dispensing equipment

Abstract

To provide dispensing equipment that reduces the workload for pharmacists while accurately providing them with necessary information about medications. [Solution] The dispensing device comprises a display device and a control device. The control device receives questions related to dispensing, inputs the questions into a custom model, and displays the answers output by the custom model on the display device. The custom model is a text generation model that has been trained on or made able to access a drug database. The drug database includes at least the name of the drug, the efficacy of the drug, and information on precautions regarding the use of the drug.

Description

Technical Field

[0001] The present invention relates to dispensing equipment.

Background Art

[0002] Patent Document 1 discloses a drug weighing device, which is a kind of dispensing equipment. The drug weighing device of Patent Document 1 performs an audit process. As specific examples of the audit process, there are an audit process for confirming that the dosage of the drug included in the prescription information meets the standard of the usual dosage, and an audit process for confirming that the simultaneous administration of the drugs included in the prescription information is not prohibited, which are described in Patent Document 1. These audit processes are performed based on the information in the drug database (drug master).

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] When a dispenser performs dispensing based on a prescription, in addition to the prescription information (drug name, dosage form, amount, usage and dosage, etc.), the dispenser needs to accurately and quickly refer to the drug information (drug efficacy, side effects, interactions, contraindications, usage and dosage, precautions, etc.) and patient information (age, gender, past medical history, allergy history, medication history, etc.) to obtain information useful for dispensing. These information are recorded in a dispersed manner in a plurality of databases such as the database of the Pharmaceuticals and Medical Devices Agency (PMDA), the drug information database independently constructed by a medical institution, an electronic medical record system, and a medication history system. However, these databases have a huge amount of information, and it is a very difficult task that requires time and effort for a dispenser to directly access each individual database to search for and obtain the necessary information.

[0005] Furthermore, while the drug weighing device disclosed in Patent Document 1 performs audit processing using a rule-based algorithm, it is not easy to establish rules for integrating and processing information from various databases. Moreover, databases may contain information in text format, and it is not easy to establish rules that correspond to all textual expressions. For this reason, it may not be possible to efficiently acquire and utilize information from databases.

[0006] The drug weighing device disclosed in Patent Document 1 performs audit processing regarding the usual dose and simultaneous administration according to the installed audit program. Performing audit processing regarding new aspects other than the usual dose and simultaneous administration requires the installation of a new audit program, which lacks flexibility.

[0007] Furthermore, the description of the problems does not preclude the existence of other problems. One aspect of the present invention does not necessarily have to solve all of the problems. It is possible to extract other problems from the description in the specification, drawings, and claims.

[0008] This invention has been made in view of the above circumstances, and its main objective is to provide a dispensing device that can efficiently provide dispensing devices with useful information for dispensing based on database information, while reducing the workload on dispensing pharmacists. Means and effects for solving the problem

[0009] The problems that this invention aims to solve are as described above, and next, the means for solving these problems and their effects will be explained.

[0010] According to the aspects of the present invention, a dispensing device having the following configuration is provided. That is, the dispensing device comprises a display device and a control device. The control device receives a question regarding dispensing, inputs the question into a custom model, and displays the answer output by the custom model on the display device. The custom model is a model that has been trained on or made able to refer to at least a drug database for text generation. The drug database includes at least information on the name of the drug, the efficacy of the drug, and points to note regarding the use of the drug.

[0011] This system allows pharmacists to efficiently obtain useful information for dispensing, such as points to note regarding dispensing. Furthermore, it enables the acquisition and utilization of information from drug databases without the need to establish complex rules. It also allows for processing equivalent to audits from various perspectives.

[0012] In the aforementioned dispensing equipment, the following configuration is preferable. That is, the display device displays prescription information. The control device displays the prescription information on the display device and also displays a question input field for inputting the question on the display device.

[0013] By displaying a field for entering questions alongside the prescription information, pharmacists can enter questions while reviewing the prescription information.

[0014] In the aforementioned dispensing device, the following configuration is preferable. That is, the custom model is a model that has been further trained on the patient database or made able to refer to the patient database, and the patient database includes at least information for identifying the patient and information about the patient's physical condition.

[0015] It is possible to obtain responses tailored to patient information.

[0016] In the aforementioned dispensing equipment, it is preferable that the patient database further includes information about the patient's past medical history, information about the patient's drug use history, or information about the patient's allergies.

[0017] This allows pharmacists to receive answers regarding the appropriateness of medication, taking into account the patient's specific circumstances and constitution.

[0018] In the aforementioned dispensing equipment, it is preferable that the drug database further includes information on appropriate dosages, information on patients to whom the medication is administered, or information on combinations of drugs.

[0019] This allows us to provide pharmacists with answers that take into account even more detailed information about the medication.

[0020] In the aforementioned dispensing equipment, the following configuration is preferable. That is, the control device displays a drug input field for inputting the name of the drug and a question input field on the display device.

[0021] This makes it easier for the custom model to understand the entered medication and question, thus facilitating the generation of more appropriate answers.

[0022] In the aforementioned dispensing equipment, it is preferable that the control device identifies the drug included in the prescription information to be dispensed and, when the drug input field is displayed, inputs the information of the identified drug into the drug input field.

[0023] Since medications are automatically entered into the medication input field, the workload for dispensing pharmacists is reduced, and human error can be eliminated.

[0024] In the aforementioned dispensing equipment, it is preferable that the control device displays the relevant section of the drug database that serves as the basis for creating the answer, or information for accessing that section, on the display device when the answer is displayed.

[0025] The dispenser can easily confirm the basis of the answer.

[0026] Note that the description of the effects does not prevent the existence of other effects. One aspect of the present invention does not necessarily have to exhibit all the effects. It is possible to extract other effects from the descriptions in the specification, drawings, and claims.

Brief Description of the Drawings

[0027] [Figure 1] Block diagram of the dispensing device and related devices according to an embodiment of the present invention. [Figure 2] Diagram showing the contents of the drug database and the patient database. [Figure 3] Example screen when displaying a selection-type question window and answering the question. [Figure 4] Example screen when displaying a fill-in-type question window and answering the question. [Figure 5] Diagram explaining the creation process of the custom model and related functions. [Figure 6] Diagram explaining a modification example by the conversion program. [Figure 7] Sequence diagram showing the process of answering the dispenser's questions.

Modes for Carrying Out the Invention

[0028] Next, embodiments of the present invention will be described with reference to the drawings.

[0029] The dispensing equipment 10 shown in Figure 1 is introduced into a dispensing pharmacy. Dispensing equipment 10 is equipment used in relation to dispensing. In this specification, "dispensing" is used in a broad sense, and "dispensing" includes not only the work of a pharmacist preparing drugs by weighing drugs etc. based on a prescription, but also prescription auditing, inquiry into prescriptions, drug inspection, medication guidance, etc. Dispensing equipment 10 in this embodiment is, for example, equipment that prepares drugs based on a prescription, such as a weighing scale or a packaging machine. A weighing scale is equipment that measures the weight of drugs (especially powders). A packaging machine is equipment that individually packages powders for each dose. However, dispensing equipment 10 is not limited to these types of equipment, and may, for example, be equipment that directs or manages dispensing based on a prescription, or equipment for prescription auditing to confirm whether the contents of a prescription are accurate or not.

[0030] The dispensing device 10 comprises a display device 12, an input device 13, a control device 14, and a communication device 15.

[0031] If the dispensing device 10 is a weighing device, the dispensing device 10 further includes an electronic balance for weighing the drug. If the dispensing device 10 is a packaging machine, the dispensing device 10 further includes a dispensing tray and packaging device for packaging the drug.

[0032] The display device 12 is a liquid crystal or organic EL display. The display device 12 displays prescription information. Prescription information includes some or all of the prescription information written on a prescription issued by a physician. The information displayed on the display device 12 is information appropriate to the use or function of the dispensing equipment 10.

[0033] The input device 13 is a keyboard, a touch panel, or a microphone. If the input device 13 is a touch panel, it is configured integrally with the display device 12. The input device 13 is a device for the pharmacist to input information to the dispensing equipment 10. The dispensing equipment 10 in this embodiment has a function to answer questions from the pharmacist, and the input device 13 is also used by the pharmacist to input questions.

[0034] The control device 14 is a computer having an arithmetic unit such as a CPU, memory, and storage. By having the arithmetic unit execute programs stored in the storage, the control device 14 can perform various processes.

[0035] The communication device 15 is either a wired communication module or a wireless communication module. The communication device 15 communicates with other devices via a local area network or the internet. By using the communication device 15, the dispensing equipment 10 can communicate with, for example, the management device 20 and the server 30.

[0036] The management device 20 is connected to the dispensing equipment 10, for example, via a local area network. The management device 20 is a computer having a processing unit such as a CPU, memory, and storage. The management device 20 stores the drug database shown in Figure 2(a) and the patient database shown in Figure 2(b). The management device 20 also stores prescription information.

[0037] As shown in Figure 2(a), the drug database is a database that associates the name of a drug with its efficacy and points to note. Points to note are things to be aware of when using the drug in question, and include information such as appropriate dosage, side effects, contraindications, incompatible drugs, administration duration limits, and allergy information. The appropriate dosage is described as the appropriate dose of the drug in question. The appropriate dosage may also be described as a calculation formula based on age or weight, etc. Side effects are described as secondary effects that may occur when taking the drug in question. Contraindications are described as patients or conditions in which the drug should not be administered, or drugs that cannot be used in combination with the drug in question. Incompatible drugs are described as drugs that should not be mixed with the drug in question (but can be used together). Contraindications and incompatible drugs can also be considered as information on drug combinations. Administration duration limits are described as the maximum number of days the drug can be administered in a single prescription. For allergies, the entry describes the component of the drug that most commonly causes allergic reactions, or, if the drug is food-derived, the food in question (especially the food that is most likely to cause allergic reactions). Note that the drug database shown in Figure 2(a) is just an example, and some of the elements described may be omitted, or other elements not described in Figure 2(a) may be added.

[0038] In the example shown in Figure 2(a), all points to note are described using unstructured data such as natural language. Alternatively, some points to note may be described using structured data. In either case, because points to note are included in unstructured data, it is not easy to handle the drug database with a rule-based algorithm.

[0039] As shown in Figure 2(b), the patient database is a database that associates patient ID, name, date of birth, weight, allergies, medical history, and medication history. The patient ID and name are information used to identify the patient. The date of birth, weight, and allergies are information about the patient's physical condition. The medical history and medication history are information about the patient's history.

[0040] The patient ID is a string of characters or numbers used to identify the patient. The name is the patient's name. The date of birth is the patient's date of birth. The patient's age can be calculated using the date of birth. The weight is the patient's weight. Weight is an important indicator for determining medication dosage and determining the appropriate amount of medication (especially for pediatric patients). Allergies are listed for medications or foods that the patient has allergic reactions to. Medical history is a list of illnesses the patient has had in the past. Medication history is a list of medications the patient has taken in the past. Note that the patient database shown in Figure 2(b) is just an example, and some of the listed elements may be omitted, or other elements not shown in Figure 2(b) may be added. The patient database, like the medication database, includes unstructured data.

[0041] Server 30 is connected to the dispensing equipment 10, for example, via the internet. The location of Server 30 is arbitrary and may be overseas. Server 30 is a computer having a processing unit such as a CPU, memory, and storage. A custom model is stored in Server 30. The custom model is a text generation model (in other words, a large-scale language model) that has been customized to be used specifically for dispensing equipment. The method for creating the custom model will be described later.

[0042] Next, with reference to Figures 3 and 4, we will explain the overview of the function that answers questions from pharmacists (hereinafter referred to as the question answering function).

[0043] Figure 3 shows the prescription screen 40 displayed on the display device 12. The prescription screen 40 is a screen for displaying prescription information. The prescription screen 40 displays prescription information belonging to the same prescription information in groups. Specifically, it displays each Rp information belonging to the same prescription information. In the prescription screen 40 of Figure 3, one Rp information (Rp number 1-1) belonging to the same prescription information (reception number 000011, hospital A, patient A) is displayed. Among the same prescription information, for example, prescription information with common usage is classified as common Rp information. The prescription screen 40 switches the Rp information displayed by switching the Rp information tab. The prescription screen 40 includes a basic information display field 41 and a drug information display field 42. The basic information display field 41 displays information such as the reception number, which is the reference number at the dispensing pharmacy, the hospital name, and the patient name (in other words, basic information of the prescription information to be dispensed, information other than the drug). The basic information display field 41 displays information common to the same prescription information. The drug information display area 42 displays information about the drugs included in the prescription information. The drug information display area 42 also displays information that identifies the Rp information. The format in which the drugs are displayed in the drug information display area 42 varies depending on the type of dispensing equipment 10. The prescription screen 40 is also a screen for selecting the prescription information to be dispensed. The control device 14 selects the specific prescription information displayed on the prescription screen 40 as the prescription to be dispensed.

[0044] As shown in Figure 3(a), the prescription screen 40 also displays a pharmacist icon 43. The pharmacist icon 43 is an example of an icon used to perform a question answering function. When the dispenser selects the pharmacist icon 43, a selection-type question window 44 pops up, as shown in Figure 3(b). The selection-type question window 44 includes a patient information input field 51, a medication input field 52, a question input field 53, and a confirmation button 54.

[0045] The patient information input field 51 allows for the input of predetermined patient information (age and weight in this example). In this specification, "input" refers not only to cases where the pharmacist directly enters numerical values ​​or characters, but also to cases where the pharmacist selects the appropriate option from a list of options. As will be described later, the control device 14 may also input appropriate values ​​into the patient information input field 51 based on the patient database information. If the prescription information includes information that needs to be entered into the patient information input field 51, the control device 14 may also input appropriate values ​​into the patient information input field 51 based on the prescription information.

[0046] The drug input field 52 allows the user to enter the name of the drug. In this embodiment, the drug input field 52 is a pull-down menu, and only drugs related to the prescription information displayed on the prescription screen 40 are displayed as selection candidates. That is, only drugs related to the prescription information to be dispensed are displayed as selection candidates. Specifically, the control device 14 extracts all drugs related to the prescription information to be dispensed based on the prescription information displayed on the prescription screen 40, and sets the extracted drugs as candidates in the pull-down menu.

[0047] Questions can be entered in the question input field 53. In this embodiment, the question input field 53 is a pull-down menu, and typical questions about the drug are displayed as selection options. Typical questions about the drug can be divided into (1) questions about the drug only and (2) questions about the combination of the drug and the patient. Questions in (1) include questions asking about side effects of a specified drug, questions asking about clinical results, questions asking about contraindications or incompatible drugs, and questions asking about the indicated diseases. Questions in (2) include questions asking about the appropriate dosage for this patient, questions asking if there are any more effective drugs for this patient, questions asking if this patient has any allergy problems, and questions asking whether this drug is appropriate considering the patient's age. Note that these questions are just examples, and questions not listed here may also be registered as typical questions. Questions about the patient only, or questions unrelated to either the drug or the patient, may also be registered.

[0048] When the pharmacist selects the decision button 54, the selection-type question window 44 is hidden, and the answer window 45 shown in Figure 3(c) pops up. The answer window 45 contains the answer to the question. Specifically, it asks about the appropriate amount of the specified medication to be used for the patient indicated in the prescription information, and the answer shows the appropriate daily amount according to the patient's characteristics and the reasoning behind it.

[0049] In the example shown in Figure 4, the configuration of the question window differs from that of the example in Figure 3. Specifically, in the example in Figure 4, a fill-in-the-blank question window 46 is displayed instead of a selection-type question window 44. The pharmacist can freely enter questions in the question input field 53 of the fill-in-the-blank question window 46. The answer window 45 is the same as in the example in Figure 3. In the example in Figure 4, the question concerns allergies to two drugs, and the answers show responses from the perspective of drug-related allergies and food-related allergies for each drug.

[0050] The dispensing device 10 has a function to display a selection-type question window 44 and a function to display a fill-in-the-blank question window 46, and the dispensing device can switch between displaying either window through the dispensing operator's operation or a prior setting. The dispensing device 10 may also be configured to display only one of the two windows, the selection-type question window 44 or the fill-in-the-blank question window 46.

[0051] Figures 3 and 4 illustrate the answers to two questions, but naturally, the custom model can answer other questions as well. For example, it can answer each of the questions listed in the explanation of question input field 53. Furthermore, because the custom model is based on a text generation model as a large-scale language model, it can generate answers to questions entered in the free-text field, depending on the content of the question. For example, even for a question with a broad meaning such as "What are the precautions for this drug?", it can generate an appropriate answer by summarizing the precautions section of the drug database or extracting characteristic parts.

[0052] Next, we will explain the process of creating custom models and related functions, referring to Figure 5. Figure 5 is a diagram showing the steps taken by the creator of the custom model, etc., rather than a flowchart that is performed automatically by the computer.

[0053] First, prepare a text generation model that will serve as the basis for your custom model. Text generation models are developed by various vendors and provided via the internet. Therefore, you can obtain a text generation model from any of these vendors. The selection of a text generation model should be based on factors such as required performance, cost, and ease of customization.

[0054] Next, prepare the drug database. Generally, manufacturers of dispensing equipment 10 have their own drug databases, as they sometimes integrate them into the dispensing equipment 10. If they do not have one, they can purchase one from an external source, or they can create a drug database from the information set that forms the basis of the drug database.

[0055] Next, customize the text generation model to generate responses based on the drug database by performing one of the following two actions on the text generation model.

[0056] The first method is fine-tuning, which involves adding the contents of the drug database to the text generation model's training data. In other words, by using the contents of the drug database as training data, the contents of the drug database are incorporated into the text generation model (the parameters of the text generation model are adjusted). This allows the text generation model to generate answers based on the contents of the drug database. During training, the contents of the drug database can be used as is, or they can be preprocessed and converted into training data. Examples of preprocessing include structuring and classifying the data. Since the points to note in the drug database are often written in text, the text data can be classified into descriptions of appropriate dosage, descriptions of side effects, etc., to create training data. This improves the learning efficiency.

[0057] The second method is RAG, which adds a drug database lookup function to the text generation model. RAG is also known as search-enhanced generation. When using RAG, it is not necessary to additionally train the text generation model with the drug database; it is sufficient to simply set the drug database as a search target when generating answers. General text generation models have the ability to search external databases. By enabling this function, the text generation model first searches for relevant information from the drug database. This search uses advanced techniques such as vector search or semantic search, so the content of the question and the database do not need to be a strict match. Next, the text generation model incorporates the relevant information obtained from the search into prompts (inputs) to its own model. As a result, the text generation model can generate answers to questions by referring to the external drug database. To make RAG easier to implement, the contents of the database may be vectorized to facilitate vector search. Database vectorization can be performed, for example, using a library related to approximate nearest neighbor search.

[0058] Next, add the template response function by doing one of the following two things. The template response function is a function that generates responses according to a template (a predetermined appropriate response flow and format). By providing the template response function, it is possible to generate organized and consistent responses. Note that the template response function is not mandatory and can be omitted. An example of a template is (1) direct response, (2) supplementary information, (3) basis for the response, and (4) follow-up. Follow-up includes general standard phrases such as "The final decision should be made by a doctor or pharmacist" or "Please let us know if you need any additional information."

[0059] The first method involves additionally training the model with questions and template-based answers (a type of fine-tuning as described above). In other words, simultaneously with or after training the drug database, typical questions and template-based answers to those questions are added for further training. This causes the text generation model to develop a tendency to generate answers that conform to templates. By matching the typical questions used for training with the questions selectable in the question input field 53, the relevance between the training content and the question content increases, making it easier to generate appropriate answers.

[0060] The second method is prompt engineering, which involves configuring the model to generate answers according to a template. Specifically, this involves automatically adding prompts such as "(1) direct answer, (2) supplementary information, (3) rationale for the answer, and (4) follow-up" when inputting a question to the text generation model. This causes the text generation model to tend to generate answers that conform to the template.

[0061] Furthermore, prompt engineering may be performed for purposes other than the template answer function. For example, the prompt may include the fact that the questioner is a pharmacist. In this case, it is clear that the questioner understands technical terms, making it easier to generate a concise answer. Additionally, a prompt may be included that takes into account the size of the display area in the answer window 45 and advises against exceeding a predetermined character limit as much as possible.

[0062] Next, add a function to output the evidence by performing one of the following two actions. The evidence is the relevant section in the drug database. Specifically, you can display a copy of the relevant section in the drug database as evidence, display information indicating the relevant section (page number or relevant item) as evidence, or display a link to the relevant section in the drug database as evidence. Also, if a certain value (e.g., appropriate amount) is calculated from a formula or condition described in the drug database, you can present the calculation process as evidence. The method for adding this function differs depending on whether you have fine-tuned the drug database or added a RAG-based referencing function for the drug database. Note that this function is not mandatory and can be omitted.

[0063] When a drug database is fine-tuned, the contents of the drug database are incorporated into the text generation model (more precisely, into the model's weights), making it difficult to uniquely identify the referenced location when generating an answer. Therefore, in this case, additional training is required to output the rationale when generating the answer. Specifically, when training the text generation model with questions and answers, the answers being trained should include the rationale, and this data should be used for training. Alternatively, when fine-tuning the drug database, individual tags can be assigned to each data point in the drug database, and the model can be trained to output these tags in the answers. While using tags takes more time to prepare the training data, it clarifies how each data point in the drug database is handled, making it easier to output the rationale.

[0064] If you add a RAG-based referencing function to the drug database, when creating prompts based on the search results in the drug database, you should configure the system to further incorporate information that identifies the section of the drug database referenced by the search into the prompt. For example, if you search the drug database and reference page X of document A, you should add the information (evidence) "page X of document A" to the model's prompt. Furthermore, prompt engineering of the text generation model should be configured to include the entered evidence in the answer. As a result, the answer generated by the text generation model will include the information (evidence) "page X of document A".

[0065] Based on the above, a custom model is created by customizing the text generation model. After the custom model is created, it is evaluated to see if it works properly, and if it is determined to work properly, the custom model is considered complete. The custom model accepts questions about drugs and generates and outputs answers based on the drug database. Similarly, for the patient database, additional learning or reference functions are added to create a custom model, just like with the drug database. The drug database and patient database may be processed simultaneously to create a single custom model, or a custom model based on the drug database and a custom model based on the patient database may be created separately. The customization based on the patient database may be omitted. In this case, the contents of the patient database can be provided to the custom model from the control device 14 as a prompt.

[0066] Next, as a related function of the custom model, a conversion program is created that converts the questions entered into the dispensing device 10 into a format for output to the custom model. By using this conversion program, as shown in Figure 6, the information entered into the selection-type question window 44 can be converted into text suitable for input to the custom model. Specifically, patient information and drug information are organized and converted into text. The questions are also replaced with questions that clearly show the relationship between patient information and drug information. If the questions are selection-type, the conversion program can be easily used to convert them by storing an example question for display in the selection-type question window 44 and a corresponding example question for input in association. Since the conversion performed by the conversion program is uniform, the conversion program can be implemented using a rule-based algorithm.

[0067] When converting data using the conversion program, you may add information that is not entered in the selection-type question window 44. For example, you may add information about the relevant patient (e.g., allergies, medical history, medication history). Alternatively, since the fill-in-the-blank question window 46 in Figure 4 does not contain any patient information, you may add further information such as the patient's date of birth and weight when converting data using the conversion program.

[0068] Next, referring to Figure 7, we will explain the process of answering the pharmacist's questions.

[0069] First, the control device 14 displays prescription information on the prescription screen 40. The control device 14 displays one prescription from among the multiple prescription information stored in the management device 20 on the prescription screen 40. The control device 14 may also display prescription information entered by the input device 13 on the prescription screen 40. Then, when specific prescription information is displayed on the prescription screen 40, the control device 14 receives a command to display a question window (selection-type question window 44 or fill-in-the-blank question window 46) (S101). As described above, the control device 14 determines whether or not the pharmacist icon 43 has been operated. The pharmacist icon 43 may be displayed on a screen other than the prescription screen 40 (in other words, when no prescription information to be dispensed has been selected). Alternatively, the command to display the question window may be given by a method other than operating the pharmacist icon 43. For example, the command to display the question window may be given by long-pressing the drug information display field 42 or by selecting from the menu screen.

[0070] Next, the control device 14 identifies patient information and drug information based on the displayed prescription information (S102). For example, as shown in Figure 3, if specific prescription information is displayed, the control device 14 identifies the patient (especially the patient ID) associated with that prescription information. Next, the control device 14 retrieves the patient information of the identified patient by accessing the patient database. Next, if specific prescription information is displayed, the control device 14 identifies the drug associated with that prescription information.

[0071] Next, the control device 14 displays a question window on the display device 12 that reflects the identified information (S103). In the example in Figure 3, the control device 14 inputs patient information into the patient information input field 51 based on the identified patient information. Also, based on the identified drug information, the control device 14 sets the drugs displayed in the pull-down menu of the drug input field 52 to only include drugs included in the prescription information to be dispensed. Note that if there is only one drug included in the prescription information, the control device 14 may input that specific drug into the drug input field 52.

[0072] After the confirmation button 54 is pressed, the control device 14 converts the information entered into the selection-type question window 44 into the model input format using the conversion program described above (S105). Next, the control device 14 uses the communication device 15 to input the questions and related information converted into the model input format into the custom model of the server 30 (S106).

[0073] On server 30, the custom model outputs an answer based on the input question and related information (S107). Next, server 30 sends the answer output by the custom model to the dispensing device 10 (S108).

[0074] The control device 14 displays the response received from the server 30 in the response window 45 (S109). This allows the device to provide answers to the pharmacist's questions.

[0075] Finally, the features and variations of the question-answering function of this embodiment will be described.

[0076] As mentioned above, drug databases sometimes contain information in text format, making processing with rule-based algorithms difficult or preventing the full utilization of the information within the database. In this respect, using a custom model (an AI model built with machine learning) allows for the capture of meaning in text and the generation of appropriate answers to questions. Therefore, pharmacists can efficiently obtain the necessary information.

[0077] Furthermore, using a general-purpose text generation model as is often results in low accuracy in responses because it may not have learned the necessary information or other less reliable information may interfere. In this respect, the custom model of this embodiment generates responses based on drug databases and patient databases, making it easier to generate appropriate responses. Therefore, pharmacists can efficiently obtain the necessary information.

[0078] In this embodiment, one custom model is generated from the text generation model. Alternatively, multiple custom models corresponding to different perspectives may be generated. Perspectives include, for example, perspectives related to the administration of drugs to patients, and perspectives related to points to note regarding individual drugs. In particular, when fine-tuning a drug database, it is often meaningful to create multiple custom models corresponding to different perspectives. When multiple custom models are generated, the appropriate custom model is selected when generating an answer. For example, if the control device 14 determines that a question relates to the perspective of drug administration to patients, it inputs the question into the corresponding custom model. This can potentially improve the accuracy of the answer even with a small amount of training data.

[0079] In this embodiment, the selection-type question window 44 and the answer window 45 are in separate windows, but it is also possible to display a single window that shows both the question and the answer.

[0080] In this embodiment, a custom model is provided on the server 30, but a custom model may also be provided on the dispensing equipment 10 or other equipment in the pharmacy.

[0081] As described above, the dispensing device 10 of this embodiment comprises a display device 12 and a control device 14. The control device 14 receives a question regarding dispensing, inputs the question into a custom model, and displays the answer output by the custom model on the display device 12. The custom model is a model that has been trained on or made able to access a drug database for text generation. The drug database includes at least information on the name of the drug, the efficacy of the drug, and points to note regarding the use of the drug.

[0082] This system allows pharmacists to efficiently obtain useful information for dispensing, such as points to note regarding dispensing. Furthermore, it enables the acquisition and utilization of information from drug databases without the need to establish complex rules. It also allows for processing equivalent to audits from various perspectives.

[0083] In the dispensing device 10 of this embodiment, the display device 12 displays prescription information. The control device 14 displays the prescription information on the display device 12 and also displays a question input field 53 for inputting the aforementioned questions on the display device 12.

[0084] By displaying a field for entering questions alongside the prescription information, pharmacists can enter questions while reviewing the prescription information.

[0085] In the dispensing device 10 of this embodiment, the custom model is a model that has been further trained on a patient database or made able to refer to said patient database, and the patient database includes at least information for identifying the patient and patient information about the patient's physical condition.

[0086] It is possible to obtain responses tailored to patient information.

[0087] In the dispensing device 10 of this embodiment, the patient database further includes information about the patient's past medical history, information about the patient's drug use history, or information about the patient's allergies.

[0088] This allows pharmacists to receive answers regarding the appropriateness of medication, taking into account the patient's specific circumstances and constitution.

[0089] In the dispensing device 10 of this embodiment, the drug database further includes information on the appropriate dosage of medication, information on the target patient for medication, or information on combinations of drugs.

[0090] This allows us to provide pharmacists with answers that take into account even more detailed information about the medication.

[0091] In the dispensing device 10 of this embodiment, the control device 14 displays a drug input field 52 for inputting the name of the drug and a question input field 53 on the display device 12.

[0092] This makes it easier for the custom model to understand the entered medication and question, thus facilitating the generation of more appropriate answers.

[0093] In the dispensing device 10 of this embodiment, the control device 14 identifies the drug included in the prescription information to be dispensed, and when the drug input field 52 is displayed, it inputs the information of the identified drug into the drug input field 52.

[0094] Since medications are automatically entered into the medication input field, the workload for dispensing pharmacists is reduced, and human error can be eliminated.

[0095] In the dispensing device 10 of this embodiment, the control device 14 causes the display device 12 to display the relevant section of the drug database that serves as the basis for creating the answer, or information for accessing that section.

[0096] The pharmacist can easily verify the basis for the answer. [Explanation of symbols]

[0097] 10. Dispensing equipment 12 Display device 13 Input device 14 Control device 20 Management device 30 servers

Claims

1. Display device and Control device and Equipped with, The control device receives questions related to dispensing, inputs those questions into a custom model, and displays the answers output by the custom model on the display device. The aforementioned custom model is a model that has been trained on or made able to access a drug database for text generation, The dispensing device is characterized in that the drug database includes at least the name of the drug, the efficacy of the drug, and information on precautions regarding the use of the drug.

2. A dispensing device according to claim 1, The aforementioned display device displays prescription information, The control device is a dispensing device characterized by displaying the prescription information on the display device and displaying a question input field for inputting the question on the display device.

3. A dispensing device according to claim 1, The aforementioned custom model is a model that has been further trained on the patient database or has been made able to access the said patient database, The dispensing device is characterized in that the patient database includes at least information for identifying a patient and information about the patient's physical condition.

4. The dispensing apparatus according to claim 3, The dispensing device is characterized in that the patient database further includes information about the patient's past medical history, information about the patient's medication use history, or information about the patient's allergies.

5. A dispensing device according to claim 1, The dispensing device is characterized in that the drug database further includes information on appropriate dosages, information on patients to whom the medication is administered, or information on combinations of drugs.

6. A dispensing device according to claim 1, The control device is a dispensing device characterized by displaying a drug input field for inputting the name of a drug and a question input field on the display device.

7. The dispensing apparatus according to claim 6, The control device is characterized by identifying the drugs included in the prescription information to be dispensed, and inputting the information of the identified drugs into the drug input field when the drug input field is displayed.

8. A dispensing device according to claim 1, The control device is characterized in that, when displaying the answer, it displays on the display device the relevant section of the drug database that serves as the basis for creating the answer, or information for accessing that section.

Images