Clinical practice-based adverse drug reaction intelligent acquisition system and method

Abstract

The invention discloses an intelligent adverse drug reaction acquisition system and method based on clinical practice, and the method comprises the steps: obtaining a medication record, a disease course text and a nursing record of a patient, and an inspection index; s2, performing identification and relation extraction on the data acquired in the S1, and constructing a time-sequenced knowledge graph of medicine-indications-test abnormality-clinical bad symptoms; training a Transform prediction model through a time sequence knowledge graph, complementing a knowledge graph relationship, and predicting a missing relationship and an entity; s3, monitoring the change of physiological indexes of the patient after medication in real time, performing dual-channel early warning, adding the detected adverse reaction into the prediction model in S3, and performing update training on the prediction model; a time sequence knowledge graph is constructed, the correlation between the medicine and the adverse reaction is comprehensively and systematically presented, powerful support is provided for clinical research and supervision, and the medicine safety monitoring level is improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent collection systems, and in particular to an intelligent collection system and method for adverse drug reactions based on clinical practice. Background Art

[0002] With the development of information technology, some electronic monitoring methods have begun to emerge, but most of these methods are simply replacements for traditional manual reporting and fail to make full use of advanced technologies such as big data and artificial intelligence to improve the efficiency and accuracy of monitoring. For example, some systems only realize the electronic recording of adverse drug reaction information, but there are still deficiencies in in-depth data analysis, real-time warning, and knowledge graph construction.

[0003] In the medical field, the monitoring and early warning of adverse drug reactions has always been a key issue. Traditional monitoring methods mainly rely on manual reporting, which is not only prone to underreporting, but also has the problem of delayed reporting, resulting in the failure to timely feedback adverse drug reaction information to relevant regulatory authorities and medical institutions. In addition, the manual reporting method is inefficient when processing large amounts of data and is difficult to meet the high requirements for drug safety in the modern medical environment. Summary of the invention

[0004] The present invention overcomes the deficiencies of the prior art and provides a drug adverse reaction intelligent collection system and method based on clinical practice.

[0005] To achieve the above object, the technical solution adopted by the present invention is: an intelligent collection method for adverse drug reactions based on clinical practice, comprising the following steps:

[0006] S1: Obtain the patient's medication records, medical history texts, nursing records, and test indicators respectively;

[0007] S2: Identify and extract relationships from the data collected by S1, and construct a temporal knowledge graph of "drugs-indications-test abnormalities-clinical adverse symptoms";

[0008] S3: Train the Transformer prediction model through the time-series knowledge graph to complete the knowledge graph relationships and predict missing relationships and entities;

[0009] S4: Real-time monitoring of changes in physiological indicators of patients after medication, and dual-channel early warning, adding the detected adverse reactions to the prediction model in S3, and updating and training the prediction model.

[0010] In a preferred embodiment of the present invention, in S1, the medication record includes the drug name, dosage, and frequency; the test indicators include liver function and blood drug concentration.

[0011] In a preferred embodiment of the present invention, in S2, the recognition and relation extraction are implemented by a BioBERT+BiLSTM-CRF model, which can extract key information such as drugs, indications, test abnormalities, and clinical adverse symptoms from unstructured electronic medical record texts.

[0012] In a preferred embodiment of the present invention, in S2, the temporal knowledge graph has timestamps that can reflect the temporal relationships between entities such as drugs, indications, test abnormalities, and clinical adverse symptoms.

[0013] In a preferred embodiment of the present invention, in S3, the prediction model can predict missing entities and relationships by complementing the relationships in the knowledge graph, enhancing the integrity of the knowledge graph.

[0014] In a preferred embodiment of the present invention, in S4, the dual-channel warning is specifically as follows:

[0015] Channel 1: Analyze the temporal change characteristics of the patient's physiological indicators after individual drug use through the LSTM-Cox model, and calculate the deviation of the current physiological indicators from the physiological indicators before the patient's drug use in real time;

[0016] Channel 2: Analyze the conflict paths between the current drug treatment plan and the patient's underlying diseases and genotypes in the knowledge graph through a graph neural network. If a conflict is detected, a new adverse reaction case is captured.

[0017] An intelligent system for collecting adverse drug reactions based on clinical practice includes a data processing module, a natural language processing module, a knowledge graph construction module, a prediction module, and an update module:

[0018] Data collection module: Obtain the patient's medication records through the hospital information system interface, including drug names, dosages, and frequencies; extract the course text and nursing records from the electronic medical records; access the laboratory system to obtain test indicators in real time, including liver function, blood drug concentration, etc.;

[0019] Natural language processing module: Used to perform entity recognition and relation extraction on the collected data, and construct a temporal knowledge graph containing "drug - indication - test abnormality - clinical adverse symptom".

[0020] Knowledge graph construction module: Based on the output of the natural language processing module, construct a temporal knowledge graph. The nodes in the graph include entities such as drugs, indications, test abnormalities, and clinical adverse symptoms, and the edges represent the relationships between entities, and the temporal changes of the relationships are recorded through timestamps;

[0021] Prediction Model Module: Train a prediction model using the constructed temporal knowledge graph. By complementing the relationships in the knowledge graph, predict missing relationships and entities to make the knowledge graph more complete;

[0022] Dual-channel Early Warning Mechanism Module:

[0023] Channel 1: By analyzing the temporal change characteristics of the physiological indicators of a patient after taking medicine, calculate the deviation degree of the current physiological indicators relative to the patient's own baseline value in real time;

[0024] Channel 2: Analyze the conflict paths between the current medication regimen and the patient's underlying diseases and genotypes in the knowledge graph through a graph neural network. If a conflict is detected, capture new adverse reaction cases;

[0025] Knowledge Graph Update Module: Add the new adverse reaction cases detected by the dual-channel early warning mechanism module to the knowledge graph and use them to update and train the prediction model, further improve the knowledge graph, and enhance the ability to identify potential drug adverse reactions.

[0026] In a preferred embodiment of the present invention, the natural language processing module uses the BioBERT+BiLSTM-CRF model for entity recognition and relationship extraction, and can extract key information such as drugs, indications, test abnormalities, and clinical adverse symptoms from unstructured electronic medical record texts.

[0027] In a preferred embodiment of the present invention, the nodes and edges in the temporal knowledge graph are both timestamped, which can reflect the temporal relationships between entities such as drugs, indications, test abnormalities, and clinical adverse symptoms.

[0028] In a preferred embodiment of the present invention, the dual-channel early warning mechanism module can analyze the conflict paths between the current medication regimen and the patient's underlying diseases and genotypes in the knowledge graph. If a conflict is detected, generate new adverse reaction cases and add them to the knowledge graph.

[0029] The present invention solves the defects in the background technology and has the following beneficial effects:

[0030] (1) The present invention provides an intelligent drug adverse reaction collection system and method based on clinical practice. By comprehensively collecting patients' medication records, medical history texts, nursing records, and test indicators, using advanced models to identify and extract key information, and constructing a temporal knowledge graph, it comprehensively and systematically presents the association between drugs and adverse reactions, provides strong support for clinical research and supervision, and improves the level of drug safety monitoring.

[0031] (2) The present invention provides an intelligent adverse drug reaction collection system and method based on clinical practice. By real-time monitoring of the changes in physiological indicators after patients take drugs, a dual-channel early warning mechanism is adopted, and multi-angle analysis is carried out through the LSTM-Cox model and graph neural network to timely discover potential adverse reactions, provide early warnings for clinicians, assist in rapid response, reduce patient risks, and ensure medication safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments recorded in the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0033] Figure 1 is a three-dimensional structure diagram of a preferred embodiment of the present invention;

[0034] Figure 2 is the drawing of the intelligent adverse drug reaction collection system of a preferred embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0035] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0036] Many specific details are set forth in the following description in order to fully understand the present invention, but the present invention can also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited by the specific embodiments disclosed below.

[0037] In the description of the present application, it should be understood that the orientation or positional relationship indicated by the terms "center", "longitudinal", "lateral", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings. These are only for the convenience of describing the present application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as limiting the scope of protection of the present application. In addition, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Therefore, the features defined with "first", "second", etc. may explicitly or implicitly include one or more of such features. In the description of the present invention, unless otherwise specified, the meaning of "a plurality of" is two or more.

[0038] In the description of the present application, it should be noted that unless otherwise clearly specified and defined, the terms "installed", "connected", "connected" should be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or an integral connection; it may be a mechanical connection or an electrical connection; it may be directly connected or indirectly connected through an intermediate medium, and it may be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to specific circumstances.

[0039] As Figure 1 , Figure 2 shown, a method for intelligent collection of adverse drug reactions based on clinical practice includes the following steps:

[0040] S1: Obtain the medication records, medical record texts, nursing records, and test indicators of patients respectively;

[0041] In the present invention, in S1, the medication records include the drug name, dosage, and frequency; the test indicators include liver function and blood drug concentration.

[0042] It should be noted that the data is obtained respectively through the interface of the hospital information system (HIS) to obtain the patient's medication records in real time, including fields such as the drug name (such as "warfarin"), dosage (such as "5mg / d"), medication frequency (such as "twice a day"), and medication start and end times. For the records with missing dosage or frequency, they are automatically completed by associating with the drug instruction library (such as Micromedex); duplicate data (such as multiple records at the same time point) is deduplicated.

[0043] Extract unstructured texts such as progress notes and nursing logs from electronic medical records (EMRs), and use regular expressions to locate key paragraphs (such as "Reaction after medication: The patient complained of dizziness and nausea"). The timestamp is marked by extracting the event time from the time description in the text (such as "3 days after surgery", "2023-10-05 14:00"), aligning it with the medication time in the HIS system, and accessing test indicators (such as liver function ALT / AST, INR value of blood drug concentration) through the laboratory information system (LIS). Build a real-time data pipeline to ensure processing thousands of concurrent data per second. For each indicator, take the average value within 3 days before medication as the baseline (such as baseline ALT = 20 U / L). If the data is insufficient, use the population average of patients of the same age and gender (such as the average ALT of healthy people = 25 U / L).

[0044] For example, a patient's medication record is "Warfarin 3 mg qd", the EMR records "Subcutaneous ecchymosis appeared on the 2nd day after medication", and the LIS detected that the INR value increased from the baseline of 1.2 to 3.5. The system aligns the three times and generates an original data table with timestamps.

[0045] S2: Identify and extract relationships from the data collected in S1 to construct a time-sequential knowledge graph of "drug - indication - test abnormality - clinical adverse symptom".

[0046] In the present invention, in S2, the identification and relationship extraction are implemented by the BioBERT + BiLSTM-CRF model. The BioBERT + BiLSTM-CRF model can extract key information such as drugs, indications, test abnormalities, and clinical adverse symptoms from unstructured EMR texts. In S2, the time-sequential knowledge graph has timestamps that can reflect the time-sequential relationships between entities such as drugs, indications, test abnormalities, and clinical adverse symptoms.

[0047] It should be noted that in the knowledge graph construction and relationship extraction stage, when designing a clinical NLP model, first, the BioBERT model based on the PubMed corpus is used to vectorize the EMR text to capture medical semantic associations. For example, the model can identify the association between "rash" and "allergic reaction". Then, the BiLSTM-CRF entity recognition technology is utilized. By inputting the word vectors output by BioBERT, four types of entities, namely drugs, indications, test abnormalities, and clinical symptoms, are identified through the bidirectional LSTM and CRF layers. For example, the drug "warfarin", the indication "atrial fibrillation anticoagulation", the test abnormality "elevated ALT", and the clinical symptom "ecchymosis" are identified from the text. Then, the relationships between entities are extracted through the multi-head attention mechanism, and the time difference attribute is added. For example, the relationship "warfarin → causes → elevated INR" is established, and the delay time is recorded as 48 hours. When constructing the temporal knowledge graph, the Neo4j graph database is used. The node attributes include the generic name of the drug, ATC code, dose range, test index type, normal range, unit, and the timestamp accurate to minutes, etc. The edge relationships are clearly defined. For example, the causal relationship and time delay between drugs and test abnormalities, etc. For example, from the text "Three days after taking warfarin, ALT rose to 80 U / L, accompanied by fatigue", entities and relationships are extracted, including entities such as warfarin (drug), elevated ALT (test abnormality), fatigue (clinical symptom), etc., and the causal relationship that warfarin causes elevated ALT and the corresponding time delay.

[0048] S3: Train the Transformer prediction model through the temporal knowledge graph to complete the relationships in the knowledge graph and predict the missing relationships and entities;

[0049] In the present invention, in S3, by completing the relationships in the knowledge graph, the prediction model can predict the missing entities and relationships to enhance the integrity of the knowledge graph.

[0050] It should be noted that during the training of the Transformer model and the knowledge graph completion stage, the knowledge graph triples are converted into a sequence format as the model input. For example, "[CLS]Warfarin[SEP]CAUSES[SEP]Elevated INR[SEP]time_lag=48h[EOS]" (The drug warfarin can cause an increase in the international normalized ratio (INR), and the time interval from the use of warfarin to the increase in INR is 48 hours). A Transformer architecture with 12 encoder layers, 8 attention heads per layer, and a hidden layer dimension of 512 is adopted, and positional encoding is generated through timestamp differences. In terms of the training strategy, a cross-entropy loss function is used, the negative sampling ratio is 1:5, the AdamW optimizer is used, the learning rate is 5e-5, and the batch size is 32. In terms of knowledge graph completion, when a new drug is introduced, the model predicts the possible associated risks based on its pharmacological classification and chemical structure, generates candidate edges, and stores them in the database after being reviewed by doctors. For example, when the new drug "rivaroxaban" is introduced, the model predicts the possible risk of "gastrointestinal bleeding" associated with it based on its pharmacological classification and chemical structure, and generates the candidate edge "rivaroxaban→CAUSES→gastrointestinal bleeding", which is added to the knowledge graph after being reviewed by doctors. For example, the model detects the lack of association between "Drug A and genotype CYP2C192", and based on the metabolic pathways of similar drugs, predicts the edge "Drug A→CONTRAINDICATED_IN→CYP2C192" with a confidence level of 0.92.

[0051] S4: Monitor the changes in the patient's physiological indicators after medication in real time, conduct dual-channel early warning, add the detected adverse reactions to the prediction model in S3, and update and train the prediction model.

[0052] In a preferred embodiment of the present invention, in S4, the dual-channel early warning is specifically as follows:

[0053] Channel 1: Analyze the temporal change characteristics of the patient's individual physiological indicators after medication through the LSTM-Cox model, and calculate the deviation degree of the current physiological indicators relative to the physiological indicators before the patient's medication in real time;

[0054] Channel 2: Analyze the conflict paths between the current medication plan and the patient's underlying diseases and genotypes in the knowledge graph through a graph neural network. If a conflict is detected, a new adverse reaction case is captured.

[0055] It should be noted that in the dual-channel early warning and dynamic update stage, Channel 1 monitors physiological indicators through LSTM-Cox. The physiological indicator sequences of the patient every 5 minutes after medication, such as heart rate, blood pressure, etc., are input. Two-layer LSTM is used to extract the time-series feature vectors, and then the risk function is calculated through the Cox risk model. When the risk value exceeds the set threshold, an alarm is triggered. For example, the risk function is h(t) = h0(t)·exp(0.5·ΔALT + 0.3·ΔHR), where ΔALT is the deviation amplitude of ALT from the baseline, and ΔHR is the heart rate change rate. When h(t) > 0.75, an alarm is triggered. Channel 2 detects conflict paths through GNN. Starting from the current medication plan, the neighborhood aggregation of GNN is used to traverse the paths connected to the patient attributes in the knowledge graph. If there are conflict paths, they are marked as conflicts. For example, starting from the patient's current medication plan "warfarin", traverse the paths connected to the patient attributes "liver cirrhosis" and "CYP2C9*3" in the knowledge graph. If there is a path "warfarin → metabolic pathway → genotype B → adverse reaction C", it is marked as a conflict. The knowledge graph update process includes an artificial review interface and incremental training. The doctor workstation displays the details of the early warning, supports one-key confirmation or rejection, and uses the new data to fine-tune the Transformer model every week. For example, when the patient's genotype is CYP2C9*3, the system detects the path "warfarin → CYP2C9 metabolism → INR increase → bleeding risk" through GNN and triggers an early warning. After the doctor adjusts the dose, the new edge "warfarin → RISK_IN → CYP2C9*3" is added to the knowledge graph.

[0056] An intelligent adverse drug reaction collection system based on clinical practice includes a data processing module, a natural language processing module, a knowledge graph construction module, a prediction module, and an update module:

[0057] Data collection module: Obtain the patient's medication records through the hospital information system interface, including drug name, dose, frequency; extract the course text and nursing records from the electronic medical record; access the laboratory system to obtain test indicators in real time, including liver function, blood drug concentration, etc.;

[0058] Natural language processing module: Used to perform entity recognition and relationship extraction on the collected data, and construct a time-series knowledge graph containing "drug - indication - test abnormality - clinical adverse symptom";

[0059] In the present invention, the natural language processing module uses the BioBERT + BiLSTM-CRF model for entity recognition and relationship extraction, and can extract key information such as drugs, indications, test abnormalities, and clinical adverse symptoms from unstructured electronic medical record texts.

[0060] Knowledge Graph Construction Module: Based on the output of the natural language processing module, a temporal knowledge graph is constructed. The nodes in the graph include entities such as drugs, indications, test abnormalities, and clinical adverse symptoms. The edges represent the relationships between entities, and the temporal changes of the relationships are recorded through timestamps.

[0061] In the present invention, both the nodes and edges in the temporal knowledge graph carry timestamps, which can reflect the temporal relationships between entities such as drugs, indications, test abnormalities, and clinical adverse symptoms.

[0062] Prediction Model Module: Use the constructed temporal knowledge graph to train a prediction model. By complementing the relationships in the knowledge graph, predict missing relationships and entities to make the knowledge graph more complete.

[0063] Dual-channel Warning Mechanism Module:

[0064] Channel 1: By analyzing the temporal change characteristics of the physiological indicators of a patient after individual drug use, calculate the deviation degree of the current physiological indicators relative to the patient's own baseline value in real time.

[0065] Channel 2: Analyze the conflict paths between the current medication regimen and the patient's underlying diseases and genotypes in the knowledge graph through a graph neural network. If a conflict is detected, a new adverse reaction case is captured.

[0066] Knowledge Graph Update Module: Add the new adverse reaction cases detected by the dual-channel warning mechanism module to the knowledge graph and use them to update and train the prediction model, further improving the knowledge graph and enhancing the ability to identify potential drug adverse reactions.

[0067] In a preferred embodiment of the present invention, the dual-channel warning mechanism module can analyze the conflict paths between the current medication regimen and the patient's underlying diseases and genotypes in the knowledge graph. If a conflict is detected, a new adverse reaction case is generated and added to the knowledge graph.

[0068] In terms of system architecture and deployment, the module interaction design includes that the data processing module uses Apache NiFi to achieve automated access and cleaning of HIS / EMR / LIS data streams, the prediction module deploys the Transformer model using PyTorch Serving, and the front-end display uses Vue.js to build a doctor dashboard to display the patient risk heat map in real time. For example, the doctor dashboard shows the patient risk situation in the form of a heat map, where red indicates high risk and green indicates safety. In terms of high-availability deployment, the back-end services are deployed using Docker containerization and managed by a Kubernetes cluster, supporting dynamic scaling. The database uses a Neo4j cluster and MongoDB to store the original data. For example, after being deployed in a certain tertiary hospital, the system processes 12,000 medication records within 24 hours, triggers 58 effective warnings, and the false alarm rate is less than 5%.

[0069] Based on the inspiration of the ideal embodiments of the present invention as above, through the above description, relevant personnel can completely make various changes and modifications without departing from the technical idea of the present invention. The technical scope of the present invention is not limited to the content in the specification, and the technical scope must be determined according to the scope of the claims.

Claims

1. An intelligent collection method for adverse drug reactions based on clinical practice, characterized in that, It includes the following steps: S1: Obtain the medication records, medical history texts, nursing records, and test indicators of the patient respectively; S2: Identify and extract relationships from the data collected in S1, and construct a time-series knowledge graph of "drug - indication - test abnormality - clinical adverse symptom"; S3: Train a Transformer prediction model through the time-series knowledge graph to complete the relationships in the knowledge graph and predict missing relationships and entities; S4: Real-time monitor the changes in the physiological indicators of the patient after medication, and conduct dual-channel early warning. Add the detected adverse reactions to the prediction model in S3 and update and train the prediction model.

2. The intelligent acquisition method for adverse drug reactions based on clinical practice according to claim 1, wherein: In the above S1, the medication records include drug names, dosages, and frequencies; the test indicators include liver function and blood drug concentration.

3. The intelligent collection method for adverse drug reactions based on clinical practice according to claim 1, characterized in that: In the above S2, the identification and relationship extraction are implemented through the BioBERT+BiLSTM-CRF model, and the BioBERT+BiLSTM-CRF model can extract key information such as drugs, indications, test abnormalities, and clinical adverse symptoms from unstructured electronic medical record texts.

4. The intelligent collection method for adverse drug reactions based on clinical practice according to claim 1, characterized in that: In the above S2, the time-series knowledge graph has timestamps that can reflect the time-series relationships between entities such as drugs, indications, test abnormalities, and clinical adverse symptoms.

5. The intelligent collection method for adverse drug reactions based on clinical practice according to claim 1, characterized in that: In the above S3, the prediction model can predict missing entities and relationships by completing the relationships in the knowledge graph, enhancing the integrity of the knowledge graph.

6. The intelligent acquisition method for adverse drug reactions based on clinical practice according to claim 1, characterized in that: In the above S4, the dual-channel early warning is specifically as follows: Channel 1: Analyze the time-series change characteristics of the physiological indicators of the patient after individual medication through the LSTM-Cox model, and calculate the deviation degree of the current physiological indicators relative to the physiological indicators before the patient's medication in real time; Channel 2: Analyze the conflict paths between the current medication plan and the patient's underlying diseases and genotypes in the knowledge graph through a graph neural network. If a conflict is detected, a new adverse reaction case is captured.

7. An intelligent adverse drug reaction collection system based on clinical practice, based on the collection method described in any one of claims 1-6, includes a data processing module, a natural language processing module, a knowledge graph construction module, a prediction module, a dual-channel early warning mechanism module, and an update module, and is characterized in that: Data collection module: Obtain the medication records of the patient, including drug names, dosages, and frequencies, through the hospital information system interface; extract the medical history text and nursing records from the electronic medical records; Access the laboratory system to obtain test indicators in real time, including liver function, blood drug concentration, etc.; Natural language processing module: Used to identify entities and extract relationships from the collected data, and construct a time-series knowledge graph containing "drug - indication - test abnormality - clinical adverse symptom"; Knowledge graph construction module: Based on the output of the natural language processing module, construct a time-series knowledge graph. The nodes in the graph include entities such as drugs, indications, test abnormalities, and clinical adverse symptoms, and the edges represent the relationships between entities, and record the time-series changes of the relationships through timestamps; Prediction model module: Use the constructed time-series knowledge graph to train a prediction model, complete the relationships in the knowledge graph, and predict missing relationships and entities to make the knowledge graph more complete; Dual-channel early warning mechanism module: Channel 1: By analyzing the temporal variation characteristics of the physiological indicators of the patient after individual medication, the deviation degree of the current physiological indicators relative to the patient's own baseline value is calculated in real time; Channel 2: By analyzing the conflict paths of the current medication plan with the patient's underlying diseases and genotypes in the knowledge graph through a graph neural network, if a conflict is detected, a new adverse reaction case is captured; Knowledge graph update module: Add the new adverse reaction cases detected by the dual-channel early warning mechanism module to the knowledge graph, and use them to update and train the prediction model, further improve the knowledge graph, and enhance the ability to identify potential drug adverse reactions.

8. An intelligent collection system for adverse drug reactions based on clinical practice according to claim 7, characterized in that: The natural language processing module uses the BioBERT+BiLSTM-CRF model for entity recognition and relationship extraction, and can extract key information such as drugs, indications, test abnormalities, and clinical adverse symptoms from unstructured electronic medical record texts.

9. The intelligent adverse drug reaction collection system based on clinical practice according to claim 7, characterized in that: The nodes and edges in the temporal knowledge graph are both timestamped, which can reflect the temporal relationships between entities such as drugs, indications, test abnormalities, and clinical adverse symptoms.

10. The intelligent collection system for adverse drug reactions based on clinical practice according to claim 7, characterized in that: The dual-channel early warning mechanism module can analyze the conflict paths of the current medication plan with the patient's underlying diseases and genotypes in the knowledge graph. If a conflict is detected, a new adverse reaction case is generated and added to the knowledge graph.

Images