Method and system for adverse drug reaction event identification based on poly-knowledge hybrid search enhancement

Abstract

The application discloses a drug adverse reaction event identification method and system based on multi-element knowledge mixed retrieval enhancement, which comprises the following steps: extracting drug entities and segmenting the medical records from the clinical medical records to be identified to obtain a drug entity set and a sentence set; for each drug entity extracted, searching the pre-constructed multi-element knowledge base for drug concept knowledge having a superior-inferior relationship with the drug entity and drug adverse reaction knowledge having an adverse reaction relationship with the drug entity; for each sentence segmented, searching the pre-constructed multi-element knowledge base for suspected drug adverse reaction events meeting a first similarity requirement and drug field text knowledge meeting a second similarity requirement; and finally, calling a large language model to identify drug adverse reaction events from the clinical medical records to be identified by taking all the searched knowledge as reference knowledge. The application can improve the accuracy and reliability of drug adverse reaction event identification.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of medical information processing, and particularly relates to a drug adverse reaction event identification method and system based on multi-element knowledge hybrid retrieval enhancement. BACKGROUND

[0002] Drug adverse reaction (ADR, Adverse Drug Reaction) refers to a harmful reaction unrelated to the purpose of drug treatment during the use of a drug.

[0003] In the process of clinical diagnosis and treatment, timely and accurate identification of drug adverse reaction events (ADE, Adverse Drug Event) from the medical records written by clinicians is crucial for patient safety, optimization of drug treatment plans, clinical new drug trial research, potential drug adverse reaction identification, and drug adverse reaction reporting. Clinical medical records contain rich patient medication information and adverse reaction related descriptions, but there is currently a lack of effective methods to accurately identify drug adverse reaction events from these unstructured texts. Traditional methods mainly rely on manual identification, which is inefficient and prone to errors. Existing automatic identification methods based on machine learning or deep learning lack rich external knowledge support and rely on large-scale high-quality labeled corpus, which has deficiencies in identification accuracy and comprehensiveness. SUMMARY

[0004] The application provides a drug adverse reaction event identification method and system based on multi-element knowledge hybrid retrieval enhancement, which improves the accuracy and reliability of drug adverse reaction event identification.

[0005] To achieve the above technical purposes, the application adopts the following technical solutions:

[0006] A drug adverse reaction event identification method based on multi-element knowledge hybrid retrieval enhancement, comprising:

[0007] Step 1: Obtain the clinical medical record to be identified, extract the drug entity and segment the medical record to obtain the drug entity set and the sentence set, respectively;

[0008] Step 2: For each extracted drug entity, search in the pre-constructed multi-element knowledge base: drug concept knowledge with hierarchical relationship, and drug adverse reaction knowledge with adverse reaction relationship;

[0009] For each segmented sentence, search in the pre-constructed multi-element knowledge base: suspected drug adverse reaction events belonging to the N1 most similar and satisfying the first similarity threshold, and drug field text knowledge belonging to the N2 most similar and satisfying the second similarity threshold;

[0010] Step 3, calling a large language model to identify adverse drug reaction events from the clinical course record to be identified, with all the knowledge retrieved in step 2 as reference knowledge.

[0011] Further, the multi-element knowledge base includes a drug concept knowledge base DRCK, an adverse drug reaction knowledge base ADRK, an adverse drug reaction event knowledge base ADEK, and a drug domain text knowledge base DDPK; the drug concept knowledge base DRCK and the adverse drug reaction knowledge base ADRK are respectively used to store drug concept knowledge and adverse drug reaction knowledge; the adverse drug reaction event knowledge base ADEK and the drug domain text knowledge base DDPK are respectively used to store suspected adverse drug reaction events and drug domain text knowledge.

[0012] Further, the drug concept knowledge base DRCK stores the hyponym-hypernym concept relationship of each drug entity in the form of triples;

[0013] The adverse drug reaction knowledge base ADRK stores the relationship between each drug entity and an adverse reaction entity in the form of triples;

[0014] The adverse drug reaction event knowledge base ADEK is used to store suspected adverse drug reaction events, and each suspected adverse drug reaction event includes a sentence description, a drug entity, an adverse reaction entity, and a label;

[0015] The drug domain text knowledge base DDPK stores drug domain text knowledge of each drug entity.

[0016] Further, the hyponym-hypernym concept relationship of each drug entity is constructed in the drug concept knowledge base DRCK using the ontology semantic relationship SubClassOf;

[0017] Using a depth-first search method, starting from a target drug entity, all hypernym drug concepts of the target drug entity are retrieved upwards and all hyponym drug concepts of the target drug entity are retrieved downwards in the drug concept knowledge base DRCK using the ontology semantic relationship SubClassOf.

[0018] Further, the types of adverse reaction entities are diseases, symptoms, signs, test results, and examination results, and the relationship between the drug entity and the adverse reaction entity is extracted from medical literature, drug instructions, and clinical research reports.

[0019] Further, in retrieving suspected adverse drug reaction events, based on the vectorized similarity between the target sentence and the sentence description in each suspected adverse drug reaction event in the adverse drug reaction event knowledge base ADEK, the N1 suspected adverse drug reaction events with the highest vector similarity and satisfying the similarity not less than the first similarity threshold are selected.

[0020] Further, in retrieving drug domain text knowledge, based on the vectorized similarity between the target sentence and the drug domain text knowledge of each drug entity in the drug domain text knowledge base DDPK, the N2 drug domain text knowledge with the highest vector similarity and satisfying the similarity not less than the second similarity threshold are selected.

[0021] Further, step 3 fills in system instructions, reference knowledge, output requirements and user input according to the preset structured prompt word framework template to call the large language model to identify adverse reaction events;

[0022] The system instruction requires the large language model to identify the adverse drug reaction event information from the clinical course record with the reference knowledge as the context;

[0023] The output requirement is used to give the output data format of the identified adverse reaction event;

[0024] The user input is the clinical course record to be identified.

[0025] A drug adverse reaction event identification system based on multi-knowledge hybrid retrieval enhancement, comprising:

[0026] A preprocessing module for extracting drug entities and segmenting course records from the clinical course record to be identified to obtain a drug entity set and a sentence set, respectively;

[0027] A multi-knowledge base module including a drug concept knowledge base DRCK, an adverse drug reaction knowledge base ADRK, an adverse drug reaction event knowledge base ADEK and a drug domain text knowledge base DDPK, respectively for storing drug concept knowledge, adverse drug reaction knowledge, suspected adverse drug reaction events and drug domain text knowledge;

[0028] A knowledge retrieval module for: (1) for each extracted drug entity, retrieving in the multi-knowledge base: drug concept knowledge having a hierarchical relationship with it, and adverse drug reaction knowledge having an adverse reaction relationship with it; (2) for each segmented sentence, retrieving in the pre-constructed multi-knowledge base: N1 suspected adverse drug reaction events with the largest similarity and satisfying the first similarity threshold, and N2 drug domain text knowledge with the largest similarity and satisfying the second similarity threshold;

[0029] The large language model module is used to identify adverse drug reaction events from the clinical course record to be identified with all the retrieved knowledge as reference knowledge.

[0030] An electronic device includes a memory and a processor, the memory stores a computer program, and the computer program is executed by the processor to make the processor implement the multi-knowledge hybrid retrieval enhanced adverse drug reaction event identification method of any one of the above.

[0031] Compared with the prior art, the present application has the following beneficial effects:

[0032] Firstly, the present application innovatively constructs a multi-knowledge base, which deeply integrates the hierarchical relationship between drug concepts, triple knowledge related to adverse drug reactions, real clinical adverse drug reaction event cases, and professional text knowledge in the field, providing a solid and rich knowledge backup for accurate identification of adverse drug reaction events.

[0033] Secondly, the present application adopts a multi-knowledge driven retrieval enhancement generation paradigm, which can deeply mine and fully utilize the relevant knowledge obtained by retrieval, fully activate the potential of the large language model, and significantly enhance its identification efficiency of adverse drug reaction events in clinical course records. In this way, the present application not only improves the accuracy of adverse drug reaction event identification, but also greatly improves the recall rate, ensuring that more potential adverse drug reaction events can be captured in time and accurately.

[0034] Thirdly, the present application can lay a solid and solid foundation for the research of downstream tasks such as clinical medication safety guarantee, promotion of new drug clinical research and test, identification and prediction of potential adverse drug reactions, and intelligent reporting of adverse drug reactions, and has important clinical application value and scientific research significance. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 is a flowchart of the adverse drug reaction event identification method described in the embodiments of the present application.

[0036] Figure 2 is an example of a drug concept knowledge base described in the embodiments of the present application.

[0037] Figure 3 is a structured prompt word framework template and an instance described in the embodiments of the present application, wherein Figure 3 (a) and Figure 3 (b) correspond to the structured prompt word framework template and the instance, respectively.

[0038] Figure 4 is the definition of the ADE JSON Schema specification described in the embodiments of the present application.

[0039] Figure 5 is the identification result of adverse drug reaction events based on the ADE JSON Shema specification according to an embodiment of the application.

[0040] Figure 6 is an example one of identifying adverse drug reaction events from clinical course records.

[0041] Figure 7 is an example two of identifying adverse drug reaction events from clinical course records.

[0042] Figure 8 is an example three of identifying adverse drug reaction events from clinical course records.

[0043] Figure 9 is an example four of identifying adverse drug reaction events from clinical course records. DETAILED DESCRIPTION

[0044] This embodiment is based on the technical solution of the application, and gives a detailed implementation and specific operation process, which further explains and describes the technical solution of the application.

[0045] The embodiment provides a drug adverse reaction event identification method based on multi-element knowledge mixed retrieval enhancement, as shown in the method. Figure 1 The method comprises the following steps:

[0046] Step 1, obtaining a clinical course record to be identified, extracting drug entities and segmenting the course record to obtain a drug entity set and a sentence set, respectively.

[0047] The embodiment specifically adopts a BiLSTM-CRF deep learning network, trains and constructs a drug entity identification model based on an annotated drug entity corpus, and thus identifies a drug entity set D from the clinical course record to be identified. The identification method is defined as follows:

[0048] ;

[0049] Among them, is an entity identification model based on BiLSTM-CRF, is a method for realizing drug entity identification, and D is the drug entity set identified from the clinical course record. The BiLSTM-CRF deep learning network is prior art, and can refer to the literature “Bidirectional LSTM-CRF Models for Sequence Tagging”, and the application will not be described in detail.

[0050] Suppose the clinical course record to be identified is The example is "patient intravenous infusion of levofloxacin after hand skin itching, rash, consider levofloxacin allergy, stop using, pay attention to self-purchased cefuroxime anti-infection treatment, review urine routine after anti-infection treatment." The identified entity set D is ["levofloxacin", "cefuroxime"].

[0051] In order to effectively retrieve relevant knowledge from the knowledge base, the embodiment adopts punctuation symbol segmentation and sliding window mode to segment the clinical course record The segmented sentence set S is obtained by sentence segmentation, and the segmentation method is defined as follows:

[0052] ;

[0053] Wherein, chunk is the segmentation method of the clinical course record; win_size represents the window size, and the default is 3; step is the step distance of the sliding window, and the default is 2; flag represents the symbol that needs to be cut, and the default is ",?;." that is, medical_record is cut according to the symbol in flag. Finally, the segmented sentence set S is obtained. Through such segmentation method, the overlapping of the segmented sentences is ensured, and the information loss is avoided, which provides support for the subsequent knowledge base retrieval.

[0054] The above Examples, the sentence set S = ["patient intravenous infusion of levofloxacin after hand skin itching, rash, consider levofloxacin allergy", "consider levofloxacin allergy, stop using, pay attention to self-purchased cefuroxime anti-infection treatment", "pay attention to self-purchased cefuroxime anti-infection treatment, review urine routine after anti-infection treatment"].

[0055] Step 2, knowledge retrieval.

[0056] On the one hand, for each extracted drug entity, the pre-constructed multi-knowledge base is retrieved: the drug concept knowledge having the upper and lower relationship with it, and the drug adverse reaction knowledge having the adverse reaction relationship with it; on the other hand, for each segmented sentence, the pre-constructed multi-knowledge base is retrieved: the suspected drug adverse reaction event belonging to the N1 largest and satisfying the first similarity threshold, and the drug field text knowledge belonging to the N2 largest and satisfying the second similarity threshold.

[0057] Firstly, the pre-constructed multi-knowledge base in the embodiment includes drug concept knowledge base DRCK, drug adverse reaction knowledge base ADRK, drug adverse reaction event knowledge base ADEK and drug field text knowledge base DDPK.

[0058] (1) Construct a drug concept knowledge base (DRCK, Drug Concept Knowledge).

[0059] The ontology semantic relationship SubClassOf is a key attribute for representing the hierarchical relationship between classes in the ontology, and has transitivity. Therefore, the embodiment uses the ontology semantic relationship SubClassOf to construct the superior and inferior concept relationship of each drug entity in the drug concept knowledge base DRCK.

[0060] Each piece of drug concept knowledge c (c ∈ DRCK) in the DRCK stores the relationship between each drug entity and an adverse reaction entity in the form of a triple, and the mode is represented as:

[0061] ;

[0062] Among them, represents a drug entity.

[0063] For example: <aspirin, subClassOf, non-steroidal anti-inflammatory drugs> represents that "aspirin is a non-steroidal anti-inflammatory drug", <non-steroidal anti-inflammatory drugs, subClassOf, anti-inflammatory drugs> represents that "non-steroidal anti-inflammatory drugs are a kind of anti-inflammatory drugs", and based on the transitivity of subClassOf, it can be inferred that "aspirin is also an anti-inflammatory drug".

[0064] The subClassOf relationship clearly defines the hierarchical classification of drug entities in different dimensions, as shown in Figure 2 , thereby supporting reasoning and querying. Moreover, the drug concept knowledge in the embodiment is stored in a graph database (such as Neo4j), with each drug entity as a node and subClassOf as an edge connecting two drug entity nodes with a subordinate relationship.

[0065] (2) Construct a drug adverse reaction knowledge base (ADRK, Adverse Drug Reaction Knowledge).

[0066] Each piece of drug adverse reaction knowledge r (r ∈ ADRK) in the ADRK is also described using triples, represented as:

[0067] ;

[0068] Among them, drug is a drug entity, has_adr represents the existence of an adverse reaction relationship, and reaction represents an adverse reaction entity, which can be a disease, a symptom, a sign, a test result, an examination result, etc.

[0069] When constructing the knowledge graph, the relationship between drug and adverse reaction entities is extracted from a large amount of medical literature, drug instructions, clinical research reports and other data sources. For example, the following knowledge can be extracted from the drug instructions:

[0070] <Aspirin, has_adr, gastric mucosa injury bleeding>

[0071] <Penicillin, has_adr, anaphylactic shock>

[0072] <Primaquine, has_adr, hemolytic anemia>

[0073] The drug adverse reaction knowledge graph is stored in a graph database (such as Neo4j), with each drug entity and adverse reaction entity as a node, and has_adr as an edge connecting the corresponding drug entity and adverse reaction entity node.

[0074] (3) Construct the drug adverse reaction event knowledge base (ADEK, Adverse Drug Event Knowledge).

[0075] In this embodiment, when constructing ADEK, suspected drug adverse reaction event information is annotated and extracted from a large number of clinical records, including positive examples and negative examples. The positive example is a record that is clearly a drug adverse reaction event, such as "patient developed a rash after taking amoxicillin"; the negative example is a record that appears to be a drug adverse reaction but is not actually so, such as "patient developed a cold during drug use, but the symptoms were not related to the drug used". Each drug adverse reaction event case includes the attribute sentence, which represents the suspected drug adverse reaction event statement; the attribute label, which represents the positive or negative example; the attribute drugs, which represents the drug entity; and the attribute reactions, which represents the adverse reaction entity. The ADEK example is shown in Table 1.

[0076] ;

[0077] For any drug adverse reaction event knowledge base represented as a (a∈ADEK), a.sentence represents the suspected adverse reaction event description statement in the case knowledge; a.drugs represents the drug entity that causes the adverse reaction in the case knowledge; a.reactions represents the adverse reaction entity caused by the drug; and a.label represents whether the current case knowledge is a positive example or a negative example.

[0078] (4) Construct the drug domain text knowledge base (DDPK, Drug Domain Paragraph Knowledge).

[0079] The drug domain text knowledge base DDPK is mainly derived from the text passages describing drugs in drug instructions, literature, and textbooks. For any drug domain text knowledge represented as p (p ∈ DDPK), an example of the knowledge is as follows:

[0080] "ibuprofen is a widely used non-steroidal anti-inflammatory drug (NSAID) that is primarily effective in relieving mild to moderate pain, such as headaches, joint pain, toothaches, and menstrual cramps, as well as effectively reducing fever symptoms caused by colds or flu. It works by inhibiting the synthesis of prostaglandins in the body to exert antipyretic, analgesic, and anti-inflammatory effects. Ibuprofen usually comes in the form of oral tablets, capsules, or liquids, making it convenient for patients of different ages to use. Although ibuprofen is effective in relieving pain and reducing fever, long-term or excessive use may cause gastrointestinal discomfort, such as stomach pain, nausea, and even serious side effects such as gastric ulcers or bleeding. Therefore, when using ibuprofen, it is necessary to follow the recommended dosage on the doctor's advice or the drug instructions to avoid self-increasing or decreasing the dosage to ensure safe and effective use of the drug."

[0081] Step 2 of this step retrieves relevant knowledge in the above-mentioned pre-constructed multi-knowledge base for each extracted drug entity and each segmented sentence.

[0082] (1) Retrieve drug concept knowledge in DRCK.

[0083] Use the subclassOf relationship for hierarchical retrieval to obtain relevant drug concept hierarchical information. During the retrieval process, by traversing the subclassOf relationship chain of the target drug entity, determine the position of the target drug entity in the concept hierarchy and the related superordinate and subordinate concepts.

[0084] Specifically, use the depth-first search (DFS) algorithm to traverse the relationship chain in DRCK, starting from the target drug entity, and tracing upwards to the top-level concept to obtain all superordinate concepts; explore all subordinate concepts downward.

[0085] Let the target drug entity to be retrieved be d (d ∈ D), and all superordinate concept entity sets of the drug concept entity d retrieved by the depth-first search algorithm be and all subordinate concept entity sets be , represented as:

[0086] ;

[0087] ;

[0088] Figure 2 ​For example, assuming that the target drug entity to be retrieved is d = "small molecule drug", then Ud = {drug, organic compound, chemical substance}, and Ld = {non-steroidal anti-inflammatory drug, aspirin, ibuprofen, antibiotic, penicillin}.

[0089] The method of retrieving a set of triples of drug concept knowledge C from the drug concept knowledge DRCK for the extracted drug entity set D is represented as:

[0090] ;

[0091] wherein the drug concept knowledge c (c e C) satisfies or .

[0092] (2) Retrieving drug adverse reaction knowledge from ADRK.

[0093] Based on the extracted drug entity set D, the adverse reaction entities related to each target drug entity are retrieved from the ADRK. The "reaction" entity corresponding to the "drug" entity can be obtained by matching. The query statement of the graph database is used for retrieval, and the query formula can be represented as: let the target drug be d (d e D), and the retrieved adverse reaction entity set be E d .

[0094] ;

[0095] For example, the adverse reaction entities corresponding to "aspirin" can be obtained as E = ["gastrointestinal bleeding", "allergic reaction", etc.].

[0096] The method of retrieving a set of triples of drug adverse reaction knowledge R from the drug adverse reaction knowledge ADRK for the drug entity set D is represented as:

[0097] ;

[0098] wherein the drug adverse reaction knowledge r (r e R) satisfies .

[0099] (3) Retrieving drug adverse reaction events from ADEK.

[0100] The vector retrieval method is used to calculate the vector similarity between the segmented sentence s (s∈S) of the clinical course record to be identified and the sentence description (i.e. a.sentence) of each suspected drug adverse reaction event a (a∈ADEK) in ADEK. First, the text vector model is used to convert the sentence s in the clinical course record into an n (n=768) dimensional vector A, and the sentence description a.sentence of the suspected drug adverse reaction event in ADEK is also converted into an n-dimensional vector B. The cosine similarity is used to calculate the cosine distance between the two vectors, and the calculation formula is as follows:

[0101] ;

[0102] Wherein, is an index for measuring the similarity of the two vectors in the direction, and the value range is between [-1, 1]. The value is closer to 1, indicating that the direction is more similar; closer to -1, the direction is more opposite; close to 0 indicates that the direction is almost orthogonal (irrelevant). According to the needs of the present application, the direction needs to be similar, and the closer to 1, the more similar the retrieved ADE case knowledge is, and the more reference significance it has.

[0103] The first similarity threshold simScore1 is set in this embodiment, and the most similar N1 drug adverse reaction events are retrieved.

[0104] The sentence s and the sentence description a.sentence of the suspected drug adverse reaction event are collectively referred to as text t, and the text t is converted into a fixed dimension w (w=768) vector is expressed as:

[0105] ;

[0106] According to the sentence set S segmented from the clinical course record, the topN1 is retrieved from ADEK, and the similarity is greater than or equal to the first similarity threshold simScore1, and the method for obtaining the drug adverse reaction event set A is:

[0107] ;

[0108] Wherein for any retrieved drug adverse reaction event a (a∈A), there is s (s∈S) that satisfies ConsineSimilarity (Va, Vs) >=simScore1, and the size of the set A is less than or equal to topN1.

[0109] (4) Retrieving drug field text knowledge in DDPK.

[0110] Similarly, using vector retrieval, the vector similarity between the segmented sentence s (s∈S) of the clinical course record to be identified and the drug domain knowledge text p (p∈DDPK) is calculated. The calculation formula is the same as that introduced in ADEK for retrieving adverse drug reaction events. The text knowledge p that satisfies the second similarity threshold simScore2 and the topN2 (default is 5) is used as the retrieval result.

[0111] Based on the sentence set S obtained from segmenting clinical course records, the top N2 similarity scores (simScore2) are retrieved from DDPK to obtain the drug domain knowledge set P, which has a similarity score greater than or equal to the second similarity threshold.

[0112] ;

[0113] For any retrieved drug domain text knowledge (p∈P), there exists s (s∈S) that satisfies ConsineSimilarity(Vp,Vs)>=simScore2, and the size of the set P is less than or equal to topN2.

[0114] Step 3: Invoke the large language model, using all the knowledge retrieved in Step 2 as reference knowledge, and identify adverse drug reaction events from the clinical course records to be identified.

[0115] This embodiment uses a pre-defined structured prompt word framework template, such as... Figure 3 As shown in (a), it is mainly divided into the following four parts:

[0116] (1) System instructions: Clearly define the system instructions, that is, use reference knowledge as context to identify adverse drug reaction event information from clinical course records.

[0117] (2) Reference Knowledge: Provides a reference knowledge template. Based on the input medical_record, it retrieves drug concept knowledge (C), adverse drug reaction knowledge (R), ADE case knowledge (A), and drug domain knowledge (P) from the multi-dimensional knowledge base (DRCK, ADRK, ADEK, DDPK). The retrieved knowledge needs to be filled into the corresponding parameters.

[0118] (3) Output requirements: Specify the data format of the output generated by the instruction, and require that it be serialized in JSON and meet the ADE JSON Schema specification.

[0119] (4) User input: The user input section requires the input of the clinical course record to be analyzed. Replace the parameter medical_record with the content of the clinical course record to be analyzed.

[0120] Fill in the system instructions, reference knowledge, output requirements and user input according to the preset structured prompt word framework template, and obtain a structured prompt word framework instance, such as Figure 3 (b) as shown.

[0121] Based on the to-be-recognized clinical medical record medical_record, and the retrieved drug concept knowledge set C, drug adverse reaction knowledge set R, drug adverse reaction event set A, and drug domain text knowledge set P, a structured prompt word framework instance is constructed and represented as:

[0122] ;

[0123] After completing the construction of the structured prompt word framework instance, a large language model LLM (such as GPT-4o, Qwen, DeepseekV3) can be called to generate, thereby identifying the drug adverse reaction event information result ADEs.

[0124] .

[0125] In order to accurately express and standardize the ADE recognition output, the present embodiment adopts JSON Schema to formulate the output specification of the structured prompt word framework. Through analysis, the drug adverse reaction events identified from the given clinical medical record have the following characteristics:

[0126] (1) May describe 0 or more drug adverse reaction event information.

[0127] (2) There will be drug entities causing adverse reactions in each drug adverse reaction event, which may be 0 or more. When it is 0, it means that there is no specific drug entity in the medical record.

[0128] (3) Each drug adverse reaction event will describe the adverse reaction entity that occurs after taking the drug, and the adverse reaction entity may be one or more.

[0129] (4) There is an explicit causal relationship between the drug entity and the adverse reaction entity in each drug adverse reaction event.

[0130] According to the above characteristics of drug adverse reaction events, the present application adopts a JSON serialization output method and formulates an output specification ADE JSON Schema, as shown in Figure 4 , described as follows:

[0131] (a) Root structure: array, containing 0 or more ADEs. If the medical record to be analyzed has no drug adverse reaction events, the recognition result is an empty array [];

[0132] (b) Each ADE structure contains three fields, respectively:

[0133] 1) sentence: string, the sentence in the medical record describing the ADE, such as "The patient developed a rash after taking aspirin";

[0134] 2) drugs: string array, the list of drug entities identified in the ADE, such as ["aspirin"];

[0135] 3) reactions: string array, the list of adverse reaction entities identified in the ADE, which are caused by drugs and have a causal relationship between them. For example, ["rash"], which is an adverse reaction caused by taking the drug aspirin.

[0136] By formulating the above ADE JSON Schema specification, the core information elements of drug adverse reaction events can be systematically and standardizedly represented, and clear and verifiable data specification constraints are provided for the target identification task of downstream methods.

[0137] Based on the clinical medical record medical_record example, the identification of drug adverse reaction events follows the ADE JSON Schema output as shown in Figure 5 .

[0138] Based on the method of the present application, 6024 clinical real medical records annotated with ADE are used as the reference set. The recognition results of the method of the present application are compared with the standard results, and the 10-fold cross-validation analysis method is used. The accuracy Precison, the recall Recall and the comprehensive evaluation index F1 are used to evaluate the effect of the method of the present application. The experimental evaluation results are shown in Table 2:

[0139] ;

[0140] As shown in Table 2, based on the method of the present application, three large language models are used for evaluation and testing, which are enire3.5-8k, DeepseekV3 and Qwen-Turbo. Among them, the recognition accuracy based on DeepseekV3 reaches 98.38%, the recall rate is 92.97%, and the comprehensive F1 value is 0.956. The experimental results show the effectiveness of the method of the present application.

[0141] Secondly, combined with specific clinical medical record case analysis, some examples of the beneficial effects of the present application are shown as follows:

[0142] (1) The method of the present application can effectively identify drug adverse reaction events, and can also infer specific drug entities combined with clinical context semantic relationship, as shown in Figure 6 .

[0143] (2) The method of the present application can effectively identify adverse drug reaction events, and when there are multiple drug entities, it can effectively analyze the causal relationship between the drug entities and the adverse reaction entities, and exclude non-adverse reaction causing drug entities, such as shown in Figure 7 .

[0144] (3) The method of the present application can effectively identify adverse drug reaction events, can understand abnormal test index values, and convert them into qualitative test results as adverse reaction entities, such as shown in Figure 8 .

[0145] (4) The method of the present application can effectively identify adverse drug reaction events, and can identify multiple adverse drug reaction events described in the medical history records, such as shown in Figure 9 .

[0146] The above examples are preferred embodiments of the present application, and those of ordinary skill in the art can make various transformations or improvements on the basis of the above examples, and these transformations or improvements should all belong to the scope of protection required by the present application without departing from the general concept of the present application.

Claims

1. A method for identifying adverse drug reaction events based on multi-dimensional knowledge hybrid retrieval enhancement, characterized in that, include: Step 1: Obtain the clinical course records to be identified, extract drug entities and segment the course records from them, and obtain the drug entity set and sentence set respectively; Step 2: For each extracted drug entity, retrieve drug concept knowledge with hierarchical relationship with it and drug adverse reaction knowledge with adverse reaction relationship with it from the pre-constructed multivariate knowledge base; For each segmented sentence, retrieve from the pre-built multivariate knowledge base: suspected adverse drug reaction events that have the highest similarity to the N1 and meet the first similarity threshold, and drug domain text knowledge that has the highest similarity to the N2 and meets the second similarity threshold; Step 3: Call the large language model and use all the knowledge retrieved in Step 2 as reference knowledge to identify adverse drug reaction events from the clinical course records to be identified; The multi-dimensional knowledge base includes a Drug Concept Knowledge Base (DRCK), an Adverse Drug Reaction Knowledge Base (ADRK), an Adverse Drug Reaction Event Knowledge Base (ADEK), and a Drug Domain Text Knowledge Base (DDPK). The DRCK and ADRK are used to store drug concept knowledge and adverse drug reaction knowledge, respectively. The ADEK and DDPK are used to store suspected adverse drug reaction events and drug domain text knowledge, respectively. The Drug Concept Knowledge Base (DRCK) stores the hierarchical relationship of each drug entity in the form of triples. The Adverse Drug Reaction Knowledge Base (ADRK) stores the relationship between each drug entity and the adverse reaction entity in a triplet format. The Adverse Drug Reaction Event Knowledge Base (ADEK) is used to store suspected adverse drug reaction events. Each suspected adverse drug reaction event includes a statement description, a drug entity, an adverse reaction entity, and a label. The Drug Domain Text Knowledge Base (DDPK) stores drug domain text knowledge for each drug entity. Specifically, the ontology semantic relation SubClassOf is used to construct the hierarchical concept relationship of each drug entity in the Drug Concept Knowledge Base DRCK; a depth-first search method is adopted, starting from the target drug entity, using the ontology semantic relation SubClassOf in the Drug Concept Knowledge Base DRCK to retrieve all the superior drug concepts of the target drug entity upwards and all the subordinate drug concepts of the target drug entity downwards. In the process of searching for suspected adverse drug reaction events, the N1 suspected adverse drug reaction events with the highest vector similarity and satisfying that the similarity is not less than the first similarity threshold are selected based on the vector similarity between the target sentence and the sentence description in each suspected adverse drug reaction event in the Adverse Drug Reaction Event Knowledge Base (ADEK). Specifically, when retrieving drug domain text knowledge, based on the vectorized similarity between the target sentence and the drug domain text knowledge of each drug entity in the Drug Domain Text Knowledge Base (DDPK), the N2 drug domain text knowledge with the highest vector similarity and satisfying a similarity of not less than the second similarity threshold are selected.

2. The method for identifying adverse drug reaction events according to claim 1, characterized in that, The types of adverse reaction entities are: diseases, symptoms, signs, test results, and examination results. The relationship between the drug entities and adverse reaction entities is extracted from medical literature, drug instructions, and clinical research reports.

3. The method for identifying adverse drug reaction events according to claim 1, characterized in that, Step 3: Fill in the system instructions, reference knowledge, output requirements and user input according to the preset structured prompt word framework template, so as to call the large language model to identify adverse reaction events; The system instructions require the large language model to identify adverse drug reaction event information from clinical course records, using reference knowledge as context. The output requirements are used to specify the output data format for the identified adverse reaction events; The user input is the clinical course record to be identified.

4. A drug adverse reaction event identification system based on multi-source knowledge hybrid retrieval enhancement, used to implement the drug adverse reaction event identification method of claim 1, characterized in that, include: The preprocessing module is used to: extract drug entities and segment the clinical course records to be identified, and obtain drug entity sets and sentence sets respectively; The multi-dimensional knowledge base module includes the Drug Concept Knowledge Base (DRCK), the Adverse Drug Reaction Knowledge Base (ADRK), the Adverse Drug Reaction Event Knowledge Base (ADEK), and the Drug Domain Text Knowledge Base (DDPK), which are used to store drug concept knowledge, adverse drug reaction knowledge, suspected adverse drug reaction events, and drug domain text knowledge, respectively. The knowledge retrieval module is used for: (1) retrieving, for each extracted drug entity, drug concept knowledge that has a hierarchical relationship with it and drug adverse reaction knowledge that has an adverse reaction relationship with it in the multivariate knowledge base; (2) retrieving, for each segmented sentence, suspected drug adverse reaction events that have the highest similarity with it and meet the first similarity threshold and drug domain text knowledge that have the highest similarity with it and meet the second similarity threshold in the pre-constructed multivariate knowledge base; The large language model module is used to identify adverse drug reaction events from clinical course records to be identified, using all retrieved knowledge as reference knowledge.

Images