to calculate the hash value of the data to be stored. Note that the method is not limited to a hash function, and other encryption methods may also be used.

[0265] The control unit 11 creates a transaction including the history ID of the clinical trial-related document and the calculated hash value of the data to be stored (step S193).The control unit 11 transmits the created transaction to any one of the nodes 31 in the blockchain 3 via the communication unit 13 (step S194).

[0266] The node 31 of the blockchain 3 receives the transaction sent from the server 1 (step S391). Based on the received transaction, the node 31 associates the hash of the data to be stored with the history ID of the clinical trial-related document and stores it (step S392).

[0267] The server 1 may store the clinical trial-related document generated by the document generation model 151 and a timestamp (e.g., the date and time of generation of the clinical trial-related document) in the blockchain 3, in association with the history ID of the clinical trial-related document. The server 1 may also store a review document of the clinical trial-related document, role information of the assistant corresponding to the review document, or prompt information including used task information, and a timestamp (e.g., the date and time of generation of the review document) in association with the history ID of the clinical trial-related document in the blockchain 3. The server 1 may also store an evaluation document of the clinical trial-related document, prompt information corresponding to the evaluation document, and a timestamp (e.g., the date and time of generation of the evaluation document) in association with the history ID of the clinical trial-related document in the blockchain 3.

[0268] Furthermore, the server 1 may store in the blockchain 3 a second clinical trial-related document that is a revision of the clinical trial-related document generated by Variation 7. In this case, the server 1 may store in the blockchain 3 the second clinical trial-related document, prompt information corresponding to the second clinical trial-related document, revision information, a timestamp (e.g., the date and time when the second clinical trial-related document was generated), etc., in association with the history ID of the clinical trial-related document.

[0269] In this way, the document information, prompt information, user information, etc. input into the document generation model 151, document review model 202, and document evaluation model 203 are stored in a tamper-resistant manner, and by storing document data obtained by reviewing, evaluating, or revising the clinical trial-related document generated by the document generation model 151 in the blockchain 3 in association with the clinical trial-related document, it is possible to track the evolution of the clinical trial-related document from the input information, thereby improving the quality of the clinical trial-related document and identifying areas for improvement that need to be made.

[0270] The server 1 may store the third clinical trial-related document created by the user and the review document of the third clinical trial-related document in the blockchain 3. In this case, the server 1 may store the third clinical trial-related document, the role information of the assistant corresponding to the third clinical trial-related document or prompt information including used task information, the review document of the third clinical trial-related document, and timestamps (e.g., the date and time of creation of the third clinical trial-related document and the date and time of creation of the review document of the third clinical trial-related document) in association with the history ID of the third clinical trial-related document in the blockchain 3.

[0271] In the above-mentioned storage process, the various data stored in the blockchain 3 may be the data itself or a hash value of the data.

[0272] Next, we will explain the process of issuing document NFTs (Non-Fungible Tokens) for the generated clinical trial-related documents. Document NFTs are digital data with an appraisal or certificate of ownership for clinical trial-related documents, and like crypto assets, they use blockchain technology for data management, making them difficult to tamper with or counterfeit.

[0273] 27 is a flowchart showing the processing steps for issuing a document NFT for a clinical trial-related document. The control unit 21 of the terminal 2 acquires the history ID of the clinical trial-related document for which the document NFT is to be issued (step S291). The control unit 21 then transmits an instruction to issue a document NFT for the clinical trial-related document to any one of the nodes 31 in the blockchain 3 via the communication unit 23 (step S292). The issuance instruction includes the history ID of the clinical trial-related document.

[0274] The node 31 of the blockchain 3 receives the document NFT issuance instruction sent from the server 1 (step S393). In response to the received issuance instruction, the node 31 issues the document NFT on the blockchain 3 (step S394). The node 31 stores the token ID, owner address, and token URI (Uniform Resource Identifier) ​​of the issued document NFT in a memory unit (step S395).

[0275] The token URI is an attribute that indicates the location of metadata for the document NFT. The location of the metadata is, for example, a uniform resource locator (URL) of the metadata (clinical trial-related document, base document, cited document, etc.). The metadata itself may be stored in an external database device in, for example, JSON (JavaScript Object Notation) format.

[0276] The node 31 transmits the issued document NFT to the terminal 2 (step S396). The control unit 21 of the terminal 2 receives the document NFT transmitted from the node 31 of the blockchain 3 via the communication unit 23 (step S293). The control unit 21 displays the received document NFT on the display unit 25 (step S294). The control unit 21 ends the processing.

[0277] In this embodiment, the document NFT issued by the blockchain 3 is sent directly to the terminal 2, but this is not limited to this and it may also be sent to the terminal 2 via the server 1.

[0278] In addition, when terminal 2 receives a user's selection operation of a document NFT, it may display a clinical trial-related document corresponding to the selected document NFT, test information or a base document corresponding to the clinical trial-related document, etc.

[0279] Specifically, based on the history ID of the clinical trial-related document corresponding to the selected document NFT, terminal 2 acquires the document data of the clinical trial-related document from generation history DB 155. Terminal 2 transmits the history ID of the clinical trial-related document to any one of nodes 31 in the blockchain 3. Based on the history ID transmitted from server 1, node 31 in the blockchain 3 acquires the test information and base document corresponding to the clinical trial-related document. Node 31 transmits the acquired test information and base document to terminal 2.

[0280] The terminal 2 displays the acquired document data of the clinical trial-related document, as well as the test information and base document sent from the node 31, on the screen in association with the document NFT.

[0281] In addition, when clinical trial-related documents, base documents, cited documents, etc. are stored as metadata of the document NFT, the terminal 2 acquires this data from the metadata and displays it on the screen. Specifically, the terminal 2 acquires the clinical trial-related documents, base documents, cited documents, etc. corresponding to the selected document NFT based on the metadata URL included in the token URI of the document NFT. The terminal 2 displays the acquired clinical trial-related documents, base documents, cited documents, etc. on the screen in association with the document NFT.

[0282] According to this embodiment, it is possible to store input data given to the document generation model 151 in the blockchain 3.

[0283] According to this embodiment, it is possible to issue document NFTs for clinical trial-related documents generated by the document generation model 151 through the blockchain 3.

[0284] (Embodiment 6) Embodiment 6 relates to a form in which clinical trial-related documents after a debate are identified based on clinical trial-related documents generated by multiple writers through the document generation model 151. Note that a description of the contents that overlap with embodiments 1 to 5 will be omitted.

[0285] FIG. 28 is an explanatory diagram illustrating the process of identifying clinical trial-related documents after a debate. In FIG. 28, there is a first writer who generates clinical trial-related documents using a document generation model 151, and a second writer who is different from the first writer, and the personalities (roles) of the first writer and the second writer may be different from each other. For example, the personality of the first writer may be "analysis" and the personality of the second writer may be "creation." The personalities of the first writer and the second writer may be the same. Furthermore, the document generation model 151 corresponding to the first writer and the document generation model 151 corresponding to the second writer may be the same or different.

[0286] Although an example of a first writer and a second writer has been described in FIG. 28, the present invention is not limited to this, and the number of writers with different roles is not particularly limited.

[0287] First, the process of generating clinical trial-related documents will be described. The server 1 acquires, via the terminal 2, test information related to the clinical trial, task information for generating the clinical trial-related documents, and role information for each of the first and second writers. The role information includes the personality of the writer. Note that the process of acquiring test information, task information, and role information is the same as in embodiment 1, and therefore will not be described here.

[0288] The server 1 acquires prompt information for the first writer and prompt information for the second writer based on the acquired test information, task information, and role information. The prompt information includes format information and content corresponding to each identified item. Note that the process of acquiring prompt information is the same as in embodiment 1, and therefore a description thereof will be omitted.

[0289] The server 1 inputs the acquired prompt information of the first writer into the document generation model 151 and generates (outputs) a clinical trial-related document by the first writer. The server 1 inputs the acquired prompt information of the second writer into the document generation model 151 and generates a clinical trial-related document by the second writer. As shown in the figure, the clinical trial-related document generated by the first writer is "clinical trial-related document A," and the clinical trial-related document generated by the second writer is "clinical trial-related document B."

[0290] For example, the content of Clinical Trial Related Document A may be: "Title: Phase 1 Study of Combination Therapy in the Treatment of Non-Small Cell Lung Cancer Phase: Phase 1 Indication: Patients with recurrent or metastatic non-small cell lung cancer who have received one or more prior therapies Study Treatment: Docetaxel 100 mg / m2 (every 3 weeks) and Bevacizumab 15 mg / kg (every 3 weeks) combination therapy."

[0291] The content of Clinical Trial Related Document B may be "Title: Clinical Trial of Drug for the Treatment of Non-Small Cell Lung Cancer Phase: Phase 1 Indication: Locally advanced or metastatic non-small cell lung cancer, previously treated patients Test Treatment: Docetaxel plus Bevacizumab."

[0292] Next, the debate process for the generated clinical trial-related document will be described. The server 1 acquires prompt information from the first writer. The prompt information from the first writer includes format information, content corresponding to each item, and debate instructions. The debate instructions are instructions to improve and revise the content of each item of the clinical trial-related document written by a writer (e.g., the first writer) by referring to the clinical trial-related document written by another writer (e.g., the second writer).

[0293] For example, the prompt information could be, "You are a writer generating a creative document. Based on the content of each item in the clinical trial document you generated, you refer to each item in the clinical trial document generated by the second writer and update the clinical trial document according to the format including title, phase, objectives, etc. citations. Contents of your clinical trial document: Title: Phase 1 study of combination therapy in the treatment of non-small cell lung cancer Phase: Phase 1 Indication: Patients with recurrent or metastatic non-small cell lung cancer who have received one or more prior therapies Study treatment: Docetaxel 100 mg / m2 (every 3 weeks) and Bevacizumab 15 mg / kg (every 3 weeks) combination therapy."

[0294] The server 1 generates an updated clinical trial-related document written by the first writer by providing the prompt information of the first writer and the clinical trial-related document written by the second writer to the document generation model 151. As shown in the figure, the generated updated clinical trial-related document written by the first writer is "clinical trial-related document AA."

[0295] As an example, the content of Clinical Trial Document AA may be: "Title: Phase 1 Study of Docetaxel and Bevacizumab Combination Therapy in the Treatment of Non-Small Cell Lung Cancer Phase: Phase 1 Indication: Patients with recurrent or metastatic non-small cell lung cancer who have received standard therapy Study Treatment: Docetaxel 100 mg / m2 (every 3 weeks) in combination with Bevacizumab 15 mg / kg (every 3 weeks)."

[0296] The server 1 generates an updated clinical trial-related document written by the second writer by providing the prompt information of the second writer and the clinical trial-related document written by the first writer to the document generation model 151. As shown in the figure, the updated clinical trial-related document generated by the second writer is "clinical trial-related document BB." Note that the generation process for clinical trial-related document BB is similar to the generation process for clinical trial-related document AA, and therefore a description thereof will be omitted.

[0297] The process of identifying a clinical trial-related document after a debate between a first writer and a second writer will now be described. The server 1 identifies a clinical trial-related document after a debate between a first writer and a second writer based on an updated clinical trial-related document by the first writer or an updated clinical trial-related document by the second writer. As shown in the figure, the identified clinical trial-related document after the debate is "clinical trial-related document C."

[0298] For example, the server 1 may randomly select a document from the updated clinical trial-related documents written by the first writer or the updated clinical trial-related documents written by the second writer, and identify the selected document as the post-debate clinical trial-related document. Alternatively, the server 1 may accept a user's designation through the terminal 2, and identify the designated document as the post-debate clinical trial-related document.

[0299] A language model may be used in the process of identifying clinical trial-related documents C. For example, by providing prompt information to the language model, such as "Please identify the post-debate clinical trial-related documents from the updated clinical trial-related documents by the first writer and the updated clinical trial-related documents by the second writer," it is possible to identify the post-debate clinical trial-related documents.

[0300] In the above process, one debate process is performed on each of the clinical trial-related documents by the first writer and the second writer, but the number of debates is not particularly limited. By performing multiple debate processes on the updated clinical trial-related documents, higher quality clinical trial-related documents can be generated.

[0301] The server 1 generates review documents from the identified post-debate clinical trial related documents through the document review model 202 by a reviewer who reviews the clinical trial related documents.

[0302] Specifically, the server 1 acquires reviewer role information including the reviewer's personality and task information for generating review documents for clinical trial-related documents. Note that the process for acquiring reviewer role information and task information is the same as in the first embodiment, and therefore a description thereof will be omitted.

[0303] The server 1 acquires prompt information to be provided to the document review model 202 based on information about the identified post-debate clinical trial-related document and the acquired review role information and task information. The prompt information includes information about the post-debate clinical trial-related document (such as name, type, or content), reviewer role information, task information, and document format.

[0304] The server 1 inputs the acquired prompt information into the document review model 202 and generates a review document of the clinical trial-related documents after the debate. As shown in the figure, the generated review document is "review document C1."

[0305] Note that the process is not limited to the process of generating review documents of clinical trial-related documents after the debate described above. For example, the server 1 may use the document review model 202 to generate a review document of the updated clinical trial-related document written by the first writer (e.g., "Clinical Trial-Related Document AA"), a review document of the updated clinical trial-related document written by the second writer (e.g., "Clinical Trial-Related Document BB"), or both review documents.

[0306] 29 and 30 are flowcharts showing the processing steps for identifying clinical trial-related documents after a debate. The control unit 21 of the terminal 2 acquires test information and task information related to the clinical trial via the input unit 24 (step S01). The control unit 21 acquires role information for the first writer and the second writer via the input unit 24 (step S02).

[0307] The control unit 21 transmits the acquired test information, task information, and role information of the first writer and the second writer to the server 1 via the communication unit 23 (step S03). The control unit 11 of the server 1 receives the test information, task information, and role information transmitted from the terminal 2 via the communication unit 13 (step S11).

[0308] The control unit 11 acquires prompt information of the first writer based on the received test information, task information, and role information of the first writer (step S12). The prompt information includes format information and content corresponding to each identified item. Note that the process of acquiring prompt information is the same as in embodiment 1, and therefore a description thereof will be omitted. The control unit 11 inputs the acquired prompt information of the first writer into the document generation model 151 (step S13), and generates a clinical trial-related document by the first writer (step S14).

[0309] The control unit 11 acquires prompt information of the second writer based on the received test information, task information, and role information of the second writer (step S15).The control unit 11 inputs the acquired prompt information of the second writer into the document generation model 151 (step S16), and generates a clinical trial-related document by the second writer (step S17).

[0310] The control unit 11 inputs the prompt information of the first writer and the clinical trial-related document written by the second writer into the document generation model 151 (step S18), and generates an updated clinical trial-related document written by the first writer (hereinafter referred to as an updated first clinical trial-related document) (step S19). The control unit 11 inputs the prompt information of the second writer and the clinical trial-related document written by the first writer into the document generation model 151 (step S20), and generates an updated clinical trial-related document written by the second writer (hereinafter referred to as an updated second clinical trial-related document) (step S21).

[0311] The control unit 11 identifies a post-debate clinical trial-related document between the first writer and the second writer based on the updated first clinical trial-related document and the updated second clinical trial-related document (step S22). For example, the control unit 11 randomly selects a document from the updated first clinical trial-related document and the updated second clinical trial-related document, and identifies the selected document as the post-debate clinical trial-related document.

[0312] The control unit 11 stores the updated first clinical trial-related document, the updated second clinical trial-related document, and the identified post-debate clinical trial-related document in the generation history DB 155 of the mass storage unit 15 (step S23). Note that the storage process for clinical trial-related documents is the same as in the first embodiment, and therefore a description thereof will be omitted.

[0313] The control unit 11 transmits the identified post-debate clinical trial-related documents to the terminal 2 via the communication unit 13 (step S24). The control unit 21 of the terminal 2 receives the post-debate clinical trial-related documents transmitted from the server 1 via the communication unit 23 (step S04). The control unit 21 displays the received post-debate clinical trial-related documents on the display unit 25 (step S05). The control unit 21 terminates the process.

[0314] The flowchart showing the processing steps for generating review documents of clinical trial-related documents after the debate is the same as that shown in FIG. 17, and therefore will not be shown here.

[0315] Next, the process of integrating the updated first clinical trial-related document and the updated second clinical trial-related document will be described.

[0316] 31 is an explanatory diagram showing an example of an integrated screen for clinical trial-related documents. The screen includes a first-writer document display field 14a, a second-writer document display field 14b, an editor document display field 14c, and an editing target selection field 14d.

[0317] The first writer document display field 14a is a display field that displays the updated first clinical trial-related document. The second writer document display field 14b is a display field that displays the updated second clinical trial-related document. The editor document display field 14c is a display field that displays the integrated clinical trial-related document that is an integration of the updated first clinical trial-related document and the updated second clinical trial-related document.

[0318] The editing target selection field 14d is a field for accepting selection of content corresponding to items to be edited. Items of clinical trial-related documents include title, phase, objectives, study design, indications, study treatment, or eligibility criteria.

[0319] Based on the document ID, the terminal 2 acquires the updated first clinical trial-related document and the updated second clinical trial-related document from the generation history DB 155 of the server 1. The terminal 2 displays the acquired updated first clinical trial-related document in the first writer document display field 14a, and displays the received updated second clinical trial-related document in the second writer document display field 14b.

[0320] As shown in the figure, for example, the first writer document display field 14a displays the contents corresponding to the respective items of title, phase, objective, and trial design in the updated first clinical trial-related document, and the second writer document display field 14b displays the contents corresponding to the respective items of title, phase, objective, and trial design in the updated second clinical trial-related document.

[0321] When terminal 2 receives a selection operation in editing target selection field 14d, it acquires content corresponding to the item to be edited from the updated first clinical trial-related document and the updated second clinical trial-related document. Terminal 2 transmits the acquired content to server 1. Based on the content transmitted from terminal 2, server 1 generates integrated content corresponding to the target item by integrating the content of the target item in the updated first clinical trial-related document with the content of the target item in the updated second clinical trial-related document.

[0322] As shown, for example, in the "Title" field, if the title of the updated first clinical trial related document is "Phase 3 clinical trial of milbekian for aspirin-controlled atrial fibrillation" and the title of the updated second clinical trial related document is "Phase 3 clinical trial comparing milbekian and aspirin for stroke prevention in patients with atrial fibrillation," the combined title is "Phase 3 clinical trial comparing milbekian and aspirin for stroke prevention in atrial fibrillation."

[0323] The integration process can be realized by a language model. Specifically, the server 1 generates integrated content corresponding to the target items by providing the language model with content corresponding to the target items of the updated first clinical trial-related document and the updated second clinical trial-related document, and prompt information including instructions for integrating the documents.

[0324] As an example, for a target item including "title," "phase," "objective," and "study design," the prompt information may be, "Updated clinical trial 1 related documents: Title: Phase 3 clinical trial of milbexian for aspirin-controlled atrial fibrillation Phase: Phase 3 Objective: XXX Study design: XXX Updated clinical trial 2 related documents: Title: Phase 3 clinical trial comparing milbexian with aspirin for stroke prevention in patients with atrial fibrillation Phase: Phase 3 Objective: XXX Study design: XXX Please create a consolidated title, phase, objective, and study design from the above documents."

[0325] The integration process can be realized using natural language processing (NLP) technology. For example, the server 1 extracts important keywords (e.g., aspirin control or atrial fibrillation) from documents and integrates the extracted keywords. The server 1 also standardizes extracted keywords with the same meaning (e.g., "phase 3" and "phase 3") into a standard term (e.g., "phase 3").

[0326] Although the integration process for each item has been described in FIG. 31, the process is not limited to this, and the integration process may be performed on the entire text of the clinical trial-related documents.

[0327] The server 1 transmits the integrated content obtained by the integration process to the terminal 2 in association with the target item. The terminal 2 receives the target item and the integrated content corresponding to the target item transmitted from the server 1. The terminal 2 displays the integrated content corresponding to the received target item in the editor document display field 14c in association with the target item.

[0328] 32 is a flowchart showing the processing steps for integrating clinical trial-related documents. Based on the document IDs, the control unit 21 of the terminal 2 retrieves the updated first clinical trial-related document and the updated second clinical trial-related document from the generation history DB 155 of the mass storage unit 15 of the server 1 via the communication unit 23 (step S41).

[0329] The control unit 21 displays the acquired updated first clinical trial-related document and updated second clinical trial-related document on the display unit 25 (step S42). Specifically, the control unit 21 displays the contents corresponding to each item, such as the title, phase, objectives, and study design, of the updated first clinical trial-related document and updated second clinical trial-related document on the display unit 25.

[0330] The control unit 21 receives a selection of content corresponding to the item to be edited from the updated first clinical trial-related document and the updated second clinical trial-related document via the input unit 24 (step S43). The control unit 21 transmits the received content to be edited to the server 1 via the communication unit 23 (step S44).

[0331] The control unit 11 of the server 1 receives the content of the edit target transmitted from the terminal 2 via the communication unit 13 (step S45). Based on the content of the edit target transmitted from the terminal 2, the control unit 11 integrates the content of the target items in the updated first clinical trial-related document with the content of the target items in the updated second clinical trial-related document using, for example, a language model (step S46). The control unit 11 transmits the integrated content of the integrated target items to the terminal 2 via the communication unit 13 (step S47).

[0332] The control unit 21 of the terminal 2 receives the target items and the integrated content corresponding to the target items transmitted from the server 1 via the communication unit 23 (step S48). The control unit 21 associates the integrated content corresponding to the target items with the received target items and displays it on the display unit 25 (step S49). The control unit 21 then ends the process.

[0333] Next, a process for generating an evaluation document of the updated first clinical trial-related document or the updated second clinical trial-related document using the document evaluation model 203 based on the clinical trial training document will be described.

[0334] 33 is an explanatory diagram showing an example of the evaluation screen for updated clinical trial-related documents. The screen includes a writer document display field 15a, a training document display field 15b, an evaluator document display field 15c, and an evaluation target selection field 15d.

[0335] The writer document display field 15a is a display field for displaying the updated first clinical trial-related document or the updated second clinical trial-related document. The training document display field 15b is a display field for displaying training documents for clinical trials. The evaluator document display field 15c is a display field for displaying evaluation documents for the updated first clinical trial-related document or the updated second clinical trial-related document.

[0336] The evaluation target selection field 15d is a field for accepting selection of content corresponding to items of the clinical trial-related document to be evaluated (title, phase, objectives, study design, indications, study treatment or eligibility criteria, etc.).

[0337] Although FIG. 33 illustrates an example of a process for generating an evaluation document for an updated first clinical trial-related document, the process can be similarly applied to an updated second clinical trial-related document.

[0338] Based on the document ID, the terminal 2 acquires the updated first clinical trial-related document from the generation history DB 155 of the server 1. The terminal 2 acquires, for example, a training document for the clinical trial that has been stored in advance in the memory unit 12 or the mass storage unit 15 of the server 1. The terminal 2 displays the acquired updated first clinical trial-related document in the writer document display field 15a, and displays the training document in the training document display field 15b.

[0339] As shown in the figure, the writer document display field 15a displays the contents of the updated first clinical trial-related document corresponding to each of the items of title, phase, objectives, and study design. The training document display field 15b displays the contents of the training document corresponding to each of the items of title, phase, objectives, and study design.

[0340] When terminal 2 receives a selection operation in evaluation target selection field 15d, it acquires content corresponding to the items to be evaluated (evaluation target items) from the updated first clinical trial-related document and training document. Evaluation target items include, for example, title, phase, or purpose. Terminal 2 acquires role information of the evaluator who will evaluate the updated first clinical trial-related document, evaluation indicators (accuracy, completeness, consistency, similarity, etc. of the content), and task information for the evaluation task of the updated first clinical trial-related document. Note that the process of acquiring evaluator role information, evaluation indicators, and task information is the same as in embodiment 3, so a description thereof will be omitted.

[0341] Terminal 2 transmits to server 1 the content corresponding to the evaluation target items in each of the updated first clinical trial-related documents and training documents that have been acquired, as well as the evaluator's role information, evaluation index, and task information. Server 1 acquires prompt information based on the content corresponding to the evaluation target items in each of the updated first clinical trial-related documents and training documents that have been sent from terminal 2, as well as the evaluator's role information, evaluation index, and task information. Note that the process of acquiring prompt information is the same as in embodiment 3, and therefore a description thereof will be omitted.

[0342] The server 1 inputs the acquired prompt information into the document evaluation model 203 and generates evaluation content corresponding to the evaluation target item. As shown in the figure, for example, in the "title" field, if the title of the updated first clinical trial related document is "Phase 3 clinical trial of milbexian for aspirin-controlled atrial fibrillation" and the title of the training document is "Phase 3 clinical trial of the new oral anticoagulant "elixaban" for patients with atrial fibrillation," the evaluation content for the title is "Accuracy: The purpose of the trial and the drug are accurately indicated. Completeness: The necessary information is included and complete."

[0343] Although the evaluation process for each item has been described in FIG. 33, the present invention is not limited to this, and evaluation process may be performed on the entire text of the clinical trial-related document.

[0344] The server 1 associates the evaluation content obtained by the evaluation process with the target item and transmits it to the terminal 2. The terminal 2 receives the target item and the evaluation content corresponding to the target item transmitted from the server 1. The terminal 2 associates the evaluation content corresponding to the received target item with the target item and displays it in the evaluator document display field 15c.

[0345] Fig. 34 is a flowchart showing the processing procedure for generating an evaluation document for an updated clinical trial-related document. Note that Fig. 34 explains an example of the processing for generating an evaluation document for an updated first clinical trial-related document, but the same can be applied to the evaluation document for an updated second clinical trial-related document.

[0346] The control unit 21 of the terminal 2 acquires the updated first clinical trial-related document from the generation history DB 155 of the mass storage unit 15 of the server 1 via the communication unit 13 based on the document ID (step S51). The control unit 21 acquires, for example, a training document for the clinical trial stored in advance in the storage unit 22 (step S52).

[0347] The control unit 21 displays the acquired updated first clinical trial-related document and training document on the display unit 25 (step S53). The control unit 21 accepts selections of content corresponding to the evaluation target items in each of the updated first clinical trial-related document and training document via the input unit 24 (step S54). The control unit 21 acquires evaluator role information, evaluation indexes, and task information (step S55). Note that the process of acquiring evaluator role information, evaluation indexes, and task information is the same as in embodiment 3, and therefore will not be described here.

[0348] The control unit 21 transmits the contents corresponding to the evaluation items in the acquired updated first clinical trial-related documents and training documents, as well as the training documents, role information, evaluation indexes, and task information, to the server 1 via the communication unit 23 (step S56). The control unit 11 of the server 1 receives the contents corresponding to the evaluation items, the training documents, role information, evaluation indexes, and task information transmitted from the terminal 2 via the communication unit 13 (step S61).

[0349] The control unit 11 acquires prompt information based on the received content corresponding to the evaluation item, the training document, the role information, the evaluation index, and the task information (step S62). Note that the process of acquiring prompt information is the same as in the third embodiment, and therefore a description thereof will be omitted.

[0350] The control unit 11 inputs the acquired prompt information into the document evaluation model 203 (step S63), and outputs (generates) evaluation contents corresponding to the evaluation target items (step S64). The control unit 11 associates the evaluation contents obtained by the evaluation process with the evaluation target items and transmits them to the terminal 2 via the communication unit 13 (step S65).

[0351] The control unit 21 receives the evaluation items and the evaluation contents corresponding to the evaluation items transmitted from the server 1 via the communication unit 23 (step S57). The control unit 21 associates the received evaluation items with the evaluation contents corresponding to the evaluation items on the display unit 25 (step S58). The control unit 21 ends the process.

[0352] According to this embodiment, it is possible to identify clinical trial related documents after debate based on clinical trial related documents generated through the document generation model 151 by multiple writers with different roles.

[0353] According to this embodiment, the document review model 202 can be used to generate review documents of clinical trial-related documents after debate.

[0354] According to this embodiment, it is possible to generate an evaluation document of an updated clinical trial-related document using the document evaluation model 203.

[0355] According to this embodiment, it is possible to generate an integrated clinical trial-related document based on the updated first clinical trial-related document and the updated second clinical trial-related document.

[0356] <Modification 9> A process of generating a clinical trial related document corresponding to a clinical trial outline created by a user through the document generation model 151 based on a clinical trial related document written by a first writer or a clinical trial related document written by a second writer will be described.

[0357] Fig. 35 is an explanatory diagram illustrating the process of generating clinical trial-related documents corresponding to a clinical trial outline. Fig. 35 shows a first writer who generates clinical trial-related documents using a document generation model 151, and a user. Note that Fig. 35 illustrates an example of the first writer, but the process can also be applied to a second writer, or both the first and second writers.

[0358] The process of generating clinical trial-related documents by the first writer is the same as that shown in FIG. 28, and therefore will not be described here.

[0359] The server 1 acquires the clinical trial outline created by the user and reference documents from the terminal 2. The clinical trial outline is a document created by the user that includes an overview of the clinical trial to be conducted. The reference documents are documents such as clinical practice guidelines or academic papers that provide information necessary for the document generation model 151 to generate clinical trial-related documents.

[0360] The server 1 provides the clinical trial-related document generated by the first writer, along with the acquired clinical trial outline and references, to the first writer, thereby generating a clinical trial-related document corresponding to the clinical trial outline through the document generation model 151.

[0361] Specifically, the server 1 generates a clinical trial-related document corresponding to the clinical trial outline by providing the document generation model 151 with a generation instruction, the clinical trial-related document written by the first writer, the acquired clinical trial outline, and references. The generation instruction is a command to generate a clinical trial document corresponding to the clinical trial outline created by the user by referencing the contents of each item in the clinical trial-related document written by the first writer and the references. As shown in the figure, the generated clinical trial-related document is "clinical trial-related document D."

[0362] FIG. 36 is a flowchart showing the processing steps for generating clinical trial related documents corresponding to a clinical trial outline.

[0363] The control unit 21 of the terminal 2 acquires the test information, role information, and task information of the first writer (step S79). The control unit 21 acquires the clinical trial outline and reference literature created by the user from the storage unit 22 (step S71). The control unit 21 transmits the acquired test information, role information, and task information of the first writer, as well as the acquired clinical trial outline and reference literature, to the server 1 via the communication unit 23 (step S72).

[0364] The control unit 11 of the server 1 receives the first writer's study information, role information, and task information, as well as the clinical trial outline and references, sent from the terminal 2 via the communication unit 13 (step S73). The control unit 11 executes the processes of steps S12 to S14. The control unit 11 inputs the generation instruction, the first writer's clinical trial document, the clinical trial outline, and references into the document generation model 151 (step S74), and generates a clinical trial-related document corresponding to the clinical trial outline (step S75).

[0365] The control unit 11 transmits the clinical trial-related documents corresponding to the generated clinical trial outline to the terminal 2 via the communication unit 13 (step S76). The control unit 21 of the terminal 2 receives the clinical trial-related documents transmitted from the server 1 via the communication unit 23 (step S77). The control unit 21 displays the received clinical trial-related documents on the display unit 25 (step S78). The control unit 21 then terminates the process.

[0366] According to this modified example, it is possible to generate clinical trial related documents corresponding to the clinical trial outline created by the user through the document generation model 151 based on clinical trial related documents written by the first writer or the second writer.

[0367] (Embodiment 7) Embodiment 7 relates to a form in which a plurality of reviewers generate review documents of the updated clinical trial-related documents generated in embodiment 6 through the document review model 202. Note that a description of the contents overlapping with embodiments 1 to 6 will be omitted.

[0368] For example, the document review model 202 may include a first reviewer who reviews clinical trial-related documents and a second reviewer who is different from the first reviewer, and the personalities (roles) of the first reviewer and the second reviewer may be different from each other. For example, the first reviewer may have a "strict" personality, and the second reviewer may have a "lenient" personality. The personalities of the first reviewer and the second reviewer may be the same. Furthermore, the document review model 202 corresponding to the first reviewer and the document review model 202 corresponding to the second reviewer may be the same or different.

[0369] 37 is a flowchart showing the processing steps for generating a review document of an updated clinical trial-related document by multiple reviewers. The control unit 11 of the server 1 retrieves the updated clinical trial-related document from the generation history DB 155 of the mass storage unit 15 based on the document ID (step S81).

[0370] The control unit 11 generates a review document by the first reviewer (hereinafter referred to as a first review document) for the acquired updated clinical trial-related document through the document review model 202 (step S82). Specifically, the control unit 11 acquires information about the clinical trial-related document, including the name, type (e.g., synopsis), or content of the updated clinical trial-related document. The control unit 11 acquires reviewer role information, including the characteristics of the first reviewer who will review the clinical trial-related document, through the terminal 2. The control unit 11 acquires task information for generating a review document for the clinical trial-related document through the terminal 2.

[0371] The control unit 11 acquires prompt information to be provided to the document review model 202 based on the acquired information about the clinical trial-related document, the reviewer's role information, and the task information. Note that the process of acquiring prompt information is the same as in embodiment 2, and therefore a description thereof will be omitted. The prompt information includes information about the updated clinical trial-related document (such as name, type, or content), the reviewer's role information, task information, and the document format. The control unit 11 inputs the acquired prompt information into the document review model 202 and generates a first review document.

[0372] The control unit 11 generates a review document by the second reviewer (hereinafter referred to as the second review document) for the acquired updated clinical trial-related document through the document review model 202 (step S83). Note that the process for generating the second review document is the same as the process for generating the first review document, and therefore a description thereof will be omitted.

[0373] The control unit 11 inputs the prompt information of the first reviewer and the second review document into the document review model 202 (step S84) and generates an updated review document by the first reviewer (hereinafter referred to as the updated first review document) (step S85). Specifically, the control unit 11 acquires the prompt information of the first reviewer. The prompt information of the first reviewer includes document format information, information about the first review document (such as the name, type, or review content of each item), and debate instructions. The debate instructions are commands to refer to the review document by another reviewer (e.g., the second reviewer), state opinions such as for or against the review content of each item in the review document by the reviewer (e.g., the first reviewer), and provide evidence.

[0374] For example, the prompt information for the first reviewer is, "You work for a pharmaceutical company and are responsible for conducting rigorous reviews. Based on the review content of each item in the first review document you generated, please refer to the review content of each item in the second review document generated by the second reviewer and debate. Based on your opinion for or against the review content of each item, please generate an updated first review document following the format including title, phase, objectives, and citations. Contents of your first review document: Clinical trial synopsis of a drug for treating non-small cell lung cancer Item Writer Reviewer Title: XXX Match Phase: Phase 1 XXX Objective: XXX XXX Study design: XXX XXX Indication: XXX XXX Study treatment: Docetaxel plus Bevacizumab XXX Control treatment: Docetaxel XXX Eligibility criteria: XXX XXX Endpoint: XXX XXX Sample size: XXX XXX Citation: XXX It could also be "XXX".

[0375] The control unit 11 inputs the second reviewer's prompt information and the first review document into the document review model 202 (step S86) and generates an updated review document by the second reviewer (hereinafter referred to as the updated second review document) (step S87). Note that the process for generating the updated second review document is the same as the process for generating the updated first review document, so a description thereof will be omitted. The control unit 11 then terminates the process.

[0376] Although the example of the first reviewer and the second reviewer has been described in FIG. 37, the present invention is not limited to this, and the number of reviewers with different roles is not particularly limited.

[0377] In the above process, one debate process is performed for each of the first review document and the second review document, but the number of debate processes is not particularly limited. By performing multiple debate processes on each review document, it is possible to generate a higher quality review document.

[0378] According to this embodiment, a plurality of reviewers with different roles can generate review documents of updated clinical trial-related documents through the document review model 202.

[0379] (Embodiment 8) Embodiment 8 relates to a form in which processing is performed to optimize a document generation model 151 according to training documents and evaluation results obtained by evaluating clinical trial-related documents. Note that a description of content that overlaps with embodiments 1 to 7 will be omitted. The optimization processing includes methods such as improving prompts, fine-tuning, adding RAG (Retrieval-Augmented Generation), or adding a machine learning algorithm. By performing optimization processing on the document generation model 151, the performance of the document generation model 151 is improved.

[0380] Prompt improvement is a technique for designing or adjusting input prompts for the document generation model 151 to output more appropriate responses. Fine tuning is a technique for further training a pre-trained document generation model 151 for a specific dataset, task, etc., to improve the performance of the document generation model 151.

[0381] Adding RAG is a technique in which the document generation model 151 acquires information from a knowledge database or the like and combines the acquired information with a task. By using adding RAG, the document generation model 151 can generate more accurate responses.

[0382] Adding a machine learning algorithm is a method of incorporating a new machine learning algorithm into an existing document generation model 151 to improve the performance or capabilities of the document generation model 151. For example, a machine learning algorithm may be added for a specific task to optimize the document generation model 151.

[0383] In this embodiment, an example of fine tuning will be described, but the present invention can be similarly applied to other optimization learning techniques.

[0384] 38 is a flowchart showing the processing procedure for fine-tuning the document generation model 151. The control unit 11 of the server 1 generates a clinical trial-related document using the document generation model 151 (step S91). Note that the process for generating the clinical trial-related document is the same as in embodiment 1, and therefore a description thereof will be omitted. Note that the updated first clinical trial-related document or the updated second clinical trial-related document in embodiment 6 may also be used.

[0385] The control unit 11 acquires the training document from the storage unit 12 or the mass storage unit 15 (step S92). The control unit 11 may acquire the training document from the terminal 2 or an external information processing device via the communication unit 13. The control unit 11 determines whether or not an editing request for the clinical trial-related document generated by the document generation model 151 has been received from the terminal 2 via the communication unit 13 (step S93).

[0386] If the control unit 11 has not received an editing request (NO in step S93), the control unit 11 proceeds to the processing of step S95 described below. If the control unit 11 has received an editing request (YES in step S93), the control unit 11 acquires editing information for the clinical trial-related document from the terminal 2 via the communication unit 13 (step S94). The editing information includes, for example, the format of the clinical trial-related document or content corresponding to the item to be edited.

[0387] The control unit 11 calculates the similarity between the acquired training document and the clinical trial-related document (step S95). Note that the similarity calculation process is the same as in embodiment 1, and therefore a description thereof will be omitted. Note that, when the control unit 11 receives editing information in the process of step S94, it calculates the similarity between the acquired training document and the clinical trial-related document to which the editing information has been added.

[0388] The control unit 11 determines whether the calculated similarity is equal to or less than a predetermined value (step S96). If the calculated similarity exceeds the predetermined value (NO in step S96), the control unit 11 fine-tunes the document generation model 151 (step S97) and returns to the processing of step S95.

[0389] The control unit 11 can evaluate the clinical trial-related document generated by the document generation model 151 by comparing the superiority or inferiority of the document with the training document, for example, using a BLEU score or the like. In this case, the control unit 11 fine-tunes the document generation model 151 based on the superiority or inferiority evaluation result obtained by the superiority or inferiority evaluation. Specifically, when a difference is confirmed in the superiority or inferiority evaluation result, the control unit 11 fine-tunes the document generation model 151 and returns to the processing of step S95.

[0390] Alternatively, the control unit 11 may evaluate the clinical trial-related documents by using a language model. Note that the evaluation process of the clinical trial-related documents based on the language model is the same as that in the first embodiment, and therefore a description thereof will be omitted.

[0391] Specifically, the control unit 11 converts clinical trial-related documents into text data (document data) for learning processing. The control unit 11 performs a conversion process to convert the converted text data into a format (tokens) that can be processed by the document generation model 151. That is, the control unit 11 performs a learning process for a tokenizer that divides the converted text data into tokens. The control unit 11 uses the divided tokens to re-train the document generation model 151 for a specific task (for example, sentence generation or question answering). Note that reinforcement learning may be performed on the document generation model 151.

[0392] If the calculated similarity is equal to or less than the predetermined value (YES in step S96), the control unit 11 ends the process.

[0393] In the above-described process, an example of evaluating clinical trial-related documents using similarity has been described, but the present invention is not limited to this. For example, when verification data for evaluating the performance of the document generation model 151 is prepared, the document generation model 151 after fine tuning may be evaluated using the verification data.

[0394] According to this embodiment, it is possible to fine-tune the document generation model 151 according to the training documents and the evaluation results obtained by evaluating clinical trial-related documents.

[0395] According to this embodiment, it is possible to fine-tune the document generation model 151 based on editing information for clinical trial-related documents.

[0396] (Embodiment 9) Embodiment 9 relates to a form in which literature data is searched from a database storing data or literature related to test information. Note that a description of the contents overlapping with Embodiments 1 to 8 will be omitted.

[0397] The server 1 can search for relevant literature data from a database that stores data or literature related to trial information. The server 1 provides the searched data or literature together with the acquired trial information related to the clinical trial and the base document related to the trial information to the document generation model 151, thereby generating clinical trial-related documents related to the clinical trial.

[0398] The process for generating clinical trial-related documents in this embodiment is based on a technology called Retrieval-Augmented Generation (RAG). RAG is a process for generating clinical trial-related documents by combining a pre-trained generative language model with information acquired from an external knowledge source (e.g., a database storing data or literature related to trial information). By utilizing trial information related to clinical trials and information such as data or literature, highly reliable clinical trial-related documents can be generated.

[0399] 39 is an explanatory diagram showing an example of a database search screen. The search screen includes a DB selection button 16a, a period input field 16b, a keyword input field 16c, a send button 16d, a search result display field 16e, a document selection check box 16f, a view button 16g, a delete button 16h, a details display field 16i, a test selection field 16j, a template selection field 16k, an attachment category selection field 16l, and an attach button 16m.

[0400] The DB selection button 16a is a button for accepting the selection of a database that stores data or literature related to trial information. Examples of databases include "CTG (ClinicalTrials.gov)" and "PubMed." "CTG" is a database that provides information about clinical trials, primarily storing clinical trial design, progress, results, and participant data. "PubMed" is a database that provides a wide range of research papers and academic articles related to medicine, life science, and health, and provides information on the latest research results, treatments, and drug effects related to trials.

[0401] The database is not limited to "CTG" and "PubMed", but may also include, for example, Scopus or Embase.

[0402] The period input field 16b is a field for accepting input of the publication date period of the data or document to be searched. The keyword input field 16c is a field for accepting input of search keywords. The send button 16d is a button for sending the entered publication date period and keywords to the server 1. The search result display field 16e is a display field for displaying the search results.

[0403] The document selection checkbox 16f is a checkbox for accepting the selection of data or documents to be attached from the multiple documents found. The view button 16g is a button for displaying detailed information about the selected document in the details display field 16i. The delete button 16h is a button for deleting data or documents to be deleted from the multiple displayed documents. The details display field 16i is a display field for displaying detailed information about the document.

[0404] The exam selection field 16j is a field for accepting the selection of the exam (study) to which data or documents are to be attached. The template selection field 16k is a field for accepting the selection of a template for the exam document to be attached. The attachment category selection field 16l is a field for accepting the selection of an attachment category for the exam to be attached. The attachment button 16m is a button for sending the selected documents, exam information (e.g., exam ID), template, and attachment category to the server 1.

[0405] When the terminal 2 receives a touch (click) operation on the DB selection button 16a, it acquires information about the database selected by the user (for example, a database ID or database name). When the terminal 2 receives an input operation in the period input field 16b, it acquires the publication date period of the document entered by the user. When the terminal 2 receives an input operation in the keyword input field 16c, it acquires the keyword entered by the user.

[0406] When the terminal 2 receives a touch operation of the send button 16d, it transmits information about the documents to be searched, including the acquired document database information (for example, the database ID or database name), the publication date period of the document, and keywords, to the server 1. The server 1 receives the document database information, the publication date period of the document, and keywords transmitted from the terminal 2.

[0407] The server 1 searches for the relevant document data from the selected database based on the received document information. Specifically, the server 1 identifies a server (e.g., a CTG database server) or platform that provides the database based on the received document database information. The server 1 obtains search results for the data or document to be searched for via the identified server or platform based on the publication date period and keywords of the received document.

[0408] The server 1 displays the search results for the acquired documents in the search result display field 16e. The search results may include one or more documents. As shown in the figure, for each document, the index, document ID, title, status, and other information are displayed in the search result display field 16e. The status indicates the progress of the document, and includes, for example, "Completed," "Not Yet Recruiting," "Recruiting," or "Terminated."

[0409] The contents of each document may include, for example, the document abstract, conditions, study type (e.g., intervention), sponsor, or update date and time.

[0410] When the terminal 2 receives a check operation on the document selection checkbox 16f, it acquires information about the document to be attached (e.g., document ID). When the terminal 2 receives a touch operation on the view button 16g, it displays detailed information about the corresponding document in the details display field 16i. The detailed information about the document includes the index, document ID, title, status, abstract, study type (e.g., experimental treatment), and conditions.

[0411] A condition refers to a medical condition or disease that is the subject of a clinical trial or study. Conditions include, for example, Lower Urinary Tract Symptoms (LUTS). LUTS refers to lower urinary tract symptoms and includes symptoms related to urination (such as urinary frequency, urgency, or urinary incontinence).

[0412] The detailed information of the document may include the official title, abbreviated title, lead sponsor, first submitted date, study start date, last updated date, study phase, etc.

[0413] When the terminal 2 receives a touch operation of the delete button 16h, it deletes the corresponding document from the plurality of documents displayed in the search result display field 16e.

[0414] When the terminal 2 receives a selection operation in the study selection field 16j, it acquires information about the study to be attached (e.g., a study ID).When the terminal 2 receives a selection operation in the template selection field 16k, it acquires a template of the study document to be attached (e.g., a synopsis, a protocol, or a statistical analysis plan).

[0415] When the terminal 2 receives a selection operation in the attachment category selection field 161, it acquires an attachment category that indicates the type of data or literature to be attached. Attachment categories for trials include, for example, past trial information (ClinicalTrials.gov), medical literature (Pubmed), model documents (Model_doc), compound profiles (Product Profiles), and treatment guidelines (Treatment Guidelines).

[0416] When the terminal 2 receives a touch operation of the attachment button 16m, it transmits the searched document data, the study ID to be attached, the template, and the attachment category to the server 1. The document data includes the document ID and detailed information of the document (index, document ID, title, status, abstract, study type, conditions, etc.).

[0417] The server 1 receives the literature data, trial information, templates, and attachment categories transmitted from the terminal 2. Based on the received trial ID, the server 1 acquires trial information related to the clinical trial from the trial information DB 152. Based on the acquired trial information, the server 1 acquires base documents related to the trial information from the base document DB 154.

[0418] The server 1 provides the retrieved literature data to the document generation model 151 together with the acquired trial information related to the clinical trial and the base document related to the trial information, thereby generating a clinical trial-related document related to the clinical trial.

[0419] Specifically, the server 1 provides the document generation model 151 with prompt information including literature data, study information (such as the name, phase, objectives, or study design of the clinical trial), writer role information, a base document (such as the name or content of the base document), a template, and an attachment category. For example, the prompt information may be, "You are a writer who generates creative documents. For the clinical trial of XXX, please extract only the main items from the instructions required for document generation, and generate a standardized synopsis for the clinical trial using the synopsis template including the title, phase, objectives, etc., citations, the study information, the base document, and the literature data below. Title: Clinical trial of a drug for treating non-small cell lung cancer Phase: Phase 1 Indication: XXX Test treatment: Docetaxel plus Bevacizumab Document ID of base document: B-001 Literature data Literature ID: NCT064225 Title: EEP in patients With... Status: NOT_YET_RECRUITING Abstract: XXX XXX."

[0420] The server 1 inputs the above-mentioned prompt information into the document generation model 151 to generate a synopsis for the clinical trial.

[0421] 40 is a flowchart showing the processing steps for searching for literature data from a database. The control unit 21 of the terminal 2 receives a selection of a database storing literature related to the test information through the input unit 24 (step S301). The control unit 21 also receives input of information related to the literature (such as the publication date and time period of the literature, keywords, etc.) through the input unit 24 (step S302).

[0422] The control unit 21 searches for relevant document data from the selected database based on the information about the document (step S303). Specifically, the control unit 21 transmits the acquired information about the document to the server 1. The server 1 identifies the server or platform that provides the database based on the document database information transmitted from the terminal 2. The server 1 obtains search results for the documents to be searched for via the identified server or platform based on the publication date and keywords of the document. The server 1 transmits the search results for the acquired documents to the terminal 2.

[0423] The control unit 21 may also directly search for the relevant document data from the server or platform that provides the selected database based on the information about the document, without going through the server 1 .

[0424] The control unit 21 displays the search results sent from the server 1 on the display unit 25 (step S304). The control unit 21 accepts a selection of a document to be attached from the searched documents at the input unit 24 (step S305). The control unit 21 accepts a selection of a target study at the input unit 24 (step S306). The control unit 21 transmits the selected document data and the study ID of the study to the server 1 via the communication unit 23 (step S307).

[0425] The control unit 11 of the server 1 receives the literature data and the study ID transmitted from the terminal 2 via the communication unit 13 (step S401). Based on the received study ID, the control unit 11 acquires study information related to the clinical trial from the study information DB 152 of the mass storage unit 15 (step S402). Based on the acquired study information, the control unit 11 searches the base document DB 154 of the mass storage unit for a base document related to the study information (step S403).

[0426] The control unit 11 acquires prompt information including the retrieved literature data, the acquired clinical trial information, and base documents related to the clinical trial information (step S404). The control unit 11 inputs the acquired prompt information into the document generation model 151 (step S405), and outputs clinical trial-related documents related to the clinical trial (step S406).

[0427] The control unit 11 stores the output clinical trial-related document in the generation history DB 155 of the mass storage unit 15 (step S407). The control unit 11 transmits the clinical trial-related document output from the document generation model 151 to the terminal 2 via the communication unit 13 (step S408). The control unit 21 of the terminal 2 receives the clinical trial-related document transmitted from the server 1 via the communication unit 23 (step S308) and displays the received clinical trial-related document on the display unit 25 (step S309). The control unit 21 ends the process.

[0428] According to this embodiment, by providing literature data retrieved from a database storing literature related to trial information to the document generation model 151 together with the trial information related to the clinical trial and the base document related to the trial information, it becomes possible to generate clinical trial-related documents related to the clinical trial.

[0429] (Embodiment 10) Embodiment 10 relates to a form in which a final clinical trial-related document is generated through debate or revision by multiple writers based on document generation models 151 provided by different companies. Note that a description of the contents that overlap with embodiments 1 to 9 will be omitted.

[0430] In this embodiment, document generation models 151 provided by different companies are used, and a final clinical trial-related document is generated by debating or revising each other. Specifically, the terminal 2 accepts selection of a first writer based on a first document generation model (first language model) 151 provided by a first company, and a second writer based on a second document generation model (second language model) 151 provided by a second company different from the first company.

[0431] The terminal 2 transmits the received first writer and second writer to the server 1. The server 1 generates a first clinical trial-related document through the first writer using a first document generation model 151 that uses the test information and the base document. The server 1 generates a second clinical trial-related document through the second writer using a second document generation model 151 that uses the test information and the base document. Note that the process of generating clinical trial-related documents is the same as in embodiment 1, so a description thereof will be omitted.

[0432] The server 1 generates a final clinical trial-related document through debate or revision by the first writer and the second writer based on the generated first clinical trial-related document and second clinical trial-related document.

[0433] In this embodiment, an example of a first document generation model 151 provided by a first company and a second document generation model 151 provided by a second company will be described, but this is not limited to this, and the present invention can also be applied to a process in which document generation models 151 provided by three or more different companies are used to debate or revise each other.

[0434] 41 is an explanatory diagram showing an example of a screen for accepting debate settings. The screen includes a test information input field 17a, a first document generation model selection field 17b, a second document generation model selection field 17c, a first writer selection field 17d, a second writer selection field 17e, and a document generation button 17f.

[0435] The test information input field 17a is a field for receiving input of test information. The first document generation model selection field 17b is a field for receiving selection of a first document generation model 151 provided by a first company. The second document generation model selection field 17c is a field for receiving selection of a second document generation model 151 provided by a second company.

[0436] The first writer selection field 17d is a field for accepting the selection of a first writer based on the first document generation model 151. The second writer selection field 17e is a field for accepting the selection of a second writer based on the second document generation model 151. The document generation button 17f is a button for generating the final clinical trial-related document.

[0437] When the terminal 2 receives an input operation in the test information input field 17a, it acquires the test information entered by the user and a template for a clinical trial-related document. The test information includes, for example, an index, a title, or an indication. The terminal 2 may also acquire the test information of the corresponding test from the test information DB 152 of the server 1 based on the test ID.

[0438] When the terminal 2 receives a selection operation in the first document generation model selection field 17b, it acquires information (e.g., model ID or model name) of the selected first document generation model 151. When the terminal 2 receives a selection operation in the second document generation model selection field 17c, it acquires information of the selected second document generation model 151.

[0439] The first document generation model 151 and the second document generation model 151 are models provided by different companies and are constructed based on different architectures, learning data, etc. For example, the first document generation model 151 is a GPT (Generative Pre-trained Transformer) model provided by OpenAI (registered trademark), Inc., and may be GPT-4.0 or GPT-4.5, etc. The second document generation model 151 may be a Gemini model provided by Google (registered trademark).

[0440] In addition to GPT and Gemini, it is also possible to use, for example, the LLAMA (Large Language Model Meta AI) language model provided by Meta (registered trademark), the Cloude language model provided by Anthropic, or various language models available through the Azure OpenAI Service provided by Microsoft (registered trademark).

[0441] When the terminal 2 receives a selection operation in the first writer selection field 17d, it acquires first writer information (for example, the writer's assistant ID) selected by the user. The first writer is based on the first document generation model 151 selected in the first document generation model selection field 17b. When the terminal 2 receives a selection operation in the second writer selection field 17e, it acquires second writer information selected by the user. The second writer is based on the second document generation model 151 selected in the second document generation model selection field 17c.

[0442] When the terminal 2 receives a touch operation of the document generation button 17f, it transmits the input test information, the selected first document generation model 151, the second document generation model 151, the first writer information, and the second writer information to the server 1. The server 1 receives the test information, the first document generation model 151, the second document generation model 151, the first writer information, and the second writer information transmitted from the terminal 2.

[0443] The server 1 acquires prompt information to be provided to the first document generation model 151 that uses the test information and the base document based on the received test information and first writer information. The server 1 inputs the acquired prompt information into the first document generation model 151 to generate a first clinical trial-related document.

[0444] The server 1 acquires prompt information to be provided to a second document generation model 151 that uses the test information and the base document based on the received test information and second writer information. The server 1 inputs the acquired prompt information into the second document generation model 151 to generate a second clinical trial-related document.

[0445] The process of obtaining prompt information and the process of generating clinical trial-related documents are the same as those in the first embodiment, and therefore a description thereof will be omitted.

[0446] The server 1 generates a final clinical trial-related document through debate or revision by the first writer and the second writer based on the generated first clinical trial-related document and second clinical trial-related document. Specifically, the server 1 obtains prompt information including the first clinical trial-related document, the second clinical trial-related document, instructions for debate or revision, and instructions for generating the final clinical trial-related document.

[0447] The debate or revision instructions may be, for example, "Please debate the differences between the first clinical trial-related document and the second clinical trial-related document." The instructions to generate the final clinical trial-related document may be, for example, "Please revise the first version that you wrote based on the first clinical trial-related document and the second clinical trial-related document to create the final clinical trial-related document."

[0448] As an example, the prompt information may be, "First clinical trial related document: XXX Second clinical trial related document: XXX Please debate the differences between the first clinical trial related document and the second clinical trial related document. Based on the first clinical trial related document and the second clinical trial related document, please create a final clinical trial related document that reflects the content of the debate. Based on the first clinical trial related document and the second clinical trial related document, please revise the first version you wrote and create a final clinical trial related document."

[0449] The server 1 inputs the acquired prompt information into the document generation model 151 to generate the final clinical trial-related document.

[0450] 42 is a flowchart showing the processing procedure for generating a final clinical trial-related document. The control unit 21 of the terminal 2 accepts input of trial information by the user through the input unit 24 (step S311). The control unit 21 accepts selection of a first document generation model 151 provided by a first company through the input unit 24 (step S312). The control unit 21 accepts selection of a second document generation model 151 provided by a second company through the input unit 24 (step S313).

[0451] The control unit 21 receives the user's selection of the first writer through the input unit 24 (step S314). The control unit 21 receives the user's selection of the second writer through the input unit 24 (step S315). The control unit 21 transmits the test information, the first document generation model 151, the second document generation model 151, the first writer information, and the second writer information to the server 1 through the communication unit 23 (step S316).

[0452] The control unit 11 of the server 1 receives the test information, the first document generation model 151, the second document generation model 151, the first writer information, and the second writer information transmitted from the terminal 2 through the communication unit 13 (step S411).

[0453] The control unit 11 generates a first clinical trial-related document based on the received test information and first writer information using the first document generation model 151 that uses the test information and the base document (step S412).The control unit 11 generates a second clinical trial-related document based on the received test information and second writer information using the second document generation model 151 that uses the test information and the base document (step S413).

[0454] The control unit 11 acquires prompt information including the first clinical trial-related document, the second clinical trial-related document, instructions for debate or revision, and instructions for generating a final clinical trial-related document (step S414). The control unit 11 inputs the acquired prompt information into the document generation model 151 (step S415), and outputs the final clinical trial-related document (step S416).

[0455] The control unit 11 stores the output final clinical trial-related document in the generation history DB 155 of the mass storage unit 15 (step S417). The control unit 11 transmits the final clinical trial-related document output from the document generation model 151 to the terminal 2 via the communication unit 13 (step S418).

[0456] The control unit 21 of the terminal 2 receives the final clinical trial-related document sent from the server 1 via the communication unit 23 (step S317). The control unit 21 displays the received final clinical trial-related document on the display unit 25 (step S318). The control unit 21 then ends the process.

[0457] Furthermore, the debate or revision process by multiple writers using document generation models 151 provided by different companies as described above can also be applied to embodiment 6. In embodiment 6, a clinical trial-related document after debate or revision is identified based on a clinical trial-related document generated by multiple writers through the document generation model 151, but unlike this embodiment, the use of the same document generation model 151 is not limited, and multiple document generation models 151 provided by different companies may be used to identify a clinical trial-related document after debate or revision.

[0458] According to this embodiment, it is possible to generate the final clinical trial-related documents through debate or revision by multiple writers based on document generation models 151 provided by different companies.

[0459] (Embodiment 11) Embodiment 11 relates to a form in which a clinical trial-related document is generated by a document generation model 151 based on review results by multiple reviewers of the first version of a clinical trial-related document. Note that a description of the contents overlapping with embodiments 1 to 10 will be omitted.

[0460] The server 1 acquires the first version of the clinical trial-related document. The first version of the clinical trial-related document may be a document generated by the document generation model 151, or may be a document created manually by a user. The server 1 uses the document review model 202 to output review results by multiple reviewers for the first version of the clinical trial-related document. The server 1 generates the clinical trial-related document using the document generation model 151 based on the review results of each reviewer that have been output.

[0461] 43 is an explanatory diagram showing an example of a display screen for review results by multiple reviewers. The screen includes a document reception field 18a, a first document review model selection field 18b, a second document review model selection field 18c, a first reviewer selection field 18d, a second reviewer selection field 18e, a writer selection field 18f, an instruction selection field 18g, a first round button 18h, a second round button 18i, a history button 16j, an item field 18k, a first review result display field 18l, a second review result display field 18m, a final document display field 18n, and a review button 18o.

[0462] The document reception field 18a is a field for receiving first-edition clinical trial-related documents and trial information. The first document review model selection field 18b is a field for receiving selection of a first document review model 202 provided by a first company. The second document review model selection field 18c is a field for receiving selection of a second document review model 202 provided by a second company.

[0463] The first reviewer selection field 18d is a field for receiving the selection of a first reviewer based on the first document review model 202. The second reviewer selection field 18e is a field for receiving the selection of a second reviewer based on the second document review model 202. The writer selection field 18f is a field for receiving the selection of a writer. The instruction selection field 18g is a field for receiving the selection of a review method. The review method includes, for example, an ethical review or a technical review.

[0464] The first round button 18h is a button for conducting a first review and is used when a reviewer reviews the first version of a clinical trial-related document. The second round button 18i is a button for conducting a second review and is used for re-evaluation after the first review. By conducting multiple reviews, the first version of a clinical trial-related document can be repeatedly evaluated, thereby improving the accuracy or validity of the document.

[0465] The history button 16j is a button for viewing the history data of the review results. The item column 18k is a display column for displaying the item names of the clinical trial-related documents. The first review result display column 18l is a display column for displaying the review results of the first reviewer. The second review result display column 18m is a display column for displaying the review results of the second reviewer.

[0466] The final document display field 18n is a display field that displays the revised clinical trial-related document generated by the document generation model 151 based on the review results of each reviewer. The review button 18o is a button for reviewing the clinical trial-related document to be reviewed.

[0467] When the terminal 2 receives the setting operation in the document reception field 18a, it acquires the trial information (trial ID, indications or trial treatment, etc.) input by the user and the uploaded first version of the clinical trial-related document. The terminal 2 displays the items of the acquired first version of the clinical trial-related document (e.g., index, literature ID, title, status, abstract, study type and conditions, etc.) in the item field 18k.

[0468] When the terminal 2 receives a selection operation in the first document review model selection field 18b, it acquires information (e.g., a model ID or a model name) of the selected first document review model 202. When the terminal 2 receives a selection operation in the second document review model selection field 18c, it acquires information of the selected second document review model 202.

[0469] The first document review model 202 and the second document review model 202 are models provided by different companies and are constructed based on different architectures, learning data, etc. For example, the first document review model 202 may be the GPT model provided by OpenAI, and the second document review model 202 may be the Gemini model provided by Google (registered trademark).

[0470] When the terminal 2 receives a selection operation in the first reviewer selection field 18d, it acquires information about the first reviewer selected by the user (e.g., the reviewer's assistant ID). The first reviewer is based on the first document review model 202 selected in the first document review model selection field 18b. When the terminal 2 receives a selection operation in the second reviewer selection field 18e, it acquires information about the second reviewer selected by the user. The second reviewer is based on the second document review model 202 selected in the second document review model selection field 18c.

[0471] When the terminal 2 receives a selection operation in the writer selection field 18f, it acquires writer information (e.g., the writer's assistant ID) selected by the user. When the terminal 2 receives a selection operation in the instruction selection field 18g, it acquires the selected review method.

[0472] When the terminal 2 receives a touch operation of the review button 18o, it transmits information about the review to the server 1. The information about the review includes the received test information and the first version of the clinical trial-related document, as well as the selected first document review model 202, second document review model 202, first reviewer information, second reviewer information, writer information, and review method, etc.

[0473] The server 1 receives information about the reviews sent from the terminal 2. Based on the received information about the reviews, the server 1 outputs the review results of the first version of the clinical trial-related document by multiple reviewers to the terminal 2. The server 1 generates the clinical trial-related document using the document generation model 151 based on the review results of each reviewer.

[0474] Specifically, the server 1 acquires prompt information to be provided to the first document review model 202 based on the received test information, review method (e.g., logical review), first version of the clinical trial-related document, and first reviewer information. The server 1 inputs the acquired prompt information into the first document review model 202, and generates a first review document of the first version of the clinical trial-related document as a review result.

[0475] Based on the received test information, review method, first-version clinical trial-related document, and second reviewer information, the server 1 acquires prompt information to be provided to the second document review model 202. The server 1 inputs the acquired prompt information into the second document review model 202, and generates a second review document of the first-version clinical trial-related document as a review result.

[0476] The process of obtaining prompt information and the process of generating review documents for clinical trial-related documents are the same as those in the second embodiment, and therefore a description thereof will be omitted.

[0477] The server 1 acquires prompt information to be provided to the document generation model 151 based on the first version of the clinical trial-related document, writer information, and the generated first and second review documents. The prompt information includes the first version of the clinical trial-related document, writer information, the first and second review documents, and instructions for generating the revised clinical trial-related document.

[0478] As an example, the prompt may be, "First version of the clinical trial document: XXX First review document: XXX Second review document: XXX You are a writer who produces creative documents. Please refer to the first review document and the second review document and create a revised clinical trial document based on the first version of the clinical trial document."

[0479] The server 1 inputs the acquired prompt information into the document generation model 151 to generate a revised clinical trial-related document. The server 1 transmits the review results of each reviewer and the revised clinical trial-related document to the terminal 2.

[0480] Terminal 2 receives the review results of each reviewer and the revised clinical trial-related document. Terminal 2 displays the review results of the first reviewer in a first review result display field 181 and the review results of the second reviewer in a second review result display field 18m. Terminal 2 displays the revised clinical trial-related document in a final document display field 18n.

[0481] As shown in the figure, the first review result display field 18l and the second review result display field 18m display the review contents (such as suggestions or improvements) corresponding to each item included in the first version of the clinical trial-related document, while the final document display field 18n displays the revised contents of each item included in the first version of the clinical trial-related document.

[0482] When the terminal 2 receives a touch operation of the first round button 18h, it displays the first review results and the revised clinical trial-related document generated based on the first review results. When the terminal 2 receives a touch operation of the second round button 18i, it conducts another review based on the first review results, and displays the second review results and the revised clinical trial-related document generated based on the second review results.

[0483] Although an example of two reviews has been described in FIG. 43, the present invention is not limited to this, and three or more reviews may be used.

[0484] Furthermore, the terminal 2 can store the first review document, the second review document, and the revised clinical trial-related document in the storage unit 22 or the mass storage unit 15 of the server 1. When the terminal 2 receives a touch operation of the history button 16j, it displays a history screen (not shown) that includes the results of reviews performed in the past, the clinical trial-related document after revision based on the review results, the creation date and time, etc.

[0485] Furthermore, terminal 2 can accept editing operations for clinical trial-related documents after past revisions on the history screen. This allows users to view past clinical trial-related documents and re-edit them as necessary, making it easier to update or improve documents. Furthermore, edited documents can be reviewed again and further revised based on the review results.

[0486] 44 is a flowchart showing the processing steps for generating a clinical trial-related document based on multiple review results. The control unit 21 of the terminal 2 accepts an upload of a first version of the clinical trial-related document via the input unit 24 (step S321). The control unit 21 accepts a selection of a first document review model 202 provided by a first company via the input unit 24 (step S322). The control unit 21 accepts a selection of a second document review model 202 provided by a second company via the input unit 24 (step S323).

[0487] The control unit 21 receives the user's selection of the first reviewer through the input unit 24 (step S324). The control unit 21 receives the user's selection of the second reviewer through the input unit 24 (step S325). The control unit 21 receives the user's selection of the writer through the input unit 24 (step S326).

[0488] The control unit 21 transmits information about the review, including the first version of the clinical trial-related document, the first document review model 202, the second document review model 202, the first reviewer information, the second reviewer information, and the writer information, to the server 1 via the communication unit 23 (step S327). The control unit 11 of the server 1 receives the information about the review transmitted from the terminal 2 via the communication unit 13 (step S421).

[0489] The control unit 11 generates a first review document of the first version of the clinical trial-related document using the first document review model 202 based on the received first version of the clinical trial-related document and the first reviewer information (step S422).The control unit 11 generates a second review document of the first version of the clinical trial-related document using the second document review model 202 based on the received first version of the clinical trial-related document and the second reviewer information (step S423).

[0490] The control unit 11 generates a revised clinical trial-related document using the document generation model 151 based on the first version of the clinical trial-related document, the writer information, and the generated first and second review documents (step S424).

[0491] The control unit 11 stores the first review document, the second review document, and the revised clinical trial-related document in the memory unit 12 or the mass memory unit 15 (step S425).The control unit 11 transmits the first review document, the second review document, and the revised clinical trial-related document to the terminal 2 via the communication unit 13 (step S426).

[0492] The control unit 21 of the terminal 2 receives the first review document, the second review document, and the revised clinical trial-related document via the communication unit 23 (step S328). The control unit 21 displays the received first review document, the second review document, and the revised clinical trial-related document on the display unit 25 (step S329). The control unit 21 then ends the process.

[0493] Furthermore, the review process by multiple reviewers on the first version of clinical trial-related documents as described above can also be applied to embodiment 7. In embodiment 7, multiple reviewers generate review documents of clinical trial-related documents through the document review model 202, but instead of using the same document review model 202 as in this embodiment, multiple document review models 202 provided by different companies may be used to generate review documents of clinical trial-related documents.

[0494] According to this embodiment, it is possible to generate a revised clinical trial-related document using the document generation model 151 based on the review results of the first version of the clinical trial-related document by multiple reviewers.

[0495] Although the base document is used in each of the above-described embodiments, the present invention is not limited to this. Clinical trial-related documents relating to clinical trials can be generated by the document generation model 151 that uses trial information relating to clinical trials without using a base document.

[0496] Specifically, the server 1 acquires trial information related to the clinical trial (e.g., trial information, writer role information, task information, etc.). The server 1 inputs the acquired trial information related to the clinical trial and prompt information including format information, etc., into the document generation model 151 without using a base document, and outputs clinical trial-related documents related to the clinical trial (protocol, patient informed consent form, statistical analysis plan, etc.).

[0497] The embodiments disclosed herein are illustrative in all respects and should not be considered limiting. The scope of the present invention is defined by the claims, not by the above meaning, and is intended to include all modifications within the meaning and scope of the claims.

[0498] The matters described in each embodiment can be combined with each other. Furthermore, the independent claims and dependent claims described in the claims can be combined with each other in any combination, regardless of the reference format. Furthermore, the claims use a format in which a claim references two or more other claims (multiple claim format), but this is not limited to this. A multiple claim (multi-multi claim) that references at least one other multiple claim may also be used.

[0499] REFERENCE SIGNS LIST 1 Information processing device (server) 11 Control unit 12 Storage unit 13 Communication unit 14 Reading unit 15 Large-capacity storage unit 151 Document generation model (language model) 152 Test information DB 153 Role information DB 154 Base document DB 155 Generation history DB 156 Format DB 157 Task information DB 1a Portable storage medium 1b Semiconductor memory 1P Control program 2 Information processing terminal (terminal) 21 Control unit 22 Storage unit 23 Communication unit 24 Input unit 25 Display unit 2P Control program 3 Blockchain system (blockchain) 31 Node 201 Base document search model 202 Document review model 203 Document evaluation model

Claims

1. A program that causes a computer to execute the following processes: acquire trial information related to a clinical trial; acquire base documents related to the acquired trial information; and generate clinical trial-related documents related to the clinical trial using a language model that uses the trial information and the base documents.

2. The program according to claim 1, wherein a review document of the generated clinical trial-related document is generated by providing prompt information including role information of a reviewer who will review the generated clinical trial-related document to the language model.

3. The program according to claim 2, wherein the program has a plurality of reviewers with different roles, and generates review documents of the clinical trial-related documents for each reviewer by providing prompt information including role information of each reviewer to the language model.

4. The program described in claim 2, which generates a second clinical trial-related document that is revised from the clinical trial-related document by providing prompt information including the reviewer's role information and revision information for the clinical trial-related document to the language model.

5. The program according to claim 1 or 2, which has a plurality of writers with different roles in generating the clinical trial-related documents, and generates the clinical trial-related documents for each writer by providing prompt information including role information of each writer to the language model.

6. The program according to claim 5, which has a plurality of writers and a plurality of reviewers, and generates review documents of clinical trial-related documents corresponding to each writer for each reviewer by providing prompt information including role information of each reviewer to the language model.

7. The program according to claim 1 or 2, which calculates the similarity between the generated clinical trial-related documents and clinical trial training documents, and outputs the calculated similarity.

8. A program as described in claim 1 or 2, which generates an evaluation document for the generated clinical trial-related document by providing the language model with prompt information including role information of an evaluator who evaluates the generated clinical trial-related document and evaluation indicators for the clinical trial-related document.

9. The program according to claim 2, which acquires a third clinical trial-related document created by a user, generates a review document for the third clinical trial-related document by providing the acquired third clinical trial-related document and prompt information including role information of the reviewer to the language model, and accepts edits to the third clinical trial-related document.

10. The program according to claim 5, further comprising: receiving input of role information of the writer including the personality of the writer; and providing prompt information including the received role information of the writer to the language model.

11. The program according to claim 2, further comprising: receiving input of role information of the reviewer including the reviewer's personality; and providing prompt information including the received role information of the reviewer to the language model.

12. The program according to claim 1 or 2, which searches for base documents that are highly similar to the clinical trial-related documents.

13. The program according to claim 12, wherein base documents similar to the clinical trial-related documents are searched for by providing prompt information including role information of a researcher searching for base documents with high similarity to the language model.

14. The program according to claim 1 or 2, wherein the clinical trial-related documents include a synopsis, a protocol, a patient consent form, a statistical analysis plan, a statistical analysis report, a clinical study report, or a paper or clinical practice guideline written based on the results of the clinical trial.

15. The program according to claim 1 or 2, wherein the prompt information for generating the clinical trial-related document includes the name, phase, and task information of the clinical trial required for generating the clinical trial-related document.

16. The program according to claim 2, wherein the prompt information for generating the review document includes the clinical trial-related document to be reviewed and reviewer role information.

17. The program described in claim 2, which outputs multiple pieces of task information for the clinical trial-related document or the review document, accepts selection of target task information from the multiple pieces of output task information, and generates the clinical trial-related document or the review document by providing prompt information including the accepted task information to the language model.

18. The program according to claim 17, which generates clinical trial-related documents from the content of a dialogue obtained by a dialogue between a user and the language model based on the task information.

19. The program according to claim 4, wherein the clinical trial-related document adopted by the user or the second clinical trial-related document is acquired as feedback information, and the acquired feedback information is stored as the base document.

20. The program according to claim 19, wherein the language model is retrained based on the feedback information.

21. The program according to claim 1 or 2, which accepts edits of the clinical trial-related document by multiple users.

22. The program according to claim 8, further comprising: acquiring user evaluation information for the clinical trial-related documents; and re-training the language model based on the evaluation documents for the clinical trial-related documents and the acquired user evaluation information.

23. The program according to claim 1 or 2, wherein the prompt information provided to the language model, the test information, and the base document are stored in a blockchain system.

24. The program described in claim 1 or 2, which issues a document NFT (Non-Fungible Token) for the generated clinical trial-related document through a blockchain system.

25. An information processing method comprising: acquiring trial information related to a clinical trial; acquiring base documents related to the acquired trial information; and generating clinical trial-related documents related to the clinical trial using a language model that uses the trial information and the base documents.

26. An information processing device including a control unit, wherein the control unit acquires test information related to a clinical trial, acquires base documents related to the acquired test information, and generates clinical trial-related documents related to the clinical trial using a language model that uses the test information and the base documents.

27. The program described in claim 1, wherein a clinical trial-related document written by the first writer and a clinical trial-related document written by the second writer are generated through the language model by a first writer and a second writer having different roles, and updated clinical trial-related documents written by the first writer and updated clinical trial-related documents written by the second writer are generated through the language model by providing the clinical trial-related document written by the second writer to the first writer and the clinical trial-related document written by the first writer to the second writer.

28. The program according to claim 27, which identifies a clinical trial-related document after a debate between the first and second writers based on an updated clinical trial-related document by the first writer or an updated clinical trial-related document by the second writer.

29. The program according to claim 28, wherein a reviewer who reviews the clinical trial-related documents after the identified debate passes the language model to generate a review document.

30. The program described in claim 27 or 28, which acquires a clinical trial outline and references created by a user, and generates a clinical trial-related document corresponding to the clinical trial outline through the language model by providing the clinical trial-related document written by the first writer or the clinical trial-related document written by the second writer, along with the acquired clinical trial outline and references, to the first writer or the second writer.

31. A program as described in claim 27 or 28, which generates an integrated clinical trial-related document by integrating the updated clinical trial-related document written by the first writer and the updated clinical trial-related document written by the second writer.

32. The program described in claim 27 or 28, which generates an evaluation document of the updated clinical trial-related document written by the first writer or the updated clinical trial-related document written by the second writer through the language model based on the updated clinical trial-related document written by the first writer or the updated clinical trial-related document written by the second writer and a clinical trial training document.

33. The program described in claim 27 or 28, which generates a review document by a first reviewer and a review document by a second reviewer through the language model by a first reviewer and a second reviewer who have different roles in reviewing updated clinical trial-related documents, and generates an updated review document by the first reviewer and an updated review document by the second reviewer through the language model by providing the review document by the second reviewer to the first reviewer and the review document by the first reviewer to the second reviewer.

34. An information processing method comprising: generating clinical trial-related documents related to a clinical trial using a language model that uses test information related to the clinical trial and base documents related to the test information; acquiring training documents related to the clinical trial; evaluating the clinical trial-related documents based on the acquired training documents; and performing processing to optimize the language model according to the evaluation results.

35. The information processing method according to claim 34, further comprising the step of optimizing the language model until the similarity between the training documents and the clinical trial-related documents is equal to or less than a predetermined value.

36. An information processing method according to claim 34 or 35, further comprising receiving editing information for the generated clinical trial-related documents, and performing processing to optimize the language model based on the received editing information.

37. An information processing method as described in claim 34 or 35, wherein a clinical trial-related document written by the first writer and a clinical trial-related document written by the second writer are generated through the language model by a first writer and a second writer having different roles, and by providing the clinical trial-related document written by the second writer to the first writer and the clinical trial-related document written by the first writer to the second writer, updated clinical trial-related documents written by the first writer and updated clinical trial-related documents written by the second writer are generated through the language model.

38. A program as described in claim 1, which accepts the selection of a database storing literature related to trial information, accepts input of information on the data or literature to be searched from the selected database, searches for relevant information from the selected database based on the accepted information on the data or literature, and generates clinical trial-related documents related to the clinical trial by providing the searched information to the language model together with the acquired trial information related to the clinical trial and base documents related to the trial information.

39. The program of claim 1, which accepts selection of a first writer based on a first language model provided by a first company and a second writer based on a second language model provided by a second company different from the first company, generates a first clinical trial-related document through the first writer using the first language model using the trial information and base documents, generates a second clinical trial-related document through the second writer using the second language model using the trial information and base documents, and generates a final clinical trial-related document through debate or revision by the first and second writers based on the generated first and second clinical trial-related documents.

40. The program described in claim 1, which acquires a first version of a clinical trial-related document, outputs review results of the first version of the clinical trial-related document by multiple reviewers using the language model, and generates a clinical trial-related document using the language model based on the review results of each reviewer that have been output.

41. A program that causes a computer to execute the process of acquiring trial information related to a clinical trial and generating clinical trial-related documents related to said clinical trial using a language model that uses the acquired trial information.

Images