A method for reporting adverse drug reactions
By conducting text analysis and machine learning processing of drug adverse reaction data, a severity rating and risk assessment model was constructed, which solved the problem that the drug adverse reaction monitoring system could not be effectively monitored and evaluated in real time, and achieved rapid identification and severity of drug adverse reactions Dividing has improved drug management capabilities and drug safety.
Patent Information
- Application Number
- CN202410522364.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-04-28
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2044-04-28
AI Technical Summary
The existing drug adverse reaction monitoring system cannot effectively monitor and evaluate risks in real time, resulting in the failure to identify drug safety issues in a timely manner, and the lack of data and poor reporting quality, which brings complexity to global drug safety monitoring.
By obtaining patient case report data for preliminary text analysis, identifying key information and semantic annotation, building a severity rating system, generating a monitoring priority list, using machine learning algorithms to mine drug adverse reaction patterns and trends, building a risk assessment model, and dynamically adjusting risk scores , generate a drug use report carrying safety labels.
It realizes rapid identification and detailed classification of adverse drug reactions, improves the accuracy and timeliness of monitoring priority evaluation, ensures timely processing of adverse drug reaction information, significantly improves drug management capabilities, and provides long-term drug safety Data support.
Smart Images

Figure CN118315082B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of information technology, and in particular to a method for reporting adverse drug reactions. Background Art
[0002] Adverse drug reactions are an important issue in modern medical practice. They refer to any harmful and undesirable drug reactions that occur during the use of drugs at normal doses. These reactions are not the therapeutic effects of drugs, but phenomena that have a negative impact on the patient's health. These reactions can range from mild to severe, or even fatal. Due to individual differences in the human body's response to drugs, it is difficult to effectively monitor the body's adverse drug reactions. Moreover, due to time and resource constraints, as well as a lack of awareness of the importance of adverse drug reactions, many adverse drug reaction events are not reported in a timely manner, resulting in a large amount of data missing. Even when adverse drug reactions are reported, the quality and completeness of the reports are often questioned, and the reports may lack key clinical information, such as the patient's medical history, co-medications, and specific adverse reaction details. It is worth mentioning that traditional data analysis methods are often inefficient and cannot cope with the needs of massive data analysis. Existing adverse drug reaction monitoring systems usually cannot provide timely risk assessments or early warnings, which means that irreversible damage may have been caused before some major drug safety issues are identified and measures are taken. There are significant differences in the establishment and implementation of adverse drug reaction monitoring systems in different regions, and this inconsistency has brought additional complexity to global drug safety monitoring. How to monitor and risk assess adverse drug reactions in real time is an urgent problem that needs to be solved. Summary of the invention
[0003] The present invention provides a method for reporting adverse drug reactions, which mainly includes:
[0004] Obtain patient case report data and conduct preliminary text analysis to identify key information in case reports and annotate them semantically, and analyze the relationship between key information in case reports and symptoms;
[0005] Construct a regional severity rating system to determine the severity of adverse drug reaction symptoms based on the patient information obtained, and divide adverse drug reaction cases into three categories: Category I, Category II, and Category III, where Category I corresponds to mild, Category II corresponds to moderate, and Category III corresponds to severe;
[0006] According to key performance indicators and business rules, evaluate the monitoring priority of each region, classify and generate a monitoring priority list, and adaptively adjust the priority strategy; based on the monitoring priority list, use historical drug safety data to build a knowledge graph of drug safety, and use machine learning algorithms to mine and predict the patterns and trends of adverse drug reactions to identify potential safety risks;
[0007] When the integrity of the adverse drug reaction report is assessed to be complete, a database of adverse drug reaction symptoms is constructed based on the adverse drug reaction report, and the medical ontology of the adverse drug reaction symptoms is identified through the medical term ontology library, and the identified medical ontology is associated and mapped with the matching score;
[0008] Based on the established adverse drug reaction symptom database, determine the matching score of each drug and adverse drug reaction symptom combination, evaluate the correlation between different drugs and adverse drug reaction symptoms, and build a risk assessment model to evaluate the drug use risks in different regions and determine the drug use risks in preset regions;
[0009] According to the output of the risk assessment model, dynamic risk scoring is performed accordingly. The risk scoring model is continuously adjusted according to the changes in real-time data and regulatory information, providing real-time dynamic risk scoring for the priority of drug adverse reaction information reporting;
[0010] The automated report generation process is triggered based on the dynamic risk score. Combined with predefined report templates and automatic review mechanisms, a drug use report with a safety label is generated and entered into the adverse drug reaction management system.
[0011] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:
[0012] The present invention provides a method for reporting adverse drug reactions. By automatically extracting and analyzing key information in patient case reports, drug-related symptoms can be quickly identified to ensure the comprehensiveness and accuracy of data collection. At the same time, the rating is used to effectively classify adverse drug reaction cases, so as to achieve a detailed division and rapid response to the severity of adverse drug reactions. The present invention also intelligently evaluates the monitoring priorities of different regions, generates a monitoring priority list, and makes adaptive adjustments based on the latest data trends, thereby enhancing the accuracy and timeliness of the prediction. A real-time dynamic score is provided for the reporting priority of adverse drug reaction information to ensure that each case is handled in a timely and accurate manner. The present invention significantly improves the management capabilities of adverse drug reactions. Effective monitoring of the body's adverse reactions to drugs not only helps to timely identify and handle individual cases, but also provides long-term data support for drug safety, which is of immeasurable value to the research and development of new drugs and the safety assessment of existing drugs. BRIEF DESCRIPTION OF THE DRAWINGS
[0013] Figure 1 The present invention is a flowchart of a method for reporting adverse drug reactions.
[0014] Figure 2 A schematic diagram of a method for reporting adverse drug reactions of the present invention.
[0015] Figure 3This is another schematic diagram of a method for reporting adverse drug reactions of the present invention. DETAILED DESCRIPTION
[0016] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0017] like Figure 1-3 In this embodiment, a method for reporting adverse drug reactions may specifically include:
[0018] Step S101, obtaining patient case report data, and performing preliminary text analysis, identifying key information in the case report and semantically annotating it, and analyzing the relationship between the key information in the case report and the symptoms.
[0019] With the patient's authorization and consent, obtain patient case report data through database query; structure patient case report data according to medical data specifications. Extract key information from structured patient case reports; identify semantic relationships between key information and determine potential connections between drug ingredients and specified symptoms. Obtain medication duration and sort medication duration data to identify the time series relationship between medication duration and symptoms in patient case reports. Determine the timeliness of the report based on the submission time of the report.
[0020] For example, there is a patient case report. The patient information is a 35-year-old male with a history of hypertension. The drug name is drug A. The drug dosage is 500 mg per day, twice a day. The duration of use is 10 days. The symptom description is that the rash began to appear on the 5th day of medication, accompanied by mild itching. The report submission time is April 5, 2023. The patient case report is structured, and the diagnosis information is extracted as hypertension. The ICD-10 code extracted from the special diagnosis code is I10. The naive Bayes classifier is used to extract key information, and the symptom description is rash and mild itching. The drug name is drug A, and the dosage is 500 mg twice a day. Word2Vec is used to determine the potential connection between drug A and rash and itching. The symptoms occurred on the 5th day of medication, and the total duration of medication is 10 days. The report submission time is April 5, 2023. The timeliness of the report is evaluated according to the current date, and the latest patient case report is given priority.
[0021] Step S102, constructing a regional severity rating system, determining the severity of adverse drug reaction symptoms based on the acquired patient information, and dividing adverse drug reaction cases into three categories: the first category, the second category, and the third category, where the first category corresponds to mild, the second category corresponds to moderate, and the third category corresponds to severe.
[0022] Obtain the patient's age, symptom manifestation, and medical history information, filter incomplete or invalid records through data preprocessing, and clean and standardize the data; Based on the drug use records and symptom manifestations, use the Apriori algorithm to determine the relationship between drug use and symptom manifestations, and identify potential adverse drug reaction cases. Based on patient information, drug use, and adverse drug reaction symptoms, use the random forest algorithm for model training, build a drug adverse reaction case classification model, determine the severity of adverse drug reaction symptoms, and classify adverse drug reaction cases into three categories: mild, moderate, and severe; through vital signs and complication data, determine the critical cases in each category and optimize the classification boundaries; use the confusion matrix evaluation method to evaluate the accuracy of the adverse drug reaction case classification model, and according to the evaluation results, use the cross-validation method to optimize the parameters of the adverse drug reaction case classification model to reduce the proportion of false positives and false negatives; based on the patient's treatment response and health status information, re-verify the classification effect of the optimized adverse drug reaction case classification model, and statistically analyze the accuracy of the adverse drug reaction case classification model prediction.
[0023] For example, a cardiovascular drug B commonly used to treat hypertension is known to cause adverse reactions such as dizziness and nausea. Data of 500 patients, including age, drug use history, and adverse drug reaction symptom records, were collected from the patient information system. The data was cleaned and standardized to ensure that all records were complete and in a uniform format. The Apriori algorithm was used to analyze the association between the use of drug B and symptoms, and it was found that 70% of the patients experienced dizziness within 24 hours after taking drug B. The random forest algorithm was used to construct an adverse drug reaction classification model based on the patient's drug B use record and symptom manifestation. The model classified adverse drug reaction cases into three categories: mild, moderate, and severe. Dizziness was classified as mild and nausea as moderate. The confusion matrix was used to evaluate the accuracy of the classification. The preliminary evaluation showed that the accuracy of the model was 85%, and the false positive rate, that is, the misclassification of non-adverse drug reaction symptoms as adverse drug reactions, was 10%, and the false negative rate, that is, the failure to identify the true adverse drug reaction symptoms, was 5%. The model was fine-tuned according to the evaluation results, and the parameters were adjusted to reduce the false positive and false negative ratios. The model was optimized by the cross-validation method, and the accuracy was increased to 92%. The optimized model was deployed in a clinical setting for real-time adverse drug reaction monitoring. In the following month, the model successfully identified 95% of dizziness symptoms and 90% of nausea symptoms and correctly classified them.
[0024] Step S103, based on key performance indicators and business rules, evaluate the monitoring priority of each area, classify and generate a monitoring priority list, and adaptively adjust the priority strategy.
[0025] Extract the current monitoring data of each region from the database. Use the ARIMA algorithm to train the model based on the historical monitoring data of each region and predict the future trend of adverse drug reaction data in each region. According to the results of trend analysis, identify the key performance indicators that affect the priority. Through the Drools business rule management system, combined with key performance indicators and business rules, evaluate the monitoring priority of each region based on the set rules for evaluating the monitoring priority of each region. Based on the evaluation results of regional priorities, generate a monitoring priority list arranged by monitoring priority to guide the allocation of actual monitoring resources. Dynamically adjust the monitoring priority list based on feedback from real-time monitoring data.
[0026] For example, there is a batch of historical adverse drug reaction data about patients' reactions to drug C, which includes data from multiple geographic regions. The patient group is 40 to 50 years old and uses drug C to treat hypertension. The historical adverse drug reaction data includes reactions such as dizziness and nausea. The time range is the data of the past 5 years, and the regions are divided into regions A, B, and C. The adverse drug reaction data of regions A, B, and C in the past 5 years are extracted from the database, including event frequency and time series changes. The ARIMA model is used to predict the future change trend of the adverse drug reaction data of regions A, B, and C respectively. The prediction shows that the growth rate of adverse drug reaction events in region A is higher. Identify key performance indicators that affect monitoring priorities, such as region A has a high event growth rate, medium event severity, and wide coverage. Based on the business rule management system Drools, set regions with high event growth rate, medium to high severity, and wide coverage as high priority monitoring regions, such as region A is classified as a high priority monitoring region, while region C is set as a low priority due to its low event growth rate and mild severity. Generate a list arranged by monitoring priority, with region A at the top of the list and region C at the bottom. The monitoring priority list is adjusted dynamically based on real-time monitoring data. If there is a sudden increase in adverse drug reaction events in area C, the priority of area C will be automatically adjusted.
[0027] Step S104, based on the monitoring priority list, using historical drug safety data, constructing a knowledge graph of drug safety, mining and predicting patterns and trends of adverse drug reactions through machine learning algorithms, and determining potential safety risks.
[0028] According to the monitoring priority list, the key adverse drug reaction data are screened out and arranged in chronological order. According to the key adverse drug reaction data, the ARIMA algorithm is used to train the model to predict the change trend of the key adverse drug reaction data; if the predicted key adverse drug reaction data shows an increasing or decreasing trend, the random forest algorithm is used to train the model according to the statistical characteristics of the time series data and the key adverse drug reaction data to predict the area or time period where the frequency of adverse drug reaction events is greater than the preset frequency threshold; if the training efficiency of the random forest model is not good, the data dimension is screened to eliminate the data dimension whose correlation with the adverse drug reaction data is lower than the preset correlation threshold. The random forest model is retrained using the screened features and the parameters of the random forest model are adjusted. The optimized random forest model is verified to obtain the results of the random forest model verification and compare them with the preset trend mining accuracy threshold to determine whether the verification results meet the requirements of prediction accuracy. If the verification results do not meet the requirements, the random forest model parameters are adjusted and retrained until the prediction accuracy requirements are met. The newly acquired adverse drug reaction data are predicted using the verified random forest model to identify the key monitoring points and potential risk factors in the report. According to the prediction report results, the original monitoring point layout is adjusted to optimize the monitoring strategy.
[0029] For example, there is a set of adverse drug reaction data about the use of drug D. The patient population is 45 to 55 years old and hypertensive patients. The historical adverse drug reaction data includes dizziness, nausea and other reactions. The time range is the data of the past 3 years. The key adverse drug reaction data, including the severity, frequency and most recent time of dizziness events, are screened from the monitoring priority list. It is found that in the past 3 years, dizziness events have occurred 150 times in patients aged 45 to 55, and the most recent one occurred in March 2023. The ARIMA model is used to predict the long-term trend of dizziness events. The prediction shows that dizziness events will gradually increase in the next year. According to the increasing trend of dizziness events, the random forest model is used to predict the area or time period where the frequency of adverse drug reaction events is greater than the preset frequency threshold. Remove data with low correlation with dizziness events, such as data of non-specific age groups. Adjust the parameters to improve the prediction accuracy and verify the optimized random forest model. The model prediction accuracy reaches 85%, which meets the preset trend mining accuracy threshold. Use the validated model to predict newly acquired adverse drug reaction data and compile a prediction report to identify, for example, the risk of dizziness in hypertensive patients aged 45 to 55 when using drug D in the summer. Adjust the layout of monitoring points based on the report results, such as increasing the intensity of monitoring for this age group in the summer.
[0030] Based on the knowledge graph of drug safety, the correlation between adverse drug reactions is analyzed, potential safety risks are predicted, and the drug instructions and dosage recommendations are adjusted according to the prediction results before the drug is approved for marketing.
[0031] Apriori algorithm is used to mine the association rules between drugs and adverse reactions. Fuzzy analytic hierarchy process is used for multi-criteria decision-making, and the comprehensive safety risk value of drugs is calculated by combining the severity and probability of adverse reactions. Before the drug is approved for marketing, the optimal dosage and administration scheme of the drug are determined; according to the comprehensive safety risk value of the drug, the corresponding medication instructions are generated.
[0032] For example, the Apriori algorithm is used, the support threshold is set to 05, and the confidence threshold is set to 8, to mine the association rules between drugs and adverse reactions. The rule is found: aspirin → gastric bleeding [support = 07, confidence = 85]. A path-based reasoning method, such as the Path Ranking Algorithm (PRA), is used. According to the semantic association path in the knowledge graph (such as aspirin-causeAdverseReaction-gastric ulcer-causeAdverseReaction-gastric bleeding), the probability that aspirin may cause gastric bleeding is predicted to be 7. The fuzzy analytic hierarchy process (FuzzyAHP) is used for multi-criteria decision-making, considering the severity of adverse reactions, the probability of occurrence and other factors, and the comprehensive safety risk value of aspirin is calculated to be 6. The safety risk values of aspirin at doses of 50mg, 100mg, and 150mg are compared, and the optimal daily dose is determined to be 100mg by minimizing the risk principle. Based on the comprehensive safety risk value of the drug, the instructions for use of aspirin are generated: "Aspirin tablets, 100 mg, oral, once a day. Adverse reactions include gastric bleeding (incidence 1%). If there is obvious gastrointestinal discomfort, please seek medical attention immediately." It is submitted to the Food and Drug Administration for review, revised and improved based on feedback, and finally issued as an approved drug instruction sheet.
[0033] Step S105, when the integrity of the adverse drug reaction report is assessed to be complete, a database of adverse drug reaction symptoms is constructed based on the adverse drug reaction report, medical ontology recognition of adverse drug reaction symptoms is performed through a medical term ontology library, and at the same time, the recognized medical ontology is associated and mapped with a matching score.
[0034] The naive Bayes classifier was used to establish the initial adverse drug reaction symptom vocabulary recognition module to identify adverse drug reaction symptom descriptions in medical reports and medical records. The extracted adverse drug reaction symptom text was converted into structured data using the TF-IDF algorithm. The medical term ontology library was used to perform association mapping with the structured adverse drug reaction symptom data. According to the matching score between nodes, the matching score between adverse drug reaction symptoms and existing cases was determined, and a correlation scoring system between diseases and adverse drug reaction symptoms was established to identify the correlation between specified adverse drug reaction symptoms and diseases. Combined with the matching score, medical literature and case reports with a correlation greater than the preset correlation threshold with the specified adverse drug reaction symptoms were screened out. The parameters of the naive Bayes classifier were adjusted through periodic feedback. The real-time update mechanism of the database was maintained, and the timeliness and dynamics of the database information were maintained through continuous data collection and analysis processing. The reliability of the adverse drug reaction symptom database was verified using feedback from medical experts and actual patient feedback.
[0035] For example, the hospital has multiple patient medical records about the use of drug H. The patient group is 45 to 55 years old hypertensive patients. Common adverse drug reaction symptoms are dizziness, nausea, and slow heart rate. The number of reports is 100 medical records. Use the naive Bayes classifier to process 100 medical record reports to identify adverse drug reaction symptom descriptions, such as dizziness, nausea, and slow heart rate. The extracted adverse drug reaction symptom text dizziness, nausea, and slow heart rate are converted into structured data through the TF-IDF algorithm. Use the medical term ontology library to map the structured adverse drug reaction symptom data to specific medical terms, such as ICD-10 codes. Calculate the matching score between the adverse drug reaction symptom dizziness and the known case drug H users. The matching score between the dizziness symptom and the drug H users is 0.8. In the range of 0 to 1, the closer to 1, the higher the correlation. According to the matching score, screen out medical literature and case reports with a correlation greater than 0.7. Regularly adjust the parameters of the naive Bayes classifier through periodic feedback to improve the accuracy of adverse drug reaction symptom recognition. Maintain a real-time update mechanism for the database, regularly add new medical records and adverse drug reaction information, and update the database monthly to reflect the latest adverse drug reaction trends. Use feedback from medical experts and actual patient feedback to verify the reliability of the adverse drug reaction symptom database.
[0036] Dependency syntactic analysis methods are used to understand the associations between medical entities, analyze the dependencies between medical entities, and determine the temporal and causal relationships between drugs and adverse reactions.
[0037] Obtain the dependency relationships between medical entities and construct a medical entity dependency graph. In the medical entity dependency graph, use the shortest path algorithm to calculate the shortest dependency path of the entity. Use regular expressions to match time phrases to extract the time information of drug use and adverse reactions. If the drug use time is earlier than the adverse reaction time, it is considered that there is a temporal relationship between the two. Extract the causal knowledge of drug-induced adverse reactions in the dependency path of drug and adverse reaction entity pairs. Use the Neo4j graph database to integrate the semantic connection, temporal relationship, and causal relationship between drugs and adverse reactions.
[0038] For example, the Stanford Core NLP tool was used to perform dependency syntactic analysis on 5,000 medical texts, obtain the dependency relationships between medical entities, and construct a medical entity dependency graph. In the medical entity dependency graph, the shortest path algorithm was used to calculate the shortest dependency path between 1,000 commonly used drug entities such as aspirin and 500 adverse reaction entities such as headache and nausea. When the shortest path length is less than or equal to 3, it is determined that there is a direct semantic connection between the drug and the adverse reaction. Regular expressions are used to match time phrases to extract the time information of drug use and adverse reaction occurrence. If the drug use time is earlier than the adverse reaction occurrence time, it is considered that there is a temporal relationship between the two. The causal relationship extraction algorithm based on Bootstrapping is used to identify causal trigger words such as "cause" and "cause" in the dependency path of drug and adverse reaction entity pairs, as well as negative words such as "not" and "non", to extract the causal knowledge of drug-induced adverse reactions, and the F1 value on the labeled data set reaches 85%. Neo4j graph database is used to integrate the semantic connection, temporal relationship and causal relationship between drugs and adverse reactions, and a knowledge graph of adverse drug reactions containing 50,000 entities and 200,000 relationships is constructed to achieve efficient management and query of drug safety information.
[0039] Combining the drug's mechanism of action and clinical knowledge, a preliminary assessment of the association between the drug and adverse reactions is conducted, and the assessment results are output for use by the matching score mapping model.
[0040] By analyzing the mechanism of action of the drug, it is determined whether the drug will cause suspected adverse reactions. Using clinical knowledge and literature reports, it is determined whether the adverse reaction has been recorded as a known side effect of the drug. After obtaining the relevance rating of the drug mechanism of action and the support rating of clinical knowledge, the comprehensive relevance score is obtained.
[0041] For example, taking the drug "aspirin" and the adverse reaction "gastric ulcer" as an example, the report data related to aspirin and gastric ulcer is obtained from the FDA's "Adverse Event Reporting System (FAERS)" database. Retrieve relevant clinical trials and case reports, such as literature on PubMed
[0042] "Effects of Aspirin on Gastric Mucosal Integrity: Systematic Review and Meta-analys is" (2019). Analyzing the mechanism of action of the drug, aspirin inhibits the cyclooxygenase (COX) enzyme and reduces the production of prostaglandins, which play an important role in protecting the gastric mucosa. Therefore, the inhibitory effect of aspirin may cause damage to the gastric mucosa and increase the risk of gastric ulcers. According to the information from the "DrugBank" database, gastric ulcer is one of the known side effects of aspirin. Literature reports show that the incidence of gastric ulcers in patients who use aspirin for a long time is significantly higher than that of non-users. The "Naranjo Algorithm" is used to assess the possibility of adverse drug reactions. A series of questions, such as whether there are other possible reasons that can explain the adverse reaction and whether the reaction disappears after the drug is discontinued, are used to score and determine the correlation between the adverse reaction and the drug. The weighted average method is used to calculate the comprehensive score of the correlation by combining the rating of the drug's mechanism of action with the support rating of clinical knowledge. For example, the mechanism rating accounts for 40% and the clinical support rating accounts for 60%. A detailed evaluation report was prepared to describe the possible association between aspirin and gastric ulcer, including the drug's mechanism of action, known side effects, literature review, NaranjoAlgorithm score, and the final comprehensive score of the association.
[0043] Step S106, based on the constructed adverse drug reaction symptom database, determine the matching score of each drug and adverse drug reaction symptom combination, evaluate the correlation between different drugs and adverse drug reaction symptoms, and build a risk assessment model to evaluate the drug use risks in different areas and determine the drug use risks in preset areas.
[0044] By integrating regional data resources, the regional characteristics of the patient's living area are obtained. According to the matching scores between nodes, the matching scores of each drug and adverse drug reaction symptom combination are determined, and the correlation between different drugs and adverse drug reaction symptoms is evaluated. According to patient information and treatment effects, the K-means clustering algorithm is used to train the model to identify the differences in drug response among different patient groups. According to the matching scores of the drug and adverse drug reaction symptom combination, patient information, drug use data, and regional characteristics of the patient's living area, the decision tree algorithm is used to train the model, build a risk assessment model, evaluate the drug use risks in different regions, and identify drugs in the specified area whose use risks are higher than the preset risk threshold. Obtain regulatory information and compare it with the drug use risk assessment results through the rule engine to identify drug use events that do not meet regulatory requirements. Convert the analysis results into charts or reports.
[0045] For example, through the electronic health record system, adverse reaction data of a drug N was collected, and 5% of patients reported stomach pain. The drug data recorded information such as the dosage and frequency of use of drug N. Patient information of 100 patients, including age, gender, weight, etc., was collected. The treatment effect of these patients after using drug N was tracked, and the treatment effect was pain relief. It was found that 80% of the patients reported pain relief. Regional data was integrated to obtain regional characteristics. The patients lived in a temperate climate area with good sanitary conditions and an average distribution of medical resources. According to the data, it was found that the matching score of drug N and stomach pain was 0.05, and 5% of the patients reported this symptom. Using the clustering algorithm, the patients were divided into two groups, one group responded well to drug N, and the other group had adverse reactions such as stomach pain, which helped to identify the patient group that was sensitive to drug N. Using the decision tree algorithm, a model was constructed to evaluate the risk of drug use in different regions by combining the matching score of drug N, patient information, drug use data, and regional characteristics. The results showed that in this temperate climate area, the risk of using drug N was lower than the preset risk threshold. We obtained the drug use guidance from the regulatory agency and compared it with our risk assessment results, and found that the use of drug N met regulatory requirements in this region. We used Tableau to convert these analysis results into charts and reports, which clearly demonstrated the risk assessment and patient group differences of drug N use.
[0046] Step S107, according to the output of the risk assessment model, a dynamic risk score is correspondingly performed, and the risk score model is continuously adjusted according to the changes in real-time data and regulatory information, so as to provide a real-time dynamic risk score for the priority of drug adverse reaction information reporting.
[0047] Obtain the latest drug use data and adverse drug reaction information, develop a dynamic risk scoring system based on the output of the risk assessment model, and perform real-time dynamic drug use risk scoring; apply the gradient descent algorithm to adjust the parameters of the risk scoring model to obtain an optimized risk scoring model. Use historical adverse drug reaction data and drug use to backtest the dynamic risk scoring system to determine the accuracy of the dynamic risk scoring system. Based on the backtest results, refine the dynamic risk scoring system and fine-tune the parameters of the risk assessment model. Obtain the latest drug use data and adverse drug reaction information, and dynamically update the risk assessment model. Adjust the risk assessment model according to changes in real-time data and regulatory information. Regularly monitor the performance of the system, evaluate the accuracy of the risk assessment model, and perform adaptive adjustments to the risk assessment model based on the new adverse drug reaction risk assessment results. Obtain the performance report of the optimized feedback mechanism, evaluate the performance of the risk assessment model in actual applications, and adjust the dynamic risk scoring system settings based on the report results.
[0048] For example, the hospital collected the usage data of drug I and related adverse drug reaction event records. The patient group was hypertensive patients aged 45 to 55 years old. The adverse drug reaction event records were dizziness, nausea, slow heart rate, etc. The data range was records from the past two years. Based on the output of the risk assessment model, such as the correlation score between drug I and dizziness and nausea, a real-time dynamic drug use risk scoring system was developed. The gradient descent algorithm was applied to adjust the parameters of the risk scoring model to optimize the model performance, such as adjusting the model to more accurately reflect the response of hypertensive patients to drug I. The dynamic risk scoring system was backtested using historical adverse drug reaction data and drug usage from the past two years. The backtest results showed that the model accurately predicted 80% of adverse drug reaction events. The dynamic risk scoring system was refined and adjusted according to the backtest results, and the parameters of the risk assessment model were fine-tuned, such as adjusting the data weights used to identify dizziness and nausea. A real-time data collection system was integrated to continuously obtain the latest drug I usage data and adverse drug reaction information. The risk assessment model was dynamically updated to reflect new data and trends. Design an automatic optimization feedback mechanism to continuously adjust the risk assessment model based on changes in real-time data and regulatory information. If new adverse drug reaction assessment results are found, automatically adjust the model parameters. Regularly monitor system performance, evaluate the accuracy of the risk assessment model, and make adaptive adjustments based on new adverse drug reaction risk patterns. Obtain performance reports on the optimization feedback mechanism to evaluate the performance of the model in actual applications. Adjust system settings based on the report results, such as enhancing the risk prediction ability for patients in specific age groups.
[0049] Based on the relationship between drug dosage and adverse drug reactions in patients' medical records, the temporal relationship between the development of adverse drug reaction symptoms and dosage changes is determined, and an adverse drug reaction symptom severity response assessment system is established, which is then applied and optimized in real time in the clinical environment.
[0050] Through the patient's medical records, we obtain data on drug dosage and adverse drug reaction symptom development, use the ARIMA algorithm to train the model, and identify the time relationship between drug dosage changes and the development of adverse drug reaction symptoms; record the delay characteristics between the development of adverse drug reaction symptoms and drug dosage changes, and determine whether the adverse drug reaction symptoms occur within a specified time after the drug dosage change. According to the data of drug dosage and adverse drug reaction symptom development, a linear regression model was constructed to determine whether the development of adverse drug reaction symptoms was linearly correlated with the change of drug dosage; if the development of adverse drug reaction symptoms was not linearly correlated with the change of drug dosage, a formula for the severity of adverse drug reaction symptoms was constructed, S(t) = S0·er(t)·t, the severity of adverse drug reaction symptoms was used to describe the nonlinear relationship between drug dosage and the development of adverse drug reaction symptoms, where S(t) represents the severity of adverse drug reaction symptoms at time t, D(t) is the drug dosage at time t, S0 is the initial severity of adverse drug reaction symptoms, and r(t) is the nonlinear response function related to the drug dosage D(t); the nonlinear response function r(t) = k·log(D(t)+1), where k is a regulating factor used to control the sensitivity of dosage changes to changes in heart rate, and the initial k value is based on the drug effect strength determined by clinical experience, and the larger the k value, the more sensitive the heart rate is to dosage changes. The existing adverse drug reaction data were used to verify and optimize the formula for the severity of adverse drug reaction symptoms, and the accuracy and reliability of the formula for the severity of adverse drug reaction symptoms were evaluated. Deploy the ADR symptom severity formula in an actual clinical setting to analyze new ADR cases in real time. Adjust treatment plans based on the output of the ADR symptom severity formula to optimize patient outcomes. Present the ADR symptom severity formula results to clinicians via charts or reports.
[0051] For example, consider a group of patients who were prescribed drug J for hypertension. The adverse drug reaction of interest is bradycardia, a known adverse drug reaction of drug J. Data from 100 patients were collected, including daily doses of drug J and heart rate, the number of beats per minute. The dose range is 25 mg to 100 mg, the normal heart rate is 60 to 100 beats per minute, and a heart rate less than 60 beats per minute is considered bradycardia. An ARIMA model was used to analyze the change in heart rate over time and identify the time delay between dose change and heart rate. It was found that the adverse drug reaction symptoms of bradycardia usually appeared within 2 days after the dose increase. A linear regression model was constructed to assess the relationship between the risk of bradycardia and dose, but the linear model could not accurately predict the risk of bradycardia. Develop a formula for the severity of adverse drug reaction symptoms, S(t) = S0·er(t)·t. Where S(t) is the heart rate, S0 is the baseline heart rate, which is preset to 70 beats per minute, and r(t) is the response function. Design r(t) = k·log(D(t)+1), where k is a regulating factor used to control the sensitivity of dose changes to heart rate changes. The initial k value is based on the drug effect strength determined by clinical experience. A larger k value means that the heart rate is more sensitive to dose changes. The initial k value = 0.05, and D(t) is the dose. A patient's original daily dose was 50 mg, which was later increased to 75 mg. The heart rate was predicted using the adverse drug reaction symptom severity formula. It was found that on the second day after the dose increase, the heart rate was expected to drop to 55 beats / minute, indicating an increased risk of bradycardia. Use cross-validation to evaluate the accuracy of the adverse drug reaction symptom severity formula in different patients and doses, adjust the k value based on actual data, and optimize the formula. Deploy the adverse drug reaction symptom severity formula in a clinical setting for real-time monitoring of bradycardia risk. Adjust the treatment plan based on the output of the adverse drug reaction symptom severity formula, such as adjusting the dose or changing the drug. Regularly collect new adverse drug reaction case data and continue to optimize the adverse drug reaction symptom severity formula. Provides clinicians with graphs and reports of adverse drug reaction symptom severity formula results to support better patient management.
[0052] Step S108, triggering the automated report generation process based on the dynamic risk score, combining the predefined report template and the automatic review mechanism, generating a drug use report with a safety label and entering it into the adverse drug reaction management system.
[0053] Generate relevant security reports based on the latest risk score information according to the results of dynamic risk scoring. Automatically obtain key information from the dynamic risk scoring system and fill in the corresponding fields of the pre-defined report template. Check the consistency and accuracy of the filled information; if the automatic audit finds problems, the exception handling process will be triggered, the exception report will be marked, and a manual review will be requested. If the audit does not find any problems, the accuracy of the report content will be evaluated and a security label will be automatically added to the report. Encode the audited report, use the code as a unique identifier, and enter it into the management system together with the report content.
[0054] For example, the hospital's dynamic risk scoring system monitors the use of drug K in real time. The patient group is 45 to 55 years old hypertensive patients, and the adverse drug reactions are dizziness, nausea, and slow heart rate. The risk score of drug K use in area A rises to 0.75, and 0.75 indicates a higher risk. The dynamic risk scoring system detects that the risk score of drug K use in area A has increased, and automatically triggers the safety report generation process. Using the predefined report template, key information is automatically filled in, the drug name is drug K, the risk score is 0.75, the affected area is area A, and the recommended measures are to strengthen monitoring and reduce the dose. The system verifies the consistency and accuracy of the filled information, including format checking and data range verification, to determine whether the risk score in the inspection report is consistent with the system record. If the system finds inconsistent or missing information, such as an incorrect entry of the drug name, it automatically marks the report as abnormal and submits it to professionals for manual review. For reports that pass the automatic review, the system automatically adds a safety label to confirm the accuracy of the report content. Generate a unique code for the report that passes the review, such as report number 2023-001A, and enter the report content together with the code into the management system. The data used for risk scoring is regularly refreshed through the database update mechanism to ensure the accuracy and timeliness of the scoring.
[0055] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A method for reporting adverse drug reactions, characterized in that: The method comprises: Obtain patient case report data and conduct preliminary text analysis to identify key information in case reports and annotate them semantically, and analyze the relationship between key information in case reports and symptoms; Construct a regional severity rating system to determine the severity of adverse drug reaction symptoms based on the patient information obtained, and divide adverse drug reaction cases into three categories: Category I, Category II, and Category III, where Category I corresponds to mild, Category II corresponds to moderate, and Category III corresponds to severe; Based on key performance indicators and business rules, evaluate the monitoring priority of each area, classify and generate a monitoring priority list, and adaptively adjust the priority strategy; Based on the monitoring priority list, we use historical drug safety data to build a knowledge graph of drug safety, and use machine learning algorithms to mine and predict patterns and trends of adverse drug reactions to identify potential safety risks. When the integrity of the adverse drug reaction report is assessed to be complete, a database of adverse drug reaction symptoms is constructed based on the adverse drug reaction report, and the medical ontology of the adverse drug reaction symptoms is identified through the medical term ontology library, and the identified medical ontology is associated and mapped with the matching score; Based on the established adverse drug reaction symptom database, determine the matching score of each drug and adverse drug reaction symptom combination, evaluate the correlation between different drugs and adverse drug reaction symptoms, and build a risk assessment model to evaluate the drug use risks in different regions and determine the drug use risks in preset regions; According to the output of the risk assessment model, dynamic risk scoring is performed accordingly. The risk scoring model is continuously adjusted according to the changes in real-time data and regulatory information, providing real-time dynamic risk scoring for the priority of drug adverse reaction information reporting; Trigger the automated report generation process based on dynamic risk scores, combine predefined report templates and automatic review mechanisms to generate drug use reports with safety labels and enter them into the adverse drug reaction management system; According to the key performance indicators and business rules, the monitoring priority of each area is evaluated, the monitoring priority list is classified and generated, and the priority strategy is adaptively adjusted, including: Extract the current monitoring data of each area from the database; Based on the historical monitoring data of each region, the ARIMA algorithm is used to train the model and predict the future trend of adverse drug reaction data in each region; Based on the results of trend analysis, identify key performance indicators that influence priorities; Through the Drools business rule management system, combined with key performance indicators and business rules, the monitoring priority of each area is evaluated based on the set rules for evaluating the monitoring priority of each area; Based on the evaluation results of regional priorities, a monitoring priority list arranged by monitoring priority is generated to guide the allocation of actual monitoring resources; Dynamically adjust the monitoring priority list based on feedback from real-time monitoring data.
2. The method according to claim 1, wherein: The patient case report data is obtained and a preliminary text analysis is performed to identify key information in the case report and annotate it semantically, and analyze the relationship between the key information in the case report and the symptoms, including: With the patient's authorization and consent, patient case report data were obtained through database query; Structural processing of patient case report data according to medical data specifications; Extract key information from structured patient case reports; Identify semantic relationships between key information and determine potential links between drug ingredients and specified symptoms; Obtain medication duration and sort medication duration data to identify the time series relationship between medication duration and symptoms in patient case reports; Determine the timeliness of the report based on the time it is submitted.
3. The method according to claim 1, wherein: The regional severity rating system is constructed to determine the severity of adverse drug reaction symptoms based on the acquired patient information, and to divide adverse drug reaction cases into three categories: the first category, the second category, and the third category, wherein the first category corresponds to mild, the second category corresponds to moderate, and the third category corresponds to severe, including: Obtain patient age, symptom manifestations, and medical history information, filter incomplete or invalid records through data preprocessing, and clean and standardize the data; Based on drug use records and symptom manifestations, the Apriori algorithm is used to determine the relationship between drug use and symptom manifestations, and identify potential adverse drug reaction cases; Based on patient information, drug use, and symptoms of adverse drug reactions, the random forest algorithm was used for model training to build a classification model for adverse drug reaction cases, determine the severity of adverse drug reaction symptoms, and classify adverse drug reaction cases into three categories: mild, moderate, and severe. Identify borderline cases in each category and optimize classification boundaries using vital signs and comorbidity data; The confusion matrix evaluation method was used to evaluate the accuracy of the adverse drug reaction case classification model. Based on the evaluation results, the parameters of the adverse drug reaction case classification model were optimized through cross-validation to reduce the proportion of false positives and false negatives. According to the patients' treatment response and health status information, the classification effect of the optimized adverse drug reaction case classification model was verified again, and the accuracy of the adverse drug reaction case classification model prediction was statistically analyzed.
4. The method according to claim 1, wherein: Based on the monitoring priority list, the historical drug safety data is used to build a knowledge graph of drug safety, and the patterns and trends of adverse drug reactions are mined and predicted through machine learning algorithms to identify potential safety risks, including: According to the monitoring priority list, key adverse drug reaction data are screened and arranged in chronological order; Based on key adverse drug reaction data, the ARIMA algorithm is used to train the model and predict the changing trend of key adverse drug reaction data; If the predicted key adverse drug reaction data show an increasing or decreasing trend, the random forest algorithm is used to train the model based on the statistical characteristics of the time series data and the key adverse drug reaction data to predict the area or time period where the frequency of adverse drug reaction events is greater than the preset frequency threshold; If the training efficiency of the random forest model is not good, data dimension screening is performed to eliminate data dimensions whose correlation with adverse drug reaction data is lower than the preset correlation threshold; Retrain the random forest model using the filtered features and adjust the random forest model parameters; Verify the optimized random forest model, obtain the verification results of the random forest model, and compare them with the preset trend mining accuracy threshold to determine whether the verification results meet the prediction accuracy requirements; If the verification result does not meet the requirements, adjust the random forest model parameters and retrain until the prediction accuracy requirements are met; Use the validated random forest model to predict newly acquired adverse drug reaction data and identify key monitoring points and potential risk factors in the reports; According to the forecast report results, adjust the original monitoring point layout and optimize the monitoring strategy; It also includes: analyzing the correlation between adverse drug reactions based on the knowledge graph of drug safety, predicting potential safety risks, and adjusting the drug's instructions for use and dosage recommendations based on the prediction results before the drug is approved for marketing. Specifically, it includes: using the Apriori algorithm to mine the association rules between drugs and adverse reactions; using the fuzzy hierarchical analysis method for multi-criteria decision-making, combining the severity of adverse reactions and the probability of occurrence factors to calculate the comprehensive safety risk value of the drug; determining the optimal dosage and administration plan of the drug before the drug is approved for marketing; and generating corresponding medication instructions based on the comprehensive safety risk value of the drug.
5. The method according to claim 1, wherein: When the integrity of the adverse drug reaction report is evaluated to be complete, a database of adverse drug reaction symptoms is constructed based on the adverse drug reaction report, a medical ontology is identified for the adverse drug reaction symptoms through a medical term ontology library, and an association mapping is performed between the identified medical ontology and the matching score, including: Using naive Bayes classifier, an initial adverse drug reaction symptom vocabulary recognition module was established to identify adverse drug reaction symptom descriptions in medical reports and records; The extracted adverse drug reaction symptom text is converted into structured data through the TF-IDF algorithm; Use the medical term ontology library to associate and map with structured adverse drug reaction symptom data; According to the matching scores between nodes, the matching scores between adverse drug reaction symptoms and existing cases are determined, a correlation scoring system between diseases and adverse drug reaction symptoms is established, and the correlation between designated adverse drug reaction symptoms and diseases is identified; Combined with the matching score, medical literature and case reports with correlation with the specified adverse drug reaction symptoms greater than the preset correlation threshold are screened out; Adjust the parameters of the Naive Bayes classifier through periodic feedback; Maintain the real-time update mechanism of the database, and keep the timeliness and dynamics of database information through continuous data collection and analysis and processing; The reliability of the adverse drug reaction symptom database was verified using feedback from medical experts and actual patient responses; It also includes: using dependency syntax analysis methods to understand the association between medical entities, analyzing the dependency relationship between medical entities, and determining the temporal and causal relationship between drugs and adverse reactions; combining the drug mechanism of action and clinical knowledge to conduct a preliminary assessment of the association between drugs and adverse reactions, and outputting the assessment results for use by the matching score mapping model; The dependency syntactic analysis method is used to understand the correlation between medical entities, analyze the dependency relationship between medical entities, and determine the temporal and causal relationship between drugs and adverse reactions, specifically including: obtaining the dependency relationship between medical entities and constructing a medical entity dependency relationship graph; in the medical entity dependency relationship graph, using the shortest path algorithm to calculate the shortest dependency path of the entity; using regular expressions to match time phrases to extract the time information of drug use and adverse reaction occurrence; if the drug use time is earlier than the adverse reaction occurrence time, it is considered that there is a temporal relationship between the two; extracting the causal knowledge of drug-induced adverse reactions in the dependency path of drug and adverse reaction entity pairs; using the Neo4j graph database to integrate the semantic connection, temporal relationship and causal relationship between drugs and adverse reactions; The combination of drug action mechanism and clinical knowledge is used to conduct a preliminary assessment of the association between the drug and adverse reactions, and the assessment results are output for use by the matching score mapping model, which specifically includes: analyzing the drug's mechanism of action to determine whether the drug will cause suspected adverse reactions; using clinical knowledge and literature reports to determine whether the adverse reactions have been recorded as known side effects of the drug; after obtaining the association rating of the drug's mechanism of action and the support rating of clinical knowledge, a comprehensive association score is obtained.
6. The method according to claim 1, wherein: Based on the established adverse drug reaction symptom database, the matching score of each drug and adverse drug reaction symptom combination is determined, the correlation between different drugs and adverse drug reaction symptoms is evaluated, and a risk assessment model is constructed to evaluate the drug use risks in different regions and determine the drug use risks in preset regions, including: By integrating regional data resources, we can obtain regional characteristics of the patient’s living area; According to the matching scores between nodes, the matching scores of each drug and adverse drug reaction symptom combination are determined to evaluate the correlation between different drugs and adverse drug reaction symptoms; Based on patient information and treatment effects, the K-means clustering algorithm is used to train the model and identify differences in drug response among different patient groups; Based on the matching scores of the drug and adverse drug reaction symptom combination, patient information and drug use data, and regional characteristics of the patient's living area, a decision tree algorithm is used to train the model and build a risk assessment model to assess the drug use risks in different regions and identify drugs with a use risk higher than the preset risk threshold in a specified area; Obtain regulatory information and compare it with drug use risk assessment results through the rule engine to identify drug use events that do not meet regulatory requirements; Convert analysis results into charts or reports.
7. The method according to claim 1, wherein: According to the output of the risk assessment model, a dynamic risk score is correspondingly performed, and the risk score model is continuously adjusted according to the changes in real-time data and regulatory information to provide a real-time dynamic risk score for the priority of drug adverse reaction information reporting, including: Obtain the latest drug use data and adverse drug reaction information, develop a dynamic risk scoring system based on the output of the risk assessment model, and perform real-time dynamic drug use risk scoring; Apply the gradient descent algorithm to adjust the parameters of the risk scoring model to obtain the optimized risk scoring model; Backtest the dynamic risk scoring system using historical adverse drug reaction data and drug usage to determine the accuracy of the dynamic risk scoring system; Based on the backtesting results, the dynamic risk scoring system is refined and the parameters of the risk assessment model are fine-tuned; Obtain the latest drug use data and adverse drug reaction information, and dynamically update the risk assessment model; Adjust risk assessment models based on real-time data and changes in regulatory information; Regularly monitor the performance of the system, evaluate the accuracy of the risk assessment model, and make adaptive adjustments to the risk assessment model based on new adverse drug reaction risk assessment results; Obtain performance reports on the optimization feedback mechanism, evaluate the performance of the risk assessment model in actual applications, and adjust the dynamic risk scoring system settings based on the report results; It also includes: determining the time relationship between the development of adverse drug reaction symptoms and dosage changes based on the relationship between drug dosage and adverse drug reactions in the patient's medical records, establishing an adverse drug reaction symptom severity response evaluation system, and applying and optimizing it in real time in a clinical environment. Specifically, it includes: obtaining drug dosage and adverse drug reaction symptom development data through the patient's medical records, using the ARIMA algorithm to train the model, and identifying the time relationship between drug dosage changes and adverse drug reaction symptom development; recording the delay characteristics between the development of adverse drug reaction symptoms and drug dosage changes, and determining whether adverse drug reaction symptoms occur within a specified time after the drug dosage change; constructing a linear regression model based on drug dosage and adverse drug reaction symptom development data to determine whether the development of adverse drug reaction symptoms is linearly related to changes in drug dosage; if the development of adverse drug reaction symptoms is not linearly related to changes in drug dosage, then constructing an adverse drug reaction symptom severity formula, S(t)=S0·er(t)·t, and the adverse drug reaction symptom severity is used to describe the relationship between drug dosage and adverse drug reaction symptoms. A nonlinear relationship between the development of adverse drug reaction symptoms, where S(t) represents the severity of adverse drug reaction symptoms at time t, D(t) is the drug dose at time t, S0 is the initial adverse drug reaction symptom severity, and r(t) is a nonlinear response function related to the drug dose D(t); the nonlinear response function r(t)=k·log(D(t)+1), where k is a regulating factor used to control the sensitivity of dose changes to heart rate changes, and the initial k value is based on the drug effect strength determined by clinical experience. The larger the k value, the more sensitive the heart rate is to dose changes; using existing adverse drug reaction data to verify and optimize the adverse drug reaction symptom severity formula, and evaluate the accuracy and reliability of the adverse drug reaction symptom severity formula; deploying the adverse drug reaction symptom severity formula in an actual clinical environment, and performing real-time analysis on new adverse drug reaction cases; adjusting the treatment plan according to the output of the adverse drug reaction symptom severity formula to optimize the patient's treatment effect; presenting the results of the adverse drug reaction symptom severity formula to clinicians through charts or reports.
8. The method according to claim 1, wherein: The automatic report generation process is triggered according to the dynamic risk score, combined with the predefined report template and the automatic review mechanism, to generate a drug use report with a safety label and enter it into the adverse drug reaction management system, including: According to the results of dynamic risk scoring, generate relevant safety reports based on the latest risk scoring information; Automatically obtain key information from the dynamic risk scoring system and fill in the corresponding fields of the pre-defined report template; Check the consistency and accuracy of the information entered; If the automatic review finds any problems, the exception handling process will be triggered, the exception report will be marked, and manual review will be requested; If the audit finds no issues, the accuracy of the report content is assessed and a safety label is automatically added to the report; Code the audited reports, use the code as a unique identifier, and enter it into the management system together with the report content.
