Electronic medical record information extraction system.

Abstract

This system provides an electronic medical record information extraction system that generates summaries from the large volume and diverse content recorded in electronic medical records and other documents, and efficiently extracts the document information necessary for processing by the generating AI. [Solution] Even in an electronic medical record environment divided into modules, the system facilitates medical staff's understanding of a patient's condition by creating an information aggregation document (summary) that enhances overview by consolidating information from each document category. Furthermore, it reduces the burden on medical staff by collecting information necessary for new document creation by the generation AI from existing documents (primary documents) and providing it to the creation of prompts for a large-scale language model, thereby facilitating the generation of text within newly created secondary documents. In an electronic medical record having multiple document categories and groups of documents belonging to each document category, the system includes a target document navigation means, a target area designation means, a target area information extraction means, a sequence registration means, and a sequence execution means.

Description

Technical Field

[0001] The present invention relates mainly to a system that generates summaries from large volumes of diverse content recorded in electronic medical records and the like, and efficiently extracts document information required for generative AI processing.

Background Art

[0002] In medical settings and the like, medical records (single documents) for a single patient require a large volume of entries, particularly for inpatient medical records. Covering a wide variety of document categories starting from disease names, personal identification information such as age, gender, and insurance type, observation records from doctors, nurses, care workers, rehabilitation staff and the like, condition assessment records, treatment plans based on the foregoing, post-treatment effect assessment records, and so on, documents for each document category are recorded in chronological order. With the aging of the population, the number of elderly people suffering from multiple chronic diseases is increasing. Treatment is required for each chronic disease, as well as for rapidly onset diseases such as pneumonia. Since multiple disease profiles overlap, documents such as the aforementioned observation records, condition assessment records, treatment plans based on the foregoing, and post-treatment effect assessment records become large in volume and complex. For the engaged medical staff, comprehensively grasping the medical condition of assigned patients has become a heavy burden. Regarding the processing of these large volumes of document data, the utilization of AI and the like is also demanded. Prior art documents related to this application include the following.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Patent Document 2

Patent Document 3

Patent Document 4

Summary of the Invention

[0004] Grasping the overall picture from the vast amount of records of patients with complex medical conditions is often difficult, and even when possible, it is extremely time-consuming, contributing to the exhaustion of medical staff. In addition to the records of the patient's condition (primary documents), there is also a huge volume of secondary documents that need to be created based on the primary documents, such as evaluations, plans, and implementation reports. Even when using large-scale language models (generative AI), it is necessary to input the patient's latest medical condition concisely as contextual information into prompts, which presents many challenges. Many current electronic medical record systems are developed using COBOL, an older language, and are not compatible with modern web standards. Even when trying to import necessary items from a primary document into a secondary document, copying the entire primary document is difficult, requiring manual copy-pasting on the monitor screen for each item. Robotic Process Automation (RPA) is used to improve work efficiency, but because it relies on a fixed screen, it becomes difficult to handle when the amount of data displayed for each item becomes large, causing the display to overflow the screen.

[0005] Most current electronic medical record systems use a modular structure based on professions such as doctors, nurses, and pharmacists. If a staff member is operating one module (for example, the nursing module) and wants to access information contained in another module (for example, the doctor module or pharmacist module), they have no choice but to close the currently running nursing module and then open the doctor module or pharmacist module separately. This process is cumbersome and significantly hinders information sharing between professions. In fact, there have been frequent reports of incidents where a doctor mistakenly entered "succinylcholine (a muscle relaxant)" instead of "succinylcholine (an anti-inflammatory drug)" due to the similar names. However, due to busy schedules and the cumbersome operation, information sharing was neglected, and a pharmacist unknowingly dispensed the incorrectly prescribed "succinylcholine" to the ward. Similarly, a nurse mixed it into an IV drip, resulting in muscle paralysis, respiratory arrest, and death of the patient. One method involves fixing the items to be extracted for each document category, creating and registering a program within the electronic medical record system in advance, and then transferring the information to the corresponding items in secondary documents as needed. While this method allows for efficient extraction, the content required for secondary documents is constantly changing. Requesting a software vendor to create and register an extraction program each time there is a change is costly and time-consuming, and is by no means easy in a hospital setting. On the other hand, advancements in generative AI have made it possible to automatically extract necessary items from plain text or structured document text. Furthermore, it has become known that providing sufficient background information (context) for document generation to the generative AI prompts enriches the generated content. Therefore, new methods are needed for extracting necessary information from single documents for use with generative AI.

[0006] This invention was made to solve the aforementioned problems of the past, and its objective is to facilitate medical staff's understanding of a patient's condition by creating an information aggregation document (summary) that enhances overview by consolidating information from each document category, even in an electronic medical record environment that is divided into modules. Furthermore, it aims to reduce the burden on medical staff by collecting the information necessary for creating new documents using generation AI from existing documents (primary documents), providing it to the creation of prompts for a large-scale language model, and facilitating the generation of text within newly created secondary documents. [Means for solving the problem]

[0007] As a means to achieve the above objective, the electronic medical record information extraction system according to claim 1 includes, in an electronic medical record having multiple document categories and groups of documents belonging to each document category, (1) tabs, bars, and buttons displayed on the electronic medical record screen (monitor screen) against A target document navigation means that displays documents belonging to the target document category from which information needs to be extracted (target documents) on a monitor screen through simulated input operations such as keyboard and mouse. (2) A means for specifying the target area (target area) in the target document displayed on the monitor screen from which information needs to be extracted. (3) A target area information extraction means for extracting monitor screen display information of the area specified by the target area designation means, (4) Sequence registration means for registering the sequences described in (1) to (3), (5) The system is equipped with a sequence execution means that selects and executes a sequence registered in (4) above.

[0008] The electronic medical record information extraction system according to claim 2 is characterized in that, in the electronic medical record information extraction system according to claim 1, the target area designation means is provided with deviation portion retrieval means that, if a part of the display area of ​​the target information deviates from the monitor screen, moves the deviated portion or expands the display area to make it displayable on the monitor screen.

[0009] The electronic medical record information extraction system according to claim 3, wherein the electronic medical record information extraction system according to claim 1 or 2, (1) a document category set registration means for registering at least one set consisting of (a) and (b), (b) a list of at least one target area within the target document, (2) The system is equipped with a document category set selection means for selecting the required document category set from the registered document category sets, (3) The system includes a document category set document content extraction means for extracting the document content of the document categories included in the selected document category set.

[0010] The electronic medical record information extraction system according to claim 4 is characterized in that, in the electronic medical record information extraction system according to claim 3, the document category set document content extraction means includes an extracted document structuring means that performs document structuring such as XML, JSON, or Markdown, rather than simply leaving the extracted document content as plain text.

[0011] In the electronic medical record information extraction system described in claim 5, Claim 3In the electronic medical record information extraction system, document The category set document content extraction means is characterized by comprising an extracted document content display means that displays a list (summary) of the extracted content.

[0012] In the electronic medical record information extraction system described in claim 6, Claim 3 In the electronic medical record information extraction system, document The system is characterized by comprising an extracted document buffer memory output means that outputs the category set document content extracted by the category set document content extraction means to a buffer memory.

[0013] The electronic medical record information extraction system according to claim 7 is characterized in that, in the electronic medical record information extraction system according to claim 6, it is further characterized by comprising an extracted document buffer memory content prompt output means that outputs the contents of the extracted document buffer memory output means to the prompt of the generating AI. [Effects of the Invention]

[0014] The electronic medical record information extraction system described in claim 1 includes a target document navigation means, a target area designation means, a target area information extraction means, a sequence registration means, and a sequence execution means. Therefore, even in an electronic medical record environment divided into modules, it is possible to create an information aggregation document (summary) that enhances overview by aggregating information from each document category, thereby facilitating medical staff to understand the patient's condition. Furthermore, by collecting information necessary for generating new documents using generative AI from existing documents (primary documents) and providing it to the large-scale language model to create prompts, the system facilitates the generation of text within newly created secondary documents, thereby reducing the burden on medical staff.

[0015] The electronic medical record information extraction system according to claim 2 includes a means for recovering deviations, so that if a portion of the display area of ​​the target information deviates from the monitor screen, the deviation can be moved or the display area can be enlarged to make it visible on the monitor screen.

[0016] The electronic medical record information extraction system according to claim 3 comprises (1)(a) a target document, (b) a list of at least one target area in the target document, and document category set registration means for registering at least one set consisting of the above (a) and (b), (2) document category set selection means for selecting a required document category set from the registered document category sets, (3) document category set document content extraction means for extracting the document content of the document categories included in the selected document category set.

[0017] The electronic medical record information extraction system according to claim 4 comprises extracted document structuring means for structuring the extracted document content into formats such as XML, JSON, and Markdown, instead of limiting the content to mere plain text.

[0018] The electronic medical record information extraction system according to claim 5 comprises extracted document content display means, so as to display a list (summary) of the extracted content.

[0019] The electronic medical record information extraction system according to claim 6 comprises extracted document buffer memory output means, so as to output the category set document content extracted by the category set document content extraction means to the buffer memory.

[0020] The electronic medical record information extraction system according to claim 7 comprises extracted document buffer memory content prompt output means, so as to output the content of the extracted document buffer memory output means to the prompt of generative AI. [Brief Description of the Drawings]

[0021] [Figure 1] This is an embodiment of the network configuration of the present invention. [Figure 2] Examples of a large number of document categories included in an electronic medical record (one document) are shown. [Figure 3] Examples of the content of each of a large number of document categories included in one document are shown. [Figure 4]This shows the set of document categories used for each job type. [Figure 5] An example of a document structuring method is shown. [Figure 6] This shows the structure of an electronic medical record system. [Figure 7] Figure 6(b) shows a typical screen transition. [Figure 8] This document describes a method for completely extracting information from a target region when the information extends beyond the display area of ​​that region. [Figure 9] This displays the extracted document content in a summary format. [Figure 10] This is an example of a user interface in a Large-Scale Language Model (LLM). [Figure 11] This is a schematic diagram illustrating how, in drug prescription audits, the content of category set documents, including drug prescriptions extracted from electronic medical records based on a document category set for pharmacists, is used to input prompts into the generating AI. [Modes for carrying out the invention]

[0022] Figure 1 shows a typical system configuration of the present invention. Large-scale language models consist of massive amounts of data, numerous CPUs (Central Processing Units), GPUs (Graphics Processing Units), and high-speed networks connecting them. Therefore, they are built on huge servers, such as cloud data centers, and provided via the web. Hospitals and similar institutions have numerous PC terminals connected via a Local Area Network (LAN) that is connected to the web. Hospitals also have servers that operate electronic medical records. In recent years, there has been an increasing trend of using mobile devices such as smartphones and tablets to access cloud services such as large-scale language models and electronic medical records while moving around within the hospital or from outside the hospital.

[0023] Servers, terminals, and mobile terminals all consist of memory for storing programs and data, recording media such as hard disks for persistently storing the programs and data as needed, a CPU for reading and processing the programs and data, a GPU for high-speed parallel processing as appropriate, and communication modules. As cloud services become more stable and affordable, there is a growing trend of migrating some or all of electronic medical records to the cloud. Conversely, there is also a movement to move some or all of large-scale language models to servers or terminals installed within hospitals that have increased processing power and memory capacity (edge ​​computing). (Figure 1 AI Server) Furthermore, the development of small-scale language models with a limited number of parameters is also progressing. It should be noted that even these small-scale models are still quite large compared to those developed before the emergence of large-scale language models, and all embodiments, including this form, are included in the present invention.

[0024] Figure 2 shows an example of the numerous document categories included in an electronic medical record (single document). Figure 3 is a list of document categories and example document contents within a single document. In the electronic medical record system, documents within each document category are managed in chronological order of creation date. Depending on the type of document, chronological order may not be necessary, and there may be only one document per document category. Each document category contains information recorded for each related item. In hospital medical records, entries for many other professions (caregivers, clinical laboratory technicians, registered dietitians, medical affairs staff, etc.) are generated. Additionally, there are instruction documents from doctors to each profession regarding tests and treatments, as well as reports on their implementation. As time passes, various types of documents accumulate, making it increasingly difficult to grasp the overall picture. Of the aforementioned document categories, the article-type document categories created by each profession can be divided into primary document categories, which involve observing and recording patients, and secondary document categories, which involve evaluations and planning based on the content of the primary category documents. As mentioned above, the objective of this invention is to create a summary of the latest information, making it easier for staff to understand the patient's condition, the considerations of each profession including doctors, and the planned examinations and treatments. Another objective of this invention is to facilitate the creation of new documents in the secondary document category group based on the contents of the primary document category group using a generation AI. In any case, it is necessary to collect the content of each item included in each document category of primary patient documents, according to the profession and circumstances.

[0025] Figure 4 shows an example of document category sets by job type. This is an example of a set of documents frequently used for each job type, along with a list of necessary items within those categories. Multiple sets can be registered and selected as needed, depending on the specific requirements of each job type. The "Comprehensive" section encompasses the aforementioned primary document group, linking the contents of each document and compiling them into a category set of document contents. This provides a list of the latest information, and by repeating this process daily and saving the results, it becomes possible to create a cross-sectional list of data from any point in the past. While the document content obtained from the "General" document category set is generally sufficient, if the document's storage capacity becomes too large, the ease of viewing the information at a glance tends to be diminished. Furthermore, the storage capacity per document on the computer being used may be exceeded. To prevent this, it is useful to register document category sets that are highly necessary for each profession, such as for pharmacists, doctors, nurses, and rehabilitation therapists (document category set registration means), and to select the appropriate document category set as needed (document category set selection means).

[0026] Figure 5 shows an example of a document structuring method. (a) The content of plain text is represented in (b) XML syntax, (c) JSON syntax, and (d) Markdown syntax. While plain text is acceptable as a prompt for generative AI, it is known that using structured syntax such as (b) to (d) reduces ambiguity and improves the accuracy of the generative AI's response. (b) to (d), HTML, or any other syntax that may become more widespread in the future may be used. Needless to say, if the electronic medical record document itself is already created using structured syntax, then that structured document can be used as is. Here, the "Patient Personal Information" document includes information that may be of little necessity depending on the profession, such as address, telephone number, and contact information, in addition to "age" and "gender," which are necessary for creating secondary documents. When structuring Figure 5, accept the document content initially, but if there is little need for certain job types, you may delete it or group it under "The following is not necessary:" or similar.

[0027] Figure 6 shows the structure of the electronic medical record system. (a) In this system, the programs for creating and editing documents for different professions, such as doctors and nurses, are integrated, ensuring consistent usability. Furthermore, since the recording and referencing of document data created for each profession are handled in an integrated electronic medical record database, nurses, for example, not only have the authority to create and refer to documents they create, but can also refer to articles created by other professions such as doctors. In contrast, (b) is modularized according to profession, such as doctors, nurses, and pharmacists. The programs used for creating and editing documents are for each module, and the recording and referencing of the created document data are performed in a database limited to that module. Therefore, when one job type tries to access an article from another job type, it is necessary to first stop the module for the current job type and then restart the module for the other job type with read-only permissions. Once you have finished accessing the data, you must stop the module in question, restart the module for your specific job category, and enter the accessed data into your document. This is because when you switch modules, the data from the previous module becomes inaccessible. Such cumbersome operations can be considered a cause of the aforementioned fatal accident. Unfortunately, the electronic medical record system with the structure shown in Figure 6(b) remains as a legacy system from 40 years ago.

[0028] To address the aforementioned problems, Robotic Process Automation (RPA) has been explored in recent years. This technology uses software robots to automate routine tasks performed by humans on PCs. By specifying the coordinates of buttons, input fields, and other elements on the screen, the system simulates mouse and keyboard operations such as clicking tabs, bars, and buttons, and entering text into input fields, on behalf of the human user. Recently, AI has become capable of analyzing the display screen, automatically acquiring the coordinates of buttons, input fields, and other elements on the screen, and automatically executing the aforementioned operations (computer use). Typically, in order to control a software program from the outside, it is necessary to define in detail, as an API (Application Programming Interface), the variables used for exchanging commands and data between the controlling software and the controlled software. However, the software being manipulated often lacks or does not disclose an API. In this case, by simulating human operations on the software, it is possible to operate the software without using an API.

[0029] Figures 71-74 show typical screen transitions in Figure 6(b). First, you specify which patient's electronic medical record you want to access. The monitor screen displays a list of inpatients and outpatients. The target patient is selected by clicking the mouse or by using arrow keys + Enter key (Figure 7-1). For the selected patient, specify the module you want to open using the same procedure (Figure 7-2). Open the selected module and specify the document you want to reference (the target document) using the same procedure (Figure 7-3). Open the target document, specify the target area, and extract the data within the target area (Figure 7-4).

[0030] Figure 8 shows that during the operation shown in Figure 7, when the information extends beyond the display area of ​​the target region, it becomes difficult to fully extract the information from the target region. In this case, the displayed content is extracted first, then the slider bar is used to move the initially hidden content to the display area, and that content is extracted. This sequence is repeated until the content of all target areas has been extracted. Alternatively, you can perform information extraction after clicking the "Full Screen" button or similar action to ensure all information in the target area is displayed.

[0031] Continue the operations in Figures 7 and 8 until all document categories and target area lists for the document category set specified in Figure 4 are completed. This allows for the extraction of necessary information from a wide variety of patient data, tailored to the specific circumstances of each profession. In the electronic medical record shown in Figure 6(a), the necessary information can be extracted using a similar procedure.

[0032] Figure 9 shows a summary of the content extracted by the operations in Figures 7 and 8 (extracted document content display means). By displaying a list of various documents from different professions necessary for handling the situation, staff can easily grasp the overall picture of the patient. When you click the "Copy to Clipboard" button in Figure 9, the contents of the extracted document displayed are output to a temporary buffer memory on the PC called the clipboard.

[0033] Figure 10 shows an example of a user interface in a Large-Scale Language Model (LLM). LLM is currently under rapid development, with numerous models being created, including ChatG PT (a registered trademark of OpenAI), Bard, LaMDA (a registered trademark of Google), and LLaMA (a registered trademark of Meta). The invention is included regardless of which LLM is used. While the user interface will naturally differ, the standard configuration, as shown in Figure 2, consists of a prompt input field for entering instructions and inquiries to the L LM, a field for displaying the response to that prompt, and a field for displaying the history of prompts and responses as a usage log.

[0034] Recently, in addition to using LLM as a standalone tool as described above, there has been an increasing trend for LLM itself to have an API (Application Programming Interface), allowing its functions to be used by external software. In this case, prompts, responses, and history are input and output to and from external software via an API, so the display format is controlled by the external software and is not limited to Figure 2. In Figure 9, clicking "Copy to Clipboard" copies the extracted document content obtained in Figure 9 to the clipboard. By inputting the contents of this clipboard as contextual information into the prompt area in Figure 10 (extracted document buffer memory contents prompt output means), it is expected that an appropriate generation AI output will be obtained according to the situation. By using a document content display means that shows the extracted content, it is possible to visually confirm whether the content provided to the generating AI is appropriate, and an improvement in the accuracy of the generating AI's responses can be expected.

[0035] If a single document, such as an electronic medical record, is built using a web-based method like HTML, then the extracted document content can be directly output to the generation AI's prompt. However, current electronic medical records and similar systems are often not web-based, making it impossible to directly output the content of category set documents to the generation AI prompt. Therefore, by first outputting the extracted document content to a buffer memory such as the "clipboard" (extracted document buffer memory output means), and then outputting it to the generation AI prompt, it becomes possible to use the generation AI even with a single document that is not web-based. Furthermore, if necessary, the output of the generating AI can be temporarily output to a similar buffer memory and then recorded as the generating AI's output document even in non-web-based single-item documents.

[0036] Figure 11 shows how the system extracts document content for a patient from the electronic medical record using a document category set for pharmacists, and uses the category set document content, which includes the drug prescription to be audited and contextual information to be considered during the audit, as contextual information for the generated AI. This makes it possible to perform drug audits, which previously required a great deal of effort from pharmacists, in a short amount of time and at low cost. By recording the audit results generated by the AI ​​in association with the drug prescription (saving the pairs of prompts and AI outputs shown in Figure 11 as related documents within the electronic medical record, thus ensuring accountability for the proper conduct of the audit), accountability can be achieved. Furthermore, the existence of this record prevents duplicate audits of the same drug prescription created before the date the record was created. It is also useful for verifying the accuracy of the AI. Furthermore, it is also useful to perform a series of operations sequentially on a specified group of documents. For example, this might occur when using a single document category set for a group of documents, such as all hospitalized patients, and performing secondary document creation processing using a generation AI.

[0037] Although embodiments have been described above, the specific configuration of the present invention is not limited to the embodiments described above, and design changes and the like that do not depart from the spirit of the invention are also included in the present invention. For example, this specification uses electronic medical records in the medical and nursing care fields as an example, but the field is not limited as long as it is a single document having multiple document categories and groups of documents belonging to each document category. The use of the output of the generation AI of the present invention may be partial during the creation of secondary documents. That is, it may be a mixture of simple transcription of item contents from the primary document and parts that are directly entered by staff.

Claims

1. An electronic medical record information extraction system characterized by comprising the following (1) to (5) in an electronic medical record having multiple document categories and groups of documents belonging to each document category. (1) A target document navigation means that displays documents belonging to the target document category from which information needs to be extracted (target documents) on the monitor screen by performing simulated input operations such as keyboard and mouse on tabs, bars, and buttons displayed on the electronic medical record screen (monitor screen), (2) A means for specifying the target area (target area) in the target document displayed on the monitor screen from which information needs to be extracted. (3) A target area information extraction means for extracting monitor screen display information of the area specified by the target area designation means, (4) Sequence registration means for registering the sequences described in (1) to (3), (5) Sequence execution means that selects and executes the sequence registered in (4) above.

2. The electronic medical record information extraction system according to claim 1, further characterized in that, if a part of the display area of ​​the target information deviates from the monitor screen in the target area designation means, the deviated portion is moved or the display area is enlarged to make it displayable on the monitor screen.

3. The electronic medical record information extraction system according to claim 1 or 2, characterized by comprising the following (1) to (3). (1) (a) a target document, (b) a list of at least one target area within the target document, and a document category set registration means for registering at least one set consisting of (a) and (b). (2) The system is equipped with a document category set selection means for selecting the required document category set from the registered document category sets, (3) Document category set document content extraction means for extracting the document content of the document categories included in the selected document category set,

4. The electronic medical record information extraction system according to claim 3, characterized in that the document category set document content extraction means includes an extracted document structuring means that performs document structuring such as XML, JSON, or Markdown, rather than simply extracting plain text from the extracted document content.

5. The electronic medical record information extraction system according to claim 3, characterized in that the document category set document content extraction means is further provided with an extracted document content display means for displaying a list (summary) of the extracted content.

6. The electronic medical record information extraction system according to claim 3, further comprising an extracted document buffer memory output means for outputting the category set document content extracted by the document category set document content extraction means to a buffer memory.

7. The electronic medical record information extraction system according to claim 6, further comprising an extracted document buffer memory content prompt output means for outputting the contents of the extracted document buffer memory output means to the prompt of the generating AI.

Images