Adverse reaction data analysis method and device, electronic equipment and storage medium

Abstract

The application provides an adverse reaction data analysis method and device, electronic equipment and a storage medium, wherein the method comprises the following steps: acquiring adverse reaction data related to a drug; based on the data type to which the adverse reaction data belongs, performing information extraction on the adverse reaction data to obtain target information of the adverse reaction data; and through an adverse reaction data analysis system, performing adverse reaction data analysis on the target information under a target dimension to obtain an analysis result. By using the application, the analysis efficiency of the adverse reaction data of the drug can be improved.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a method, apparatus, electronic device, and storage medium for adverse reaction data analysis. Background Technology

[0002] Currently, adverse reaction data analysis is a crucial part of the drug development process.

[0003] However, existing adverse reaction data analysis methods not only lack the underlying data support for the entire drug development lifecycle, but also still rely on manual experience to analyze and integrate the massive amount of adverse event data reported by various parties. As a result, even after investing a lot of manpower and energy, it is still impossible to meet users' high-precision analysis needs for adverse drug reactions.

[0004] Therefore, existing methods for analyzing adverse reaction data suffer from low analytical efficiency. Summary of the Invention

[0005] This invention provides an adverse reaction data analysis method, apparatus, electronic device, and storage medium to address the shortcomings of existing adverse reaction data analysis systems that cannot meet users' high-precision analysis needs for adverse drug reactions.

[0006] This invention provides a method for adverse reaction data analysis, comprising:

[0007] Obtain data on adverse drug reactions;

[0008] Based on the data type of the adverse reaction data, information is extracted from the adverse reaction data to obtain the target information of the adverse reaction data;

[0009] The adverse reaction data analysis system is used to analyze the target information under the target dimension to obtain the analysis results.

[0010] According to the adverse reaction data analysis method provided by the present invention, the adverse reaction data analysis system includes at least one of a time analysis unit, a drug analysis unit, and an enterprise analysis unit;

[0011] The step of using an adverse reaction data analysis system to perform adverse reaction data analysis on the target information under the target dimension to obtain analysis results includes:

[0012] Determine the target dimension, which includes at least one of the following: time dimension, drug dimension, and enterprise dimension;

[0013] Extract the information to be analyzed that is associated with the target dimension from the target information;

[0014] By using at least one of the time analysis unit, the drug analysis unit, and the enterprise analysis unit, adverse reaction data analysis is performed on the information to be analyzed under the target dimension to obtain the analysis results.

[0015] According to the adverse reaction data analysis method provided by the present invention, the step of extracting the information to be analyzed associated with the target dimension from the target information includes:

[0016] Obtain the target object, which is based on at least one description of patient information, drug information, enterprise information, indication information, reporting time, and medical terminology information;

[0017] Extract the information to be analyzed from the target information that is associated with both the target dimension and the target object.

[0018] According to the adverse reaction data analysis method provided by the present invention, the step of performing adverse reaction data analysis on the information to be analyzed under a target dimension through at least one of the time analysis unit, the drug analysis unit, and the enterprise analysis unit to obtain analysis results includes:

[0019] When the target dimension is the time dimension, the time analysis unit analyzes the type and / or frequency of adverse reaction events corresponding to the target time in the information to be analyzed, and obtains the analysis results.

[0020] When the target dimension is the drug dimension, the drug analysis unit analyzes the type and / or frequency of adverse reaction events corresponding to the target drug in the information to be analyzed to obtain the analysis results;

[0021] When the target dimension is the enterprise dimension, the enterprise analysis unit analyzes the target drugs associated with the target enterprise in the information to be analyzed, as well as the types and / or frequency of adverse reaction events associated with the target drugs, to obtain the analysis results.

[0022] According to the adverse reaction data analysis method provided by the present invention, the step of analyzing the type and / or frequency of adverse reaction events corresponding to the target drug in the information to be analyzed through the drug analysis unit to obtain the analysis results includes:

[0023] Based on the causal relationship between each target drug and each adverse reaction event in the information to be analyzed, the type and / or frequency of adverse reaction events corresponding to each target drug are statistically analyzed from the information to be analyzed.

[0024] The drug analysis unit compares the types and / or frequency of adverse events corresponding to each target drug, and determines the analysis results based on the comparison results.

[0025] According to the adverse reaction data analysis method provided by the present invention, the acquisition of drug-related adverse reaction data includes:

[0026] Acquire drug information data and extract information text from the drug information data to perform adverse reaction association classification on the information text. Then, based on the adverse reaction association classification results, filter adverse reaction data from the drug information data according to information sources; and,

[0027] Obtain adverse reaction data from sources of clinical adverse reaction reports and / or post-marketing adverse reaction databases.

[0028] According to the adverse reaction data analysis method provided by the present invention, the step of extracting information from the adverse reaction data based on the data type to which the adverse reaction data belongs, to obtain the target information of the adverse reaction data, includes:

[0029] When the data type of the adverse reaction data includes text, entity extraction is performed on the text in the adverse reaction data to obtain the adverse reaction entity;

[0030] The target information is determined based on the adverse reaction entities and the relationships between them in the text.

[0031] The present invention also provides an adverse reaction data analysis device, comprising:

[0032] Acquisition unit, used to acquire adverse reaction data related to drugs;

[0033] The extraction unit is used to extract information from the adverse reaction data based on the data type to which the adverse reaction data belongs, and to obtain the target information of the adverse reaction data;

[0034] The analysis unit is used to perform adverse reaction data analysis on the target information under the target dimension through the adverse reaction data analysis system, and obtain the analysis results.

[0035] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the adverse reaction data analysis method as described above.

[0036] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the adverse reaction data analysis method as described above.

[0037] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the adverse reaction data analysis method as described above.

[0038] The adverse reaction data analysis method, device, electronic device, storage medium, and computer program product provided by this invention acquire drug-related adverse reaction data, providing underlying data support for drug development throughout its entire lifecycle for subsequent adverse reaction data analysis; and through the adverse reaction data analysis system, perform adverse reaction data analysis on the target information extracted from the information at the target dimension, enabling rapid multi-dimensional and systematic combined analysis of adverse reaction data, meeting users' high-precision analysis needs for drug adverse reactions, and thus solving the problem of low efficiency in existing adverse reaction data analysis in one stop. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in this invention 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 some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0040] Figure 1 This is a flowchart illustrating the adverse reaction data analysis method provided by the present invention;

[0041] Figure 2 This is a schematic diagram of the adverse reaction data analysis device provided by the present invention;

[0042] Figure 3 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0044] Establishing pharmacovigilance during clinical trials, strengthening quality management of rapid reporting of safety data during clinical trials, monitoring rapid reporting of safety data during clinical trials, identifying safety signals, assessing and controlling safety risks, and establishing pharmacovigilance covering the entire life cycle of drugs are important tasks of regulatory science.

[0045] The China Hospital Pharmacovigilance System (CHPS) is an active monitoring system established by the National Center for Adverse Drug Reaction Monitoring to achieve post-marketing adverse drug reaction monitoring. Through the cooperation mechanism established by CASSA, CHPS was established to open up data channels between data sources, and by establishing risk analysis models, drug safety issues can be verified.

[0046] However, CHPS currently only monitors adverse events following drug marketing, and its data sources are concentrated in a few top-tier hospitals within CASSA with strong research capabilities and high levels of information technology, resulting in a single data source for analysis. This leads to a narrow monitoring scope for the data analysis system built upon this foundation, directly impacting the effectiveness and timeliness of data analysis on adverse drug reactions.

[0047] Furthermore, the massive amount of adverse event data reported by various parties still relies on manual experience for analysis and integration, which, despite the significant investment of manpower and effort, still cannot meet users' high-precision analysis needs for adverse drug reactions.

[0048] Based on this, embodiments of the present invention provide an adverse reaction data analysis method, which is applied to the data analysis scenario of adverse drug reactions. For example, it can directly perform data analysis on the original report of suspected unexpected serious adverse reaction (SUSAR) to meet users' high-precision analysis needs for adverse drug reactions and improve the efficiency of data analysis for adverse drug reactions.

[0049] Figure 1 This is a flowchart illustrating the adverse reaction data analysis method provided by the present invention, as shown below. Figure 1 As shown, the adverse reaction data analysis method provided in this embodiment of the invention may include the following steps:

[0050] Step 110: Obtain adverse reaction data related to the drug;

[0051] Step 120: Based on the data type of the adverse reaction data, extract information from the adverse reaction data to obtain the target information of the adverse reaction data;

[0052] Step 130: Using the adverse reaction data analysis system, perform adverse reaction data analysis on the target information under the target dimension to obtain the analysis results.

[0053] Specifically, adverse drug reaction data refers to data related to adverse reaction events, which can include adverse drug reactions (ADRs) and adverse drug events (ADEs). An adverse drug reaction is a harmful or unexpected reaction to a qualified drug when used at the normal dosage and administration; an adverse drug event is any unfortunate medical or health event that occurs during drug treatment, and such event is not necessarily causally related to the drug treatment.

[0054] Adverse reaction data can include various modalities such as text, charts (images and tables), and audio and video. Adverse reaction data can be acquired through external devices, blockchain-based acquisition, polling, request-based acquisition, or direct acquisition from pre-stored data. Taking text in adverse reaction data as an example, it can be directly input by the user, or it can be obtained by transcribing collected audio, or it can be obtained by capturing images through image acquisition devices such as scanners, mobile phones, and cameras and performing OCR (Optical Character Recognition) on the images, or it can be obtained from the Internet through web crawling. This embodiment of the invention does not specifically limit the method of acquiring adverse reaction data.

[0055] Given that current monitoring primarily focuses on adverse event monitoring of post-marketed drugs, and the data sources are often limited, the data analysis systems built upon this foundation have a narrow monitoring scope and lack comprehensive underlying data support for the entire drug development lifecycle. In this embodiment of the invention, the adverse event data sources include at least two, for example, data can be obtained from two or more data sources. Furthermore, these at least two data sources can include data sources for monitoring adverse events of post-marketed drugs, patient medical records, and data sources for monitoring adverse events of drugs in clinical trials, thereby covering adverse event monitoring throughout the entire drug lifecycle. This broadens the monitoring scope of the data analysis system built upon adverse event data, thereby improving the effectiveness and timeliness of data analysis for adverse drug reactions.

[0056] In some embodiments, obtaining adverse reaction data related to the drug, i.e., step 110 specifically includes:

[0057] Step 111: Obtain drug information data and extract information text from the drug information data to classify the information text according to adverse reaction associations. Then, based on the adverse reaction association classification results, filter adverse reaction data from the information sources in the drug information data.

[0058] And, in step 112, obtain adverse reaction data from sources of clinical adverse reaction reports and / or post-marketing adverse reaction databases.

[0059] It is understood that steps 111 and 112 here are parallel steps, and the embodiments of the present invention do not specifically limit the execution order of steps 111 and 112.

[0060] Specifically, the data source for pharmaceutical information data is information sources, such as information literature related to the clinical trial phase or post-marketing application phase of a drug. Pharmaceutical information data can specifically come from announcements or news from pharmaceutical companies or regulatory authorities, or from literature or conference information related to medicine, pharmacy, biology, health science, or nursing, etc. This embodiment of the invention does not specifically limit this.

[0061] To acquire pharmaceutical information data, web scraping technology can be used to crawl data from pharmaceutical company websites, public accounts, news websites, and other sources. Automated detection and alerts can be implemented to achieve real-time monitoring of pharmaceutical information data, ensuring the availability and quality of the data sources. Furthermore, web scraping programs written in Python can be used with multi-threading technology to crawl target websites and obtain pharmaceutical information data.

[0062] Given that drug information data may include information from different directions, such as adverse drug reaction events, drug marketing and sales, it is necessary to filter the drug information data to obtain adverse reaction data.

[0063] To obtain adverse reaction data, the information text in the drug information data can be extracted first. Here, information text can be understood as textual information in the drug information data. The information text can be in any natural language, such as English, Chinese, Japanese, etc., but this embodiment of the invention does not specifically limit this.

[0064] Information text can be obtained by parsing the text in the drug information data, or by performing speech recognition on the speech, or by performing OCR (Optical Character Recognition) on the images. This embodiment of the invention does not specifically limit the specific methods used.

[0065] In some embodiments, considering that the drug information data may contain chart-like data, step 111 involves extracting the information text from the drug information data, including:

[0066] Step 111-1: Based on the object detection model, perform chart detection on the drug information data. The object detection model is trained on a set of pre-labeled sample data where the chart area covers the title area.

[0067] Step 111-2: If the chart position information in the drug information data is detected, then extract the information text from the chart in the drug information data based on the chart position information.

[0068] Specifically, to extract the information text from the charts in pharmaceutical information data, a trained object detection model can be used to perform chart detection on the pharmaceutical information data. Specifically, the pharmaceutical information data can be input into the trained object detection model, which will then perform chart detection. If the chart location information is detected within the pharmaceutical information data, this chart location information is output. This chart location information includes not only the location of the area occupied by the chart within the pharmaceutical information data, but also the location of the area occupied by the chart title; that is, the chart location information can include both the location of the chart area and the location of the title area.

[0069] The target detection model can be trained in the following way:

[0070] Obtain a set of sample documents. Each sample document in the set is pre-annotated with a chart area and a title area. The pre-annotated chart area can cover the title area associated with the chart. In other words, each sample document in the set is pre-annotated with a chart position and a title position, and the pre-annotated chart position is bound to the corresponding title position.

[0071] For example, if the sample document contains a chart A, when annotating the sample document, the area occupied by the title corresponding to chart A is marked as the title area, and the area occupied by the title area and chart A is marked as the chart area. Thus, the annotated chart area covers the title area.

[0072] Subsequently, the initial object detection model can be trained based on the pre-labeled sample document set. During the training process, the initial object detection model continuously learns the position information of the chart area and title area in the sample documents. The object detection model trained in this way can accurately predict the charts and their titles that may be contained in the drug information data and obtain the chart position information.

[0073] After obtaining the chart location information from the drug information data, we can analyze the chart area location information A and the title area location information B contained within the chart location information to determine the unique title corresponding to each chart (for example, the unique correspondence between the two objects can be determined if location information A and location information B meet the following conditions: ① Location information B is not within the invalid area of ​​the document page; ② Location information A covers location information B; ③ Location information B is not within the invalid area of ​​the chart contained in location information A; ④ The distance between the title and the chart meets a preset threshold; ⑤ The title content contains preset title keywords). Then, the title content is used as an index to initially filter out chart location information unrelated to adverse reaction events (for example, based on preset keywords related to drug listing and drug sales, the title content is traversed to achieve initial filtering of non-target charts). This eliminates the need to extract text from all charts in the drug information data, thereby reducing the workload of information text extraction and improving the efficiency of information text extraction.

[0074] At this point, the information text extracted from the remaining charts may include the following two situations:

[0075] 1) For charts in plain text format, the information text can be accurately obtained by directly parsing the content of the chart source file and combining it with the chart position information;

[0076] 2) For charts in image format, the chart can be located based on its position information, and OCR technology can be used to extract information text from the located chart.

[0077] After obtaining the information text from the drug information data, although charts unrelated to adverse reaction events were initially filtered out based on the chart titles, it cannot be guaranteed that all the information text extracted from the remaining charts is related to adverse reaction events. Therefore, in order to improve the accuracy of subsequent data analysis on adverse drug reactions, the information text can be further classified according to adverse reaction associations to filter adverse reaction data from the drug information data.

[0078] Adverse reaction association classification of information texts can be achieved through machine learning techniques, such as using a trained text classification model, to identify and filter adverse reaction data from drug information data. Understandably, the text classification model here is used to determine whether the information text is associated with adverse reactions.

[0079] To improve the accuracy and efficiency of text classification models, a multilingual BERT large-scale pre-trained model combined with fine-tuning can be used to train the classifier. This model can handle text in multiple languages ​​simultaneously without requiring a custom model for each language.

[0080] Furthermore, the text classification model can be trained through the following steps:

[0081] First, the crawled sample text is preprocessed, including removing stop words, punctuation marks and other useless information, and the sample text is converted into a token sequence for model processing.

[0082] Then, in the data annotation stage, a semi-supervised method is used. Specific documents are pre-filtered using keywords such as "adverse reaction," "adverse event," and "adverse reaction." These documents are then manually verified to identify positive and negative samples. The remaining negative samples are randomly selected from the remaining documents.

[0083] Subsequently, the mBERT model was used as the base model, and fine-tuning was employed to adjust and optimize the model to suit the text classification task. Specifically, a fully connected layer was added on top of the mBERT model, and the cross-entropy loss function was used to measure the model's loss during training, while stochastic gradient descent was used for optimization.

[0084] After multiple rounds of training, a highly efficient and accurate text classification model can be obtained. This model can classify adverse reaction associations in information texts and obtain the adverse reaction association classification results. Based on this, adverse reaction data from information sources can be filtered from drug information data according to the adverse reaction association classification results.

[0085] In addition to adverse reaction data from information sources, adverse reaction data from clinical adverse reaction reports and / or post-marketing adverse reaction databases can also be obtained, thus obtaining adverse reaction data from at least two data sources.

[0086] Clinical adverse reaction reports can be reports provided by the company to research institutions and regulators during clinical trials, including at least SUSAR (Suspected Unexpected Serious Adverse Reaction) reports, DSUR (Development Safety Update Report) reports, and SAE (Serious Adverse Event) reports.

[0087] Post-marketing adverse reaction databases can reflect adverse event information and medication errors of marketed drugs. These databases may include: the FDA Adverse Event Reporting System (FAERS), the European Union Drug Regulating Authorities Pharmacovigilance (EMA Eudra Vigilance) database, and the Japanese Adverse Drug Event Report (JADER) database.

[0088] As the US pharmacovigilance database, FAERS monitors the safety of all marketed drugs and therapeutic biological products. The public can access information through a "dashboard" (which is categorized and summarized based on key fields such as report type, region, patient age, and severity) and "quarterly data files" (which contain raw report data extracted from FAERS within a specified time frame).

[0089] The data elements presented in the "Quarterly Data File" include:

[0090] ① General Information: Case Report ID (ASCII), FAERS Case Report ID, Case Report Version, Pharmaceutical Company Unique Case Report ID, Regulatory Agency Case Report ID, Date Sent to FDA, Pharmaceutical Company First Receipt Date (ASCII), FDA First / Last Receipt Date, Report Type (Urgent, Non-Urgent, Direct), Report Format (Self-Initiated Report, Study Report, Unknown, XML), Whether the Pharmaceutical Company Submitted Electronically (ASCII), Severity (XML), Whether the Rapid Reporting Standards are Met (XML);

[0091] ② Patient information: patient age and age group, patient gender, weight;

[0092] ③ Drug Information: Drug Unique Identifier (ASCII), Drug Name (primary and secondary suspected drugs in ASCII, suspected concomitant drugs in XML), Active Substance Name, Route of Administration, Dosage, Frequency, Cumulative Dosage, Dosage Form, Concomitant Drugs, Start and End Dates of Drug Use, Treatment Duration, Drug Indications in MedDRA Preferred Terms, Drug Use Behavior (Increasing or Decreasing Dosage, etc. in XML), De-excitation Effect (ASCII), Re-excitation Effect (ASCII), Batch Number, Expiry Date (ASCII), New Drug Application Number;

[0093] ④ Event Information: MedDRA preferred term "adverse drug reaction", MedDRA version (XML), date of occurrence (XML), severity, country of occurrence, and outcome (XML);

[0094] ⑤ Report source information: Reporter type, reporter's country, pharmaceutical company / organization code that sent the report, report source (ASCII), whether the pharmaceutical company was notified of the voluntary report (ASCII).

[0095] Understandably, by filtering adverse reaction data from drug information data in step 111 and obtaining adverse reaction data from clinical adverse reaction reports and / or post-marketing adverse reaction databases in step 112, adverse reaction data from at least two data sources can be obtained, which can broaden the monitoring scope of the data analysis system and thus improve the effectiveness and timeliness of data analysis for adverse drug reactions.

[0096] To enable data analysis of adverse reaction data, information can be further extracted from the adverse reaction data obtained in step 110 to obtain target information. This target information can be understood as information related to adverse reaction data analysis, such as drug information, patient information, adverse reaction details, and reporting time within the adverse reaction data.

[0097] The data types of adverse reaction data can include text, images, and audio / video, etc. Different methods can be used to extract information from adverse reaction data for different data types. These information extraction methods can include at least one of entity extraction, entity relation extraction, keyword extraction, speech recognition, and OCR. For example, for text data types, Named Entity Recognition (NER) or keyword extraction can be used to extract target information; for image data types, OCR technology can be used to extract text from the image first, and then target information can be extracted from the text; alternatively, image recognition can be used to directly extract target information from the image; for audio / video data types, target information can be extracted directly from the audio and video data, or the audio data can be transcribed into speech to obtain transcribed text, and then target information can be extracted from the transcribed text; or OCR technology can be applied to each frame of the video data to extract text, and then target information can be extracted from the extracted text. This embodiment of the invention does not specifically limit these methods.

[0098] In some embodiments, based on the data type of the adverse reaction data, information extraction is performed on the adverse reaction data to obtain the target information of the adverse reaction data. Specifically, step 120 includes:

[0099] Step 121: If the adverse reaction data belongs to a data type that includes text, entity extraction is performed on the text in the adverse reaction data based on a pre-trained language model to obtain adverse reaction entities.

[0100] Step 122: Determine target information based on adverse reaction entities and the relationships between adverse reaction entities in the text.

[0101] Specifically, entity extraction, also known as named entity recognition (NER), refers to the automatic identification of named entities from text information. When the adverse reaction data includes text, in order to obtain the target information, entity extraction can be performed on the text in the adverse reaction data based on a pre-trained language model to obtain the adverse reaction entities.

[0102] Entity extraction from text in adverse reaction data can be divided into the following two cases:

[0103] 1) For plain text that does not contain charts, entity extraction can be performed on the text based on a trained NER model or a pre-trained language model to obtain adverse reaction entities.

[0104] 2) For text containing charts, in order to effectively utilize the chart information in the text to enhance the structure of entity extraction and thus improve the completeness and accuracy of entity extraction, a pre-trained language model can be used to extract entities from text containing charts to obtain adverse reaction entities.

[0105] In some embodiments, in step 121, for text containing charts, entity extraction is performed on the text in the adverse reaction data based on a pre-trained language model to obtain adverse reaction entities, specifically including:

[0106] Step 121-1: Obtain the table information from the adverse reaction data, perform entity masking on the table information, and obtain the cloze test template.

[0107] Step 121-2: Based on the cloze test template and the text in the adverse reaction data, determine the target text sequence;

[0108] Step 121-3: Based on the pre-trained language model, entity extraction is performed on the target text sequence to obtain adverse reaction entities;

[0109] The pre-trained language model is obtained by pre-training the initial language model based on entity alignment, a pre-defined entity filling task, and a cloze test task; entity alignment is obtained by aligning the text and related table information in the sample data.

[0110] Specifically, for adverse reaction data where the text contains tables, the trained object detection model described in step 111-1 can be used first to obtain the table information of the tables in the adverse reaction data. This table information reflects the content of the table, which may include the content inside the table and the table title. The table information can be obtained by locating the table based on its position.

[0111] The text to be detected can be input into a trained object detection model, which outputs the chart position information of the text. Then, by analyzing the chart position information and the title position information, a one-to-one binding is established between the chart and the title. After binding, the title content is obtained based on the title position. Finally, the chart positions are filtered as needed using the title content as an index. After obtaining the target chart positions corresponding to the title content, the key charts can be extracted from the text. Here, the key charts are specifically the table information, including the content inside the table and the table title.

[0112] After obtaining the table information from the text, entity masking can be performed on the table information to obtain a cloze test template. The cloze test template can be used to describe the relationships between entities in the table information.

[0113] In some embodiments, step 121-1 involves performing entity masking on the table information to obtain a cloze test template, specifically including:

[0114] Steps 121-11: Obtain the layout information corresponding to the table information;

[0115] Steps 121-12: Based on the layout information, determine the entity relationships between cell entities in the table information; wherein, the entity relationship includes at least one of the following: same column entity relationship, same row entity relationship, and same row and column entity relationship;

[0116] Steps 121-13: Fill the cell entities into the task template corresponding to the entity relationship to obtain the template text;

[0117] Steps 121-14: In the template text, mask the target entities in the cell entities to obtain the cloze test template.

[0118] Specifically, the layout information corresponding to the table information can characterize the layout position of each cell entity in the table. The layout information can specifically include the table's header row cells, header column cells, and the correspondence between other cells and the header row cells and header column cells. Here, "other cells" refers to cells other than the header row cells and header column cells. The layout information can determine whether any two or more entities in the alignment entity have an entity relationship.

[0119] The entity relationships between cells in the table information mainly consider the following two types of relationships:

[0120] 1) In a single column or row, the header cell has an entity relationship with other cells, namely, a same-column entity relationship or a same-row entity relationship. A same-column entity relationship or a same-row entity relationship usually refers to the relationship between two cells. Any non-header row cell in the same row has a same-row entity relationship with the header row cell of that row, and any non-header column cell in the same column has a same-column entity relationship with the header column cell of that column.

[0121] 2) A relationship exists between another cell and its corresponding header row cell and header column cell, i.e., a row-column entity relationship. A row-column entity relationship usually refers to the relationship between three cells.

[0122] For example, the layout information corresponding to the table information can be displayed as shown in Table 1 below. The first row of the table is the header row, and the first column is the header column.

[0123] Table 1

[0124] Median OS 10.7 months 10.5 months 0.975 Median PFS 5.0 months 4.4 months 0.986 …… …… …… ……

[0125] Based on the layout information in Table 1, we can see that there is a same-column entity relationship between cell entities “P-value” and “0.975”, a same-line entity relationship between cell entities “Median PFS” and “4.4months”, and a same-column entity relationship between cell entities “Median OS”, “10.7months”, and “Sorafenib-ravastatin”.

[0126] Once the entity relationships between cell entities are obtained, the cell entities can be filled into the task template corresponding to the entity relationships to obtain the template text.

[0127] The preset task template is text used to describe the entity relationships between cell entities. In some embodiments, the task template may be displayed as shown in Table 2 below:

[0128] Table 2

[0129] 2 Ent1 is associated with Ent2. 3 <![CDATA[ Ent1 and Ent2 may be related to Ent3 .]]>

[0130] In Table 2, there are two types of task templates, corresponding to text describing the relationship between two entities and text describing the relationship between three entities, respectively. Ent1, Ent2, and Ent3 are placeholders for the cell entities that have entity relationships.

[0131] After determining the task template, cell entities can be filled into the task template corresponding to the entity relationship to obtain the template text. For example, two cell entities "P-value" and "P-value" that have an entity relationship in the same column can be filled into the task template corresponding to the number of entities 2, and the resulting template text can be expressed as "P-value is associated with P-value".

[0132] For example, the three cell entities "Median OS", "10.7months" and "Sorafenib-ravastatin" that have a row and column relationship can be filled into the task template corresponding to the number of entities 3. The resulting template text can be expressed as "Median OS and 10.7months may be related to Sorafenib-ravastatin".

[0133] Then, the target entity in the cell is masked within the template text to obtain the cloze test template. Here, the target entity can be a random entity to be filled into the template text. During the masking process, a new special term [SOE] can be introduced to replace the masked entity.

[0134] For example, a target entity "Median OS" can be randomly selected from the template text "Median OS and 10.7 months may be related to Sorafenib-ravastatin", and then masked. The resulting cloze test template can be represented as "[SOE]and 10.7 months may be related to Sorafenib-ravastatin". The resulting cloze test template can be represented as X. C .

[0135] It should be noted that the embodiments of the present invention do not specifically limit the number of template texts included in the cloze test template, for example, it may include 3 or 5 template texts.

[0136] After obtaining the cloze test template, the target text sequence can be determined based on the cloze test template and the text in the adverse reaction data. In one embodiment, the target text sequence can be represented in the following form:

[0137] X = [CLS]X C [SEP]X E [SEP]

[0138] Where X is the target text sequence; X C It is a cloze test template; X E[CLS] and [SEP] are text in the adverse reaction data; [CLS] and [SEP] are special markers.

[0139] Subsequently, based on the pre-trained language model, entity extraction can be performed on the target text sequence to obtain adverse reaction entities. Specifically, the target text sequence can be input into the pre-trained language model to obtain the adverse reaction entities output by the pre-trained language model.

[0140] The pre-trained language model can be obtained through the following steps:

[0141] 1. Obtain sample documents. Sample documents include text information and associated table information. Only documents containing both text information and associated table information can be used as sample documents. Sample documents can be obtained from databases containing a large amount of relevant documents in medicine, biology, health science, or nursing, such as PubMed and PubTab.

[0142] 2. After obtaining the text and table information from the sample document, entity alignment can be performed on the text and table information. Entity alignment can be used to determine whether two or more entities from different information sources point to the same object in the real world, and to group named entities with the same referent together to obtain aligned entities.

[0143] Understandably, the alignment entity originates from both textual and tabular information and points to the same object in the real world.

[0144] Entity alignment can be achieved through string matching, which involves matching the cell content in the table information with the terms in the text information to obtain the aligned entities. String matching methods can include the following two:

[0145] 1) Taking English as an example, retrieve the table and its cell contents, along with related text information. Convert each word in the cell (except stop words and punctuation marks) to the same root word, and attempt to find the corresponding position of each word in the text information. Converting to the same root word here is to improve matching efficiency. For example, if the word in the cell is plural and the corresponding word in the text information is singular, root word conversion can further improve matching efficiency and accuracy.

[0146] 2) Enumerate all possible phrases in the text information and compare each possible phrase with the cell content. Based on the overlap ratio of words appearing in the cell and phrase, score each phrase to obtain a score for each phrase in the text information, and retain the phrase with the highest score as the alignment entity. For example, if the cell content is "Sequenced genomefalciparum" and the phrase in the text information is "Sequenced genome plasmodium falciparum", its corresponding score is 0.75. Furthermore, a score threshold can be set to identify phrases with scores higher than the preset threshold as successfully matched alignment entities; for example, the score threshold could be 0.5.

[0147] 3. Based on entity alignment, pre-defined entity filling tasks, and cloze test tasks, the initial language model is pre-trained to obtain a pre-trained language model.

[0148] Two self-supervised tasks were designed during pre-training: Entity Infilling (EI) and Table Cloze Test (TCT).

[0149] Compared to existing masked language models, the entity completion task (EI) masks aligned entities in text and requires the model to recover them. Meanwhile, the cloze test (TCT) converts multiple aligned entities in a table layout into a text segment with missing entities, requiring the model to extract the correct entities from the text to fill in the gaps. Through these two tasks, knowledge of tables is effectively integrated into the language model.

[0150] The pre-trained language model obtained in this way can make full use of entity information in the table information related to text information, so as to achieve good processing accuracy in downstream NLP tasks (such as information extraction, relation extraction, classification, etc.).

[0151] A pre-trained language model may include an input layer, a semantic information extraction layer, a prediction layer, and a pointer layer. The semantic information extraction layer may be a Bidirectional Long Short-Term Memory (BiLSTM) network, the prediction layer may be a Conditional Random Field (CRF) network, and the pointer layer may be an attention-based pointer network. Both the prediction layer and the pointer layer are connected to the semantic information extraction layer, and the prediction layer and the pointer layer operate in parallel.

[0152] Using a pre-trained language model, entity extraction is performed on the target text sequence to obtain adverse reaction entities. This can specifically include the following steps:

[0153] 1) Input the target text sequence into the input layer, and use the input layer to encode the target text sequence into a semantic vector sequence;

[0154] 2) Input the encoded semantic vector sequence into the semantic information extraction layer, and output the semantic information of each word in the semantic vector sequence in the surrounding text;

[0155] 3) Input the semantic information of each word in the semantic vector sequence into the prediction layer, and solve complex situations such as entity intersection or nesting through global optimization to obtain the first entity output by the prediction layer based on the entity filling task;

[0156] 4) Input the semantic information of each word in the semantic vector sequence into the pointer layer, find the corresponding position of the missing entity in the text information for the cloze template, and obtain the second entity output by the pointer layer based on the cloze task;

[0157] 5) Based on this, the first entity and the second entity can be merged to obtain the adverse reaction entity. If the pointer layer can find the corresponding entity in the text but the prediction layer does not predict it, the second entity output by the pointer layer will also be used as the final output entity. Experiments show that the entity correction method based on Pointer Network can effectively improve the prediction performance of CRF.

[0158] Having obtained the adverse reaction entities, and considering that a patient may receive multiple medications in a clinical trial, resulting in multiple adverse reaction events, it is also necessary to extract the correspondences between these adverse reaction entities. A relation extraction model can be built on a pre-trained language model to further extract the relationships between adverse reaction entities in the text.

[0159] Alternatively, the relationships between adverse reaction entities in the text can be obtained through the following steps:

[0160] Encode the target text sequence and each entity to obtain a semantic vector sequence;

[0161] Based on the semantic vector sequence, the relationships between entities are extracted.

[0162] Specifically, the target text sequence and the adverse reaction entities obtained from entity extraction are first input into the model. A pre-trained language model is then used to encode the target text sequence into a semantic vector sequence. The adverse reaction entities can be added to the input text as special segments, similar to the PURE model approach.

[0163] Subsequently, a relation classification model is constructed on the segments corresponding to the adverse reaction entities, and relations are extracted for each adverse reaction entity to determine whether a specific relationship exists between the given entities.

[0164] The resulting target information may include adverse reaction entities and the relationships between them. For example, adverse reaction entities include, but are not limited to: ① drug information (drug name, dosage, administration method, route of administration), ② suitability, ③ adverse reaction / event information, ④ SMQ of the adverse reaction, ⑤ report type / severity / causation, ⑥ patient information (age, gender, age at the time of adverse reaction / event), ⑦ test results, ⑧ treatment details, ⑨ ICSR number, and ⑩ report submission time.

[0165] In addition, the correspondence between the entities “ICSR number-reporting time-indication-drug information-adverse reaction-adverse reaction SMQ-report type / severity / causation-age” can be obtained.

[0166] It should be noted that, considering that the ICSR number is the same for the same patient, if the same patient submits multiple SUSAR reports, only the version number will change. Therefore, the ICSR number plus the version number can be used as the unique code for the SUSAR report.

[0167] In some embodiments, after obtaining the target information, the target information can also be standardized, specifically including:

[0168] 1) Match the drug name, company name, and indication in the target information with the pre-built standard drug dictionary, company dictionary, and indication dictionary to obtain the standard name.

[0169] 2) Standardize the sources of correlation assessment between drugs and reactions / events in the target information, and the description of the correlation assessment results between drugs and reactions / events.

[0170] The assessment sources include companies or researchers, which typically refer to the sources of adverse reaction data. For example, companies in the assessment sources may be described as "Company," "Sponsor," "Pharmaceutical Company," or "AMH" in different reports; researchers in the assessment sources may be described as "Reporter," "Investigator's Judgment," or "Primary Source Reporter" in different reports.

[0171] The assessment results (causal relationship) include: definitely related, very likely related, possibly related, uncertain / to be evaluated or cannot be evaluated, possibly unrelated, unrelated, etc.

[0172] The evaluation results are described as follows in different reports: "definitely", "certainly", "Related", "related", "relevant", "definite", "strong correlation", etc.; "very likely to be related" is described as "very likely", "high probability", "perfect correlation"; "possibly related" is described as "possibly", "certain probability", "association", "relational", "Moderate correlation"; "uncertain / to be evaluated" or "cannot be evaluated" is described as "cannot be evaluated" or "uncertain"; "possibly irrelevant" is described as "possibly irrelevant" or "weak correlation"; and "irrelevant" is described as "irrelevant", "irrelevant", or "no correlation".

[0173] 3) Link the acquired indication information with the MedDRA dictionary (medical-dictionary of-regulatory-activities). The MedDRA dictionary is an international medical terminology set for clinical validation used by regulatory agencies and the regulated biopharmaceutical industry, and is used for data entry, retrieval, evaluation, and presentation throughout the pre-market and post-marketing regulatory process.

[0174] The MedDRA dictionary uses a five-level structure to categorize and associate terms: 1. SOC (System Organ Class); 2. HLGT (High Level Group Term); 3. HLT (High Level Term); 4. PT (Preferred Term); 5. LLT (Lowest Level Term). PT and SOC are commonly used dimensions in data analysis. In SUSAR reports, LLT codes are usually annotated next to indications. Based on the hierarchical correspondence in the dictionary, PT and SOC codes and their corresponding terms can be obtained.

[0175] 4) Integrate the acquired adverse event information with the MedDRA dictionary and the SMQ dictionary. The SMQ dictionary is a standard MedDRA analysis query that refers to a set of terms related to a specific medical condition or area of ​​concern, from one or more SOCs, designed to assist in case identification.

[0176] The logic of integrating adverse event information with the MedDRA dictionary is the same as that of the indication information. Adverse event information is also integrated with the SMQ dictionary. In the SUSAR report, each adverse event is marked with an LLT code. Based on the MedDRA dictionary, the PT code can be obtained through the LLT code. Furthermore, based on the existing logic of the SMQ dictionary, the corresponding SMQ information can be obtained through the LLT code or the PT code.

[0177] Based on this, the adverse reaction data analysis system can be used to analyze the adverse reaction data under the target dimension according to the target information obtained in step 120, and the analysis results can be obtained.

[0178] With the accumulation of adverse reaction reports, dynamic and multi-dimensional analysis has become a major demand scenario, in addition to reviewing individual reports. Currently, achieving this type of analysis requires a significant amount of manual time to collect and summarize reports, followed by unified data cleaning of the master table. This method consumes substantial manpower and time. However, an adverse reaction data analysis system can quickly meet this need by performing adverse reaction data analysis on target information within the target dimensions. Such a system can provide a one-stop solution to the problems of existing adverse reaction data analysis methods failing to meet users' high-precision analysis needs for adverse drug reactions and exhibiting low analysis efficiency.

[0179] In some embodiments, the adverse reaction data analysis system includes at least one of a time analysis unit, a drug analysis unit, and an enterprise analysis unit. The adverse reaction data analysis system performs adverse reaction data analysis on the target information under the target dimension to obtain the analysis results. Specifically, step 130 includes:

[0180] Step 131: Determine the target dimension, which includes at least one of the following: time dimension, drug dimension, and enterprise dimension;

[0181] Step 132: Extract the information to be analyzed that is related to the target dimension from the target information;

[0182] Step 133: Through at least one of the time analysis unit, drug analysis unit, and enterprise analysis unit, adverse reaction data analysis is performed on the information to be analyzed under the target dimension to obtain the analysis results.

[0183] Specifically, the target dimension can be any single dimension among time dimension, drug dimension, and enterprise dimension; it can also include at least two dimensions, for example, the target dimension can include any two dimensions among time dimension, drug dimension, and enterprise dimension; it can also include all three dimensions at the same time, which can be flexibly selected according to the needs of the scenario. This embodiment of the invention does not make specific limitations in this regard.

[0184] Once the target dimensions are determined, the information to be analyzed associated with those target dimensions can be extracted from the target information. For example, when the target dimensions include time and drug dimensions, the information to be analyzed for any target drug within any target time period can be extracted from the target information; when the target dimensions include time, drug, and company dimensions, the information to be analyzed for any target drug under any target company within any target time period can be extracted from the target information.

[0185] In some embodiments, extracting the information to be analyzed that is associated with the target dimension from the target information, i.e., step 132 specifically includes:

[0186] Step 132-1: Obtain the target object, which is based on at least one description from the following: patient information, drug information, enterprise information, indication information, reporting time, and medical terminology information.

[0187] Step 132-2: Extract the information to be analyzed from the target information that is related to both the target dimension and the target object.

[0188] Specifically, it is possible not only to extract information to be analyzed from the target dimension, but also to extract information related to both the target dimension and the target object simultaneously. The target object here can be based on at least one description from patient information, drug information, company information, indication information, reporting time, and medical terminology information. Medical terminology information may include, for example, SOC (Systemic Organ Classification), HLT (Higher Level Terminology), PT (Preferred Terminology), SMQ, etc.

[0189] For the extraction of information to be analyzed that is associated with both the target dimension and the target object, the information to be analyzed associated with the target dimension can be extracted first, and then the information to be analyzed associated with the target object can be extracted; alternatively, the information to be analyzed associated with the target object can be extracted first, and then the information to be analyzed associated with the target dimension can be extracted; of course, the information to be analyzed that is associated with both the target dimension and the target object can be extracted simultaneously. This embodiment of the invention does not specifically limit this.

[0190] For example, when the target dimension includes a time dimension, the first target information can be extracted from the target information based on at least one of patient information, drug information, company information, indication information, and medical terminology information; then, the first target information within any time period can be extracted from the first target information, thereby obtaining the information to be analyzed that is associated with both the target dimension and the target object.

[0191] After obtaining the information to be analyzed, adverse reaction data analysis under the target dimension can be performed through at least one of the time analysis unit, drug analysis unit, and enterprise analysis unit to obtain the analysis results.

[0192] In some embodiments, at least one of the time analysis unit, drug analysis unit, and enterprise analysis unit is used to perform adverse reaction data analysis on the target dimension of the information to be analyzed, and the analysis results are obtained. Specifically, step 133 includes:

[0193] Step 133-1: When the target dimension is time, the time analysis unit is used to analyze the type and / or frequency of adverse events corresponding to the target time in the information to be analyzed, and the analysis results are obtained.

[0194] Step 133-2: When the target dimension is the drug dimension, the drug analysis unit is used to analyze the type and / or frequency of adverse reaction events corresponding to the target drug in the information to be analyzed, and the analysis results are obtained.

[0195] Step 133-3: When the target dimension is the enterprise dimension, the target drug associated with the target enterprise in the information to be analyzed, as well as the type and / or frequency of adverse reaction events associated with the target drug, are analyzed through the enterprise analysis unit to obtain the analysis results.

[0196] Specifically, the time analysis unit in the adverse reaction data analysis system can be used to perform time-dimensional adverse reaction data analysis on the information to be analyzed. When the target dimension is time, the system can report the total number of SUSARs, the total number of drugs involved, the type of adverse reaction event, and / or the number of occurrences within any given time frame. Furthermore, the data can be further specified based on any or a combination of dimensions such as patient information, drug name, company name, and indication information. The target time in the information to be analyzed can be entered by the user through the client of the adverse reaction data analysis system.

[0197] For example, it is possible to count the types of adverse events during the target period from June 7, 2021 to April 7, 2023. These types may include death, life-threatening events, hospitalization / prolonged hospitalization, permanent / significant loss of function, teratogenicity / birth defects, other important medical events, and the number of occurrences of each type of adverse event. It is also possible to count the number of occurrences of each type of adverse event per year during this target period.

[0198] Furthermore, the statistical results can be analyzed, and the results can be visualized using tables, time stacked charts, pie charts, etc.

[0199] The drug analysis unit in the adverse reaction data analysis system can be used to perform adverse reaction data analysis on a drug-level basis on the information to be analyzed. When the target dimension is the drug dimension, the total number of SUSARs reported for any target drug, the type of adverse event, and / or the number of occurrences can be obtained at the drug dimension. Furthermore, the data can be further restricted based on any or a combination of dimensions such as patient information, indication information, and reporting time.

[0200] For example, it is possible to obtain the types and frequency of adverse events reported for the target drug "fluzoparib capsules".

[0201] In some embodiments, the drug analysis unit analyzes the type and / or frequency of adverse reaction events corresponding to the target drug in the information to be analyzed to obtain the analysis results. Specifically, step 133-2 includes:

[0202] Based on the causal relationship between each target drug and each adverse reaction event in the information to be analyzed, the type and / or frequency of adverse reaction events corresponding to each target drug are statistically analyzed from the information to be analyzed.

[0203] The drug analysis unit compares the types and / or frequency of adverse events corresponding to each target drug, and determines the analysis results based on the comparison results.

[0204] Specifically, based on the relationship between drug, adverse event, assessment source, and assessment result, the causal relationship between each target drug and each adverse reaction event can be obtained. This allows for the statistical analysis of the types and / or frequency of adverse reactions corresponding to each target drug from the information to be analyzed. For example, the types and / or frequency of adverse reactions corresponding to target drug A, target drug B, target drug C, and so on can be obtained.

[0205] Based on this, the types and / or frequency of adverse events corresponding to each target drug are compared, and the analysis results are determined based on the comparison results.

[0206] In practical applications, the drug analysis unit in the adverse reaction data analysis system can provide an entry point for "Drug Analysis". After entering the "Drug Analysis" page, click the "Drug Comparison" button to enter the "Drug Comparison" page. The drugs entered on the interface will be automatically entered into the drug search box by default. You can also enter the drugs to be compared in the drug search box and enter suggestions, such as comparing five drugs.

[0207] Furthermore, it can limit the comparison of SUSAR events for drugs across different indications, reporting times, genders, and age ranges.

[0208] The resulting comparison results may include drug name, total number of SUSARs, number of clinical trials, number of events at each risk level, PT type and its number, and SMQ type and its number.

[0209] Subsequently, the analysis results can be determined based on the comparison results, and the analysis results can be visualized using tables, time stacked charts, pie charts, etc.

[0210] The enterprise analysis unit in the adverse reaction data analysis system can be used to perform enterprise-level adverse reaction data analysis on the information to be analyzed. When the target dimension is enterprise-level, the analysis results are determined based on the target drugs associated with the target enterprise in the information to be analyzed, as well as the types and / or frequency of adverse reaction events associated with the target drugs.

[0211] For example, at the enterprise level, the total number of SUSARs reported by any target enterprise, the type and number of adverse events, and the name and number of drugs corresponding to the adverse events can be obtained. Furthermore, the data can be further restricted based on patient information, drug name, indication information, reporting time, SOC, HLT, PT, SMQ, or any combination of these dimensions.

[0212] In practical applications, the analysis results may include the proportion and number of drugs reported by the target company for SUSAR under different times, genders, ages, countries, drugs, indications, and adverse reactions; it may also include the types of adverse events (PTs) reported by the target company for different times and different drugs.

[0213] For example, the left side of the page can be used for search terms, such as "time" (default is all time), "drug" (default is the drug with the most reports), and "PT type". The right side of the page displays the PT type (pie chart) and the reporting time - reporting quantity (line chart) corresponding to the target drug.

[0214] The adverse reaction data analysis method provided in this invention uses an adverse reaction data analysis system to perform structured analysis on drug-related adverse reaction data, obtain target information of the adverse reaction data, broaden the monitoring scope based on data analysis, and improve the effectiveness and timeliness of data analysis for adverse drug reactions.

[0215] Furthermore, by matching the target information with MedDRA and SMQ dictionaries, users can quickly perform multi-dimensional, systematic, and visual combined analysis of adverse reaction data (such as the original SUSAR report) when they only need to provide the adverse reaction data. This satisfies users' retrieval and statistical needs for adverse reactions / events, thus solving the problem of low efficiency in existing adverse reaction data analysis in one stop.

[0216] Based on the above embodiments, an adverse reaction data analysis method is provided, including:

[0217] S1, obtain drug information data.

[0218] S2, based on the object detection model, performs chart detection on the drug information data. The object detection model is trained on a sample data set with pre-labeled chart areas covering the title area. If the chart position information in the drug information data is detected, the information text is extracted from the charts in the drug information data based on the chart position information.

[0219] S3, perform adverse reaction association classification on the information text, and filter adverse reaction data from the information source from the drug information data based on the adverse reaction association classification results;

[0220] In addition, adverse reaction data from sources of clinical adverse reaction reports and / or post-marketing adverse reaction databases.

[0221] S4. When the adverse reaction data belongs to a data type that includes text, entity extraction is performed on the text in the adverse reaction data based on a pre-trained language model to obtain adverse reaction entities; based on the adverse reaction entities and the relationships between adverse reaction entities in the text, target information is determined.

[0222] In S4, based on a pre-trained language model, entity extraction is performed on the text in the adverse reaction data to obtain adverse reaction entities, specifically including:

[0223] S41, Obtain the table information from the adverse reaction data, perform entity masking on the table information to obtain the cloze test template; Based on the cloze test template and the text in the adverse reaction data, determine the target text sequence; Based on the pre-trained language model, extract entities from the target text sequence to obtain adverse reaction entities; The pre-trained language model is obtained by pre-training the initial language model based on aligned entities, a preset entity filling task, and a cloze test task; Aligned entities are obtained by aligning the text and related table information in the sample data.

[0224] S5. Determine the target dimension, which includes at least one of the following dimensions: time dimension, drug dimension, and enterprise dimension; Obtain the target object, which is based on at least one of the following descriptions: patient information, drug information, enterprise information, indication information, reporting time, and medical terminology information.

[0225] Extract the information to be analyzed from the target information that is related to both the target dimension and the target object;

[0226] When the target dimension is time, the time analysis unit in the adverse reaction data analysis system is used to analyze the type and / or frequency of adverse reaction events corresponding to the target time in the information to be analyzed, and the analysis results are obtained.

[0227] When the target dimension is the drug dimension, based on the causal relationship between each target drug and each adverse reaction event in the information to be analyzed, the type and / or frequency of adverse reaction events corresponding to each target drug are statistically determined from the information to be analyzed; through the drug analysis unit in the adverse reaction data analysis system, the type and / or frequency of adverse reaction events corresponding to each target drug are compared, and the analysis results are determined based on the comparison results.

[0228] When the target dimension is the enterprise dimension, the enterprise analysis unit in the adverse reaction data analysis system is used to analyze the type and / or frequency of adverse reaction events corresponding to the target drug in the information to be analyzed, and the analysis results are obtained.

[0229] The adverse reaction data analysis device provided by the present invention is described below. The adverse reaction data analysis device described below and the adverse reaction data analysis method described above can be referred to in correspondence.

[0230] Figure 2 This is a schematic diagram of the adverse reaction data analysis device provided by the present invention, as shown below. Figure 2 As shown, the adverse reaction data analysis device provided in this embodiment of the invention includes:

[0231] Acquisition unit 210 is used to acquire adverse reaction data related to drugs;

[0232] Extraction unit 220 is used to extract information from the adverse reaction data based on the data type to which the adverse reaction data belongs, and obtain the target information of the adverse reaction data;

[0233] The analysis unit 230 is used to perform adverse reaction data analysis on the target information under the target dimension through the adverse reaction data analysis system, and obtain the analysis results.

[0234] The adverse reaction data analysis device provided in this invention acquires drug-related adverse reaction data, providing underlying data support for drug development throughout its entire lifecycle for subsequent adverse reaction data analysis; and through the adverse reaction data analysis system, it performs adverse reaction data analysis on the target information extracted from the information at the target dimension, enabling rapid multi-dimensional and systematic combined analysis of adverse reaction data, meeting users' high-precision analysis needs for drug adverse reactions, and thus solving the problem of low efficiency in existing adverse reaction data analysis in one stop.

[0235] Based on the above embodiments, the acquisition unit 210 is specifically used for:

[0236] Obtain drug information data and extract information text from the drug information data to perform adverse reaction association classification on the information text, and then filter adverse reaction data from the information source based on the adverse reaction association classification results;

[0237] In addition, adverse reaction data from sources of clinical adverse reaction reports and / or post-marketing adverse reaction databases.

[0238] Based on the above embodiments, the acquisition unit 210 is further specifically used for:

[0239] Based on the object detection model, the drug information data is subjected to chart detection. The object detection model is trained on a set of pre-labeled sample data where the chart area covers the title area.

[0240] If the chart position information in the drug information data is detected, the information text is extracted from the chart in the drug information data based on the chart position information.

[0241] Based on the above embodiments, the extraction unit 220 is specifically used for:

[0242] When the adverse reaction data belongs to a data type that includes text, entity extraction is performed on the text in the adverse reaction data based on a pre-trained language model to obtain adverse reaction entities.

[0243] The target information is determined based on the adverse reaction entities and the relationships between them in the text.

[0244] Based on the above embodiments, the extraction unit 220 is further specifically used for:

[0245] Obtain the table information from the adverse reaction data, perform entity masking on the table information, and obtain a cloze test template;

[0246] Based on the cloze test template and the text in the adverse reaction data, determine the target text sequence;

[0247] Based on a pre-trained language model, entity extraction is performed on the target text sequence to obtain the adverse reaction entity;

[0248] The pre-trained language model is obtained by pre-training the initial language model based on entity alignment, a preset entity filling task, and a cloze test task; the entity alignment is obtained by aligning the text and related table information in the sample data.

[0249] Based on the above embodiments, the extraction unit 220 is further specifically used for:

[0250] Obtain the layout information corresponding to the table information;

[0251] Based on the layout information, determine the entity relationships between cell entities in the table information; wherein, the entity relationships include at least one of the following: same-column entity relationship, same-row entity relationship, and same-row-column entity relationship;

[0252] The cell entity is filled into the task template corresponding to the entity relationship to obtain the template text;

[0253] In the template text, the target entity in the cell entity is masked to obtain the cloze test template.

[0254] Based on the above embodiments, the adverse reaction data analysis system includes at least one of a time analysis unit, a drug analysis unit, and an enterprise analysis unit. The analysis unit 230 is specifically used for:

[0255] Determine the target dimension, which includes at least one of the following: time dimension, drug dimension, and enterprise dimension;

[0256] Extract the information to be analyzed that is associated with the target dimension from the target information;

[0257] By using at least one of the time analysis unit, the drug analysis unit, and the enterprise analysis unit, adverse reaction data analysis is performed on the information to be analyzed under the target dimension to obtain the analysis results.

[0258] Based on the above embodiments, the analysis unit 230 is further specifically used for:

[0259] Obtain the target object, which is based on at least one description of patient information, drug information, enterprise information, indication information, reporting time, and medical terminology information;

[0260] Extract the information to be analyzed from the target information that is associated with both the target dimension and the target object.

[0261] Based on the above embodiments, the analysis unit 230 is further specifically used for:

[0262] When the target dimension is the time dimension, the time analysis unit analyzes the type and / or frequency of adverse reaction events corresponding to the target time in the information to be analyzed, and obtains the analysis results.

[0263] When the target dimension is the drug dimension, the drug analysis unit analyzes the type and / or frequency of adverse reaction events corresponding to the target drug in the information to be analyzed to obtain the analysis results;

[0264] When the target dimension is the enterprise dimension, the enterprise analysis unit analyzes the target drugs associated with the target enterprise in the information to be analyzed, as well as the types and / or frequency of adverse reaction events associated with the target drugs, to obtain the analysis results.

[0265] Based on the above embodiments, the analysis unit 230 is further specifically used for:

[0266] Based on the causal relationship between each target drug and each adverse reaction event in the information to be analyzed, the type and / or frequency of adverse reaction events corresponding to each target drug are statistically analyzed from the information to be analyzed.

[0267] The drug analysis unit compares the types and / or frequency of adverse events corresponding to each target drug, and determines the analysis results based on the comparison results.

[0268] Figure 3 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 3 As shown, the electronic device may include a processor 310, a communications interface 320, a memory 330, and a communication bus 340, wherein the processor 310, communications interface 320, and memory 330 communicate with each other via the communication bus 340. The processor 310 can call logical instructions in the memory 330 to execute a data analysis method, which includes: acquiring adverse reaction data related to the drug; extracting information from the adverse reaction data based on its data type to obtain target information; and performing adverse reaction data analysis on the target information under a target dimension using an adverse reaction data analysis system to obtain analysis results.

[0269] Furthermore, the logical instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0270] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the adverse reaction data analysis method provided by the above methods. The method includes: acquiring adverse reaction data related to drugs; extracting information from the adverse reaction data based on the data type to which the adverse reaction data belongs, to obtain target information of the adverse reaction data; and performing adverse reaction data analysis on the target information under the target dimension through an adverse reaction data analysis system to obtain analysis results.

[0271] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements an adverse reaction data analysis method provided by the methods described above. This method includes: acquiring adverse reaction data related to a drug; extracting information from the adverse reaction data based on the data type to which the adverse reaction data belongs, to obtain target information of the adverse reaction data; and performing adverse reaction data analysis on the target information under a target dimension using an adverse reaction data analysis system to obtain analysis results.

[0272] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0273] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0274] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for analyzing adverse reaction data, characterized in that, include: Obtain data on adverse drug reactions; Based on the data type of the adverse reaction data, information is extracted from the adverse reaction data to obtain the target information of the adverse reaction data; The adverse reaction data analysis system is used to perform adverse reaction data analysis on the target information under the target dimension to obtain the analysis results. The acquisition of drug-related adverse reaction data includes: Acquire drug information data and extract information text from the drug information data to perform adverse reaction association classification on the information text, and then filter adverse reaction data from the drug information data based on the adverse reaction association classification results; and acquire adverse reaction data from clinical adverse reaction reports and / or post-marketing adverse reaction databases. The extraction of information text from the drug information data includes: Based on the object detection model, chart detection is performed on drug information data to obtain chart position information; the chart position information includes the position information of the chart area and the position information of the title area; the object detection model is trained on a sample document set that is pre-labeled with chart position and title position, and the pre-labeled chart position is bound to the corresponding title position; Based on the spatial relationship between the chart area and the title area, and determining that there is a correspondence between the chart area and the title area, the title content corresponding to the title area is used as an index to initially filter out charts that are unrelated to adverse reaction events; information text is then extracted from the remaining charts after the initial filtering.

2. The adverse reaction data analysis method according to claim 1, characterized in that, The adverse reaction data analysis system includes at least one of a time analysis unit, a drug analysis unit, and an enterprise analysis unit. The step of using an adverse reaction data analysis system to perform adverse reaction data analysis on the target information under the target dimension to obtain analysis results includes: Determine the target dimension, which includes at least one of the following: time dimension, drug dimension, and enterprise dimension; Extract the information to be analyzed that is associated with the target dimension from the target information; By using at least one of the time analysis unit, the drug analysis unit, and the enterprise analysis unit, adverse reaction data analysis is performed on the information to be analyzed under the target dimension to obtain the analysis results.

3. The adverse reaction data analysis method according to claim 2, characterized in that, The step of extracting the information to be analyzed that is associated with the target dimension from the target information includes: Obtain the target object, which is based on at least one description of patient information, drug information, enterprise information, indication information, reporting time, and medical terminology information; Extract the information to be analyzed from the target information that is associated with both the target dimension and the target object.

4. The adverse reaction data analysis method according to claim 2, characterized in that, The method involves performing adverse reaction data analysis on the information to be analyzed under a target dimension using at least one of the time analysis unit, the drug analysis unit, and the enterprise analysis unit to obtain analysis results, including: When the target dimension is the time dimension, the time analysis unit analyzes the type and / or frequency of adverse reaction events corresponding to the target time in the information to be analyzed, and obtains the analysis results. When the target dimension is the drug dimension, the drug analysis unit analyzes the type and / or frequency of adverse reaction events corresponding to the target drug in the information to be analyzed to obtain the analysis results; When the target dimension is the enterprise dimension, the enterprise analysis unit analyzes the target drugs associated with the target enterprise in the information to be analyzed, as well as the types and / or frequency of adverse reaction events associated with the target drugs, to obtain the analysis results.

5. The adverse reaction data analysis method according to claim 4, characterized in that, The process involves using the drug analysis unit to analyze the type and / or frequency of adverse reaction events corresponding to the target drug in the information to be analyzed, and obtaining the analysis results, including: Based on the causal relationship between each target drug and each adverse reaction event in the information to be analyzed, the type and / or frequency of adverse reaction events corresponding to each target drug are statistically analyzed from the information to be analyzed. The drug analysis unit compares the types and / or frequency of adverse events corresponding to each target drug, and determines the analysis results based on the comparison results.

6. The adverse reaction data analysis method according to any one of claims 1 to 5, characterized in that, The step of extracting information from the adverse reaction data based on its data type to obtain target information includes: When the adverse reaction data belongs to a data type that includes text, entity extraction is performed on the text in the adverse reaction data based on a pre-trained language model to obtain adverse reaction entities. The target information is determined based on the adverse reaction entities and the relationships between them in the text.

7. An adverse reaction data analysis device, characterized in that, include: Acquisition unit, used to acquire adverse reaction data related to drugs; The extraction unit is used to extract information from the adverse reaction data based on the data type to which the adverse reaction data belongs, and to obtain the target information of the adverse reaction data; The analysis unit is used to perform adverse reaction data analysis on the target information under the target dimension through the adverse reaction data analysis system, and obtain the analysis results; The acquisition unit is specifically used for: acquiring drug information data, extracting information text from the drug information data, classifying the information text for adverse reaction association, and then, based on the adverse reaction association classification results, filtering adverse reaction data from the drug information data for information sources; and acquiring adverse reaction data from clinical adverse reaction reports and / or post-marketing adverse reaction databases. The acquisition unit is further configured to: perform chart detection on the drug information data based on the target detection model to obtain chart position information; the chart position information includes the position information of the chart area and the position information of the title area; The target detection model is trained on a set of sample documents with pre-labeled chart and title positions, and the pre-labeled chart positions are bound to the corresponding title positions. Based on the spatial relationship between the chart area and the title area, and after determining that there is a correspondence between the chart area and the title area, the title content corresponding to the title area is used as an index to initially filter out charts that are not related to adverse reaction events. Information text is extracted from the remaining charts after the initial filtering.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the adverse reaction data analysis method as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the adverse reaction data analysis method as described in any one of claims 1 to 6.

Images