Electronic medical chart system using large language model

Abstract

To provide a system that assists medical persons in creating order documents and recording documents in electronic medical charts to support efficient medical activities, using large language models (LLM) in which a large amount of the latest medical knowledge is learned.SOLUTION: The flow of an electronic medical chart system that uses a large language model used by doctors, nurses, and other staff via terminals connected to a LAN, includes electronic medical chart item extraction means for extracting information on items required for processing documents in an arbitrarily designated document category from electronic medical charts, and electronic medical chart content transcription means for providing the extracted electronic medical chart content as a prompt for the large language model.SELECTED DRAWING: Figure 10

Description

[Technical field]

[0001] The present invention relates to an electronic medical record system that uses a large-scale language model to provide high functionality and an improved user interface. [Background technology]

[0002] Computer systems (electronic medical records) are becoming widely used in the medical field to efficiently record medical instructions (orders) and records. In recent years, there has been remarkable progress in artificial intelligence (AI), and in particular, large-scale language models (LLMs), which can learn from large amounts of literature and data and enable inquiries and responses in natural language, have been attracting attention. Prior art documents relevant to this application include the following: [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Special Publication No. 2023-523644 [Patent Document 2] Patent Publication No. 2023-73095 [Patent Document 3] Patent No. 5484317 [Patent Document 4] Patent No. 7313757

[0004] [Non-Patent Document 1] Inside the Mind of ChatGPT Stephen Wolfram Hayakawa Shinsho 009 Summary of the Invention [Problem to be solved by the invention]

[0005] When providing medical care, examination plans are made based on the patient's symptoms and examination findings, and treatment plans are made based on the examination results, and the treatment is then carried out and recorded. Since a wide variety of medical professionals, including doctors and nurses, are involved, it goes without saying that electronic medical records are useful for efficient medical activities. However, medical science advances rapidly, and new, highly effective drugs and treatments are constantly being developed. Updating to the latest medical knowledge places a heavy burden on medical professionals. The preparation of medical instructions (orders) for tests and procedures, and the recording of their implementation are all done manually by medical professionals. This requires a huge amount of work, resulting in long working hours. In addition, differences in the knowledge and skill levels of medical professionals lead to inconsistencies in the effectiveness of treatment.

[0006] The present invention has been made to solve such conventional problems, and its purpose is to provide a system that uses a large-scale language model that has been trained with a large amount of the latest medical knowledge to assist medical professionals in creating order documents and record documents for electronic medical records, thereby supporting efficient medical activities. [Means for solving the problem]

[0007] As a means for achieving the above-mentioned objective, the electronic medical record system described in claim 1 is characterized in that, in an electronic medical record system that also uses a large-scale language model, it is equipped with an electronic medical record item extraction means that extracts information on items necessary for processing documents of an arbitrarily specified document category from the electronic medical record, and is equipped with a transcription means for transcribing the electronic medical record content into a prompt that provides the extracted electronic medical record content as a prompt for the large-scale language model.

[0008] The electronic medical record system according to claim 2 is characterized in that, in the electronic medical record system according to claim 1, it is provided with an answer content transcription means for transcribing the answer content to the prompt into the corresponding document in the electronic medical record.

[0009] The electronic medical record system of claim 3 is characterized in that, in the electronic medical record system of claim 1 or 2, it has an electronic medical record additional document creation means for creating at least one of an additional order document, record document, or disease name registration document to the electronic medical record as necessary in response to the large-scale language model.

[0010] In the electronic medical record system according to claim 4, in the electronic medical record system according to claim 1 or 2, the large-scale language model has a linking means with an external system.

[0011] The electronic medical record system of claim 5 is characterized in that in the electronic medical record system of claim 1 or 2, the electronic medical record item extraction means has an electronic medical record extraction item management means for each document category that manages a list of electronic medical record items required for each document category.

[0012] The electronic medical record system of claim 6 is characterized in that, in the electronic medical record system of claim 1 or 2, the large-scale language model is subjected to reinforcement learning using medical-specific learning data, and has a reference data limiting means for referring only to the medical-specific learning data when providing an answer.

[0013] The electronic medical record system of claim 7 is characterized in that, in the electronic medical record system of claim 1 or 2, it has a batch processing prompt creation means for inputting multiple prompt contents in a batch in the prompt of the large-scale language model.

[0014] The electronic medical record system of claim 8 is characterized in that, in the electronic medical record system of claim 1 or 2, the large-scale language model has a document content recommendation means for inputting patient information obtained by the electronic medical record item extraction means for documents of a specified document category as prompt content and requesting recommended content as an answer.

[0015] In the electronic medical record system described in claim 9, in the electronic medical record system described in claim 1 or 2, the electronic medical record system is characterized in that it has an API for each electronic medical record module, and the large-scale language model calls the electronic medical record module via the API to perform processing for necessary processing.

[0016] The electronic medical record system of claim 10 is characterized in that, in the electronic medical record system of claim 9, it has a series of processing group management means for registering the necessary processing as a series of coherent series of processing groups and enabling it to be specified as a series of processing groups for the large-scale language model. Effect of the Invention

[0017] The electronic medical record system described in claim 1 is equipped with an electronic medical record item extraction means, and in an electronic medical record system that also uses a large-scale language model, information on items required for processing documents of an arbitrarily specified document category is extracted from the electronic medical record. Since a means for transcribing the contents of the electronic medical record into a prompt is provided, the extracted contents of the electronic medical record are provided as a prompt for the large-scale language model.

[0018] The electronic medical record system according to claim 2 includes a means for transcribing the answer content to the electronic medical record, so that the answer content to the prompt is transcribed to the corresponding document in the electronic medical record.

[0019] The electronic medical record system described in claim 3 is provided with an electronic medical record additional document creation means, and in response to the large-scale language model, creates at least one of an order document, a record document, and a disease name registration document additional to the electronic medical record as necessary.

[0020] In the electronic medical record system according to claim 4, the large-scale language model has a linking means with an external system.

[0021] The electronic medical record system according to claim 5 includes an electronic medical record extraction item management means for each document category, so that the electronic medical record item extraction means manages a list of electronic medical record items required for each document category.

[0022] In the electronic medical record system according to claim 6, a reference data limiting means is provided, so that the large-scale language model is reinforced learning using medical-specific learning data, and only the medical-specific learning data is referred to when providing an answer.

[0023] The electronic medical record system according to claim 7 includes a batch processing prompt creating means, so that the contents of a plurality of prompts are inputted in a batch in the prompts of the large-scale language model.

[0024] The electronic medical record system described in claim 8 is equipped with a document content recommendation means, so that in a large-scale language model, for documents of a specified document category, patient information obtained by the electronic medical record item extraction means is input as prompt content, and recommended content is obtained as an answer.

[0025] In the electronic medical record system according to claim 9, the electronic medical record system includes an API for each electronic medical record module, and the large-scale language model performs necessary processing by calling the electronic medical record module via the API.

[0026] In the electronic medical record system according to claim 10, since it is provided with a series of process group management means, necessary processes can be registered as a series of integrated series of process groups, and can be designated as a series of process groups for the large-scale language model. [Brief description of the drawings]

[0027] [Figure 1] FIG. 2 is a diagram illustrating a hardware configuration of the present invention. [Diagram 2] This is an example of a user interface for a large-scale language model. [Diagram 3] FIG. 1 is a diagram showing examples of various document categories of an electronic medical record. [Figure 4] This is an example of extending a large-scale language model using an external collaboration system. [Diagram 5] 1 is an example of data to be additionally trained on a large-scale language model. [Figure 6] This is an example of correcting drug prescription orders in electronic medical records using a large-scale language model. [Figure 7] This is an example of verifying the disease name corresponding to a drug prescription order in an electronic medical record. [Figure 8] This is an example of verifying missing disease names and registering additional disease names in the electronic medical record. [Figure 9] This is an explanatory diagram for referencing the blood test and X-ray test history from the electronic medical record and ordering recommended tests. [Figure 10] This figure shows the process of referencing an electronic medical record using a large-scale language model, placing any necessary additional orders, correcting the content based on the results, and recording it in the electronic medical record. [Figure 11] FIG. 13 is a diagram of an electronic medical record extraction item management means for each document category. [Figure 12] FIG. 13 illustrates an example of a batch processing prompt. [Figure 13] This figure shows an example in which symptoms, findings, test findings, and disease name are input through a prompt, and the LLM responds with a recommended drug prescription example. [Figure 14] This is a conceptual diagram in which the LLM takes the lead in referencing extracted items in the electronic medical record and creating order documents for tests and procedures. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0028] FIG. 1 is a diagram showing an example of a hardware configuration according to the present invention. Because large-scale language models require huge amounts of data and computational resources, they are usually stored on the cloud and connected to a local area network (LAN) within the medical institution via a router over the Internet. The medical institution has an electronic medical record server. Doctors, nurses, and other staff access electronic medical records and large-scale language models via terminals connected to the LAN. It is also possible to build part or all of the electronic medical record system on the cloud, or in the case of a lightweight version, to install part or all of the large-scale language model on-site at the medical institution.

[0029] Figure 2 is an example of a user interface for a large-scale language model (LLM). Development of LLM is currently progressing rapidly, and numerous models have been developed, including ChatGPT (a registered trademark of OpenAI), Bard, LaMDA (a registered trademark of Google), and LLaMA (a registered trademark of Meta). Naturally, the user interface will differ, but typically, as shown in Figure 2, it consists of a box for entering prompts to give instructions and inquiries to the LLM, a box for displaying answers to those prompts, and a box for displaying the history of prompts and answers as a usage log.

[0030] Figure 3 shows examples of various document categories in an electronic medical record. The electronic medical records contain chronological records of various observation records for each patient, as well as order documents for tests and procedures, created by staff from a range of professions.

[0031] Figure 4 shows an example of a dedicated LLM for a large-scale language model, using Function Calling. In LLM, a general-purpose language model (substrate learning model) is constructed by learning large amounts of data, but knowledge in a specific field is not necessarily deep. For this reason, specialized models are constructed by intensively learning additional knowledge about specific fields (see below the drug-specific LLM in the figure). In addition, the general-purpose substrate learning model LLM is known to be weak in numerical calculations and logical operations. To compensate for this weakness, there are dedicated modules such as Wolfram Alpha (Non-Patent Document 1) (numerical logic processing system in the figure), both of which respond to prompt queries from the LLM and return answers.

[0032] In addition, there tends to be a time lag in acquiring knowledge about recent events. For example, a system could be used to query today's weather from a web search engine and retrieve the response. As one method for solving such problems, function calling has been proposed. This allows LLMs to use the functionality of other search engines, databases, or LLMs specialized for specific purposes via an API (Application Programming Interface). Taking the present invention as an example, the description of the efficacy and effectiveness of a drug is stable, but the drug price fluctuates from year to year. Current drug price information can be learned by the board or a dedicated LLM, but recording it in a dedicated database and searching it will be more responsive and less costly since the LLM learning will need to be updated less frequently. In addition, if dedicated LLMs or the like can be used in parallel, response times can be expected to be reduced. Furthermore, as described below, if it is possible to reference information from and write to the electronic medical record system, the electronic medical record system can be controlled directly from the LLM using Function Calling.

[0033] FIG. 5 shows an example of data to be additionally trained on a large-scale language model in the present invention. The interview form for each drug contains dozens of pages of information, including the drug name, dosage, indications, etc. These statements are based on data used for Ministry of Health, Labour and Welfare approval. Disease data comes from medical textbooks and treatment guidelines for each disease created by various medical societies. The tests include blood tests, physiological tests such as electrocardiograms, and explanatory books on X-rays, CT scans, MRIs, etc. All of these are strictly supervised by specialists such as doctors and nurses, so they are extremely reliable. In the case of the substrate LLM, some of the large amount of training data contains documents that contain malicious or discriminatory content, which is causing problems by returning malicious or discriminatory responses. Furthermore, "hallucinations," which cause people to respond to false information as if it were true, reduce the reliability of the answers. These problems can be avoided by limiting the range of learning data to be referenced as answers to highly reliable learning data as shown in this figure (reference data limiting means).

[0034] Figure 6 shows an example of correcting a drug prescription order in an electronic medical record using a large-scale language model. The drug prescription is recorded in the drug prescription document in the electronic medical record. The prescription contents are transcribed into the LLM prompt as shown in the figure (means for transcribing electronic medical record contents into a prompt). LLM will answer your questions by referring to the information written on the interview form, etc. If the doctor determines that the correction is appropriate, the correction is transcribed into the drug prescription document in the electronic medical record (means of transcribing the response into the electronic medical record). In addition, "transcription into electronic medical record" includes not only transcribing the response content into the document in question, but also creating an additional document as necessary and transcribing the response content into the newly created document.

[0035] Figure 7 shows an example of verifying the disease name corresponding to a drug prescription order in an electronic medical record. The contents of the drug prescription and the list of disease names are extracted from the electronic medical record (electronic medical record item extraction means) and transcribed into the prompt (means for transcribing electronic medical record contents into the prompt: reprinted). The response stated that the drug indication data would be collated with applicable drugs for each disease, and that no additional disease name would be required.

[0036] Figure 8 shows an example of verifying missing disease names and registering additional disease names in the electronic medical record. The prompt is the same as in Figure 7. This diagram points out that there are no disease names that are suitable for Opalmon. In addition, a list of diseases for which Opalmon is suitable is provided, and the doctor is asked to select a disease name, which is then transcribed into the electronic medical record (means of transcribing answers into the electronic medical record: reprinted), and the disease name is registered. This will reduce the assessment of medications and tests due to missing disease names (refusal of medical fees by insurers) which can sometimes exceed several million yen per month.

[0037] FIG. 9 is an explanatory diagram showing how test orders are recommended by referring to the histories of blood tests and radiology tests in the electronic medical records. Take the example of drug prescriptions for rheumatoid arthritis. Guidelines for the treatment of rheumatoid arthritis state that blood tests should be conducted at least once every few months to check for anemia, liver dysfunction, and kidney dysfunction caused by bone marrow suppression, which are side effects of Rheumatrex. It is also recommended that a chest X-ray be taken about once every six months to check for the side effect of interstitial pneumonia. The electronic medical record item extraction means extracts the dates of recent blood tests and chest X-rays, and if there is a long interval between tests, recommends the test and creates an order for it (electronic medical record additional document creation means). If side effects appear or worsening trends are observed, it is recommended that the drug be discontinued.

[0038] Figure 10 shows the process of using a large-scale language model as shown in Figures 6 to 9 to refer to electronic medical record items, making any necessary additional orders, correcting the content based on the results, and transcribing it into the electronic medical record. Select the document of interest (in this example, a drug prescription). The electronic medical record items required for LLM processing of the document are extracted from the electronic medical record and transcribed into the prompt. The electronic medical record items required for LLM processing vary depending on the document category. Although the prompt can be transcribed manually each time, as shown in Figure 11, if the electronic medical record extraction items are registered by document category (electronic medical record extraction item management means for each document category), the electronic medical record extraction items can be obtained easily and completely, and the necessary parts can be transcribed into the prompt (means for transcribing electronic medical record contents into the prompt). Once a response is received from the LLM, the content of the document is revised (a means of transcribing the response content into the electronic medical record) or, if necessary, additional documents such as blood test orders or radiology test orders are created (a means of creating additional documents into the electronic medical record).

[0039] FIG. 11 is an explanatory diagram of the electronic medical record extraction item management means for each document category. Items to be transcribed into the prompts are managed by the document category in which they appear. As shown in Figures 6 to 9, the extracted content can be transcribed into a prompt as needed. When ordering an MRI, check to see if the items shown in the figure are listed, and if they are not listed, ask the patient or their family to confirm. If there are any intracranial aneurysm clips, pacemakers, or metal objects inserted in the body near the target area, the examination will be canceled.

[0040] Figure 12 is an example of a bulk processing prompt. In the examples so far, the prompt and answer are carried out one step at a time, but if the processing routine is determined, the processing requests for each step may be sent to the prompt all at once. If human intervention is required along the way, a decision can be made or additional documentation can be created, and the process can continue from there.

[0041] Figure 13 shows an example in which symptoms, findings, test results, and disease name are inputted through a prompt, and the LLM responds with a recommended drug prescription example. Until now, the contents of the electronic medical record have been transcribed into a prompt and then queried to the LLM, but as a generative AI, the LLM can also generate recommended content. In Figure 13, we are requesting a recommended drug prescription based on the patient's attributes such as age and gender, symptoms and findings, test result data, and disease name. If the doctor deems it appropriate, he or she can simply transcribe it into the electronic medical record. If necessary, a recommendation may be made for a chronological document of multiple prescriptions, such as intravenous drips, rehabilitation, lifestyle advice, and dietary instructions (a so-called clinical path), which can then be transcribed into the electronic medical record.

[0042] Figure 14 is a conceptual diagram in which the LLM takes the lead in referencing extracted items in the electronic medical record and creating order documents for tests and procedures. In the examples in Figures 6 to 12, a prompt was created starting from a document in the electronic medical record document category and then the LLM was called. In other words, the electronic medical record category document side was the main focus, with LLM being utilized appropriately via prompts. In contrast, in this diagram, the API of each module in the electronic medical record is mainly called from the LLM side, and a series of operations are performed, including obtaining data on items that need to be referenced, such as symptoms and findings, creating test orders where necessary, estimating and registering the disease name based on the obtained test results, and creating order documents for treatment and tests based on the estimated disease name. Using this system, electronic medical record modules can be freely combined, making it easy to respond to a variety of pathological conditions. Furthermore, each module of the electronic medical record must be available from external systems such as LLM via an API. It goes without saying that verification of eligibility to use each electronic medical record module API is mandatory.

[0043] Although the embodiments have been described above, the specific configuration of the present invention is not limited to the above-described embodiments, and the present invention includes design changes and the like that do not deviate from the gist of the invention. For example, although the medical field has been used as an example, a similar system can be used in nursing care. ChatGPT and other systems have been used as examples of LLM, but the same is true for other systems, and the present invention also includes systems that will be newly developed in the future.

Claims

1. An electronic medical record system that also uses a large-scale language model, comprising an electronic medical record item extraction means for extracting information on items required for processing documents of an arbitrarily specified document category from the electronic medical record, and a transcription means for transcribing the electronic medical record contents into a prompt for providing the extracted electronic medical record contents as a prompt for the large-scale language model.

2. 2. The electronic medical record system according to claim 1, further comprising an answer content transcription means for transcribing answers to said prompts into said documents in the electronic medical record.

3. The electronic medical record system according to claim 1 or 2, characterized in that it has an electronic medical record additional document creation means for creating at least one of an additional order document, record document, or disease name registration document to the electronic medical record as necessary in response to the large-scale language model.

4. 3. The electronic medical record system according to claim 1, wherein the large-scale language model has a linking means with an external system.

5. 3. The electronic medical record system according to claim 1, further comprising: electronic medical record extraction item management means for managing a list of electronic medical record items required for each document category, said electronic medical record item extraction means comprising: a document category-by-document category management means for managing a list of electronic medical record items required for each document category.

6. The electronic medical record system according to claim 1 or 2, characterized in that the large-scale language model is subjected to reinforcement learning using medical-specific learning data, and has a reference data limiting means for referring only to the medical-specific learning data when answering an inquiry.

7. 3. The electronic medical record system according to claim 1, further comprising batch processing prompt creation means for inputting a plurality of prompt contents collectively in the prompts of the large-scale language model.

8. The electronic medical record system according to claim 1 or 2, characterized in that in the large-scale language model, a document content recommendation means is provided which inputs patient information obtained by the electronic medical record item extraction means as prompt content for documents of a specified document category, and requests recommended content as an answer.

9. The electronic medical record system according to claim 1 or 2, characterized in that the electronic medical record system is provided with an API for each electronic medical record module, and the large-scale language model calls the electronic medical record module via the API to perform necessary processing.

10. 10. The electronic medical record system according to claim 9, further comprising a process group management means for registering the necessary processes as a series of process groups, and enabling the processes to be specified as a process group for the large-scale language model.

Images