Medical instrument adverse event identification method and system based on real world data

Abstract

The invention discloses a medical instrument adverse event recognition method and system based on real world data, and relates to the technical field of monitoring analysis, and the method comprises the steps: data input and paragraph analysis, obtaining of medical text data input by a target user, preprocessing of the medical text data, and obtaining of medical instrument adverse events. Performing text structure and paragraph analysis on the processed medical text data, and identifying text content data corresponding to different stages; carrying out adverse event entity identification, carrying out entity identification on the analyzed text content data, and confirming medical entity data corresponding to a target user; adverse event extraction: extracting perioperative period adverse events according to the medical entity data corresponding to the target user; and similarity matching and filtering: matching the perioperative period adverse events with preset dictionary set entries so as to confirm a similarity evaluation coefficient, comparing the similarity evaluation coefficient with a preset similarity evaluation threshold, and outputting an identification result based on a comparison result. The method has the effect of improving recognition.

Description

Technical Field

[0001] The present application relates to the field of monitoring and analysis technology, and in particular to a method and system for identifying adverse drug and device events based on real-world data. Background Art

[0002] With the rapid development of artificial intelligence technology, natural language processing (NLP) technology has been increasingly used in various fields, especially in the medical field, showing great potential. Medical data, especially electronic health records (EHR), contains a large amount of patient information and clinical records. These data are of great significance for medical research, clinical decision support and medical quality management. Real World Data (RWD), as an important part of medical data, not only covers electronic medical records, but also includes multi-source heterogeneous data such as surgical records, nursing logs, imaging reports, and patient follow-up records. Moreover, RWD has the advantages of covering a wide population, a long time span, and being able to reflect real clinical practice. It is suitable for dynamic monitoring and long-term tracking of adverse events. However, medical text data in real-world environments are usually unstructured and contain a large number of medical terms, abbreviations, and complex semantic relationships, which makes it extremely challenging to extract valuable information from these data. In perioperative scenarios, the complexity of real-world data is particularly prominent: the description of surgical records and postoperative care texts is significantly different, and the same event may be scattered in data of different dimensions such as medical records, anesthesia reports, and nursing assessments. In addition, there are large differences in data structure and terminology systems among different medical institutions. In addition, real-world data generally have problems such as information fragmentation, non-standardization, and complex dynamic correlations in time series, which further increase the difficulty of identifying adverse events.

[0003] In recent years, pre-trained language models (such as BERT, ERNIE, GPT, etc.) have made breakthrough progress in natural language processing tasks. These models can capture rich language features and semantic relationships through unsupervised learning on large-scale text data, thus performing well in tasks such as text classification, named entity recognition, and relationship extraction. In the medical field, pre-trained language models are also widely used in the analysis and processing of electronic medical records, including disease diagnosis, drug recommendations, and adverse reaction monitoring.

[0004] In medical quality management, the monitoring and management of perioperative adverse events is a key link. Perioperative adverse events refer to adverse events that occur before and after surgery, which may affect the patient's recovery or even endanger his life. Therefore, automated identification technology based on real-world data can continuously analyze massive clinical texts to discover potential adverse events that are difficult to detect by manual review, providing data-driven decision support for quality improvement.

[0005] In related technologies, the monitoring of perioperative adverse events mainly relies on manual review of electronic medical records, which results in manual review requiring a large amount of manpower and time, and is inefficient. Manual review is also easily affected by subjective factors, resulting in missed detections and false detections. Different reviewers may have different judgment criteria for adverse events, which affects the reliability of the results and reduces the recognition efficiency. There is room for improvement. Summary of the invention

[0006] In response to the deficiencies in the prior art, the present application provides a method and system for identifying drug-device adverse events based on real-world data.

[0007] In a first aspect, the present application provides a method for identifying adverse drug-device events based on real-world data, comprising the following steps: Data input and paragraph analysis: obtaining medical text data input by the target user, preprocessing the medical text data, and performing text structure and paragraph analysis on the processed medical text data, thereby identifying text content data corresponding to different stages; Adverse event entity recognition: using the UIE model to perform entity recognition on the parsed text content data, and then confirm the medical entity data corresponding to the target user; Adverse event extraction: extract perioperative adverse events based on the medical entity data corresponding to the target user; Similarity matching and filtering, matching the perioperative adverse events with the preset dictionary set entries, and then confirming the similarity evaluation coefficient, and comparing the similarity evaluation coefficient with the preset similarity evaluation threshold, and outputting the recognition result based on the comparison result.

[0008] Preferably, the text content data corresponding to the different stages include preoperative text content data, intraoperative text content data, postoperative text content data and discharge text content data.

[0009] Preferably, the medical text data is preprocessed, and the processed medical text data is analyzed for text structure and paragraphs, specifically including: Acquire medical text data input by a target user, and pre-process the medical text data using regular expressions, wherein the pre-processing includes removing non-Chinese characters, abnormal spaces, and format symbols unique to medical texts; The preprocessed medical text data is input into the preset ERNIE-Doc model for paragraph analysis. The steps of paragraph analysis are as follows: The medical text data is segmented using a sliding window mechanism while retaining contextual semantic coherence; Establish a bidirectional attention mechanism to capture the semantic correlation between paragraphs of medical text data; Building a paragraph classifier based on the semantic associations between the medical text data paragraphs, and parsing the medical text data into preoperative text content data, intraoperative text content data, postoperative text content data, and discharge text content data; The preoperative text content data, intraoperative text content data, postoperative text content data and discharge text content data are classified and stored in a relational database according to the diagnosis and treatment stage, and a stage-paragraph mapping index is established.

[0010] Preferably, the UIE model is used to perform entity recognition on the parsed text content data, thereby confirming the medical entity data corresponding to the target user, specifically including: Extracting text content data corresponding to each stage from the relational database and inputting them into a preset UIE model for multi-task information extraction, thereby constructing a medical entity extraction template, wherein the medical entity extraction template includes an operation time entity, an operation name entity, and a clinical reaction entity, wherein the clinical reaction entity includes a positive clinical reaction and a negative clinical reaction; A multi-granularity attention mechanism is used to perform boundary detection and semantic disambiguation on medical entities. The extracted medical entity data are then classified and stored according to the diagnosis and treatment stage, and an entity-timeline association relationship is established.

[0011] Preferably, the perioperative adverse events are extracted according to the medical entity data corresponding to the target user, specifically including: Analyze the temporal relationship between the admission time entity and the surgery time entity, and establish a postoperative time calculation model, and then calculate the standard observation period based on the end time of the surgery; During the standard observation period, a postoperative adverse event feature template is defined, and then perioperative adverse events are generated based on the postoperative adverse event feature template. The postoperative adverse event feature template includes symptom keywords, symptom onset time, symptom duration and related diagnosis and treatment operations.

[0012] Preferably, the perioperative adverse events are matched with preset dictionary set entries to determine a similarity evaluation coefficient, and the similarity evaluation coefficient is compared with a preset similarity evaluation threshold, and a recognition result is output based on the comparison result, specifically including: Constructing a multi-dimensional dictionary set, wherein the dictionary set includes a standard adverse event dictionary, a symptom synonym dictionary, and a drug-side effect association library; Matching the perioperative adverse events with the preset dictionary set entries, and then confirming the literal similarity, semantic similarity and temporal weight factor between the perioperative adverse events and the preset dictionary set entries; A similarity evaluation coefficient is determined based on the literal similarity, semantic similarity and temporal weight factor between the perioperative adverse events and the preset dictionary set entries, and the similarity evaluation coefficient is compared with a preset similarity evaluation threshold, and a recognition result is output based on the comparison result.

[0013] In the second aspect, the present application provides a drug-device adverse event identification system based on real-world data, including: The data input and paragraph analysis module is used to obtain the medical text data input by the target user, pre-process the medical text data, and perform text structure and paragraph analysis on the processed medical text data, thereby identifying the text content data corresponding to different stages; An adverse event entity recognition module is used to perform entity recognition on the parsed text content data using a UIE model, thereby confirming the medical entity data corresponding to the target user; Adverse event extraction module, used to extract perioperative adverse events based on the medical entity data corresponding to the target user; The similarity matching and filtering module is used to match the perioperative adverse events with the preset dictionary set entries, thereby confirming the similarity evaluation coefficient, and comparing the similarity evaluation coefficient with the preset similarity evaluation threshold, and outputting the recognition result based on the comparison result.

[0014] In a third aspect, the present application provides a computer-readable storage medium storing instructions, which, when executed on a computer, enables the computer to execute any one of the above-mentioned methods for identifying adverse drug and device events based on real-world data.

[0015] In summary, this application includes the following beneficial technical effects: The present application provides a method for identifying adverse drug and device events based on real-world data, by identifying text content data corresponding to different stages, and using the UIE model to perform entity recognition on the parsed text content data, thereby confirming the medical entity data corresponding to the target user, extracting perioperative adverse events based on the medical entity data corresponding to the target user, matching the perioperative adverse events with the entries in a preset dictionary set, and then confirming the similarity evaluation coefficient, and comparing the similarity evaluation coefficient with the preset similarity evaluation threshold, and outputting the identification result based on the comparison result, thereby realizing automated and precise identification and screening of adverse events, improving the efficiency and accuracy of medical data processing, and providing reliable support for medical decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0017] Figure 1 This is a flow chart of a method for identifying adverse drug and device events based on real-world data in an embodiment of the present application.

[0018] Figure 2 This is a schematic diagram of a drug-device adverse event identification system based on real-world data in an embodiment of the present application. DETAILED DESCRIPTION

[0019] The following is combined with Figure 1-Figure 2 This application is described in further detail.

[0020] Example 1

[0021] The embodiment of the present application discloses a method for identifying adverse drug and device events based on real-world data.

[0022] Reference Figure 1 , a method for identifying adverse drug and device events based on real-world data, comprising the following steps: Data input and paragraph analysis: obtaining medical text data input by the target user, preprocessing the medical text data, and performing text structure and paragraph analysis on the processed medical text data, thereby identifying text content data corresponding to different stages; Adverse event entity recognition: using the UIE model to perform entity recognition on the parsed text content data, and then confirm the medical entity data corresponding to the target user; Adverse event extraction: extract perioperative adverse events based on the medical entity data corresponding to the target user; Similarity matching and filtering, matching the perioperative adverse events with the preset dictionary set entries, and then confirming the similarity evaluation coefficient, and comparing the similarity evaluation coefficient with the preset similarity evaluation threshold, and outputting the recognition result based on the comparison result.

[0023] By adopting the above technical solution, when processing medical text data, the medical text data input by the user is obtained through data input and paragraph parsing, and preprocessed, including noise removal, standardized format and parsing text structure, to identify the text content corresponding to different stages, and the UIE model is used to perform entity recognition on the parsed text to extract key information, such as drug names, surgical steps, patient reactions, etc., to confirm the medical entity data corresponding to the user. Then, based on these entity data, perioperative adverse events are extracted to ensure that events that are critical to patient safety are identified. The extracted adverse events are matched with the preset dictionary set entries for similarity, and the similarity evaluation coefficient is calculated. By comparing with the preset similarity threshold, it is determined whether the event is within the scope of concern. Finally, the recognition result is output according to the comparison result, thereby realizing the automatic and precise identification and screening of adverse events, improving the efficiency and accuracy of medical data processing, and providing reliable support for medical decision-making.

[0024] Furthermore, the text content data corresponding to the different stages include preoperative text content data, intraoperative text content data, postoperative text content data and discharge text content data.

[0025] It should be noted that the medical text data is preprocessed, and the text structure and paragraph analysis of the processed medical text data specifically includes: Acquire medical text data input by a target user, and pre-process the medical text data using regular expressions, wherein the pre-processing includes removing non-Chinese characters, abnormal spaces, and format symbols unique to medical texts; The preprocessed medical text data is input into the preset ERNIE-Doc model for paragraph analysis. The steps of paragraph analysis are as follows: The medical text data is segmented using a sliding window mechanism while retaining contextual semantic coherence; Establish a bidirectional attention mechanism to capture the semantic correlation between paragraphs of medical text data; Building a paragraph classifier based on the semantic associations between the medical text data paragraphs, and parsing the medical text data into preoperative text content data, intraoperative text content data, postoperative text content data, and discharge text content data; The preoperative text content data, intraoperative text content data, postoperative text content data and discharge text content data are classified and stored in a relational database according to the diagnosis and treatment stage, and a stage-paragraph mapping index is established.

[0026] Specifically, the ERNIE-Doc model is used for multi-level paragraph parsing, which can more accurately identify and distinguish the content of the four stages: before surgery, during surgery, after surgery, and discharge. It can effectively cope with the complex structure of medical texts and significantly reduce the rates of missed detection and false detection.

[0027] It should be noted that the UIE model is used to perform entity recognition on the parsed text content data, thereby confirming the medical entity data corresponding to the target user, specifically including: Extracting text content data corresponding to each stage from the relational database and inputting them into a preset UIE model for multi-task information extraction, thereby constructing a medical entity extraction template, wherein the medical entity extraction template includes an operation time entity, an operation name entity, and a clinical reaction entity, wherein the clinical reaction entity includes a positive clinical reaction and a negative clinical reaction; A multi-granularity attention mechanism is used to perform boundary detection and semantic disambiguation on medical entities. The extracted medical entity data are then classified and stored according to the diagnosis and treatment stage, and an entity-timeline association relationship is established.

[0028] Specifically, using the UIE model to perform entity recognition on the parsed medical text content data can significantly improve the efficiency and accuracy of medical data processing. The UIE model has a strong ability to extract general information and can accurately identify key medical entities such as disease names, surgery types, drug names, adverse reactions, time nodes, etc. from unstructured medical texts. It can efficiently process complex medical texts and avoid the limitations of traditional rule matching methods, especially when facing medical data with strong semantic diversity and context dependence. In addition, the UIE model can continuously optimize the recognition accuracy and adapt to different scenarios and data types through deep learning technology. Its automated recognition capability not only reduces the time cost of manual processing, but also reduces the risk of human error, providing more reliable basic data support for clinical decision-making. At the same time, the UIE model can quickly respond to the processing needs of large-scale medical data, and provide medical institutions with efficient solutions in the monitoring and management of perioperative adverse events, thereby further improving patient safety and medical quality.

[0029] It should be noted that the perioperative adverse events are extracted based on the medical entity data corresponding to the target user, including: Analyze the temporal relationship between the admission time entity and the surgery time entity, and establish a postoperative time calculation model, and then calculate the standard observation period based on the end time of the surgery; During the standard observation period, a postoperative adverse event feature template is defined, and then perioperative adverse events are generated based on the postoperative adverse event feature template. The postoperative adverse event feature template includes symptom keywords, symptom onset time, symptom duration and related diagnosis and treatment operations.

[0030] It should be noted that the perioperative adverse events are matched with the preset dictionary set entries to determine the similarity evaluation coefficient, and the similarity evaluation coefficient is compared with the preset similarity evaluation threshold, and the recognition result is output based on the comparison result, specifically including: Constructing a multi-dimensional dictionary set, wherein the dictionary set includes a standard adverse event dictionary, a symptom synonym dictionary, and a drug-side effect association library; Matching the perioperative adverse events with the preset dictionary set entries, and then confirming the literal similarity, semantic similarity and temporal weight factor between the perioperative adverse events and the preset dictionary set entries; A similarity evaluation coefficient is determined based on the literal similarity, semantic similarity and temporal weight factor between the perioperative adverse events and the preset dictionary set entries, and the similarity evaluation coefficient is compared with a preset similarity evaluation threshold, and a recognition result is output based on the comparison result.

[0031] Specifically, the similarity evaluation coefficient between perioperative adverse events and preset dictionary entries is confirmed by combining literal similarity, semantic similarity and temporal weight factors, which significantly improves the accuracy and reliability of adverse event identification. Literal similarity helps identify directly matching terms to ensure basic consistency; semantic similarity identifies potential semantic relevance and captures more complex language variants by understanding context and word meaning relationships; the temporal weight factor considers the chronological order and importance of events, making the evaluation more dynamic and clinically relevant. The similarity evaluation coefficient calculated based on the above factors can comprehensively reflect the degree of association between adverse events and dictionary entries. Comparing the coefficient with the preset similarity threshold helps to accurately screen out events that need attention and reduce false positives and negatives, thereby not only improving the accuracy of perioperative adverse event monitoring, but also optimizing the event management process, improving medical quality and patient safety, and providing more comprehensive support for medical decision-making.

[0032] Example 2

[0033] The embodiments of the present application also disclose a drug-device adverse event identification system based on real-world data.

[0034] Reference Figure 2 , a drug-device adverse event identification system based on real-world data, including: The data input and paragraph analysis module is used to obtain the medical text data input by the target user, pre-process the medical text data, and perform text structure and paragraph analysis on the processed medical text data, thereby identifying the text content data corresponding to different stages; An adverse event entity recognition module is used to perform entity recognition on the parsed text content data using a UIE model, thereby confirming the medical entity data corresponding to the target user; Adverse event extraction module, used to extract perioperative adverse events based on the medical entity data corresponding to the target user; The similarity matching and filtering module is used to match the perioperative adverse events with the preset dictionary set entries, thereby confirming the similarity evaluation coefficient, and comparing the similarity evaluation coefficient with the preset similarity evaluation threshold, and outputting the recognition result based on the comparison result.

[0035] The above contents are merely examples and explanations of the concept of the present invention. Those skilled in the art may make various modifications or additions to the specific embodiments described or replace them in a similar manner. As long as they do not deviate from the concept of the invention, they should all fall within the protection scope of the present invention.

[0036] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0037] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to specific implementation methods. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well.

Claims

1. A method for identifying adverse drug and device events based on real-world data, characterized in that: The following steps are involved: Data input and paragraph analysis: obtaining medical text data input by the target user, preprocessing the medical text data, and performing text structure and paragraph analysis on the processed medical text data, thereby identifying text content data corresponding to different stages; Adverse event entity recognition: using the UIE model to perform entity recognition on the parsed text content data, and then confirm the medical entity data corresponding to the target user; Adverse event extraction: extract perioperative adverse events based on the medical entity data corresponding to the target user; Similarity matching and filtering, matching the perioperative adverse events with the preset dictionary set entries, and then confirming the similarity evaluation coefficient, and comparing the similarity evaluation coefficient with the preset similarity evaluation threshold, and outputting the recognition result based on the comparison result.

2. According to claim 1, a method for identifying adverse drug and device events based on real-world data is characterized in that: The text content data corresponding to the different stages include preoperative text content data, intraoperative text content data, postoperative text content data and discharge text content data.

3. The method for identifying adverse drug and device events based on real-world data according to claim 2, characterized in that: Preprocessing the medical text data and performing text structure and paragraph analysis on the processed medical text data, specifically including: Acquire medical text data input by a target user, and pre-process the medical text data using regular expressions, wherein the pre-processing includes removing non-Chinese characters, abnormal spaces, and format symbols unique to medical texts; The preprocessed medical text data is input into the preset ERNIE-Doc model for paragraph analysis. The steps of paragraph analysis are as follows: The medical text data is segmented using a sliding window mechanism while retaining contextual semantic coherence; Establish a bidirectional attention mechanism to capture the semantic correlation between paragraphs of medical text data; Building a paragraph classifier based on the semantic associations between the medical text data paragraphs, and parsing the medical text data into preoperative text content data, intraoperative text content data, postoperative text content data, and discharge text content data; The preoperative text content data, intraoperative text content data, postoperative text content data and discharge text content data are classified and stored in a relational database according to the diagnosis and treatment stage, and a stage-paragraph mapping index is established.

4. The method for identifying adverse drug and device events based on real-world data according to claim 1, characterized in that: The UIE model is used to perform entity recognition on the parsed text content data, thereby confirming the medical entity data corresponding to the target user, specifically including: Extract the text content data corresponding to each stage from the relational database and input it into the preset UIE model for multi-task information extraction, and then construct a medical entity extraction template, which includes an operation time entity, an operation name entity and a clinical reaction entity, and the clinical reaction entity includes a positive clinical reaction and a negative clinical reaction; A multi-granularity attention mechanism is used to perform boundary detection and semantic disambiguation on medical entities. The extracted medical entity data are then classified and stored according to the diagnosis and treatment stage, and an entity-timeline association relationship is established.

5. The method for identifying adverse drug and device events based on real-world data according to claim 4, characterized in that: Perioperative adverse events are extracted based on the medical entity data corresponding to the target user, including: Analyze the temporal relationship between the admission time entity and the surgery time entity, and establish a postoperative time calculation model, and then calculate the standard observation period based on the end time of the surgery; During the standard observation period, a postoperative adverse event feature template is defined, and then perioperative adverse events are generated based on the postoperative adverse event feature template. The postoperative adverse event feature template includes symptom keywords, symptom onset time, symptom duration and related diagnosis and treatment operations.

6. The method for identifying adverse drug and device events based on real-world data according to claim 5, characterized in that: The perioperative adverse events are matched with the preset dictionary set entries, and then the similarity evaluation coefficient is determined, and the similarity evaluation coefficient is compared with the preset similarity evaluation threshold, and the recognition result is output based on the comparison result, specifically including: Constructing a multi-dimensional dictionary set, wherein the dictionary set includes a standard adverse event dictionary, a symptom synonym dictionary, and a drug-side effect association library; Matching the perioperative adverse events with the preset dictionary set entries, and then confirming the literal similarity, semantic similarity and temporal weight factor between the perioperative adverse events and the preset dictionary set entries; A similarity evaluation coefficient is determined based on the literal similarity, semantic similarity and temporal weight factor between the perioperative adverse events and the preset dictionary set entries, and the similarity evaluation coefficient is compared with a preset similarity evaluation threshold, and a recognition result is output based on the comparison result.

7. A drug-device adverse event identification system based on real-world data, applied to a drug-device adverse event identification method based on real-world data as described in any one of claims 1 to 6, characterized in that: include: The data input and paragraph analysis module is used to obtain the medical text data input by the target user, pre-process the medical text data, and perform text structure and paragraph analysis on the processed medical text data, thereby identifying the text content data corresponding to different stages; An adverse event entity recognition module is used to perform entity recognition on the parsed text content data using a UIE model, thereby confirming the medical entity data corresponding to the target user; Adverse event extraction module, used to extract perioperative adverse events based on the medical entity data corresponding to the target user; The similarity matching and filtering module is used to match the perioperative adverse events with the preset dictionary set entries, thereby confirming the similarity evaluation coefficient, and comparing the similarity evaluation coefficient with the preset similarity evaluation threshold, and outputting the recognition result based on the comparison result.

8. A computer-readable storage medium, characterized in that: Instructions are stored, and when the instructions are executed on a computer, the computer is caused to execute a method for identifying adverse drug-device events based on real-world data as described in any one of claims 1 to 6.

Images