Report draft generation system and report draft generation method

Abstract

This involves consolidating information scattered across multiple information systems and quickly creating draft reports. [Solution] The report draft generation system 1 comprises a data collection unit 24 that collects multiple types of data from multiple systems, a GMP platform 2 that centrally stores the multiple types of data collected by the data collection unit 24, and a report draft generation unit 15 that references the GMP platform 2 and item data mapping definition information 12, maps the data from the GMP platform 2 to each item using a large-scale language model 17, generates content to be described in the item, and generates a report draft.

Description

Technical Field

[0001] The present invention relates to a report draft generation system and a report draft generation method.

Background Art

[0002] GMP (Good Manufacturing Practice), also translated as Good Manufacturing Practice, indicates the standards to be observed in each process of manufacturing and quality control in order to ensure the quality and safety of pharmaceuticals and the like. Although the specific content of GMP varies by country or region, it commonly aims to "ensure that products consistently meet quality standards." In Japan, it is defined as an ordinance of the Ministry of Health, Labour and Welfare, the Ordinance on Standards for Manufacturing Control and Quality Control of Pharmaceuticals and Quasi-Drugs, Ministry of Health, Labour and Welfare Ordinance No. 179 of December 24, 2004 (hereinafter, "GMP Ordinance"). In the United States, it corresponds to 21 CFR Part 210 / 211 of the FDA, and in Europe, it corresponds to EudraLex Volume 4. These standards are designed to centralize the "final determination of the feasibility of manufacturing and sales" in the quality assurance department.

[0003] It is a core operation for the quality assurance department in pharmaceutical manufacturing to confirm manufacturing records, test records, deviation records, etc. and take responsibility for shipment determination. The Japanese GMP Ordinance specifies the following obligations.

[0004] Article 12 (Quality Control): It is necessary to confirm manufacturing records and test records and verify that the quality conforms to the predetermined standards. Article 13 (Shipment Determination): Before shipment, it is necessary to review manufacturing records, test records, deviations, changes, CAPA, etc. and establish a system in which the quality responsible person determines the feasibility of shipment. Article 17 (Quality Assurance System): The quality assurance department is independent from the manufacturing department and has the ultimate responsibility and decision-making authority regarding quality.

[0005] Therefore, the quality assurance department must comprehensively review various records to ensure compliance with standards. The items that the quality assurance department checks in practice are as follows. The quality assurance department will scrutinize the following records in order to comply with the GMP regulations.

[0006] Manufacturing execution system data: Did the product come from the work procedure? Were there any deviations or omissions in the records? Test data: Does it conform to the specifications? Are there any signs of fraud, such as retesting or recalculation? Deviation records, CAPA records, change records: Are there any quality impact events in the lot in question, or have they been appropriately corrected?

[0007] Patent Document 1 describes an invention comprising: a search unit 121 that searches for related documents corresponding to a string of characters included in a document creation command entered by a document creator; a generation unit 122 that generates a draft document based on the document creation command and the searched related documents; and an extraction unit 123 that extracts items to be checked by a document verifier within the draft document based on the content of the draft document generated by the document creator from the draft document. [Prior art documents] [Patent Documents]

[0008] [Patent Document 1] Japanese Patent Publication No. 2025-062312 [Overview of the project] [Problems that the invention aims to solve]

[0009] Traditionally, manufacturing-related data has been a mix of exported data from various systems and paper documents, requiring significant effort to integrate and structure this data. Specifically, it is essential to locate the storage locations of paper documents, collect the originals, transcribe the necessary sections, and digitize them. Data collection requires cooperation not only from the quality assurance department but also from the manufacturing and quality control departments, necessitating cross-departmental coordination and resource allocation. As a result, when problems occur, necessary information cannot be accessed immediately, making it difficult to pinpoint the cause. Furthermore, because data collection, structuring, and transcription are reliant on manual processes, the overall effort and resources required are enormous, leading to a decline in operational efficiency.

[0010] The collected manufacturing-related data is formatted into graphs, lists, and tables, including comparisons between measured values ​​and specification values. However, because it is necessary to consider the results and transcribe them into a predetermined format, data analysis and documentation still require a significant amount of effort.

[0011] In the current situation where data is dispersed across multiple information systems, judgment errors can occur due to insufficient or incomplete collection of necessary information. While interlocks and check logic can eliminate misjudgments in judgments that are completed within the system, human error can occur in judgments that use information from outside the system. In addition, the process of detecting omissions during judgment also relies on human intervention, so complete prevention has not been achieved.

[0012] In document creation, it is common practice to revise existing documents each year while referring to them. While this helps reduce the burden of creating new documents, it presents challenges in ensuring that the latest information is not overlooked and maintaining consistency.

[0013] Lot verification Checking all records using checklists requires switching between electronic screens and paper documents, resulting in considerable effort. Furthermore, verifying deviations or changes within a lot is time-consuming because related data is siloed across multiple systems.

[0014] Annual Review In the annual review process, the biggest burden lies in collecting necessary data from multiple data sources. While the data formatting procedure itself can be defined in advance, a significant amount of effort is spent on aggregating and verifying the consistency of the original data, leading to a decrease in work efficiency.

[0015] According to the invention described in Patent Document 1, a new draft document can be generated based on related documents. However, there is no description whatsoever regarding the appropriate selection of related documents, which are the information necessary for creating the draft document.

[0016] Therefore, the present invention aims to create a draft report based on information distributed across multiple information systems. [Means for solving the problem]

[0017] To solve the aforementioned problems, the present invention provides a report draft generation system comprising: a data collection unit that collects multiple types of data from multiple systems; and an information infrastructure that centrally stores the multiple types of data collected by the data collection unit. A verification unit checks the consistency of each data collected by the data collection unit and notifies the system from which the data collection unit collected the information of the consistency information; a verification unit targets data for which consistency is satisfied, associates each item included in the draft report with the type of data in the information infrastructure related to the content to be described in that item, and provides item data mapping definition information including prompt template identifiers and format definition IDs associated with each item. Refer to the aforementioned information infrastructure Then, configure the input based on the prompt template identifier and format definition ID, The system is characterized by comprising: a report draft generation unit that generates a draft report by mapping data from the information infrastructure to each item using a large-scale language model and generating content to be described in the item.

[0018] The present invention's method for generating a draft report comprises the steps of: a data collection unit collecting multiple types of data from multiple systems; and an information infrastructure centrally storing the data collected by the data collection unit. The verification unit verifies the consistency of each piece of data collected by the data collection unit and notifies the system from which the data collection unit collected the information of the consistency information, The report draft generation department, The data that satisfies consistency requirements will be used as the subject of the draft report, and each item included in the draft report will be associated with the type of data in the information infrastructure related to the content to be described in that item, and item data mapping definition information including the prompt template identifier and format definition ID associated with each item will be used. Refer to the aforementioned information infrastructure Then, configure the input based on the prompt template identifier and format definition ID, The method is characterized by comprising the steps of: using a large-scale language model to map data from the information infrastructure to each item, generating content to be described in the item, and generating a draft report. Other means will be described within the descriptions of embodiments for carrying out the invention.

Advantages of the Invention

[0019] According to the present invention, it becomes possible to create a draft report based on information distributed among a plurality of information systems.

Brief Description of the Drawings

[0020] [Figure 1] It is a configuration diagram of a draft report generation system according to this embodiment. [Figure 2A] It is a flowchart of data collection processing from a manufacturing execution system. [Figure 2B] It is a flowchart of data collection processing from a test record system. [Figure 2C] It is a flowchart of data collection processing from a quality management system. [Figure 3] It is a flowchart of draft report generation processing. [Figure 4] It is a flowchart of visualization processing. [Figure 5A] It is a schematic diagram showing a manufacturing record database. [Figure 5B] It is a detailed item diagram showing a manufacturing record database. [Figure 6A] It is a schematic diagram showing a test record database. [Figure 6B] It is a detailed item diagram showing a test record database. [Figure 7A] It is a schematic diagram showing a quality management database. [Figure 7B] It is a schematic diagram showing change information. [Figure 7C] It is a schematic diagram showing a deviation record. [Figure 7D] It is a schematic diagram showing CAPA information. [Figure 8] It is a schematic diagram showing item data mapping definition information.

Modes for Carrying Out the Invention

[0021] Hereafter, embodiments for carrying out the present invention will be described in detail with reference to the figures. Figure 1 is a diagram showing the configuration of the report draft generation system 1 according to this embodiment. The draft report generation system 1 comprises a draft report generation unit 15, a visualization unit 16, and a large-scale language model 17. The draft report generation system 1 is composed of a GMP platform 2, legal and regulatory guideline information 11, item data mapping definition information 12, a past review report database 13, and GMP ministerial ordinance information 14. However, it is not limited to this, and the draft report generation system 1 may also generate draft reports by calling an external server that embodies the large-scale language model. The report draft generation system 1 of this embodiment is implemented on-premises, but a system for generating report drafts may be implemented in various forms, such as cloud SaaS provision, and is not limited to this.

[0022] The GMP Platform 2 is an information infrastructure comprising a manufacturing record database 21, a test record database 22, a quality control database 23, a data acquisition unit 24, and a review unit 25. The GMP Platform 2 connects to multiple external information sources related to pharmaceuticals, centralizes and structures the necessary data, and stores it in the manufacturing record database 21, the test record database 22, and the quality control database 23.

[0023] The data collection unit 24 acquires multiple types of data from various external information sources, namely the manufacturing execution system 31, the test record system 32, and the quality control system 33, and normalizes them into a common format. The data collection unit 24 unifies and structures the multiple types of data collected from the manufacturing execution system 31 and stores them in the manufacturing record database 21. The verification unit 25 verifies the data collected by the data collection unit 24 from the manufacturing execution system 31 by comparing it with the data to determine whether there are any defects or omissions. If there are defects or omissions in the data collected from the manufacturing execution system 31, the verification unit 25 sends the result of the determination to the manufacturing execution system 31 as a flag. This reduces human error and improves the reliability of the judgment.

[0024] The data collection unit 24 centralizes and structures multiple types of data collected from the test record system 32 and stores them in the test record database 22. The verification unit 25 verifies the data collected by the data collection unit 24 from the test record system 32 to determine whether there are any deficiencies or omissions. If the verification unit 25 finds any deficiencies or omissions in the data collected from the test record system 32, it sends the result of this determination to the test record system 32 as a flag. This reduces human error and improves the reliability of the judgment.

[0025] The data collection unit 24 centralizes and structures multiple types of data collected from the quality control system 33 and stores them in the quality control database 23. The verification unit 25 verifies the data collected by the data collection unit 24 from the quality control system 33 and determines whether or not there are any defects or omissions. If the verification unit 25 finds any defects or omissions in the data collected from the quality control system 33, it sends the result of the determination to the quality control system 33 as flag information. Upon receiving this flag information, the quality control system 33 can reduce human error and improve the reliability of its judgment by informing the manager of the presence of defects or omissions. This allows the GMP platform 2 to centrally store multiple types of data from unified and structured external information sources. Furthermore, the verification unit 25 can verify the integrity of the data, reducing data deficiencies and omissions.

[0026] The data collection unit 24 centrally structures the data necessary for decision-making and stores it in the manufacturing record database 21, the test record database 22, and the quality control database 23. This streamlines the data collection process and facilitates information tracing, contributing to improved product quality. Furthermore, since data is collected automatically from multiple systems without human intervention, the cause of problems can be quickly investigated when they occur.

[0027] Judgment and Review When data is distributed across multiple systems, there is a risk of judgment errors due to insufficient or incomplete collection of necessary information. While interlocks and check logic can eliminate misjudgments in judgments that are completed within a single system, human error is more likely to occur when using information from outside the system.

[0028] Therefore, the verification unit 25 automatically compares and verifies the data collected by the data collection unit 24 and transmits the verification results as flag information to the relevant system. This reduces human error in the data, thereby improving the reliability of the draft report.

[0029] Legal and Regulatory Guideline Information 11 is a set of guidelines that summarizes the technical and organizational requirements that must be followed at each stage of pharmaceutical research and development, manufacturing, distribution, and post-market safety management. Legal and Regulatory Guideline Information 11 translates the mandatory provisions stipulated by laws such as the Pharmaceuticals and Medical Devices Act into concrete operational levels, and is used by authorities as a basis for review, inspection, and administrative guidance. In practice, it is treated as a legally binding standard.

[0030] The item data mapping definition information 12 is definition information that maps various data to each item that makes up the draft report, and will be explained in Figure 8 below. The past review report database 13 is a database that stores draft reports previously generated by the draft report generation unit 15.

[0031] GMP Ministerial Ordinance Information 14 formally refers to the "Ministerial Ordinance Concerning Standards for Manufacturing and Quality Control of Pharmaceuticals and Quasi-drugs" (Ministry of Health, Labour and Welfare Ordinance No. 179 of 2004), and a series of official documents issued by the Ministry of Health, Labour and Welfare and the PMDA, including notifications, Q&A, and explanatory materials on amendments. The main text of the ordinance is based on Article 14, Paragraph 2, Item 4 of the Pharmaceuticals and Medical Devices Act and legally stipulates the technical requirements for manufacturing and quality control (GMP) that manufacturing facilities must comply with. It consists of general provisions, standards for pharmaceutical manufacturing facilities, product-specific standards such as sterile preparations and biological products, and standards for quasi-drugs, and covers requirements such as validation, quality risk management, CAPA, and data integrity.

[0032] The GMP ministerial ordinance has been amended periodically since its enactment in 2004, and most recently, a major revision aimed at strengthening consistency with PIC / S GMP was implemented on August 1, 2021. This revision explicitly includes requirements such as "development of a pharmaceutical quality system (PQS)," "introduction of quality risk management," "thorough implementation of corrective / preventive actions (CAPA)," "prevention of cross-contamination," and "transition from standards to procedures," aiming to bridge the gap with international standards and prevent fraudulent manufacturing.

[0033] Therefore, GMP Ministerial Ordinance Information 14 includes the full text of the ordinance (all articles and supplementary provisions), the revised ordinance and a comparison table of the old and new versions, enforcement notices, administrative communications, article-by-article commentary, Q&A, inspection focus materials published by the PMDA, and presentation slides. Manufacturers and distributors must review and reflect this information as needed to maintain a quality system that can withstand domestic and international GMP inspections.

[0034] The report draft generation unit 15 inputs normalized data and predetermined prompts into the large-scale language model 17 and automatically generates various report drafts, such as annual review reports. The visualization unit 16 visualizes the data collected and centralized from external information sources on the user interface using tables, graphs, etc. This allows the user to easily understand the situation regarding external information sources.

[0035] Analysis, discussion, and documentation The collected data will be organized in appropriate formats, including graphs, lists, and tables, which will include comparisons between measured values ​​and standard values. Discussions based on the results will be reviewed as needed and transcribed into the designated format.

[0036] Therefore, at the stage of creating forms and drafting documents, the visualization unit 16 generates forms and graphs using the data. Furthermore, the report draft generation unit 15 creates a draft document using the large-scale language model 17. This reduces the man-hours required to create reports.

[0037] As described above, this system integrates data from various business systems and regulatory information sources, automatically generates draft reports using a large-scale language model 17, and visualizes the results, thereby significantly reducing the amount of work required for document creation in quality assurance operations and improving traceability.

[0038] By structuring and centrally managing business data scattered across various processes and systems, it becomes easier to perform analysis, extract data, and utilize it from a field perspective, supporting the root cause analysis of quality problems and improving efficiency. When users perform lot verification and annual verification, they can quickly access and verify the necessary information. A large-scale language model 17, which is a well-trained generative AI, can generate draft reports.

[0039] The flowcharts shown in Figures 2A to 2C all consist of the common steps: "Collection → Normalization → Storage → Verification → Flag Notification". Here, only the differences between individual systems will be discussed later.

[0040] Figure 2A is a flowchart of the data collection process from the manufacturing execution system 31. First, the data collection unit 24 collects manufacturing records from the manufacturing execution system 31 (step S20). The data collection unit 24 unifies and structures the data of multiple types of manufacturing records and stores it in the manufacturing record database 21 (step S21).

[0041] The verification unit 25 compares and verifies the data collected by the data collection unit 24 from the manufacturing execution system 31 to determine whether or not there are any defects or omissions (step S22). If there are any defects or omissions in the data collected from the manufacturing execution system 31, the verification unit 25 sends the verification result to the manufacturing execution system 31 as flag information (step S23), and the process returns to step S20. This allows the operators of the manufacturing execution system 31 to notice and correct human errors, thereby improving the reliability of the manufacturing record database 21.

[0042] Figure 2B is a flowchart of the data acquisition process from the test recording system 32. First, the data acquisition unit 24 collects test records from the test record system 32 (step S30). The data acquisition unit 24 unifies and structures the data from multiple types of test records and stores it in the test record database 22 (step S31).

[0043] The verification unit 25 compares and verifies the data collected by the data acquisition unit 24 from the test record system 32 to determine whether there are any defects or omissions (step S32). If there are any defects or omissions in the data collected from the test record system 32, the verification unit 25 sends the verification result to the test record system 32 as flag information (step S33), and the process returns to step S30. This allows the operator of the test record system 32 to notice and correct human errors, thereby improving the reliability of the test record database 22.

[0044] Figure 2C is a flowchart of the data collection process from the quality control system 33. First, the data collection unit 24 collects quality control data from the quality control system 33 (step S40). The data collection unit 24 unifies and structures multiple types of quality control data and stores them in the quality control database 23 (step S41).

[0045] The verification unit 25 compares and verifies the data collected by the data collection unit 24 from the quality control system 33 to determine whether or not there are any defects or omissions (step S42). If there are any defects or omissions in the data collected from the quality control system 33, the verification unit 25 sends the verification result to the quality control system 33 as flag information (step S43), and the process returns to step S40. This allows operators of the quality control system 33 to notice and correct human errors, thereby improving the reliability of the quality control database 23.

[0046] Figure 3 is a flowchart of the report draft generation process. This flowchart illustrates the sequence of steps for automatically generating a draft annual review report. When this process is initiated, the draft report generation unit 15 prompts the large-scale language model 17 to create a draft annual review report based on the item data mapping definition information 12 (step S10).

[0047] Next, the large-scale language model 17 uses search extension generation to search for the data required for each item of the draft annual review report from various databases of the GMP platform 2, legal and regulatory guideline information 11, past review report database 13, and GMP ministerial ordinance information 14 (step S11). The GMP platform 2 stores data obtained from the Manufacturing Execution System (MES) 31, the Laboratory Information Management System (LIMS) 32, and the Quality Management System (QMS) 33.

[0048] Next, the large-scale language model 17 generates a draft annual review report based on the input data using its RAG (Retrieval-Augmented Generation) function (step S12). After the draft annual review report is generated, the process ends.

[0049] Figure 4 is a flowchart of the visualization process. When this process is started, the visualization unit 16 inputs a prompt to the large-scale language model 17 instructing it to create a graph relating to the desired system (step S50).

[0050] Next, the large-scale language model 17 searches for the data necessary for the desired graph from various databases of the GMP platform 2 using search extension generation (step S51).

[0051] Next, the large-scale language model 17 generates the desired graph based on the input data using the RAG function (step S52). After the graph is generated, the process ends.

[0052] Figure 5A is a schematic diagram showing the manufacturing record database 21. The manufacturing record database 21 stores process information 211, manufacturing records 212, and equipment inspection records 213. Process information 211 holds the conditions and control values ​​for each manufacturing process. Manufacturing records 212 holds the operation history and work records for each batch. Equipment inspection records 213 holds the results of equipment inspections. These three types of data are automatically collected and normalized by the data collection unit 24 and then supplied to the generation process of the annual review report draft by the large-scale language model 17. This diagram visually organizes the smallest unit of MES-related data in the GMP platform 2.

[0053] Figure 5B is a detailed item diagram showing the manufacturing record database 21. The manufacturing record database 21 stores information such as the name of the manufactured product, manufacturing lot, manufacturing date, expiration date, work start date, work end date, output, yield, raw material name, instruction values / input quantities, instruction notes, work record (control values), and approver. This multifaceted quality data is normalized and integrated into the GMP platform 2 via the data collection unit 24 and used in the generation process of the annual review report draft by the large-scale language model 17, thereby contributing to the efficiency of quality control operations and improved traceability.

[0054] Figure 6A is a schematic diagram showing the test record database 22. This section schematically illustrates the internal structure of the test record database 22. The test record database 22 has a structure that hierarchically holds the main information handled in laboratory information management operations in six records.

[0055] Specifically, test item 221 registers the test item information. Non-conformance rate 222 aggregates the non-conformance rate for each product and process. Test standard 223 is reference information that defines what constitutes a pass or fail in each test, and is a record for centrally managing the acceptable range and judgment logic for each measurement target within the system. Test conditions 224 is information that holds the parameters necessary when conducting each test, such as temperature, time, and reagent concentration, and represents a set of settings that support highly reproducible test execution and accurate data acquisition.

[0056] Test results 225 are records that document the numerical data and judgment results output by the analytical instrument, and serve as the basis for evaluating the quality conformity of the lot through comparison with test standards 223. Equipment inspection records 226 are information that stores the maintenance history and calibration results of analytical instruments and test equipment, and include maintenance and inspection records to ensure the performance assurance of the equipment and the reliability of the data.

[0057] This multifaceted data structure allows for the cross-sectional extraction of quality trends by cross-referencing with process data and regulatory information. Furthermore, the generation of draft annual review reports using a large-scale language model 17 enables highly accurate tracking of deviation factors and evaluation of the appropriateness of corrective actions.

[0058] Figure 6B is a detailed item diagram showing the test record database 22. The test record database 22 stores the test record data. The product name indicates the name of the product being tested. The lot number indicates the identification number of the manufacturing unit to which the product being tested belongs. The test items refer to the characteristics to be evaluated or the analytical items to be measured. The standard defines the upper and lower limits and judgment logic set as the pass / fail criteria for each test item.

[0059] The test results record the actual measured values ​​obtained, and if they fall outside the specifications, they are treated as "OOS (Out Of Specification)". The test conditions specify parameters such as temperature, time, and reagent concentration to accurately reproduce the test, and the judgment results record a "conformity" or "non-conformity" judgment based on a comparison with the specifications.

[0060] The stability test results include data to confirm long-term quality changes. The test date is recorded as the actual date the test was conducted. The approval date is recorded as the date the quality manager officially approved the results. The approver field records the name of the person who evaluated and approved the test results, clearly ensuring traceability and accountability for the entire test.

[0061] Figure 7A is a schematic diagram showing the quality control database 23. Figure 7A schematically shows the internal structure of the quality control database 23. The core of the quality control database 23 consists of four elements: deviation records 231, CAPA information 232, non-conformance rates 233, and change information 234.

[0062] Deviation record 231 is a record that maintains detailed information about quality deviation events that occurred in the manufacturing process in chronological order, and consistently records the event summary, date of occurrence, and results of cause analysis. CAPA information 232 stores data on the planning, implementation status, and effectiveness verification of corrective and preventive actions to be taken in response to deviation events, and holds information for managing the recurrence prevention process.

[0063] The non-conformance rate 233 is a record constructed to statistically understand the frequency and trends of non-conformities that occur at the product and process level. The change information 234 records a detailed log to ensure traceability by linking a series of change histories, from proposed changes to quality systems and manufacturing processes to implementation and completion confirmation, using a common change ID.

[0064] These records are interconnected using a common identifier to enable consistent tracking of the quality assurance process, from the occurrence of abnormal events to subsequent corrective actions, statistical evaluations, and change management.

[0065] This allows for consistent management of the entire change lifecycle—from registering change proposals and assessing impacts to tracking implementation status and validating completion—on a single database. This enhances the traceability and auditability of change history through matching with other manufacturing record databases 21 and test record databases 22.

[0066] Figure 7B is a schematic diagram showing change information 234. Change Information 234 is a record that centrally manages the lifecycle of various changes implemented to quality systems and manufacturing processes. This record first clearly states the background and objectives as the reason for the change, followed by a detailed description of the specific changes. Furthermore, it stores a risk assessment that evaluates the impact of the change on product quality and regulatory compliance, and maintains a traceable record of who considered and approved the change, when, and under what procedures as part of the approval process. Finally, it records the progress of the change's implementation, from the planning stage to completion. This makes it possible to consistently manage the entire history from change proposal to completion confirmation on a single database.

[0067] Figure 7C is a schematic diagram showing deviation record 231. Figure 7C schematically shows the detailed record structure of the deviation record 231 stored in the quality control database 23 and its relationship to surrounding tables. The deviation record 231 uses the deviation ID as the primary key and includes items such as the date of deviation occurrence, location of deviation occurrence, content of deviation, scope of impact, cause analysis, risk assessment, corrective action, preventive action, approval process, stakeholder review, follow-up, and recurrence prevention measures. It consistently maintains the history from the occurrence of the deviation to its cause analysis and risk assessment, corrective and preventive actions, and follow-up and recurrence prevention.

[0068] Each record is cross-referenced via a common product identifier or batch identifier with the CAPA information 232 shown in Figure 7D and the change information 234 shown in Figure 7B, enabling bidirectional tracking of deviation cause analysis results and corrective effects. This allows GMP Platform 2 to comprehensively evaluate the frequency of deviation events, the implementation status of corrective actions, and their impact on final quality indicators on a single database, significantly improving audit readiness and traceability.

[0069] Figure 7D is a schematic diagram showing CAPA information 232. This CAPA Information 232, issued on June 9, 2025, by Ichiro Tanaka of the Quality Assurance Department, is a record that designates the Manufacturing Department and the Quality Control Department as relevant departments, under CAPA number "CAPA-2025-001". Following the detection of an off-spec pH value in product X of lot number 1234, the need for recurrence prevention based on customer complaints was clearly stated. After analyzing the root causes as deficiencies in raw material acceptance inspection and the lack of work procedures, corrective actions are planned, including procedure revision, personnel training, and product recall of the affected lot. Furthermore, preventive measures include a review of all inspection procedures and strengthening of the annual training program. These are scheduled to be completed by June 15, 2025, and a system has been established to follow up on the effectiveness of the corrective actions on July 15. Hanako Sato, Manager of the Quality Assurance Department, has been appointed as the person in charge, and Taro Yamada, Manager of Quality Control, has been appointed as the approver, ensuring the appropriate implementation of the measures and compliance with audits.

[0070] Figure 8 is a schematic diagram showing the item data mapping definition information 12. Figure 8 shows an overview of the item data mapping definition information 12 in this platform. The item data mapping definition information 12 functions as a set of metadata that maps each item constituting a form such as an annual review report to fields in multiple data sources such as the manufacturing record database 21, the test record database 22, and the quality control database 23.

[0071] MES shows information from the manufacturing record database 21. LIMS shows information from the test record database 22. QMS shows information from the quality control database 23.

[0072] The relevant systems for the first line, "Results of critical process control and final product quality control," are MES, LIMS, and QMS. The specific relevant data are process information 211 from the manufacturing record database 21, test items 221 from the test record database 22, and deviation records 231 from the quality control database 23.

[0073] The relevant systems for the second line, "All batches that were non-conforming to established standards and their investigations," are MES, LIMS, and QMS. The specific relevant data are manufacturing records 212 from the manufacturing record database 21, test standards 223 from the test record database 22, and deviation records 231 and CAPA information 232 from the quality control database 23.

[0074] The relevant systems for the third line, "Regarding the results of testing and inspection at the time of acceptance of raw materials and supplies," are LIMS and QMS. The specific relevant data are test items 221 and non-conformity rates 222 in the test record database 22, and deviation records 231 in the quality control database 23.

[0075] The relevant system in the fourth line, "All significant deviations or nonconformities, related investigations, and the effectiveness of the corrective and preventive actions taken as a result," is the QMS. The specific relevant data are the deviation records 231 and CAPA information 232 in the quality control database 23.

[0076] The relevant system for "all changes made to processes or analytical methods" in the fifth line is the QMS. The specific relevant data is change information 234 in the quality control database 23.

[0077] The relevant system for the sixth line, "Regarding the plan for changes to approved items," is the QMS. The specific relevant data is change information 234 in the quality control database 23.

[0078] The relevant system for the seventh line, "Regarding the results of stability monitoring and all undesirable trends," is LIMS. The specific relevant data are the test conditions 224, test items 221, and test results 225 from the test record database 22.

[0079] The relevant systems in the eighth line, "All returns, quality information and recalls related to quality, and the cause investigations conducted at the time," are the QMS and LIMS. The specific relevant data are test item 221 in the test record database 22 and CAPA information 232 in the quality control database 23.

[0080] The relevant system for the ninth line, "Regarding the appropriateness of the corrective actions previously taken in relation to the conduct of the tests and inspections," is the QMS. The specific relevant data are the deviation record 231 and change information 234 in the quality control database 23.

[0081] The relevant systems in line 11, "Qualification status of relevant equipment and utilities," are MES and LIMS. The specific relevant data are equipment inspection record 213 from the manufacturing record database 21 and equipment inspection record 226 from the test record database 22. Furthermore, the related systems for "Post-Market Commitment" in line 10 and "Management of Contractors" in line 12 are not registered, and no related data is registered.

[0082] Each mapping row is accompanied by a prompt template identifier referenced by the report draft generation unit 15 and a format definition ID to be passed to the large-scale language model 17. Therefore, the item data mapping definition information 12 is configured to allow consistent management in a machine-readable format from data extraction to language model input. This makes it possible to flexibly adapt the automatic report generation logic simply by updating the mapping definition, even if the report layout or legal regulatory requirements change, thus achieving both maintainability and regulatory compliance.

[0083] The report draft generation system 1 structures and centrally manages business data scattered across various processes and systems, facilitating analysis from a field perspective, data extraction, and utilization. Therefore, the report draft generation system 1 can support the root cause analysis and efficiency improvements of quality problems. The report draft generation system 1 significantly reduces the documentation workload for quality assurance operations while simultaneously ensuring data traceability across all processes.

[0084] The report draft generation system 1 has the function of taking in a wide variety of data scattered across various processes and multiple business systems on the manufacturing site, formatting it into a predetermined common format, and managing it centrally. As a result, on-site personnel can view and analyze information such as equipment operating status, quality inspection results, and history of deviation events, which previously could only be obtained by accessing each system individually, from a single interface. This makes it possible to quickly extract necessary information from a vast amount of data and pursue root causes while visualizing correlations, thereby enabling efficient early detection of quality problems and the planning of countermeasures.

[0085] Furthermore, the analysis results are automatically output as a draft report, significantly reducing the time and effort required for report creation while functioning as a foundation to support information sharing and decision-making throughout the organization. These mechanisms enable both increased productivity and traceability in quality assurance operations.

[0086] The structure and effects of the present invention are described below.

[0087] [1] A data collection unit (24) that collects multiple types of data from multiple systems, The aforementioned data collection unit (24) has an information infrastructure (GMP platform 2) that centrally stores multiple types of data, A report draft generation unit (15) generates a report draft by referencing the information infrastructure (GMP platform 2) and item data mapping definition information (12), and mapping the data of the information infrastructure (GMP platform 2) to each item using a large-scale language model (17), thereby generating the content to be described in the item, and generating a report draft. A report draft generation system (1) characterized by comprising the following:

[0088] By centrally storing data acquired from multiple systems and having a large-scale language model directly reference this integrated data to automatically generate draft reports, transcription errors and information omissions can be fundamentally suppressed, while significantly reducing the time required for report creation. Furthermore, since the data sources are clearly indicated in the written report, traceability of the basis for decisions and the content of the report is ensured.

[0089] [2] The aforementioned item data mapping definition information (12) associates each item included in the draft report with the type of data in the information infrastructure (GMP platform 2) related to the content to be described in that item. The report draft generation system (1) according to feature 1.

[0090] The item data mapping definition information allows each item in the report to be linked to the data type in the information infrastructure in a machine-readable format. This means that even if the report layout or regulatory requirements change, operation can continue simply by updating the mapping table, minimizing maintenance burden and improving the transparency of the extraction logic.

[0091] [3] The information infrastructure (GMP platform 2) further includes a verification unit (25) that verifies the integrity of the data collected by the data collection unit (24). The report draft generation system (1) according to feature 1.

[0092] By incorporating a verification unit into the information infrastructure to automatically check the consistency of collected data, abnormal values ​​and missing information can be blocked from passing to downstream processes, dramatically increasing the reliability of draft reports.

[0093] [4] The verification unit (25) notifies the system from which the data collection unit (24) collected the information of information regarding the consistency of the information collected by the data collection unit (24). The report draft generation system (1) according to feature 3.

[0094] Because the review unit immediately feeds back any deficiencies it detects to the original system, a closed loop of data collection → correction → re-collection is formed, allowing field operators to quickly notice and correct errors, and continuously improving data quality.

[0095] [5] The aforementioned large-scale language model (17) has a search extension generation function, The data obtained from the aforementioned information infrastructure (GMP Platform 2) is used as context to generate a draft report. The report draft generation system (1) according to feature 1.

[0096] Because a large-scale language model with search and extended generation capabilities constantly incorporates the latest data from the information infrastructure to generate text, it is possible to obtain highly accurate draft reports that quickly reflect specialized knowledge while minimizing hallucination.

[0097] [6] The aforementioned multiple systems include a manufacturing execution system (31), The data collection unit (24) collects data from the manufacturing execution system (31) and stores it centrally in the manufacturing record database (21) of the information infrastructure (GMP platform 2). The report draft generation system (1) according to feature 1.

[0098] By integrating process conditions and work history automatically acquired from the manufacturing execution system into the manufacturing record database, complete traceability at the batch level is established, allowing for cross-sectional analysis of manufacturing site anomalies and quality results to quickly identify the root cause.

[0099] [7] The aforementioned multiple systems include a quality control system (33), The data collection unit (24) collects data from the quality management system (33) and stores it centrally in the quality management database (23) of the information infrastructure (GMP platform 2). The report draft generation system (1) according to feature 1.

[0100] Integrating deviations, CAPA, and change history from the quality management system into a quality control database automates the evaluation of corrective action validity and trend analysis of quality events, facilitating a risk-based approach and continuous improvement.

[0101] [8] The aforementioned multiple systems include a test recording system, The data collection unit (24) collects data from the test record system (32) and stores it centrally in the test record database (22) of the information infrastructure (GMP platform 2). The report draft generation system (1) according to feature 1.

[0102] By importing test standards and measurement values ​​from the test record system and integrating them into the test record database, it becomes possible to prevent errors in conformity judgments while enabling time-series matching with manufacturing data, allowing for early detection of abnormal trends.

[0103] [9] The system includes a visualization unit (16) that visualizes the data stored in the aforementioned information infrastructure (GMP platform 2) as a table or graph. The report draft generation system (1) according to claim 1.

[0104] Because there is a visualization department that visualizes integrated data as tables and graphs, those in charge can instantly grasp the situation from a single dashboard, improving the speed and quality of decision-making and promoting cross-departmental collaboration.

[0105]

[10] The aforementioned item data mapping definition information (12) is, Includes a prompt template identifier and format definition ID associated with each item, The report draft generation system (1) according to claim 1.

[0106] By including prompt template identifiers and format definition IDs in the mapping definition, report types can be added and formats modified without coding, significantly reducing validation work during regulatory revisions.

[0107]

[11] The data collection unit (24) collects multiple types of data from multiple systems, The information infrastructure includes the step of centrally storing the data collected by the data collection unit, The report draft generation unit (15) refers to the information infrastructure (GMP platform 2) and item data mapping definition information (12), and uses a large-scale language model (17) to map the data of the information infrastructure (GMP platform 2) to each item, generating content to be described in the item and generating a report draft. A method for generating a draft report, characterized by comprising the following features.

[0108] Reports can be generated in various forms, including on-premise implementations and cloud SaaS offerings.

[0109] (modified version) The present invention is not limited to the embodiments described above, and includes various modifications. For example, the embodiments described above are described in detail to make the present invention easier to understand, and are not necessarily limited to those having all the configurations described. It is possible to replace parts of the configuration of one embodiment with the configuration of another embodiment, and it is also possible to add configurations from other embodiments to the configuration of one embodiment. Furthermore, it is possible to add, delete, or replace parts of the configuration of each embodiment with other configurations.

[0110] Each of the above configurations, functions, processing units, and processing means may be implemented in part or in whole by hardware, such as an integrated circuit. Each of the above configurations and functions may also be implemented in software by a processor interpreting and executing a program that implements each function. Information such as programs, tables, and files that implement each function can be stored in a recording device such as memory, a hard disk, or an SSD (Solid State Drive), or on a recording medium such as a flash memory card or a DVD (Digital Versatile Disk).

[0111] In each embodiment, the control lines and information lines shown are those deemed necessary for explanation and do not necessarily represent all control lines and information lines in the actual product. In practice, it can be assumed that almost all components are interconnected. [Explanation of Symbols]

[0112] 1. Report Draft Generation System 11. Legal and Regulatory Guideline Information 12-item data mapping definition information 13 Past Review Report Database 14. GMP Ministerial Ordinance Information 15. Report Draft Generation Department 16 Visualization Section 17. Large-scale language models 2 GMP Platform 21 Manufacturing Record Database 211 Process information 212 Manufacturing Records 213 Equipment Inspection Record 22 Examination Record Database 221 test items 223 Test Standards 224 Test Conditions 226 Equipment Inspection Record 23 Quality Control Database 231 Deviation Record 232 CAPA Information 233 Nonconformity rate 234 Change Information 24 Data Collection Department 31 Manufacturing Execution System 32 Test Recording System 33 Quality Management System

Claims

1. A data collection unit that collects multiple types of data from multiple systems, An information infrastructure that centrally stores multiple types of data collected by the aforementioned data collection unit, A verification unit that checks the consistency of each piece of data collected by the data collection unit and notifies the system from which the data collection unit collected the information of the consistency, A report draft generation unit generates a report draft by targeting data that satisfies consistency requirements, associating each item included in the report draft with the type of data in the information infrastructure related to the content to be described in that item, and referencing item data mapping definition information including the prompt template identifier and format definition ID associated with each item and the information infrastructure, configuring input based on the prompt template identifier and format definition ID, and using a large-scale language model to map the data in the information infrastructure to each item and generate the content to be described in that item, thereby generating a report draft. A report draft generation system characterized by comprising the following features.

2. The aforementioned large-scale language model has search extension generation capabilities. The data obtained from the aforementioned information infrastructure is used as context to generate a draft report. The report draft generation system according to feature 1.

3. The aforementioned multiple systems include a manufacturing execution system, The data collection unit collects data from the manufacturing execution system and stores it centrally in the manufacturing record database of the information infrastructure. The report draft generation system according to feature 1.

4. The aforementioned multiple systems include a quality control system, The data collection unit collects data from the quality management system and stores it centrally in the quality management database of the information infrastructure. The report draft generation system according to feature 1.

5. The aforementioned multiple systems include a test recording system, The data collection unit collects data from the test record system and stores it centrally in the test record database of the information infrastructure. The report draft generation system according to feature 1.

6. The system includes a visualization unit that visualizes the data stored in the aforementioned information infrastructure as a table or graph. The report draft generation system according to claim 1.

7. The data collection unit collects multiple types of data from multiple systems, The information infrastructure includes the step of centrally storing the data collected by the data collection unit, The verification unit verifies the consistency of each piece of data collected by the data collection unit and notifies the system from which the data collection unit collected the information of the consistency information, The report draft generation unit selects data that meets consistency requirements as the target of the report draft, associates each item included in the report draft with the type of data in the information infrastructure related to the content to be described in that item, and, by referring to the item data mapping definition information, which includes the prompt template identifier and format definition ID associated with each item, and the information infrastructure, configures the input based on the prompt template identifier and format definition ID, and generates the report draft by mapping the data in the information infrastructure to each item using a large-scale language model to generate the content to be described in that item, A method for generating a draft report, characterized by comprising the following features.

Images