Intelligent risk detection method and system after drug marketing

Abstract

The invention provides an intelligent risk detection method and system after drug marketing, and relates to the technical field of intelligent detection of risks after drug marketing, and the method comprises the steps: collecting multi-source data in a drug life cycle after drug marketing, and carrying out the standardization processing of the multi-source data, and obtaining standard data; performing feature extraction processing on the standard data to obtain a medicine feature set, and performing fusion analysis on the medicine feature set by adopting a deep learning algorithm to obtain a risk portrait representing the overall safety condition of the medicine; performing drug risk detection based on the risk portrait to obtain a potential risk mode and a potential risk level after the drug appears on the market; and then verifying the potential risk mode and the potential risk level by adopting a support vector machine, and obtaining a risk detection result after the drug is sold on the market based on a verification result, so that full-process intelligent detection can be realized, and an accurate risk judgment basis is provided for medication safety.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligently detecting post-marketing risks of drugs, and in particular to a method and system for intelligently detecting post-marketing risks of drugs. BACKGROUND

[0002] In the whole life cycle of a drug after it is marketed, production environment stability, compliance of circulation and transportation conditions, and patient clinical medication reactions all directly affect the safety of the use of the drug; moreover, regulatory agencies need to know in real time the risk hidden dangers in each link from production to use of the drug, and pharmaceutical companies also need to identify potential risks in a timely manner to avoid safety accidents and protect the health of the medication population, and therefore there is an urgent need for a set of post-marketing risk detection solutions for drugs that can cover multiple links and accurately identify risks to meet the needs of comprehensive and efficient monitoring of the safety of drugs.

[0003] Currently, in the existing risk detection process of drugs after they are marketed, for the production link, data such as temperature and humidity, cleanliness, etc. are usually recorded and stored separately, and whether there is an environmental anomaly is determined by periodically manually reviewing the data; for the circulation and transportation link, paper or electronic reports provided by the transportation party, such as transportation routes and temperature records, are relied on, and whether the transportation conditions meet the preset standards is checked manually; for the clinical link, patient adverse reaction text records or abnormal physiological index data fed back by hospitals are relied on, and single cases or scattered data are counted to preliminarily determine whether there is a medication-related anomaly.

[0004] However, the most important disadvantage of the prior art is that risk identification is limited to a single link, and it is difficult to find potential risks across links; at the same time, the analysis method relies on manual statistics and simple data checking, and lacks the ability to deeply mine multi-source data, and cannot accurately extract risk characteristics and form complete risk logic, and therefore it is difficult to efficiently and accurately identify potential risks of drugs after they are marketed, and cannot meet the needs of comprehensive and accurate drug safety monitoring. SUMMARY

[0005] The present application aims to provide a method and system for intelligently detecting post-marketing risks of drugs to solve the problem of low accuracy of drug safety monitoring caused by inaccurate drug risk detection in the prior art.

[0006] To solve the above technical problems, in a first aspect, the present application provides a method for intelligently detecting post-marketing risks of drugs, comprising:

[0007] After a drug is marketed, multi-source data in the life cycle of the drug are collected, the multi-source data are standardized to obtain standard data;

[0008] The standard data is processed by feature extraction to obtain a drug feature set. A deep learning algorithm is then used to perform fusion analysis on the drug feature set to obtain a risk profile representing the overall safety status of the drug.

[0009] Based on the aforementioned risk profile, drug risk detection is conducted to obtain the potential risk patterns and potential risk levels after the drug is launched on the market.

[0010] Support vector machines are used to verify the potential risk patterns and potential risk levels, and the post-marketing risk detection results are obtained based on the verification results.

[0011] Optionally, the step of performing feature extraction processing on the standard data to obtain a drug feature set, and then using a deep learning algorithm to perform fusion analysis on the drug feature set to obtain a risk profile representing the overall safety status of the drug, includes:

[0012] Extract the characteristics of temperature and humidity data variation within a set time window, the number of abnormalities in cleanliness data exceeding a preset threshold, the transportation characteristics corresponding to transportation data, the adverse reaction characteristics corresponding to text data, and the abnormal duration characteristics corresponding to physiological indicator data from the standard data to form a drug feature set.

[0013] The first sub-network of the deep learning model performs correlation analysis on the change amplitude feature, the abnormality frequency feature and the transportation feature in the drug feature set to obtain the correlation between environment and transportation. The second sub-network of the deep learning algorithm performs correlation analysis on the adverse reaction feature and the abnormal duration feature to obtain the abnormal influencing factors. The deep learning model is constructed based on the deep learning algorithm.

[0014] By using the fusion layer of a deep learning model, the correlation and the abnormal influencing factors are fused and analyzed to obtain a risk profile.

[0015] Optionally, the first sub-network employing a deep learning model performs correlation analysis on the variation amplitude features, the abnormality frequency features, and the transportation features in the drug feature set to obtain the correlation between the environment and transportation. The second sub-network employing a deep learning algorithm performs correlation analysis on the adverse reaction features and the abnormality duration features to obtain the abnormal influencing factors, including:

[0016] The first sub-network of the deep learning model synchronously analyzes and processes the change amplitude feature, the transportation feature and the anomaly frequency feature according to the time dimension to obtain the association set. The association set is then processed by rule extraction to obtain the association relationship between the environment and transportation.

[0017] The second sub-network using a deep learning algorithm performs interval classification processing on the adverse reaction features based on a preset first interval division rule to obtain first interval features corresponding to different frequency intervals. Based on a preset second interval division rule, the abnormal duration features are also processed to obtain second interval features corresponding to different duration intervals.

[0018] The first interval features are associated with the second interval features to obtain a feature association set. Key features are extracted from the feature association set to obtain the abnormal influencing factors.

[0019] Optionally, the step of fusing and analyzing the correlation and the abnormal influencing factors to obtain a risk profile includes:

[0020] The association rules in the association relationship are matched with the association features in the abnormal influencing factors over time to obtain the combination of association features corresponding to different time intervals.

[0021] Risk information extraction processing is performed on the combination of the associated features to obtain environmental anomaly information, transportation anomaly information, and medication anomaly information;

[0022] The environmental anomaly information, the transportation anomaly information, and the medication anomaly information are structured and integrated to obtain a risk dataset;

[0023] The degree classification process is performed on the combination of associated features to obtain the degree classification result;

[0024] Based on the severity classification results, a profiling framework is constructed on the risk dataset to obtain a risk profile.

[0025] Optionally, the step of using a support vector machine to verify the potential risk patterns and potential risk levels, and obtaining post-marketing risk detection results based on the verification results, includes:

[0026] Risk pattern samples, corresponding risk level samples, and actual risk records are extracted from the historical drug risk database. The risk pattern samples, the risk level samples, and the actual risk records are then correlated and analyzed to form a reference validation dataset.

[0027] The potential risk patterns are structurally transformed to obtain features to be verified, the potential risk levels are standardized to obtain levels to be verified, and the features to be verified and the levels to be verified are combined to obtain a dataset to be verified.

[0028] A support vector machine is used to compare and analyze the dataset to be verified with the reference verification dataset to obtain the verification matching results;

[0029] Based on the verification matching results and the preset verification pass conditions, the potential risk patterns and potential risk levels are comprehensively analyzed to obtain the risk detection results after the drug is marketed.

[0030] Optionally, the step of using a support vector machine to compare and analyze the dataset to be verified with the reference verification dataset to obtain verification matching results includes:

[0031] The dataset to be verified and the reference verification dataset are split into multiple feature groups to be verified and multiple sample feature groups.

[0032] Support vector machine is used to compare the feature group to be verified with the sample feature group for each feature dimension to obtain the matching degree value of each feature pair. The matching degree values ​​of all feature pairs are combined and processed to obtain the overall matching degree.

[0033] Target samples whose overall matching degree reaches a preset matching threshold are selected from the reference verification dataset;

[0034] The verification level in the dataset to be verified is numerically compared with the risk level sample of the target sample to obtain the level matching degree. The verification feature in the dataset to be verified is feature-matched with the actual risk record of the target sample to obtain the feature correlation degree.

[0035] The overall matching degree, the grade matching degree, and the feature correlation degree are integrated and processed to obtain the verification matching result.

[0036] Optionally, based on the risk profile, the drug risk detection is performed to obtain the potential risk patterns and potential risk levels after the drug is marketed, including:

[0037] From the correlation between the environment and transportation in the risk profile, abnormal correlation combinations whose correlation strength values ​​exceed a preset range are selected;

[0038] In the risk profile, match the abnormal influencing factors corresponding to the abnormal association combination, and bind the abnormal association combination with the corresponding abnormal influencing factors to form a potential risk pattern;

[0039] The number of abnormal occurrences and the number of patients in the abnormal association combination within a specified monitoring period are statistically analyzed. The number of abnormal occurrences and the number of patients are then combined to calculate a grade reference value.

[0040] Based on a preset risk level mapping table, the level reference value is mapped to the corresponding potential risk level.

[0041] Secondly, this application provides a post-marketing risk intelligent detection system for pharmaceuticals, comprising:

[0042] The data acquisition module is used to collect multi-source data throughout the drug's lifecycle after it is launched on the market, and to standardize the multi-source data to obtain standard data.

[0043] The analysis module is used to perform feature extraction processing on the standard data to obtain a drug feature set, and to use a deep learning algorithm to perform fusion analysis on the drug feature set to obtain a risk profile representing the overall safety status of the drug.

[0044] The detection module is used to perform drug risk detection based on the risk profile to obtain the potential risk patterns and potential risk levels of the drug after it is launched on the market.

[0045] The verification module is used to verify the potential risk patterns and potential risk levels using a support vector machine, and to obtain the post-marketing risk detection results based on the verification results.

[0046] Thirdly, this application provides an electronic device, comprising:

[0047] Memory, used to store computer programs;

[0048] A processor is configured to execute the computer program to implement the steps of a post-marketing risk intelligent detection method for pharmaceuticals as described in the first aspect above.

[0049] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps of the intelligent risk detection method for post-marketing pharmaceuticals as described in the first aspect above.

[0050] The technical solution of this application has the following beneficial effects:

[0051] This application first covers and standardizes key data throughout the entire lifecycle of a drug, from production and distribution to clinical use, to eliminate differences in format, units, and standards among data from different sources, thereby avoiding subsequent analysis biases caused by data clutter; then, it extracts features reflecting the core dimensions of drug safety from the standardized data to eliminate redundant information and focus on key safety indicators.

[0052] Furthermore, deep learning enables the deep fusion of multiple features, thereby overcoming the limitations of single feature analysis. The structured profile can intuitively and systematically present the overall safety status of drugs, solving the problem that "fragmented data cannot reflect overall safety" in traditional analysis. This further provides comprehensive feature and profile support for subsequent risk detection. Then, relying on the risk profile that covers the logic of multiple links, potential risk logic across links can be accurately identified. At the same time, the severity of risks can be quantified by classifying the risk level to avoid risk identification being limited to a superficial judgment of "whether there is an abnormality".

[0053] Subsequently, the reliability of the potential risk patterns and levels obtained from the initial detection is verified by the support vector machine algorithm to eliminate misjudgments caused by accidental data fluctuations or local biases, thereby improving the accuracy of risk assessment. Finally, the final detection results output based on the verification results directly provide a reliable basis for regulatory decisions and pharmaceutical companies' risk management, further ensuring the rigor of risk assessment.

[0054] Furthermore, this application first clarifies the scope of core safety feature extraction and precisely focuses on three key dimensions: stability of the drug production environment, compliance of distribution and transportation, and correlation with clinical response, thereby avoiding the blindness and redundancy of feature extraction. Then, by using a twin network to specifically analyze the environmental, transportation, and clinical links, deep correlation of features in different dimensions can be achieved, thereby improving the accuracy of identifying the internal correlation logic of each link. Finally, by integrating the two types of key analysis results through a fusion layer, the final risk profile not only covers anomalies in a single link but also reflects the cross-link correlation logic of "production, distribution, and clinical", further enriching the comprehensiveness and depth of the profile and providing a more accurate and relevant foundation for subsequent risk detection based on the profile.

[0055] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0056] To more clearly illustrate the technical solutions of the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0057] Figure 1 A flowchart illustrating a post-marketing risk intelligent detection method for pharmaceuticals provided in this application embodiment;

[0058] Figure 2 This application provides a schematic diagram illustrating a specific implementation of a post-marketing risk intelligent detection method for pharmaceuticals.

[0059] Figure 3 This application provides a schematic diagram of the structure of a post-marketing risk intelligent detection system for pharmaceuticals.

[0060] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0061] In post-marketing safety monitoring of drugs, current testing methods still have significant limitations: they cannot detect risks across different stages; at the same time, most methods that rely on manual data review and statistical anomaly analysis are difficult to accurately extract key safety features from massive amounts of data, and cannot present the overall safety status of drugs. This limits risk identification to a single stage and makes it difficult to meet the needs for efficient monitoring of drug safety throughout its entire life cycle.

[0062] To address the aforementioned issues, this application proposes an intelligent risk detection method for post-marketing drugs. This method employs the following technical concept: First, it collects multi-source data from the entire lifecycle of a drug—production, distribution, and clinical use—and standardizes the data to eliminate data format differences. Next, it extracts key safety features from the standardized data and uses deep learning technology to fuse and analyze these features, forming a risk profile that intuitively reflects the overall safety status of the drug. Then, it identifies potential risk patterns and risk levels based on this profile. Finally, it uses a support vector machine to verify the identified risks, ensuring the accuracy of the results. This method solves the data fragmentation problem by integrating data from multiple stages, and uses intelligent algorithms to replace manual analysis, improving the depth and efficiency of the analysis. Furthermore, based on the risk profile and verification process, it achieves accurate identification and confirmation of risks across different stages, effectively filling the gaps in comprehensiveness and accuracy of existing detection methods.

[0063] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0064] The core of this application is to provide a method for intelligent risk detection after drug marketing, and a flowchart of one specific implementation is shown below. Figure 1 As shown, the method includes:

[0065] S101. After the drug is launched on the market, collect multi-source data throughout the drug's life cycle, and standardize the multi-source data to obtain standard data.

[0066] Among them, the drug life cycle refers to the entire process from the completion of the drug production process to the clinical use of the drug by the patient. Multi-source data refers to key safety-related data from different sources within this stage. Multi-source data includes temperature and humidity data of the production environment, cleanliness data of the production environment, transportation data, as well as text data and physiological indicator data fed back by patients after clinical use of the drug.

[0067] Furthermore, standardization refers to the process of unifying the format, converting units, and correcting outliers of collected multi-source data. The purpose is to eliminate the differences between data from different sources and make them consistent enough for direct subsequent analysis. For example, standardization includes: data format unification, data unit unification, outlier correction, data dimension unification, data type conversion, and data time alignment.

[0068] In step 101, the format of the multi-source data is first unified. For example, quantitative data such as temperature and humidity data, cleanliness data, physiological index data, and transportation time are converted into a unified format, such as: time-data type-value-unit. Text data is converted into a unified format: patient identification-medication time-symptom description. Then, the units of the quantitative data are unified. If different units exist for the same type of data, they are converted into standard units using the corresponding unit conversion formula.

[0069] Next, outliers in the data are corrected. When data that exceeds the reasonable range is detected, normal data before and after the data are retrieved, and linear interpolation is used to calculate the corrected data to replace the original outlier and obtain the standard data.

[0070] In this embodiment of the invention, comprehensive coverage of key information related to drug safety is achieved by collecting multi-source data, thereby avoiding information loss caused by focusing only on data from a single link; then, standardization processing of multi-source data can ensure data consistency and accuracy; finally, the obtained standard data can directly provide a high-quality, analyzable data foundation for subsequent feature extraction steps.

[0071] S102. The standard data is processed by feature extraction to obtain a drug feature set. A deep learning algorithm is used to perform fusion analysis on the drug feature set to obtain a risk profile representing the overall safety status of the drug.

[0072] Among them, the risk profile contains risk-related information throughout the entire life cycle of drug production, distribution, and clinical practice. This risk-related information includes abnormal information at each stage and the corresponding patterns between environmental anomalies, transportation conditions, and clinical hazards.

[0073] In one specific implementation, such as Figure 2 As shown, step S102 includes:

[0074] Step 1021: Extract the variation range characteristics of temperature and humidity data within a set time window, the abnormal number of times cleanliness data exceeds a preset threshold, the transportation characteristics corresponding to transportation data, the adverse reaction characteristics corresponding to text data, and the abnormal duration characteristics corresponding to physiological indicator data from the standard data to form a drug feature set.

[0075] Among them, the set time window refers to a fixed time interval preset for analyzing the changing trends of temperature and humidity data, used to focus on the temperature and humidity fluctuations within a specific period; the change amplitude characteristic refers to the difference between the maximum and minimum values ​​of temperature and humidity data within the set time window, reflecting the degree of temperature and humidity fluctuation; the preset threshold refers to the upper limit of cleanliness qualification set according to pharmaceutical production standards, and when the cleanliness data exceeds this value, it is judged as abnormal; the abnormality frequency characteristic refers to the number of times the cleanliness data exceeds the preset threshold, reflecting the stability of the cleanliness of the production environment;

[0076] Furthermore, transportation characteristics refer to the correlation information between transportation temperature, transportation time and transportation route in transportation data, reflecting the impact of transportation on drug safety; adverse reaction characteristics refer to the number of times patients report the same adverse reaction per unit time, used to reflect the frequency of clinical adverse reactions; abnormal duration characteristics refer to the duration of abnormal physiological indicators or adverse reactions in patients, reflecting the severity of clinical abnormalities.

[0077] For example, in the implementation of a certain hypoglycemic drug, firstly, a time window of 2 hours is set, and temperature and humidity data for two time periods, 9:00-11:00 and 11:00-13:00, are selected from the standard data of the day the drug was produced; then, a formula is used to calculate the variation characteristics. ,in, Indicates the range of temperature and humidity changes. This indicates the maximum temperature and humidity within a given time period. This represents the minimum temperature and humidity values ​​within a given time period. For example, the maximum temperature during the 9:00-11:00 time period is 24℃, and the minimum temperature is 22℃. Substituting these values ​​into the formula yields... =24℃-22℃=2℃, the maximum temperature during the 11:00-13:00 period is 23℃ and the minimum temperature is 22℃. Substituting into the formula, we get... =23℃-22=1℃, to obtain the variation range characteristics of the two time periods.

[0078] With a cleanliness threshold set at ≤3500 particles / m³ for particles ≥0.5μm in diameter, five cleanliness tests were conducted daily. One test result showed 3800 particles / m³, exceeding the threshold, thus identifying one instance of anomaly. Further analysis of transportation data revealed that the batch of medicines departed at 7:00 AM and arrived at 2:00 PM, using the formula... Calculate the transportation time, where t represents the transportation time. Indicates arrival time. The departure time is given. Substituting the data, we get t = 14 - 7 = 7 hours. Combined with the transportation route, which is from a production center in a certain area to a medical institution in a certain area, and the transportation temperature is maintained at 5-6℃, we can form the transportation characteristics.

[0079] Subsequently, the patient feedback data for the day was analyzed, yielding 8 feedback entries, of which "fatigue" was reported 5 times. Then, a formula was used... Calculate the adverse reaction repetition frequency, where f represents the adverse reaction repetition frequency and N represents the number of times the same adverse reaction occurs per unit time. Indicates a unit of time, for example. Assuming a day duration of 1, substituting the data yields f = 5 / 1 = 5 times / day, thus identifying the adverse reaction characteristics. The duration of abnormal physiological indicators for the three patients was recorded as 2 hours, 2.5 hours, and 3 hours, respectively, allowing for the calculation of the average duration of abnormality. =(2+2.5+3) / 3=2.5 hours, thus obtaining the abnormal duration characteristics, and integrating the above five types of characteristics into a drug characteristic set.

[0080] Step 1022: Using the first sub-network of the deep learning model, the variation amplitude feature, the abnormality frequency feature, and the transportation feature in the drug feature set are correlated and analyzed to obtain the correlation between the environment and transportation. Using the second sub-network of the deep learning algorithm, the adverse reaction feature and the abnormal duration feature are correlated and analyzed to obtain the abnormal influencing factors. The deep learning model is constructed based on the deep learning algorithm.

[0081] The deep learning model refers to an analysis model built based on deep learning algorithms. It includes a first sub-network for processing the correlation analysis of production environment and circulation and transportation-related features, a second sub-network for processing the correlation analysis of clinically relevant features, and a fusion layer for integrating the analysis results of different sub-networks. Furthermore, the embodiments of this application do not specifically limit the network structure of the first sub-network, the second sub-network, and the fusion layer. Abnormal influencing factors refer to key factors related to drug risk formed after the correlation of clinical features.

[0082] It should be understood that step 1022 may specifically include the following steps:

[0083] Step a1: Using the first sub-network of the deep learning model, the change amplitude feature, the transportation feature, and the anomaly frequency feature are analyzed and processed synchronously according to the time dimension to obtain the association set. The association set is then processed by rule extraction to obtain the association relationship between the environment and transportation.

[0084] Among them, the association relationship refers to the correspondence between environmental characteristics and transportation characteristics that are stable and repeatable extracted from the association set. This relationship can reflect the inherent connection between the two in the dimension of drug safety.

[0085] For example, the first sub-network of the deep learning model is started, and the change range features of 2℃ and 1℃, the number of anomalies of 1 time, and the transportation features are synchronously associated according to the daily time association dimension to obtain the association set "single day: change range of 2℃, 1 anomaly, transportation temperature of 5-6℃ and duration of 7 hours; single day: change range of 1℃, 1 anomaly, transportation temperature of 5-6℃ and duration of 7 hours"; then the association set is extracted to obtain the association relationship between environment and transportation "when cleanliness is abnormal once, the transportation temperature is mostly 5-6℃ and the transportation duration is 7 hours".

[0086] Step a2: The second sub-network of the deep learning algorithm performs interval classification processing on the adverse reaction features based on the preset first interval division rule to obtain the first interval features corresponding to different frequency intervals. Based on the preset second interval division rule, the abnormal duration features are performed interval classification processing to obtain the second interval features corresponding to different duration intervals.

[0087] For example, the second sub-network is started, and the first interval division rule is preset as 1-3 times / day, 4-6 times / day, and more than 7 times / day, and the second interval division rule is 1-2 hours, 2-4 hours, and more than 4 hours; then the "fatigue" frequency of 5 times / day is classified into the 4-6 times / day interval to obtain the first interval feature "fatigue: 4-6 times / day"; then the abnormal duration of 2.5 hours is classified into the 2-4 hour interval to obtain the second interval feature "abnormal physiological indicators: 2-4 hours".

[0088] Step a3: Associate the first interval features with the second interval features to obtain a feature association set, and extract key features from the feature association set to obtain abnormal influencing factors.

[0089] For example, the association identification yields the feature association set "fatigue 4-6 times / day and abnormal physiological indicators for 2-4 hours"; finally, the key features are extracted to obtain the abnormal influencing factor "fatigue frequency 4-6 times / day and abnormal physiological indicators lasting for 2-4 hours".

[0090] Step 1023: Through the fusion layer of the deep learning model, the correlation and the abnormal influencing factors are fused and analyzed to obtain a risk profile.

[0091] Risk profiling refers to a comprehensive safety profile that is presented in the form of structured data, covering risk information across all stages of drug production, distribution, and clinical use.

[0092] Step 1023 may specifically include the following steps:

[0093] Step b1: Perform time matching between the association rules in the association relationship and the association features in the abnormal influencing factors to obtain the association feature combinations corresponding to different time intervals.

[0094] In step b1, the time correlation dimension is determined to be a single day. Then, based on the time correlation dimension, the correlation between the environment and transportation and abnormal influencing factors are matched to obtain the single-day correlation feature combination.

[0095] Step b2: Perform risk information extraction processing on the associated feature combination to obtain environmental anomaly information, transportation anomaly information, and medication anomaly information.

[0096] Risk information refers to content related to environmental anomalies, transportation anomalies, and clinical anomalies extracted from combinations of associated features.

[0097] Step b3: The environmental anomaly information, the transportation anomaly information, and the medication anomaly information are structured and integrated to obtain a risk dataset.

[0098] For example, from the daily associated feature combination, environmental abnormal information "cleanliness abnormal once", transportation abnormal information "transportation temperature 5-6℃", and medication abnormal information "fatigue 4-6 times / day, physiological indicators abnormal for 2.5 hours" are extracted and integrated into a risk dataset "single day: cleanliness abnormal once, temperature 5-6℃, fatigue 4-6 times / day and indicators abnormal for 2.5 hours".

[0099] Step b4: Perform degree classification processing on the associated feature combination to obtain the degree classification result.

[0100] For example, when the degree of impact includes slight, moderate, and significant, the combination can be determined to have a moderate impact based on a preset mapping relationship, thus obtaining a degree classification result.

[0101] Step b5: Based on the severity classification results, perform profile framework construction on the risk dataset to obtain a risk profile.

[0102] For example, based on the same profiling framework corresponding to different levels of grading results, the risk dataset and the grading results are mapped together into the profiling framework to obtain a risk profile, which includes environmental, transportation, clinical, and grading modules; or, for another example, based on different profiling frameworks corresponding to different levels of grading results, the data is mapped to the environmental, transportation, and clinical modules of the medium-grade profiling framework to obtain a risk profile.

[0103] This application effectively filters out core information related to drug safety by extracting key features from the entire process of drug production, distribution, and clinical use, thus avoiding interference from redundant data in subsequent analysis. Next, by utilizing a dual-network deep learning model to process production and transportation features and clinical features separately, it achieves precise correlation analysis of features from different stages, thereby accurately uncovering the correlation patterns between the environment and transportation, as well as abnormal influencing factors at the clinical level. Finally, a fusion layer integrates the results of the two types of analysis to construct a structured risk profile, comprehensively presenting the overall safety status of the drug.

[0104] S103. Based on the risk profile, conduct drug risk detection to obtain the potential risk patterns and potential risk levels after the drug is marketed.

[0105] Among them, the potential risk model refers to the structured description of chain risks formed by binding abnormal association combinations with corresponding abnormal influencing factors, which clarifies the path of risk occurrence.

[0106] In one specific implementation, step S103 includes:

[0107] Step 1031: Filter out abnormal association combinations whose association strength values ​​exceed the preset range from the correlation between environment and transportation in the risk profile.

[0108] The preset range refers to the normal range of correlation strength values ​​set based on historical drug safety data and production and transportation standards. When the correlation strength value exceeds this range, the corresponding correlation relationship is judged as abnormal.

[0109] In step 1031, the correlation coefficient method is first used to calculate the degree of correlation between environmental features and transportation features in the risk profile to obtain the correlation strength value; then, a preset range is determined based on the correlation strength values ​​of multiple safe batches of the drug in the past; then, the correlation strength value of the current environmental and transportation relationship is compared with the preset range, and the combination of environmental features and transportation features corresponding to the relationship that exceeds the range is judged as an abnormal correlation combination.

[0110] For example, to create a risk profile for a certain hypoglycemic drug, the Pearson correlation coefficient method was used to calculate the correlation strength between "number of cleanliness anomalies" and "transportation temperature." Sample data from the past 7 days were selected, yielding cleanliness anomalies of 1, 2, 2, 1, 2, 0, and 1 times, and transportation temperatures of 7℃, 8℃, 8℃, 7℃, 8℃, 6℃, and 7℃. The Pearson correlation coefficient formula is as follows: Where n is the number of samples, such as n=7, x is the number of cleanliness tests, and y is the transportation temperature. The calculated value is r=0.7. The preset range is 0.2-0.5. Since 0.7 exceeds the preset range, the combination of "cleanliness abnormality twice and transportation temperature 8℃" is selected as the abnormal association combination.

[0111] Step 1032: Match the abnormal influencing factors corresponding to the abnormal association combination in the risk profile, and bind the abnormal association combination with the corresponding abnormal influencing factors to form a potential risk pattern.

[0112] In step 1032, the abnormal influencing factors corresponding to the abnormal association combination within the same time interval are matched in the risk profile according to the time dimension corresponding to the abnormal association combination; then the abnormal association combination is bound to the matched abnormal influencing factors to clarify the chain risk transmission path from abnormal production environment to deviation of transportation conditions to adverse clinical reactions, thereby forming a potential risk pattern describing the chain risk.

[0113] For example, based on the abnormal association combination "cleanliness abnormal 2 times and transportation temperature 8℃", the abnormal influencing factor "fatigue 6 times / day and blood sugar fluctuation 3 hours" is matched and bound to form a potential risk pattern "cleanliness abnormal 2 times and transportation temperature 8℃, fatigue 6 times / day and blood sugar fluctuation 3 hours".

[0114] Step 1033: Count the number of abnormal occurrences and the number of patients in the abnormal association combination within the specified monitoring period, and calculate the level reference value by combining the number of abnormal occurrences and the number of patients.

[0115] The designated monitoring period refers to a fixed time range preset for statistically analyzing the occurrence of abnormal correlation combinations, ensuring the timeliness and representativeness of the statistical results.

[0116] In step 1033, firstly, a specified monitoring period for statistical analysis is determined, which can be set based on the common cycles of drug risk occurrence; then, within the specified monitoring period, the number of abnormal occurrences of abnormal association combinations in potential risk patterns is counted; simultaneously, the number of patients associated with this chain risk is counted from the clinical module of the risk profile; and then, a formula is used... Calculate the grade reference value, where R is the grade reference value, a is the weight of the number of abnormal occurrences, b is the weight of the number of patients, N is the number of abnormal occurrences, and M is the number of patients. It should be understood that a and b are preset weight coefficients and their sum is 1, which are used to balance the influence of the number of abnormal occurrences and the number of patients.

[0117] For example, first set a specified monitoring period of 7 days, and count the occurrence of abnormal combinations 4 times and 12 associated patients, and a=0.4, b=0.6, and then use the above formula R=a×N+b×M to calculate the grade reference value R=0.4×4+0.6×12=8.8.

[0118] Step 1034: Based on the preset risk level mapping table, map the level reference value to the corresponding potential risk level.

[0119] The preset risk level mapping table refers to a pre-defined table that corresponds to the reference values ​​of the risk level and the risk level, which is used to quickly determine the severity of the risk.

[0120] In step 1034, a preset risk level mapping table based on industry drug risk standards and clinical safety requirements is retrieved. This table clarifies the risk level corresponding to different reference value ranges. Then, the calculated reference value is compared with the range in the preset risk level mapping table to determine the potential risk level corresponding to the reference value. The resulting potential risk pattern and potential risk level will be used for verification operations in subsequent steps.

[0121] This application can accurately locate potential risks deviating from normal ranges in the production and transportation process by screening abnormal association combinations, thus avoiding the omission of potential risk points across stages. Subsequently, by forming a potential risk model, it clarifies the chain risk path from the environment and transportation to the clinical setting, making the risk transmission logic clear and traceable, thereby solving the problem of "knowing the abnormality but not the cause" in traditional risk detection. Furthermore, by quantifying and integrating the number of abnormalities with the number of patients, it calculates a level reference value, providing objective data support for risk level determination, thereby avoiding bias caused by subjective judgment. Finally, the determined potential risk level intuitively reflects the severity of drug risk, providing a clear basis for subsequent risk verification and control decisions.

[0122] S104. Using a support vector machine, the potential risk patterns and potential risk levels are verified, and the post-marketing risk detection results are obtained based on the verification results.

[0123] In one specific implementation, S104 includes:

[0124] Step 1041: Extract risk pattern samples, corresponding risk level samples, and actual risk records from the drug historical risk database. Perform correlation analysis on the risk pattern samples, the risk level samples, and the actual risk records to form a reference validation dataset.

[0125] Among them, the historical drug risk database refers to a database that stores information related to past drug risks, and this historical drug risk database includes risk pattern samples, risk level samples, and actual risk records; the reference validation dataset refers to a structured dataset formed by associating risk pattern samples, risk level samples, and actual risk records, which is used to provide a comparison benchmark for the data to be validated.

[0126] In step 1041, historical risk data of the same type as the current drug are screened from the historical risk database of drugs, and risk pattern samples, corresponding risk level samples and actual risk records are extracted from them; then, data association technology is used to bind the risk pattern samples, risk level samples and actual risk records corresponding to the same historical risk event to form a structured reference verification dataset.

[0127] For example, after a certain hypoglycemic drug is launched, firstly, risk data of similar hypoglycemic drugs in the past 5 years are screened from the historical drug risk database to obtain 25 risk pattern samples, 25 risk level samples, and 25 actual risk records. The screening criteria are that the data covers different production batches and different transportation areas and has been filed and confirmed by regulatory agencies. Then, a data association algorithm is used to bind each risk pattern sample with its corresponding risk level sample actual risk record to form a reference validation dataset containing 25 structured records.

[0128] Step 1042: Perform a structured transformation on the potential risk pattern to obtain the feature to be verified, perform a standardized transformation on the potential risk level to obtain the level to be verified, and combine the feature to be verified and the level to be verified to obtain the dataset to be verified.

[0129] In step 1042, the potential risk patterns are structurally transformed, and the risk transmission paths described in text are decomposed into quantifiable features to form features to be verified. Then, the potential risk levels are standardized and transformed, and the qualitative or graded risk levels are converted into unified levels to be verified according to the preset level-value correspondence rules. Subsequently, the features to be verified and the levels to be verified are combined to form a dataset to be verified that can be used for comparison.

[0130] For example, the potential risk patterns obtained—cleanliness abnormalities twice, transportation temperature of 8°C, fatigue six times a day, and blood glucose fluctuations for three hours—are structurally transformed and broken down into four features to be verified: cleanliness abnormality frequency twice, transportation temperature of 8°C, fatigue frequency of six times, and blood glucose fluctuation duration of three hours per day. Then, according to the preset level standardization rules: low risk corresponds to 1 to 3, medium risk corresponds to 4 to 7, and high risk corresponds to 8 to 10, the potential risk level of high risk is converted into a verification level of 10, and the four features to be verified are combined with the verification level of 10 to form a verification dataset.

[0131] Step 1043: Using a support vector machine, the dataset to be verified is compared and analyzed with the reference verification dataset to obtain the verification matching result.

[0132] Step 1043 may specifically include the following steps:

[0133] Step c1: Split the dataset to be verified and the reference verification dataset to obtain multiple feature groups to be verified and multiple sample feature groups.

[0134] For example, the dataset to be verified is split into five feature groups to be verified: 2 instances of cleanliness abnormality, 8℃ transportation temperature, 6 instances of fatigue, 3 hours of daily blood glucose fluctuation, and a grade value of 10. At the same time, the 25 samples in the reference verification dataset are split one by one. Taking one sample as an example, it is split into five sample feature groups: 2 instances of cleanliness abnormality, 9℃ transportation temperature, 7 instances of fatigue, 4 hours of daily blood glucose fluctuation, and a grade value of 10.

[0135] Step c2: Using a support vector machine, the feature groups to be verified corresponding to each feature dimension are compared with the sample feature groups item by item to obtain the matching degree value of each feature pair. The matching degree values ​​of all feature pairs are then combined and processed to obtain the overall matching degree.

[0136] In this context, the feature group to be verified corresponding to a feature dimension and the sample feature group constitute a feature pair; and the comprehensive operation can refer to the weighted operation of the weight of each feature component and the corresponding matching degree value.

[0137] For example, first set the weights of the feature components, such as: the number of cleanliness abnormalities and transportation temperature each account for 0.15, the frequency of fatigue and the duration of blood sugar fluctuations each account for 0.2, and the grade dimension accounts for 0.3.

[0138] Then, the support vector machine was used to calculate the matching degree value of each feature pair. The matching degree value of cleanliness 2 times and cleanliness 2 times was 1; the matching degree value of transport temperature 8℃ and transport temperature 9℃ was 0.8 because the values ​​were close; the matching degree value of fatigue 6 times a day and fatigue 7 times a day was 0.7; the matching degree value of blood sugar fluctuation 3 hours and blood sugar fluctuation 4 hours was 0.6; and the matching degree value of grade 10 and grade 10 was 1.

[0139] Then the formula was used. Calculate the overall matching degree, where, This represents the overall matching degree, where n represents the number of feature components, e.g., n=5. This represents the weight of the i-th feature component. This represents the matching degree value of the i-th feature component, which is then substituted into the data to calculate the overall matching degree. =0.83.

[0140] Step c3: Select target samples from the reference verification dataset whose overall matching degree reaches the preset matching threshold.

[0141] The preset matching threshold refers to the critical value for determining whether a sample is a target sample.

[0142] For example, if the preset matching threshold is set to 0.8, the sample is determined to be the target sample because the overall matching degree of the sample is 0.83, which meets the threshold.

[0143] Step c4: Compare the levels to be verified in the dataset to be verified with the risk level samples of the target sample to obtain the level matching degree; compare the features to be verified in the dataset to be verified with the actual risk records of the target sample to obtain the feature correlation degree.

[0144] For example, risk level sample 10 and actual risk records are extracted from the target sample, and no new feedback is received within a week after the transportation temperature is adjusted to 5 to 6 degrees Celsius; then the level to be verified 10 is compared with risk level sample 10. Since the values ​​are the same, the level matching degree is 1; then the fatigue and blood glucose fluctuation in the feature to be verified are compared with the fatigue and blood glucose fluctuation in the actual risk record. Since they are both the same type of clinical response, the feature correlation degree is 0.9.

[0145] Step c5: Integrate the overall matching degree, the grade matching degree, and the feature correlation degree to obtain the verification matching result.

[0146] Among them, the verification matching result refers to the multi-dimensional comparison information set formed by integrating the overall matching degree, the level matching degree, and the feature correlation degree.

[0147] For example, the overall matching degree of 0.83, the level matching degree of 1, and the feature correlation degree of 0.9 are integrated to form the verification matching result.

[0148] Step 1044: Based on the verification matching results and the preset verification pass conditions, a comprehensive analysis is performed on the potential risk patterns and the potential risk levels to obtain the post-marketing risk detection results of the drug.

[0149] The preset verification pass conditions refer to the criteria for determining whether the potential risk patterns and levels are effective. This embodiment does not specifically limit the content of these conditions, which can be set according to the actual situation. The risk detection result refers to the final confirmed drug risk conclusion based on comprehensive analysis.

[0150] In step 1044, the preset verification pass conditions are retrieved, which specify the minimum standards that the overall matching degree, grade matching degree, and feature correlation degree must reach. The verification matching results are then compared with the preset verification pass conditions to determine the conformity of the potential risk pattern and potential risk level. If the pass conditions are met, the potential risk pattern is determined to be effective and the potential risk level to be accurate. Based on this, the post-marketing risk detection results of the drug are formed. If the conditions are not met, the data from the previous steps need to be re-verified until a reliable result is obtained.

[0151] For example, the preset verification pass conditions are: overall matching degree not less than 0.8, grade matching degree not less than 0.7, and feature correlation degree not less than 0.8. Then, the verification matching results are compared with the pass conditions. Since the overall matching degree of 0.83 meets the requirement of not less than 0.8, the grade matching degree of 1 meets the requirement of not less than 0.7, and the feature correlation degree of 0.9 meets the requirement of not less than 0.8, it is determined that the potential risk pattern and potential risk level both meet the requirements. Based on this, the risk detection results are obtained: the potential risk pattern of a certain hypoglycemic drug is effective, the high risk level is accurate, and it is recommended to immediately control the transportation temperature between 2 and 6 degrees Celsius and increase the frequency of production cleanliness testing from once a day to twice a day.

[0152] This application provides an objective comparison benchmark for the content to be verified by constructing a reference verification dataset based on historical risk data, which can avoid subjective verification without basis. Then, by converting potential risk patterns and levels, unstructured information is transformed into a quantifiable dataset, which can ensure the accuracy and operability of subsequent comparisons. Subsequently, by using support vector machines to perform multi-dimensional comparisons, verification is carried out from multiple aspects such as overall similarity, level consistency, and feature correlation, and the key dimensions of risk verification are fully covered, thereby avoiding the bias caused by single-dimensional comparisons. Finally, the objectivity and standardization of the verification conclusions are ensured by judging according to preset conditions.

[0153] Figure 3 This is a schematic diagram illustrating a specific implementation of a post-marketing risk intelligent detection system for pharmaceuticals, as provided in this application. (Refer to...) Figure 3 The system may include:

[0154] The data acquisition module 31 is used to collect multi-source data during the drug's life cycle after the drug is launched on the market, and to standardize the multi-source data to obtain standard data.

[0155] The analysis module 32 is used to perform feature extraction processing on the standard data to obtain a drug feature set, and to use a deep learning algorithm to perform fusion analysis on the drug feature set to obtain a risk profile representing the overall safety status of the drug.

[0156] The detection module 33 is used to perform drug risk detection based on the risk profile to obtain the potential risk patterns and potential risk levels after the drug is launched on the market.

[0157] The verification module 34 is used to verify the potential risk patterns and potential risk levels using a support vector machine, and to obtain the post-marketing risk detection results based on the verification results.

[0158] This application provides an intelligent post-marketing risk detection system for drugs, which is used to implement the aforementioned intelligent post-marketing risk detection method for drugs. Therefore, the specific implementation of the intelligent post-marketing risk detection system for drugs can be found in the embodiment section of the intelligent post-marketing risk detection method for drugs mentioned above. The specific implementation can be referred to the description of the corresponding embodiments, and will not be repeated here.

[0159] like Figure 4 This application also provides an electronic device, including: a memory 41 for storing a computer program; and a processor 42 for executing the computer program to implement the steps of any of the above-described intelligent risk detection methods for post-marketing drug products.

[0160] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of any of the above-described intelligent risk detection methods for post-marketing drug products.

[0161] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, random access memory, portable hard drives, magnetic disks, or optical disks.

[0162] Embodiments of the present invention also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the embodiments of the intelligent risk detection method for post-marketing of a drug.

[0163] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0164] The above provides a detailed description of the intelligent risk detection method and system for post-marketing drug products provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A method for intelligent risk detection after drug marketing, characterized in that, include: After a drug is launched on the market, multi-source data is collected throughout the drug's lifecycle, and the multi-source data is standardized to obtain standard data. The standard data is processed by feature extraction to obtain a drug feature set. A deep learning algorithm is then used to perform fusion analysis on the drug feature set to obtain a risk profile representing the overall safety status of the drug. Based on the aforementioned risk profile, drug risk detection is conducted to obtain the potential risk patterns and potential risk levels after the drug is launched on the market. Support vector machines are used to verify the potential risk patterns and potential risk levels, and the post-marketing risk detection results are obtained based on the verification results.

2. The method according to claim 1, characterized in that, The standard data is processed by feature extraction to obtain a drug feature set. A deep learning algorithm is then used to fuse and analyze the drug feature set to obtain a risk profile representing the overall safety status of the drug, including: Extract the characteristics of temperature and humidity data variation within a set time window, the number of abnormalities in cleanliness data exceeding a preset threshold, the transportation characteristics corresponding to transportation data, the adverse reaction characteristics corresponding to text data, and the abnormal duration characteristics corresponding to physiological indicator data from the standard data to form a drug feature set. The first sub-network of the deep learning model performs correlation analysis on the change amplitude feature, the abnormality frequency feature and the transportation feature in the drug feature set to obtain the correlation between environment and transportation. The second sub-network of the deep learning algorithm performs correlation analysis on the adverse reaction feature and the abnormal duration feature to obtain the abnormal influencing factors. The deep learning model is constructed based on the deep learning algorithm. By using the fusion layer of a deep learning model, the correlation and the abnormal influencing factors are fused and analyzed to obtain a risk profile.

3. The method according to claim 2, characterized in that, The first sub-network employing a deep learning model performs correlation analysis on the variation amplitude features, the abnormality frequency features, and the transportation features in the drug feature set to obtain the correlation between the environment and transportation. The second sub-network employing a deep learning algorithm performs correlation analysis on the adverse reaction features and the abnormality duration features to obtain the abnormal influencing factors, including: The first sub-network of the deep learning model synchronously analyzes and processes the change amplitude feature, the transportation feature and the anomaly frequency feature according to the time dimension to obtain the association set. The association set is then processed by rule extraction to obtain the association relationship between the environment and transportation. The second sub-network using a deep learning algorithm performs interval classification processing on the adverse reaction features based on a preset first interval division rule to obtain first interval features corresponding to different frequency intervals. Based on a preset second interval division rule, the abnormal duration features are also processed to obtain second interval features corresponding to different duration intervals. The first interval features are associated with the second interval features to obtain a feature association set. Key features are extracted from the feature association set to obtain the abnormal influencing factors.

4. The method according to claim 2, characterized in that, The process of fusing and analyzing the correlations and the abnormal influencing factors to obtain a risk profile includes: The association rules in the association relationship are matched with the association features in the abnormal influencing factors over time to obtain the combination of association features corresponding to different time intervals. Risk information extraction processing is performed on the aforementioned combination of related features to obtain environmental anomaly information, transportation anomaly information, and medication anomaly information; The environmental anomaly information, the transportation anomaly information, and the medication anomaly information are structured and integrated to obtain a risk dataset; The degree classification process is performed on the combination of associated features to obtain the degree classification result; Based on the severity classification results, a profiling framework is constructed on the risk dataset to obtain a risk profile.

5. The method according to claim 1, characterized in that, The step of employing a support vector machine to validate the potential risk patterns and levels, and obtaining post-marketing risk detection results based on the validation results, includes: Risk pattern samples, corresponding risk level samples, and actual risk records are extracted from the historical drug risk database. The risk pattern samples, the risk level samples, and the actual risk records are then correlated and analyzed to form a reference validation dataset. The potential risk patterns are structurally transformed to obtain features to be verified, the potential risk levels are standardized to obtain levels to be verified, and the features to be verified and the levels to be verified are combined to obtain a dataset to be verified. A support vector machine is used to compare and analyze the dataset to be verified with the reference verification dataset to obtain the verification matching results; Based on the verification matching results and the preset verification pass conditions, the potential risk patterns and potential risk levels are comprehensively analyzed to obtain the risk detection results after the drug is marketed.

6. The method according to claim 5, characterized in that, The step of using a support vector machine to compare and analyze the dataset to be verified with the reference verification dataset to obtain verification matching results includes: The dataset to be verified and the reference verification dataset are split into multiple feature groups to be verified and multiple sample feature groups. Support vector machine is used to compare the feature group to be verified with the sample feature group for each feature dimension to obtain the matching degree value of each feature pair. The matching degree values ​​of all feature pairs are combined and processed to obtain the overall matching degree. Target samples whose overall matching degree reaches a preset matching threshold are selected from the reference verification dataset; The verification level in the dataset to be verified is numerically compared with the risk level sample of the target sample to obtain the level matching degree. The verification feature in the dataset to be verified is feature-matched with the actual risk record of the target sample to obtain the feature correlation degree. The overall matching degree, the grade matching degree, and the feature correlation degree are integrated and processed to obtain the verification matching result.

7. The method according to claim 1, characterized in that, Based on the risk profile, drug risk detection is performed to obtain the potential risk patterns and potential risk levels after the drug is marketed, including: From the correlation between the environment and transportation in the risk profile, abnormal correlation combinations whose correlation strength values ​​exceed a preset range are selected; In the risk profile, match the abnormal influencing factors corresponding to the abnormal association combination, and bind the abnormal association combination with the corresponding abnormal influencing factors to form a potential risk pattern; The number of abnormal occurrences and the number of patients in the abnormal association combination within a specified monitoring period are statistically analyzed. The number of abnormal occurrences and the number of patients are then combined to calculate a grade reference value. Based on a preset risk level mapping table, the level reference value is mapped to the corresponding potential risk level.

8. A post-marketing risk intelligent detection system for pharmaceuticals, characterized in that, include: The data acquisition module is used to collect multi-source data throughout the drug's lifecycle after it is launched on the market, and to standardize the multi-source data to obtain standard data. The analysis module is used to perform feature extraction processing on the standard data to obtain a drug feature set, and to use a deep learning algorithm to perform fusion analysis on the drug feature set to obtain a risk profile representing the overall safety status of the drug. The detection module is used to perform drug risk detection based on the risk profile to obtain the potential risk patterns and potential risk levels of the drug after it is launched on the market. The verification module is used to verify the potential risk patterns and potential risk levels using a support vector machine, and to obtain the post-marketing risk detection results based on the verification results.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the intelligent risk detection method for post-marketing pharmaceuticals as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables the intelligent risk detection method for drugs after marketing as described in any one of claims 1 to 7.

Images