Medication rationality identification method and device, storage medium and electronic equipment

Abstract

The invention relates to a drug use reasonability identification method and device, a storage medium and electronic equipment, and relates to the technical field of data processing, and the method comprises the steps: obtaining a drug use knowledge graph which comprises a corresponding relation between diseases and treatment drugs and a use relation between the treatment drugs; clinical diagnosis information and medication list information of a patient are obtained, target information is screened from the medication knowledge map according to the clinical diagnosis information and the medication list information, and the target information is used for judging whether the medication list information is reasonable for the clinical diagnosis information or not; and according to the target information and the fine-tuned large model, a drug use rationality result corresponding to the drug use list information is determined, and the fine-tuned large model is a model which combines graph data and a large language model technology and is subjected to fine-tuning training. The method and the device have the effect of reducing the manual auditing cost for identifying the drug use rationality.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method, device, storage medium and electronic device for identifying rationality of medication. Background Art

[0002] Rational medication refers to the behavior of doctors selecting appropriate medications, dosages, and courses of treatment according to the patient's condition and treatment needs during the medical process. This behavior is intended to ensure that patients receive the best treatment results while avoiding unnecessary waste and risks. With the continuous development of medical technology and the continuous increase in medical needs, the demand for medical insurance reimbursement is also increasing.

[0003] At present, the method usually adopted to identify the rationality of drug use is: the medical insurance department organizes professionals to conduct manual review to compare the drug usage in the medical insurance reimbursement materials with the clinical guidelines, so as to determine the rationality of drug use. However, once the volume of medical insurance reimbursement materials is large, the manual review cost of identifying the rationality of drug use under this method is high. Summary of the invention

[0004] In order to reduce the manual review cost of identifying the rationality of medication, the present application provides a medication rationality identification method, device, storage medium and electronic device.

[0005] In a first aspect of the present application, a method for identifying rationality of medication is provided, which specifically comprises: Obtaining a medication knowledge graph, wherein the medication knowledge graph includes a correspondence between diseases and therapeutic drugs and a usage relationship between therapeutic drugs; Obtaining the patient's clinical diagnosis information and medication list information, and filtering target information from the medication knowledge graph based on the clinical diagnosis information and the medication list information, wherein the target information is used to determine whether the medication list information is reasonable for the clinical diagnosis information; According to the target information and the fine-tuned large model, the medication rationality result corresponding to the medication list information is determined. The fine-tuned large model is a model that combines graph data and large language model technology and is fine-tuned and trained.

[0006] By adopting the above technical solution, the medication knowledge graph is obtained, and then based on the correspondence between different diseases and therapeutic drugs in the medication knowledge graph and the usage relationship between different therapeutic drugs, the relationship between clinical diagnosis information and therapeutic drugs in the medication knowledge graph is determined, and then combined with the usage relationship between two drugs in the medication list information in the medication knowledge graph, the target information is screened out, that is, the basis for determining whether the medication list information is reasonable for clinical diagnosis information. Finally, based on the target information, that is, the basis for determination, the large model after fine-tuning is used to determine whether the medication list information is reasonable for clinical diagnosis information, thereby realizing automatic identification of the rationality of medication in the medication list information, without the need for manual review based on clinical guidelines, thereby reducing the cost of manual review for identifying the rationality of medication.

[0007] Optionally, the obtaining of the medication knowledge graph specifically includes: Obtaining medical content, including descriptions of different diseases and corresponding therapeutic drugs and instructions for use of different therapeutic drugs; Desensitizing the medical content to obtain processed results; Extracting at least one knowledge graph triple from the processed result through a preset target big model, wherein the knowledge graph triple is a set consisting of diseases, therapeutic drugs and the corresponding relationship between the two or a set consisting of different therapeutic drugs and the corresponding usage relationship between different therapeutic drugs, and the target big model is a model combining graph data and large language model technology; Each of the knowledge graph triples is merged into an initial knowledge graph in a preset graph database to obtain a medication knowledge graph.

[0008] By adopting the above technical solution, the acquired medical content is desensitized to obtain the processed results, thereby removing the personal privacy data. Then, the target large model is used to identify the relationship between entities in the processed results, and the knowledge graph triples are extracted. Finally, the knowledge graph triples are merged into the initial knowledge graph to update and optimize the initial knowledge graph, so that the target information subsequently screened from the medication knowledge graph is more comprehensive.

[0009] Optionally, before determining the medication rationality result corresponding to the medication list information according to the target information and the fine-tuned large model, the method further includes: Generating fine-tuning training samples according to the medication knowledge graph; Fine-tune the preset target large model through the fine-tuning training samples to obtain a fine-tuned large model and store it in a preset database, wherein the target large model is a model that combines graph data and large language model technology; When it is necessary to identify the rationality of medication, the fine-tuned large model is called from the database.

[0010] By adopting the above technical solution, the target large model is fine-tuned by fine-tuning the training samples, and the fine-tuned large model is obtained and stored in a preset database, so that the model has a stronger ability to understand or process the relationship between diseases and therapeutic drugs, and between therapeutic drugs, so that the subsequent fine-tuned large model can more accurately identify the rationality of drug use.

[0011] Optionally, the filtering of target information from the medication knowledge graph according to the clinical diagnosis information and the medication list information specifically includes: Constructing a first graph query statement according to the clinical diagnosis information; Through the first graph query statement, query the correct treatment drug set and the contraindication treatment drug set corresponding to the disease in the clinical diagnosis information from the medication knowledge graph; Constructing a second graph query statement according to each actual therapeutic drug in the medication list information; According to the second graph query statement, the target usage relationship between each of the actual therapeutic drugs is queried from the medication knowledge graph, and the correct therapeutic drug set, the contraindicated therapeutic drug set and each of the target usage relationships are determined as target information.

[0012] By adopting the above technical solution, based on the first graph query statement, entities associated with the disease entity in the clinical diagnosis information, that is, the set of correct treatment drugs and the set of contraindicated treatment drugs are found from the medication knowledge graph; then based on the second graph query statement, the usage relationship between two actual treatment drugs is found from the medication knowledge graph, and finally the target information is determined based on this, so as to provide a judgment basis for the subsequent fine-tuned large model to judge the rationality of medication.

[0013] Optionally, determining the medication rationality result corresponding to the medication list information according to the target information and the fine-tuned large model specifically includes: Splicing the target information, the clinical diagnosis information and the medication list information to obtain a splicing result; Construct prompt task information, and concatenate the prompt task information with the concatenation result to obtain a final prompt instruction; The final prompt instruction is input into the fine-tuned large model to obtain at least one judgment result, and based on each of the judgment results, the medication rationality result corresponding to the medication list information is determined. The final prompt instruction is used to judge the rationality of the medication list information based on the target information.

[0014] By adopting the above technical scheme, the target information, clinical diagnosis information and medication list information are spliced ​​to obtain the splicing result, so as to integrate the judgment object and judgment basis for judging the rationality of medication. Furthermore, the prompt task information and the splicing result are spliced ​​again to obtain the final prompt instruction. Finally, the final prompt instruction is input into the fine-tuned large model. Through this final prompt instruction, the rationality of medication in the medication list information can be better guided to be judged according to the judgment basis, that is, the target information, so that the result of the rationality of medication is more accurate.

[0015] Optionally, the prompt task information includes a first determination task, a second determination task, a third determination task and a fourth determination task, wherein the first determination task is a task for determining whether the actual therapeutic drug is effective, the second determination task is a task for determining whether the actual therapeutic drug is a contraindicated drug for the disease in the clinical diagnosis information, the third determination task is a task for determining whether there are mutual side effects between the actual therapeutic drugs, and the fourth determination task is a task for determining whether the dosage of the actual therapeutic drug is reasonable, and the final prompt instruction is input into the fine-tuned large model to obtain at least one determination result, specifically including: Inputting the final prompt instruction into the fine-tuned large model, if the actual therapeutic drug does not exist in the correct therapeutic drug set, determining that the corresponding actual therapeutic drug is invalid as the determination result of the first determination task; If the actual therapeutic drug exists in the set of contraindicated therapeutic drugs, determining that the corresponding actual therapeutic drug is a contraindicated drug for the disease in the clinical diagnosis information is a determination result of the second determination task; If the target usage relationship between the actual therapeutic drugs is that there are mutual side effects, then determining that there are mutual side effects between the corresponding actual therapeutic drugs is a determination result of the third determination task; If the dosage of the actual therapeutic drug in the correct therapeutic drug set is inconsistent with the dosage of the corresponding correct therapeutic drug, the unreasonable dosage of the corresponding actual therapeutic drug is determined as the determination result of the fourth determination task.

[0016] By adopting the above technical solution, if a single actual therapeutic drug is not in the correct therapeutic drug set, it means that it is not the symptomatic drug for the disease in the clinical diagnosis information, and the corresponding actual therapeutic drug is determined to be invalid. If a single actual therapeutic drug is in the contraindicated therapeutic drug set, the corresponding actual therapeutic drug is determined to be a contraindicated drug for the disease in the clinical diagnosis information; if the target use relationship between the actual therapeutic drugs is considered to have mutual side effects, it means that the corresponding actual therapeutic drugs cannot be used together; further, the dosage of the actual therapeutic drugs in the correct therapeutic drug set is inconsistent with the dosage of the corresponding correct therapeutic drugs, which means that the dosage of the corresponding actual therapeutic drugs in the medication list information is unreasonable for disease diagnosis and treatment.

[0017] Optionally, determining the medication rationality result corresponding to the medication list information according to each of the determination results specifically includes: If there is a target determination result among the determination results, it is determined that the medication rationality result corresponding to the medication list information is unreasonable, and one of the following exists in the target determination result: a single actual therapeutic drug is ineffective, a single actual therapeutic drug is a contraindicated drug, a single actual therapeutic drug has a mutual side effect with other actual therapeutic drugs, or the dosage of a single actual therapeutic drug is unreasonable; If the target determination result does not exist in the determination results, the medication rationality result corresponding to the medication list information is determined to be reasonable.

[0018] By adopting the above technical scheme, if there is a target judgment result in each judgment result, it means that the medication list information has contraindications for the treatment of the patient's disease, the medication list information is not completely applicable to the patient's disease, there are mutual side effects between the actual therapeutic drugs, or there is unreasonable use of the actual therapeutic drugs, then the result of the rationality of medication is determined to be unreasonable; if there is no target judgment result in each judgment result, it means that all the actual therapeutic drugs in the medication list information are neither contraindicated drugs, nor have mutual side effects between them, and can be used at the same time, and all the actual therapeutic drugs are effective for the diseases in the clinical diagnosis information, and the dosage is reasonable, then the result of the rationality of medication is determined to be reasonable, thereby realizing a multi-dimensional evaluation of the rationality of the medication list information, making the identification of rationality of medication more accurate.

[0019] Optionally, the target large model is Graph+LLM.

[0020] By adopting the above technical solutions, the target large model combined with graph database technology can help the large language model better understand the relationship between entities and improve its own expression and reasoning capabilities. Through graph database technology, the relationship and semantic information between entities can also be extracted from large-scale text data, so that the target large model can more accurately extract entities and the relationship between entities from medical content.

[0021] In a second aspect of the present application, a device for identifying rationality of medication is provided, which specifically includes: A graph acquisition module, used to acquire a medication knowledge graph, wherein the medication knowledge graph includes a correspondence between diseases and therapeutic drugs and a usage relationship between therapeutic drugs; An information screening module, used to obtain the patient's clinical diagnosis information and medication list information, and screen target information from the medication knowledge graph based on the clinical diagnosis information and the medication list information, wherein the target information is used to determine whether the medication list information is reasonable for the clinical diagnosis information; The rational determination module is used to determine the rationality result of medication corresponding to the medication list information based on the target information and the fine-tuned large model. The fine-tuned large model is a model that combines graph data and large language model technology and is fine-tuned and trained.

[0022] By adopting the above technical solution, the graph acquisition module obtains the medication knowledge graph, and then the information screening module obtains the patient's clinical diagnosis information and medication list information, and screens the target information from the medication knowledge graph based on the clinical diagnosis information and medication list information. Finally, the rational judgment module determines the medication rationality result corresponding to the medication list information based on the target information and the fine-tuned large model.

[0023] In a third aspect of the present application, a computer-readable storage medium is provided, in which a computer program is stored. When the computer program is loaded and executed by a processor, the method steps described in any one of the first aspects are performed.

[0024] In a fourth aspect of the present application, an electronic device is provided, specifically comprising: A processor, a memory, and a computer program stored in the memory and capable of running on the processor, wherein the processor is used to load and execute the computer program stored in the memory so that the electronic device performs the method as described in any one of the first aspects.

[0025] In summary, the present application includes at least one of the following beneficial technical effects: based on the correspondence between different diseases and therapeutic drugs in the medication knowledge graph and the usage relationship between different therapeutic drugs, the relationship between therapeutic drugs in the medication knowledge graph and clinical diagnosis information, as well as the relationship between therapeutic drugs in the medication list information, is determined, and then the target information is screened out, that is, the basis for determining whether the medication list information is reasonable for the clinical diagnosis information. Finally, based on the target information, that is, the basis for determination, the large model is fine-tuned to determine whether the medication list information is reasonable for the clinical diagnosis information, thereby realizing automatic identification of the rationality of medication in the medication list information, without the need for manual review based on clinical guidelines, thereby reducing the cost of manual review for identifying the rationality of medication. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a flow chart of a method for identifying rationality of medication provided in an embodiment of the present application; Figure 2 It is a flow chart of another method for identifying rationality of medication provided in an embodiment of the present application; Figure 3 It is a structural schematic diagram of a device for identifying rationality of medication provided in an embodiment of the present application; Figure 4 It is a structural schematic diagram of another device for identifying rationality of medication provided in an embodiment of the present application.

[0027] Explanation of the accompanying drawings: 11. Graph acquisition module; 12. Information screening module; 13. Reasonable judgment module; 14. Model fine-tuning module. DETAILED DESCRIPTION

[0028] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this application, not all of the embodiments.

[0029] In the description of the embodiments of the present application, words such as "illustrative", "for example" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "illustrative", "for example" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "illustrative", "for example" or "for example" is intended to present related concepts in a concrete way.

[0030] In the description of the embodiments of the present application, the term "and / or" is only a kind of association relationship describing the associated objects, indicating that there may be three kinds of relationships, for example, A and / or B, which can represent: A exists alone, B exists alone, and A and B exist at the same time. In addition, unless otherwise specified, the meaning of the term "multiple" refers to two or more. For example, multiple systems refer to two or more systems, and multiple screen terminals refer to two or more screen terminals. In addition, the terms "first" and "second" are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the indicated technical features. Thus, the features defined as "first" and "second" can explicitly or implicitly include one or more of the features. The terms "include", "comprise", "have" and their variations all mean "including but not limited to", unless otherwise specifically emphasized.

[0031] See also Figure 1 The embodiment of the present application discloses a flowchart of a method for identifying the rationality of medication, which can be implemented by a computer program or run on a medication rationality identification device based on the von Neumann system. The computer program can be integrated into an application or run as an independent tool application, specifically including: S101: Obtain a medication knowledge graph, which includes a correspondence between diseases and therapeutic drugs and a usage relationship between therapeutic drugs.

[0032] Specifically, a knowledge graph is a structured semantic knowledge base used to describe concepts and their relationships in the real world. The basic unit of a knowledge graph is a triple composed of "entity-relationship-entity". The form of the triple is concise and intuitive, capable of clearly expressing the semantic associations between entities, thereby clearly expressing various types of knowledge. The triple can be expressed as: <entity, relationship, entity>. Exemplarily, <A, father-son, B>, where A is one of the entities, B is one of the entities, and "father-son" represents the association between the two entities. In the embodiments of the present application, a medication knowledge graph is a structured semantic network specifically representing medication-related knowledge, characterizing entities such as medications and diseases and the relationships between the entities. Specifically, the medication knowledge graph includes the correspondence between diseases and treatment medications, and the usage relationships between different treatment medications. Among them, the correspondence is divided into the applicable relationship and the contraindication relationship between diseases and treatment medications. The applicable relationship indicates that the treatment medication is effective for treating the disease, and the contraindication relationship indicates that the treatment medication is not applicable to the disease. In addition, the usage relationships between treatment medications are divided into having mutual side effects and not having mutual side effects. Having mutual side effects means that using two treatment medications simultaneously will produce side effects. Not having mutual side effects means that using two treatment medications simultaneously will not produce side effects. Exemplarily, the usage relationship between cephalosporin and amoxicillin is having mutual side effects, and using them simultaneously will cause an allergic reaction. In other embodiments, the medication knowledge graph can also characterize entities such as medications, diseases, symptoms, treatment plans, etc. and the relevant relationships between the entities.

[0033] Furthermore, in the embodiments of the present application, the execution subject of a medication rationality recognition method is a server. The server is connected to the terminal through a wireless network. A client or a small program related to medication recognition is installed in the terminal. The server is the background server of the client or the small program. The terminal can be a smart phone or a personal computer. The server can be an independent physical server, or a server cluster or a distributed system composed of multiple physical servers. The specific application scenario is as follows: When a person needs to perform a medication rationality analysis on various medications in the medication list information of a patient, they open the client or the small program in the terminal and click the "Start Button" for medication rationality recognition on the opened interface. In response to this operation, the server first obtains the medication knowledge graph.

[0034] Furthermore, a feasible way to obtain a medication knowledge graph is: Through web crawling technology, obtain medical knowledge Q&A with high timeliness and the drug instructions of various different treatment medications. The medical knowledge Q&A is the Q&A about different diseases and the corresponding treatment drug instructions, and the drug instructions contain the usage instructions of the corresponding treatment medications. Finally, determine the obtained content as medical content. Among them, the treatment drug instructions are mainly the instructions for symptomatic drugs and contraindicated drugs for diseases.

[0035] Further, the medical content is desensitized to remove the personal privacy data therein and obtain the processed result, wherein desensitization is a data protection technology that aims to hide or blur sensitive information to prevent unauthorized access and leakage. Specifically, desensitization is performed by hash desensitization. In other embodiments, desensitization can also be performed by encryption desensitization. Then, the processed result is input into a preset target large model, and at least one knowledge graph triple is extracted from the processed result through this target large model. Finally, each extracted knowledge graph triple is merged into the initial knowledge graph in the preset graph database to update the initial knowledge graph and obtain the medication knowledge graph. It should be noted that the graph database can be a Neo4j graph database, and in other embodiments, it can be a TigerGraph graph database. In addition, the target large model can be Graph+LLM, and in other embodiments, it can also be GNN-BERT. Among them, Graph+LLM combines graph database technology (Graph) with a large language model (LLM). Graph database technology can help the large language model better understand the relationship between entities and improve its own expression and reasoning capabilities. Through graph database technology, the relationships and semantic information between entities can also be extracted from large-scale text data.

[0036] Among them, the target large model is a model that combines graph data and large language model technology. It combines graph data and language models to improve the model's ability to process or identify complex relational data, such as different entities and the relationships between different entities. Therefore, the target large model can identify different entities and different relationships in the processed results and convert them into triples. Graph data is a data structure composed of vertices and edges, each vertex represents an entity, and each edge represents the association relationship between two entities. In addition, the knowledge graph triple is a set consisting of diseases, therapeutic drugs, and the corresponding relationship between the two, or a set consisting of different therapeutic drugs and the corresponding usage relationships between different therapeutic drugs. Exemplarily, the extracted medication-related knowledge graph triple is: <drug a, applicable, disease b>.

[0037] S102: Obtain the patient's clinical diagnosis information and medication list information, and filter target information from the medication knowledge graph based on the clinical diagnosis information and medication list information. The target information is used to determine whether the medication list information is reasonable for the clinical diagnosis information.

[0038] Specifically, after obtaining the medication knowledge graph, the image of the prescription is extracted from the database storing the patient's medical records, prescriptions and other material images, and the patient's clinical diagnosis information and medication list information are identified from the prescription image by optical character recognition (OCR) technology, and the clinical diagnosis information includes the patient's disease information or symptoms. The medication list information includes the patient's actual treatment drugs and the corresponding usage and dosage information. In other embodiments, the patient's clinical diagnosis information and medication list information can also be identified from the prescription image through a preset multimodal large model. In another embodiment, the prescription image can also be obtained by sending it through the patient's terminal. In another embodiment, the patient's clinical diagnosis information and medication list information can be directly obtained from the medical system or medical information storage database. It should be noted that the patient can be a patient with medical insurance reimbursement needs or a patient without medical insurance reimbursement needs.

[0039] Furthermore, according to the obtained clinical diagnosis information and medication list information, target information is filtered from the medication knowledge graph, and the target information is used to determine whether the medication list information is reasonable for the clinical diagnosis information. In the embodiment of the present application, a feasible screening method is: since the medication knowledge graph is stored in the Neo4j graph database, the first graph query statement is constructed according to the patient's disease or symptoms in the clinical diagnosis information through the cypher query language corresponding to the Neo4j graph database. For example, if the patient's disease is "pneumonia", then the corresponding first graph query statement is: MATCH (d:Disease {name:"Pneumonia"})-[r:TREATS]-(m:Medication) RETURN d.name AS Disease, m.name AS Medication; "MATCH (d:Disease{name:"Pneumonia"})" means to find a node with the label Disease, and the name attribute value of the node is "Pneumonia"; "-[r:TREATS]-" means to match the relationship from a disease node to a therapeutic drug node; "(m:Medication)" indicates the other end of the specified relationship, that is, the node with the Medication label; "RETURN d.name AS Disease,m.name AS Medication" indicates the columns of the query result, including the disease name (d.name) and the therapeutic drug name (m.name).

[0040] Furthermore, through this first graph query statement, the correct set of therapeutic drugs corresponding to the disease in the clinical diagnosis information is queried from the medication knowledge graph, that is, each therapeutic drug that has an applicable relationship with the disease, and the set of contraindicated therapeutic drugs, that is, each therapeutic drug that has a contraindicated relationship with the disease.

[0041] Furthermore, according to each actual therapeutic drug in the medication list information, a second graph query statement is constructed, and specifically according to each actual therapeutic drug, a corresponding second graph query statement is constructed. Through the second graph query statement, the target usage relationship between the corresponding actual therapeutic drugs is queried from the medication knowledge graph, and finally only one of the correct therapeutic drug set, the contraindicated therapeutic drug set, and each target usage relationship can be determined as the target information. In other embodiments, the correct therapeutic drug set, the contraindicated therapeutic drug set, and each target usage relationship can also be determined as the target information.

[0042] In one embodiment, at least one triple is extracted from the acquired medication knowledge graph, and these triples are determined as fine-tuning training samples. In other embodiments, the knowledge graph triple extracted from the medical content can also be determined as fine-tuning training samples. The fine-tuning training sample refers to a data set used to fine-tune the target large model during the machine learning or deep learning model training process. Further, the fine-tuning training sample is divided into a training set and a test set according to a preset ratio, and the preset ratio can be 7:3 or 6:4. The target large model is trained by the training set and the test set to obtain a fine-tuned large model and store it in a preset database, so that the model has a stronger ability to understand or process the relationship between the disease and the therapeutic drug, and between the therapeutic drug and the therapeutic drug. During the training process, the model parameters are optimized by the reverse gradient algorithm until the model converges. This is a prior art and will not be repeated here. Finally, when it is necessary to identify the rationality of medication, the fine-tuned large model is called from the database.

[0043] S103: Determine the medication rationality result corresponding to the medication list information based on the target information and the fine-tuned large model, where the fine-tuned large model is a model that combines graph data and large language model technology and is fine-tuned and trained.

[0044] Specifically, after the target information is determined, the result of the rationality of medication corresponding to the medication list information is determined in combination with the fine-tuned large model called. In an embodiment of the present application, a feasible determination method is: splicing and combining the target information, clinical diagnosis information and medication list information to obtain a splicing result. Exemplarily, the target message is a set of correct treatment drugs. The correct treatment drug set is spliced ​​with the clinical diagnosis information and the medication list information to obtain a splicing result, and then the prompt task information is constructed. By constructing the prompt task information, the fine-tuned large model is guided to perform the tasks that need to be performed when identifying the rationality of medication. The prompt task information can be expressed as follows: If you are a doctor expert, please strictly follow medical common sense. I need you to judge whether the actual treatment drugs in the {medication list information} are effective based on the {disease in the clinical diagnosis information}. Please refer to the {correct treatment drug set} for the judgment basis.

[0045] Furthermore, the prompt task information is spliced ​​with the splicing result to obtain the final prompt instruction, and finally the final prompt instruction is input into the fine-tuned large model. The fine-tuned large model executes the relevant tasks in the prompt task information according to the final prompt instruction to obtain the determination result related to whether the actual therapeutic drugs are reasonable. The corresponding determination result contains the result of whether each actual therapeutic drug is reasonable. If there is an actual therapeutic drug in the determination result that is ineffective for the disease, the result of the rationality of the medication corresponding to the medication list information is determined to be unreasonable; if each actual therapeutic drug in the determination result is effective for the disease, the result of the rationality of the medication corresponding to the medication list information is determined to be reasonable.

[0046] See also Figure 2 The present application embodiment discloses a flowchart of another method for identifying the rationality of medication, which can be implemented by a computer program or run on a medication rationality identification device based on the von Neumann system. The computer program can be integrated into an application or run as an independent tool application, specifically including: S201: Obtain a medication knowledge graph, which includes the correspondence between diseases and therapeutic drugs and the usage relationship between therapeutic drugs.

[0047] S202: Obtain the patient's clinical diagnosis information and medication list information.

[0048] S203: Construct a first graph query statement according to the clinical diagnosis information.

[0049] S204: Using the first graph query statement, query the medication knowledge graph for a set of correct treatment drugs and a set of contraindicated treatment drugs corresponding to the disease in the clinical diagnosis information.

[0050] S205: Construct a second graph query statement according to each actual therapeutic drug in the medication list information.

[0051] S206: According to the second graph query statement, the target usage relationship between each actual therapeutic drug is queried from the medication knowledge graph, and the correct therapeutic drug set, the contraindicated therapeutic drug set and each target usage relationship are determined as target information.

[0052] For details, please refer to steps S101-S102, which will not be described in detail here. In the embodiment of the present application, the obtained correct treatment drug set, the contraindication treatment drug set and each target usage relationship are determined as target information.

[0053] S207: splicing the target information, clinical diagnosis information and medication list information to obtain a splicing result.

[0054] S208: Construct prompt task information, and concatenate the prompt task information with the concatenation result to obtain a final prompt instruction.

[0055] S209: Input the final prompt instruction into the fine-tuned large model to obtain at least one judgment result, and determine the medication rationality result corresponding to the medication list information based on each judgment result.

[0056] Specifically, the correct treatment drug set, the contraindicated treatment drug set, the target usage relationship, the clinical diagnosis information and the medication list information are spliced ​​to obtain the splicing result, and then the corresponding prompt task information is constructed. In an embodiment of the present application, a feasible construction method is: receiving the task description information sent by the terminal and determining it as the prompt task information, the prompt task information includes a first judgment task, a second judgment task, a third judgment task and a fourth judgment task. The first judgment task is the task of determining whether the actual treatment drug is effective, the second judgment task is the task of determining whether the actual treatment drug is a contraindicated drug for the disease in the clinical diagnosis information, the third judgment task is the task of determining whether there are mutual side effects between the actual treatment drugs, and the fourth judgment task is the task of determining whether the dosage of the actual treatment drug is reasonable.

[0057] For example, the prompt task information can be expressed as follows: If you are a medical expert, please strictly follow medical common sense. I need you to judge whether the actual treatment drugs in the {medication list information} are reasonable based on the {disease in the clinical diagnosis information}. Please refer to the {correct treatment drug set}, {contraindicated treatment drug set}, and {target use relationship between actual treatment drugs in the medication list information} for judgment. The rationality of medication is mainly judged from the following aspects: (1) Whether the actual therapeutic drugs in the medication list information are effective for the disease; (2) Whether the actual therapeutic drugs in the medication list information are contraindicated drugs for the disease; (3) Whether there are any side effects between the actual therapeutic drugs in the medication list information; (4) Whether the actual dosage of therapeutic drugs in the medication list information is reasonable.

[0058] Furthermore, the constructed prompt task information is combined with the splicing result to obtain the corresponding final prompt instruction, and the final prompt instruction is input into the fine-tuned large model, so that the fine-tuned large model refers to the disease, the correct treatment drug set, the contraindicated treatment drug set and the target usage relationship in the clinical diagnosis information, and executes according to the four judgment tasks. The specific process is: determine whether a single actual treatment drug is in the correct treatment drug set. If not, then determine that the corresponding actual treatment drug is invalid and use it as the judgment result of the first judgment task; if it is, determine that the corresponding actual treatment drug is effective and use it as the judgment result of the first judgment task. After the first judgment task is executed, the corresponding judgment result contains the result of whether each actual treatment drug is effective for the disease.

[0059] Further, it is determined whether a single actual therapeutic drug is in the set of contraindicated therapeutic drugs. If so, the corresponding actual therapeutic drug is determined to be a contraindicated drug for the disease in the clinical diagnosis information, and is used as the determination result of the second determination task. After the second determination task is completed, the corresponding determination result contains the result of whether each actual therapeutic drug is a contraindicated drug. Further, it is determined whether the target usage relationship between all actual therapeutic drugs is that there are mutual side effects. If so, the existence of mutual side effects between the corresponding actual therapeutic drugs is used as the determination result of the third determination task; otherwise, the absence of mutual side effects between the corresponding actual therapeutic drugs is used as the determination result of the third determination task.

[0060] Furthermore, it is determined whether the dosage of the actual therapeutic drugs in the correct therapeutic drug set is consistent with the corresponding correct therapeutic drugs, that is, whether the dosage of the correct therapeutic drugs in the correct therapeutic drug set that is the same as the actual therapeutic drugs is consistent. If not, the unreasonable dosage of the corresponding actual therapeutic drugs is used as the judgment result of the fourth judgment task; otherwise, the reasonable dosage of the corresponding actual therapeutic drugs is used as the judgment result of the fourth judgment task. After the fourth judgment task is executed, the corresponding judgment result contains the result of whether the dosage of each actual therapeutic drug is reasonable. Finally, the judgment result corresponding to each judgment task is obtained.

[0061] Further, according to each judgment result, the medication rationality result corresponding to the medication list information of this patient is determined. A feasible determination method is: if there is a target judgment result in each judgment result, the target judgment result contains one of the following: a single actual therapeutic drug is ineffective, a single actual therapeutic drug is a contraindicated drug, a single actual therapeutic drug has a mutual side effect with other actual therapeutic drugs, or the dosage of a single actual therapeutic drug is unreasonable. Exemplarily, if there is a result that an actual therapeutic drug is ineffective for a disease in the judgment result corresponding to the first judgment task, the judgment result corresponding to the first judgment task is regarded as the target judgment result, indicating that the medication list information is not completely applicable to the patient's disease, and the medication rationality result is determined to be unreasonable; if there is a result that an actual therapeutic drug is a contraindicated drug for a disease in the judgment result corresponding to the second judgment task, the judgment result corresponding to the second judgment task is regarded as the target judgment result, indicating that the medication list information is contraindicated for the treatment of the patient's disease, and the medication rationality result is determined to be unreasonable, and there is no need to evaluate the judgment results of the remaining judgment tasks, and so on. Furthermore, if this target judgment result does not exist in each judgment result, it means that all actual therapeutic drugs in the medication list information are neither contraindicated drugs nor have any mutual side effects between them, and they can be used at the same time. Moreover, all actual therapeutic drugs are effective for the diseases in the clinical diagnosis information and the dosage is reasonable. Then the result of the rationality of medication is determined to be reasonable.

[0062] In other embodiments, based on the rational medication list information of historical patients with the same disease type as this patient, the number of occurrences of each therapeutic drug in the patient's medication list information in all rational medication list information is determined, and the first number of therapeutic drugs is selected from each therapeutic drug in descending order of the number of occurrences as the target drug, and the remaining therapeutic drugs are determined as the remaining drugs, and the first weight of each target drug is calculated, and the first weight is the ratio of the number of occurrences of each target drug to the sum of the number of occurrences of all target drugs. Further, the number of co-occurrences of the remaining drugs that co-appear with each target drug in the rational medication list information is counted, and the second weight of each remaining drug is calculated, that is, the ratio of the number of co-occurrences of a single remaining drug corresponding to the target drug to the sum of the number of co-occurrences of all corresponding remaining drugs. Finally, for a single target drug, the product of the first weight and the corresponding second weights is calculated and summed to obtain the sum of the products. The larger the sum of the products, the more reasonable the combination of the remaining drugs and the target drug is for the patient's disease. The sum of the products corresponding to each target drug is summed to obtain the final sum of the products. If the final sum of the products is greater than the preset sum of products threshold, it means that the rationality of each therapeutic drug in the patient's medication list information is high, thereby achieving successful verification of the rationality of medication results.

[0063] The implementation principle of a method for identifying the rationality of medication in an embodiment of the present application is: based on the correspondence between different diseases and therapeutic drugs in the medication knowledge graph and the usage relationship between different therapeutic drugs, determine the relationship between therapeutic drugs in the medication knowledge graph and clinical diagnosis information, as well as the relationship between therapeutic drugs in the medication list information, and then filter out the target information, that is, the basis for determining whether the medication list information is reasonable for the clinical diagnosis information. Finally, based on the target information, that is, the basis for determination, the large model after fine-tuning is used to determine whether the medication list information is reasonable for the clinical diagnosis information, thereby realizing automatic identification of the rationality of medication in the medication list information, without the need for manual review based on clinical guidelines, thereby reducing the manual review cost for identifying the rationality of medication.

[0064] The following is an embodiment of the device of the present application, which can be used to execute the embodiment of the method of the present application. For details not disclosed in the embodiment of the device of the present application, please refer to the embodiment of the method of the present application.

[0065] See also Figure 3 , is a schematic diagram of the structure of the device for identifying the rationality of medication provided in the embodiment of the present application. The device for identifying the rationality of medication can be implemented as all or part of the device through software, hardware or a combination of both. The device includes a graph acquisition module 11, an information screening module 12 and a rational determination module 13.

[0066] A graph acquisition module 11 is used to acquire a medication knowledge graph, which includes the correspondence between diseases and therapeutic drugs and the usage relationship between therapeutic drugs; The information screening module 12 is used to obtain the patient's clinical diagnosis information and medication list information, and screen the target information from the medication knowledge graph based on the clinical diagnosis information and medication list information. The target information is used to determine whether the medication list information is reasonable for the clinical diagnosis information; The rational determination module 13 is used to determine the rationality result of medication corresponding to the medication list information based on the target information and the fine-tuned large model. The fine-tuned large model is a model that combines graph data and large language model technology and is fine-tuned and trained.

[0067] Optionally, the atlas acquisition module 11 is specifically used for: Obtain medical content, including descriptions of different diseases and corresponding treatment drugs, as well as instructions for use of different treatment drugs; Desensitize the medical content and obtain the processed results; Through the preset target big model, at least one knowledge graph triple is extracted from the processed results. The knowledge graph triple is a set consisting of diseases, therapeutic drugs and the corresponding relationship between the two, or a set consisting of different therapeutic drugs and the corresponding usage relationship between different therapeutic drugs. The target big model is a model that combines graph data and large language model technology. Each knowledge graph triple is merged into the initial knowledge graph in the preset graph database to obtain the medication knowledge graph.

[0068] Optional, such as Figure 4 As shown, the device also includes a model fine-tuning module 14, which is specifically used to: Generate fine-tuning training samples based on the medication knowledge graph; Fine-tune the preset target large model by fine-tuning the training samples, obtain the fine-tuned large model and store it in a preset database. The target large model is a model that combines graph data and large language model technology. When it is necessary to identify the rationality of medication, the fine-tuned large model is called from the database.

[0069] Optionally, the information screening module 12 is specifically used for: Constructing a first graph query statement according to the clinical diagnosis information; Through the first graph query statement, the correct treatment drug set and the contraindication treatment drug set corresponding to the disease in the clinical diagnosis information are queried from the medication knowledge graph; Constructing a second graph query statement according to each actual therapeutic drug in the medication list information; According to the second graph query statement, the target usage relationship between each actual therapeutic drug is queried from the medication knowledge graph, and the correct therapeutic drug set, the contraindicated therapeutic drug set and each target usage relationship are determined as target information.

[0070] Optionally, the reasonable determination module 13 is specifically used for: The target information, clinical diagnosis information and medication list information are spliced ​​to obtain a splicing result; Construct prompt task information, and concatenate the prompt task information with the concatenation result to obtain the final prompt instruction; The final prompt instruction is input into the fine-tuned large model to obtain at least one judgment result, and based on each judgment result, the medication rationality result corresponding to the medication list information is determined. The final prompt instruction is used to judge the rationality of the medication list information based on the target information.

[0071] Optionally, the reasonable determination module 13 is specifically used for: The final prompt instruction is input into the fine-tuned large model. If the actual therapeutic drug does not exist in the correct therapeutic drug set, the corresponding actual therapeutic drug is determined to be invalid as the judgment result of the first judgment task; If the actual therapeutic drug exists in the set of contraindicated therapeutic drugs, determining that the corresponding actual therapeutic drug is a contraindicated drug for the disease in the clinical diagnosis information is a determination result of the second determination task; If the target usage relationship between the actual therapeutic drugs is that there are mutual side effects, then determining that there are mutual side effects between the corresponding actual therapeutic drugs is a determination result of the third determination task; If the dosage of the actual therapeutic drug in the correct therapeutic drug set is inconsistent with the dosage of the corresponding correct therapeutic drug, the unreasonable dosage of the corresponding actual therapeutic drug is determined as the determination result of the fourth determination task.

[0072] Optionally, the reasonable determination module 13 is specifically used for: If there is a target determination result among the determination results, the medication rationality result corresponding to the medication list information is determined to be unreasonable, and one of the following exists in the target determination result: a single actual therapeutic drug is ineffective, a single actual therapeutic drug is a contraindicated drug, a single actual therapeutic drug has a mutual side effect with other actual therapeutic drugs, or the dosage of a single actual therapeutic drug is unreasonable; If the target determination result does not exist in each determination result, the medication rationality result corresponding to the medication list information is determined to be reasonable.

[0073] It should be noted that the device for identifying the rationality of medication provided in the above embodiment only uses the division of the above functional modules as an example when executing the method for identifying the rationality of medication. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the device for identifying the rationality of medication provided in the above embodiment and the embodiment of the method for identifying the rationality of medication belong to the same concept. The implementation process is detailed in the method embodiment and will not be repeated here.

[0074] The embodiment of the present application further discloses a computer-readable storage medium, and the computer-readable storage medium stores a computer program, wherein when the computer program is executed by a processor, a method for identifying rationality of medication use of the above embodiment is adopted.

[0075] Among them, the computer program can be stored in a computer-readable medium, the computer program includes computer program code, the computer program code can be in the form of source code, object code, executable file or certain middleware, etc. The computer-readable medium includes any entity or device that can carry the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the computer-readable medium includes but is not limited to the above-mentioned components.

[0076] Among them, through this computer-readable storage medium, a method for identifying rationality of medication in the above embodiment is stored in a computer-readable storage medium, and is loaded and executed on a processor to facilitate the storage and application of the above method.

[0077] An embodiment of the present application also discloses an electronic device, in which a computer program is stored in a computer-readable storage medium. When the computer program is loaded and executed by a processor, the above-mentioned method for identifying rationality of medication is adopted.

[0078] The electronic device may be a desktop computer, a laptop computer, a cloud server or other electronic device, and the electronic device includes but is not limited to a processor and a memory. For example, the electronic device may also include input and output devices, a network access device, and a bus.

[0079] Among them, the processor can adopt a central processing unit (CPU). Of course, according to actual usage, other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. can also be adopted. The general-purpose processor can adopt a microprocessor or any conventional processor, etc., and this application does not impose any restrictions on this.

[0080] Among them, the memory can be an internal storage unit of the electronic device, such as a hard disk or memory of the electronic device, or it can be an external storage device of the electronic device, such as a plug-in hard disk, a smart memory card (SMC), a secure digital card (SD) or a flash memory card (FC) equipped on the electronic device. Moreover, the memory can also be a combination of an internal storage unit and an external storage device of the electronic device. The memory is used to store computer programs and other programs and data required by the electronic device. The memory can also be used to temporarily store data that has been output or is to be output, and this application does not impose any restrictions on this.

[0081] Among them, through this electronic device, a method for identifying rationality of medication in the above embodiment is stored in the memory of the electronic device, and is loaded and executed on the processor of the electronic device for easy use.

[0082] The above is only an exemplary embodiment of the present disclosure and cannot be used to limit the scope of the present disclosure. That is, any equivalent changes and modifications made according to the teachings of the present disclosure are still within the scope of the present disclosure. This application is intended to cover any variation, use or adaptive change of the present disclosure, which follows the general principles of the present disclosure and includes common knowledge or customary technical means in the technical field that are not recorded in the present disclosure. The description and examples are only regarded as exemplary, and the scope and spirit of the present disclosure are defined by the claims.

Claims

1. A method for identifying rationality of medication use, It is characterized in that The method comprises: Obtaining a medication knowledge graph, wherein the medication knowledge graph includes a correspondence between diseases and therapeutic drugs and a usage relationship between therapeutic drugs; Obtaining the patient's clinical diagnosis information and medication list information, and filtering target information from the medication knowledge graph based on the clinical diagnosis information and the medication list information, wherein the target information is used to determine whether the medication list information is reasonable for the clinical diagnosis information; According to the target information and the fine-tuned large model, the medication rationality result corresponding to the medication list information is determined. The fine-tuned large model is a model that combines graph data and large language model technology and is fine-tuned and trained.

2. The method for identifying rationality of medication according to claim 1, It is characterized in that The obtaining of the medication knowledge graph specifically includes: Obtaining medical content, including descriptions of different diseases and corresponding therapeutic drugs and instructions for use of different therapeutic drugs; Desensitizing the medical content to obtain processed results; Extracting at least one knowledge graph triple from the processed result through a preset target big model, wherein the knowledge graph triple is a set consisting of diseases, therapeutic drugs and the corresponding relationship between the two or a set consisting of different therapeutic drugs and the corresponding usage relationship between different therapeutic drugs, and the target big model is a model combining graph data and large language model technology; Each of the knowledge graph triples is merged into an initial knowledge graph in a preset graph database to obtain a medication knowledge graph.

3. The method for identifying rationality of medication according to claim 1, It is characterized in that Before determining the medication rationality result corresponding to the medication list information according to the target information and the fine-tuned large model, the method further includes: Generating fine-tuning training samples according to the medication knowledge graph; Fine-tune the preset target large model through the fine-tuning training samples to obtain a fine-tuned large model and store it in a preset database, wherein the target large model is a model that combines graph data and large language model technology; When it is necessary to identify the rationality of medication, the fine-tuned large model is called from the database.

4. The method for identifying rationality of medication according to claim 1, It is characterized in that The step of screening target information from the medication knowledge graph according to the clinical diagnosis information and the medication list information specifically includes: Constructing a first graph query statement according to the clinical diagnosis information; Through the first graph query statement, query the correct treatment drug set and the contraindication treatment drug set corresponding to the disease in the clinical diagnosis information from the medication knowledge graph; Constructing a second graph query statement according to each actual therapeutic drug in the medication list information; According to the second graph query statement, the target usage relationship between each of the actual therapeutic drugs is queried from the medication knowledge graph, and the correct therapeutic drug set, the contraindicated therapeutic drug set and each of the target usage relationships are determined as target information.

5. The method for identifying rationality of medication according to claim 4, It is characterized in that Determining the medication rationality result corresponding to the medication list information according to the target information and the fine-tuned large model specifically includes: Splicing the target information, the clinical diagnosis information and the medication list information to obtain a splicing result; Construct prompt task information, and concatenate the prompt task information with the concatenation result to obtain a final prompt instruction; The final prompt instruction is input into the fine-tuned large model to obtain at least one judgment result, and based on each of the judgment results, the medication rationality result corresponding to the medication list information is determined. The final prompt instruction is used to judge the rationality of the medication list information based on the target information.

6. The method for identifying rationality of medication according to claim 5, It is characterized in that The prompt task information includes a first determination task, a second determination task, a third determination task and a fourth determination task, wherein the first determination task is a task for determining whether the actual therapeutic drug is effective, the second determination task is a task for determining whether the actual therapeutic drug is a contraindicated drug for the disease in the clinical diagnosis information, the third determination task is a task for determining whether there are mutual side effects between the actual therapeutic drugs, and the fourth determination task is a task for determining whether the dosage of the actual therapeutic drug is reasonable. The final prompt instruction is input into the fine-tuned large model to obtain at least one determination result, specifically including: Inputting the final prompt instruction into the fine-tuned large model, if the actual therapeutic drug does not exist in the correct therapeutic drug set, determining that the corresponding actual therapeutic drug is invalid as the determination result of the first determination task; If the actual therapeutic drug exists in the set of contraindicated therapeutic drugs, determining that the corresponding actual therapeutic drug is a contraindicated drug for the disease in the clinical diagnosis information is a determination result of the second determination task; If the target usage relationship between the actual therapeutic drugs is that there are mutual side effects, then determining that there are mutual side effects between the corresponding actual therapeutic drugs is a determination result of the third determination task; If the dosage of the actual therapeutic drug in the correct therapeutic drug set is inconsistent with the dosage of the corresponding correct therapeutic drug, the unreasonable dosage of the corresponding actual therapeutic drug is determined as the determination result of the fourth determination task.

7. The method for identifying rationality of medication according to claim 6, It is characterized in that Determining the rationality of medication corresponding to the medication list information according to each of the determination results specifically includes: If there is a target determination result among the determination results, it is determined that the medication rationality result corresponding to the medication list information is unreasonable, and one of the following exists in the target determination result: a single actual therapeutic drug is ineffective, a single actual therapeutic drug is a contraindicated drug, a single actual therapeutic drug has a mutual side effect with other actual therapeutic drugs, or the dosage of a single actual therapeutic drug is unreasonable; If the target determination result does not exist in the determination results, the medication rationality result corresponding to the medication list information is determined to be reasonable.

8. The method for identifying rationality of medication according to claim 2 or 3, Features: The target large model is Graph+LLM.

9. A device for identifying the rationality of medication, It is characterized in that include: A graph acquisition module (11) is used to acquire a medication knowledge graph, wherein the medication knowledge graph includes a correspondence between diseases and therapeutic drugs and a usage relationship between therapeutic drugs; An information screening module (12) is used to obtain the patient's clinical diagnosis information and medication list information, and screen target information from the medication knowledge graph based on the clinical diagnosis information and the medication list information, wherein the target information is used to determine whether the medication list information is reasonable for the clinical diagnosis information; A rational determination module (13) is used to determine the rationality result of medication corresponding to the medication list information based on the target information and the fine-tuned large model, wherein the fine-tuned large model is a model that combines graph data and large language model technology and is fine-tuned and trained.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, It is characterized in that When the processor loads and executes the computer program, the method according to any one of claims 1 to 8 is adopted.

Images