Hospital adverse event risk identification method and system based on causal reasoning

Abstract

The invention provides a hospital adverse event risk identification method and system based on causal reasoning, and the method comprises the steps: responding to an operation request of a user for reporting a hospital adverse event based on the optimization of medical advice data processing and a medical quality management process in an operation risk identification scene, inputting the detailed information of the event through a formatted form, and carrying out the operation of the event. Through multi-node, multi-auditor and multi-department cooperative auditing, an adverse event form reported by a user is subjected to approval, flow and summary analysis, medical advice data is subjected to structured storage by adopting visual enhancement OCR, dynamic knowledge graph verification and a time sequence causal reasoning model, a time window division algorithm is adopted, and medical advice data is obtained. The causal discovery algorithm constructs an unplanned re-operation association network, early warning of adverse events of operation complications is carried out, and compared with a traditional mode, the medical advice processing efficiency, the transcription error rate and the unplanned operation detection rate are all greatly improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of medical information technology, in particular to a hospital adverse event risk identification method and system based on causal reasoning. BACKGROUND

[0002] Medical safety is the core pillar of the medical system and an important measure of the quality of medical services. The management and reporting of hospital adverse events are important issues in the field of medical safety management. In this context, medical institutions have also actively explored and practiced various patient safety management methods and technologies in order to reduce medical risks and reduce the occurrence of adverse events, and to protect the safety of patients' lives and health. How to effectively manage and report adverse events has become the focus of attention of medical institution managers and medical staff.

[0003] In traditional hospital management, there are many limitations in identifying the risk of unplanned reoperation. On the one hand, it mainly relies on the clinical experience and subjective judgment of medical staff, which lacks systematicness and comprehensiveness, and is prone to omissions or misjudgments when facing complex patient conditions and various types of surgery. On the other hand, although some hospitals have established adverse event reporting systems, these systems mostly focus on post-reporting and handling, and have weak proactive identification capabilities for preoperative and intraoperative potential risks, making it difficult to discover and warn potential risk factors that may cause unplanned reoperation in a timely manner before surgery or during surgery, resulting in many potential risks that cannot be effectively controlled and avoided, and ultimately leading to a high incidence of unplanned reoperation, which seriously affects medical quality and medical safety. SUMMARY

[0004] The purpose of the present application is to provide a hospital adverse event risk identification method and system based on causal reasoning, which aims to solve the problem of low transcription efficiency and high error rate caused by the heterogeneity of medical order data, to achieve automatic and accurate identification of unplanned reoperation, and to reduce the cost and risk of manual screening.

[0005] In order to achieve the above-mentioned purpose, the present application provides a hospital adverse event risk identification method based on causal reasoning, comprising:

[0006] Step S100, in response to the operation request of the user reporting the hospital adverse event, inputting the detailed information of the event through the formatted form;

[0007] Step S200, approving, streaming and summarizing the adverse event form reported by the user through multi-node, multi-auditor and multi-department collaborative review;

[0008] Step S300, collect the medical order data image, adopt Retinex algorithm to correct the light, then through the mixed OCR recognition, JSON structured translation is carried out, the standardized data formed is checked by knowledge graph compliance, and the formatted medical order data is generated after automatic correction and enters the database;

[0009] Step S400, identify the unplanned reoperation according to the unplanned reoperation adverse event early warning model, the model is based on the corrected diagnosis and treatment and operation data, the correlation network of unplanned reoperation is identified and constructed through causal algorithm, and the output result is JSON structured translation, the standardized data formed is checked by knowledge graph compliance, and the formatted operation adverse event early warning information is generated after automatic correction and enters the database;

[0010] Step S500, report and audit the operation adverse event early warning information, and carry out closed loop continuous improvement operation based on the adverse event.

[0011] Further, in step S100, the form type of the hospital adverse event is bound with the responsible department, and the report is carried out according to the direct administrative department, intelligent attribution department, hospital area, clinical special group and related collaborative department of the event.

[0012] Further, in step S100, the formatted form includes: event content, patient basic information, event special content, event basic situation and event classification.

[0013] Further, in step S200, the audit time limit is set based on the attribution of the formatted form, and the case is audited, copied, invalidated and summarized according to the auditors, multi-department and condition branches in the corresponding nodes.

[0014] Further, in step S300, it also includes:

[0015] Image preprocessing, denoising and color space conversion are carried out on the obtained diagnosis and treatment and operation data image;

[0016] Multi-scale Retinex processing, a plurality of Gaussian kernels are adopted, the reflection components under different scales are fused through weighted average, and the global brightness and local details are balanced;

[0017] Post-processing optimization, dynamic range compression and detail enhancement are carried out on the corrected image.

[0018] Further, in step S300, it also includes:

[0019] 12-layer Transformer mixed ResNet-50 network is adopted, and the global text layout features of the medical order image are captured through hierarchical window attention mechanism;

[0020] Based on the medical order text features obtained by dynamic knowledge graph verification, the text errors are automatically corrected, and the dynamic knowledge graph includes a dynamic layer established by hospital internal rules and a static layer established by pharmacopoeia and diagnosis and treatment specifications.

[0021] Further, in step S400, further comprising:

[0022] A sliding window mechanism is adopted to divide the multiple surgeries in the hospitalization period into surgery windows according to the occurrence frequency based on the surgery time stamp;

[0023] Input includes surgery operation, complication, patient basic disease diagnosis and treatment and surgery structured data, generates a causal association matrix, establishes a confounding factor management, and reduces the false association probability;

[0024] The surgery causal association matrix is embedded into the causal strength weight output by the algorithm to construct a non-planned reoperation association network architecture, and a causal discovery algorithm is used to identify non-planned reoperation features.

[0025] Further, in step S400, further comprising:

[0026] Based on the non-planned reoperation features obtained by dynamic knowledge graph verification, the text errors are automatically corrected, and the dynamic knowledge graph includes a dynamic layer established by hospital internal rules and a static layer established by diagnosis / surgery specifications.

[0027] In another aspect, the application also provides a hospital adverse event risk identification system based on causal reasoning, comprising:

[0028] The data acquisition layer obtains medical order data by photographing, and obtains patient diagnosis and treatment and surgery data through the HIS system, and obtains the reported hospital adverse data form through the hospital adverse event;

[0029] The intelligent processing layer adopts Retinex algorithm for illumination correction, then performs JSON structured translation through mixed OCR identification, and forms standardized data which is verified by knowledge graph compliance, automatically corrected to generate formatted medical order data into the database; the non-planned reoperation is identified according to the non-planned reoperation adverse event early warning model, the model is based on the corrected diagnosis and treatment and surgery data, the non-planned reoperation association network is constructed by the causal algorithm, and the output result is JSON structured translation, the standardized data is verified by knowledge graph compliance, automatically corrected to generate formatted surgery adverse event early warning information into the database;

[0030] The application layer reports and audits the surgery adverse event early warning information, and performs closed-loop continuous improvement operation based on the adverse event.

[0031] The application has the following technical progress and beneficial effects:

[0032] The application provides a hospital adverse event risk identification method and system based on causal reasoning, which is based on optimization of medical order data processing and surgical risk identification scene in the medical quality management process, responds to the operation request of the user reporting the hospital adverse event, inputs the detailed information of the event through the formatted form, approves, transmits and analyzes the adverse event form reported by the user through multi-node, multi-auditor and multi-department collaborative review, adopts visual enhanced OCR, dynamic knowledge graph verification and time sequence causal reasoning model, stores the medical order data in a structured manner, adopts time window division algorithm, constructs an unplanned reoperation correlation network through a causal discovery algorithm, and performs surgical complication adverse event early warning, and compared with the traditional method, the medical order processing efficiency, transcription error rate and unplanned surgery detection rate are greatly improved. BRIEF DESCRIPTION OF DRAWINGS

[0033] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0034] Figure 1 It is a method flow chart of a hospital adverse event risk identification method based on causal reasoning according to an embodiment of the present application.

[0035] Figure 2 It is a medical order data image data processing architecture diagram according to an embodiment of the present application.

[0036] Figure 3 It is a data processing method diagram of an unplanned reoperation adverse event early warning model according to an embodiment of the present application.

[0037] Figure 4 It is a data method diagram of dynamic knowledge graph verification according to an embodiment of the present application. DETAILED DESCRIPTION

[0038] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0039] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments, but it is to be understood that "some embodiments" can be the same subset or different subsets as each other and as other subsets of all possible embodiments, and can be combined with each other in a manner not inconsistent with the application.

[0040] If the application file contains similar descriptions of "first / second", the following explanations are added: In the following description, the terms "first, second, third" refer to similar objects only and do not represent a specific order or sequence for the objects, and it is understood that "first, second, third" can be interchanged in a specific order or sequence as allowed, so that the application embodiments described herein can be implemented in an order other than that illustrated or described herein.

[0041] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the application belongs. The terms used herein are only for the purpose of describing the embodiments of the application and are not intended to limit the application.

[0042] Embodiment one

[0043] Figures 1-4 The method flow chart of a hospital adverse event risk identification method based on causal reasoning according to one embodiment of the application is shown. As shown in the figure, Figure 1 The hospital adverse event risk identification method based on causal reasoning according to the embodiment includes:

[0044] Step S100, in response to the operation request of the user reporting the hospital adverse event, the detailed information of the event is input through the formatted form.

[0045] Step S200, the adverse event form reported by the user is approved, streamed and analyzed by multi-node, multi-auditor and multi-department collaborative review.

[0046] Step S300, collect the medical order data image, use Retinex algorithm for light correction, then through mixed OCR identification, JSON structured translation is carried out, the standardized data formed is verified by knowledge graph compliance, and the error is automatically corrected to generate formatted medical order data into the database;

[0047] Step S400, according to the non-planned reoperation adverse event early warning model, the non-planned reoperation is identified, the model is based on the corrected diagnosis and treatment and operation data, the non-planned reoperation association network is constructed by causal algorithm identification, and the output result is JSON structured translation, the standardized data formed is verified by knowledge graph compliance, and the error is automatically corrected to generate formatted operation adverse event early warning information into the database;

[0048] Step S500, report and review the operation adverse event early warning information, and continuously improve the operation based on the adverse event.

[0049] Further, in step S100, the form type of the hospital adverse event is bound to the responsible department, and the report is made according to the directly managed administrative department, the intelligent attribution department, the hospital area, the clinical specialist group and the related collaborative department.

[0050] Further, in step S100, the formatted form includes event content, patient basic information, event special content, event basic situation and event classification.

[0051] Further, in step S200, based on the review time limit of the formatted form, the case is audited, copied, invalidated and statistically summarized according to the reviewer, multi-department and condition branch at the corresponding node.

[0052] Further, in step S300, it also includes:

[0053] Image preprocessing, denoising and color space conversion are performed on the obtained diagnosis and treatment and operation data images;

[0054] Multi-scale Retinex processing, using multiple Gaussian kernels, fusing the reflection components under different scales through weighted average, balancing global brightness and local details;

[0055] Post-processing optimization, dynamic range compression and detail enhancement are performed on the corrected images.

[0056] Further, in step S300, it also includes:

[0057] A 12-layer 1-Transformer hybrid ResNet-50 network is used to capture the global text layout features of the medical order image through a hierarchical window attention mechanism, and 100,000 medical order samples from 8 first-class hospitals are used for training;

[0058] Based on the dynamic knowledge graph, the obtained medical order text features are automatically corrected, and the dynamic knowledge graph includes a dynamic layer established by hospital internal rules and a static layer established by pharmacopoeia and diagnosis and treatment specifications.

[0059] Further, in step S400, it also includes:

[0060] A sliding window mechanism is used, and the operation time stamp is used as the basis to divide the multiple operations in the hospitalization period according to the occurrence frequency, and the time window is divided into: 0-24h, 24-72h, 72h-7d after operation.

[0061] The input includes surgical operation, complication, patient basic disease diagnosis and treatment and surgical structured data, a causal association matrix is generated, confounding factor management is established, and false association probability is reduced.

[0062] A non-planned reoperation association network architecture is constructed, the surgical causal association matrix is embedded into the causal strength weight output by the algorithm, and a causal discovery algorithm is used to identify the characteristics of non-planned reoperation. The model is trained using 3000 surgical cases (including 450 non-planned reoperation cases).

[0063] Further, in step S400, it also includes:

[0064] Based on the non-planned reoperation characteristics obtained by the dynamic knowledge graph verification, the text errors are automatically corrected, and the dynamic knowledge graph includes a dynamic layer composed of hospital internal rules and a static layer composed of diagnosis / surgery specifications.

[0065] Embodiment two

[0066] In another aspect, the application also provides a hospital adverse event risk identification system based on causal reasoning, comprising:

[0067] A data acquisition layer obtains medical order data by taking pictures, and obtains patient diagnosis and treatment and surgery data through a HIS system, and obtains a hospital adverse event report form through a hospital adverse event;

[0068] An intelligent processing layer uses a Retinex algorithm for illumination correction, and then performs JSON structured translation after mixed OCR recognition, and the standardized data formed is verified by a knowledge graph compliance, and the corrected error is automatically generated into a formatted medical order data into a database; a non-planned reoperation is identified according to a non-planned reoperation adverse event early warning model, the model is based on the corrected diagnosis and treatment and surgery data, a non-planned reoperation association network is constructed by a causal algorithm, and the output result is JSON structured translation, the standardized data formed is verified by a knowledge graph compliance, and the corrected error is automatically generated into a formatted surgical adverse event early warning information into a database;

[0069] An application layer reports and audits surgical adverse event early warning information, and performs closed-loop continuous improvement operation based on the adverse event.

[0070] Although the embodiments of the application have been shown and described above, it should be understood that the above embodiments are exemplary and should not be construed as limiting the application, and those skilled in the art can make changes, modifications, replacements and variations to the above embodiments within the scope of the application. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the application shall be included in the protection scope of the application.

Claims

1. A risk identification method for hospital adverse events based on causal reasoning, characterized by: include: Step S100, responding to the user's request to report a hospital adverse event, and inputting detailed information of the event through a formatted form; Step S200: Through multi-node, multi-reviewer, and multi-department collaborative review, the adverse event form reported by the user is reviewed, circulated, and summarized and analyzed; Step S300: Acquire medical order data images, perform illumination correction using the Retinex algorithm, and then perform hybrid OCR recognition followed by JSON structured translation. The resulting standardized data is then verified for compliance with the knowledge graph, and errors are automatically corrected before generating formatted medical order data that is entered into the database. Step S400: Identify unplanned reoperations based on an unplanned reoperation adverse event warning model. The model, based on the revised diagnosis and surgical data, uses a causal algorithm to identify and construct an unplanned reoperation association network. The output results are then translated into JSON structure. The resulting standardized data is verified for compliance with the knowledge graph, and after automatically correcting errors, formatted surgical adverse event warning information is generated and entered into the database. Step S500: Report and review warning information of adverse surgical events, and perform closed-loop continuous improvement operations based on the adverse events.

2. The risk identification method for hospital adverse events based on causal reasoning according to claim 1, characterized in that: In step S100, the form type of the hospital adverse event is bound to the responsible department, and the event is reported to the relevant collaborative departments according to the direct administrative management department, intelligent attribution department, hospital district, clinical specialty group, and other related departments.

3. The risk identification method for hospital adverse events based on causal reasoning according to claim 2, characterized in that: In step S100, the formatted form includes: event content, patient basic information, event-specific content, event basic situation and event classification.

4. The risk identification method for hospital adverse events based on causal reasoning according to claim 3 is characterized in that: In step S200, the review time limit is set based on the attribution of the formatted form, and the case is reviewed, copied, invalidated, and summarized and counted at the corresponding node according to the reviewer, multiple departments, and conditional branches.

5. The risk identification method for hospital adverse events based on causal reasoning according to claim 4 is characterized in that: In step S300, the following is also included: Image preprocessing: denoising and color space conversion of acquired diagnostic and surgical data images; Multi-scale Retinex processing uses multiple Gaussian kernels to fuse reflection components at different scales through weighted averaging to balance global brightness and local details; Post-processing optimization: dynamic range compression and detail enhancement are performed on the corrected image.

6. The risk identification method for hospital adverse events based on causal reasoning according to claim 5, characterized in that: In step S300, the following is also included: A 12-layer Transformer hybrid ResNet-50 network is used to capture the global text layout features of medical order images through a hierarchical window attention mechanism. The text features of medical orders obtained by verification based on the dynamic knowledge graph are automatically corrected for text errors. The dynamic knowledge graph includes a dynamic layer established by the hospital's internal rules and a static layer established by the pharmacopoeia and diagnosis and treatment specifications.

7. The risk identification method for hospital adverse events based on causal reasoning according to claim 6, characterized in that: In step S400, the following is also included: A sliding window mechanism is used to divide the surgical windows according to the frequency of occurrence of multiple surgeries during the hospitalization period based on the surgical timestamp. Input includes surgical procedures, complications, diagnosis and treatment of patients' underlying diseases, and structured surgical data to generate a causal association matrix, establish confounding factor management, and reduce the probability of false associations; An unplanned reoperation association network architecture was constructed, and the surgical causal association matrix was used as the core node to embed the causal strength weight output by the algorithm, and a causal discovery algorithm was constructed to identify the characteristics of unplanned reoperation.

8. The risk identification method for hospital adverse events based on causal reasoning according to claim 7, characterized in that: In step S400, the following is also included: The unplanned reoperation features obtained by verification based on the dynamic knowledge graph are automatically corrected for text errors. The dynamic knowledge graph includes a dynamic layer composed of internal hospital rules and a static layer composed of diagnosis / surgical specifications.

9. A risk identification system for hospital adverse events based on causal reasoning, characterized in that: include: The data collection layer takes photos to obtain medical order data, obtains patient diagnosis and surgery data through the HIS system, and obtains the reported hospital adverse data forms through hospital adverse events; The intelligent processing layer uses the Retinex algorithm for illumination correction, followed by hybrid OCR recognition and JSON structured translation. The resulting standardized data passes the knowledge graph compliance check, automatically corrects errors, and generates formatted medical order data for entry into the database. Unplanned reoperations are identified based on the unplanned reoperation adverse event warning model. The model, based on the corrected diagnosis and surgical data, uses a causal algorithm to identify and construct an unplanned reoperation association network, and performs JSON structured translation on the output results. The resulting standardized data passes the knowledge graph compliance check, automatically corrects errors, and generates formatted surgical adverse event warning information for entry into the database. At the application layer, early warning information on adverse surgical events is reported and reviewed, and closed-loop continuous improvement operations are performed based on the adverse events.

Images