Intelligent medical record quality control method based on rule and large language model hybrid engine
By using an intelligent medical record quality control method based on a hybrid engine of rules and large language models, the problems of low efficiency and poor accuracy in medical record quality control in existing technologies are solved. This method achieves automated and in-depth quality control of all medical records, improving the efficiency and accuracy of quality control and adapting to the needs of multi-institutional medical networks.
Patent Information
- Application Number
- CN202511472124.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-15
- Publication Date
- 2026-01-27
AI Technical Summary
Existing medical record quality control methods suffer from problems such as low efficiency, high cost, inconsistent results, and inability to deeply understand medical semantics, making it difficult to meet the modern medical demand for comprehensive, efficient, and precise control over medical record quality.
An intelligent medical record quality control method based on a hybrid engine of rules and large language models is adopted. The medical record data is transformed into structured objects through post-structured processing. The method combines pure rule quality control, pure large language model quality control and hybrid quality control paths, adaptively selects quality control strategies, and constrains the output of the large language model through a dynamic prompt word construction method.
It achieves automated and in-depth quality control of all medical records, covering both form and content checks, improving quality control efficiency and accuracy, reducing computing resource consumption, and possessing flexibility and scalability to adapt to multi-institutional medical networks.
Smart Images

Figure CN121415967A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of natural language processing technology, specifically to an intelligent medical record quality control method based on a hybrid engine of rules and large language models. Background Technology
[0002] Medical records, as core medical documents documenting the entire patient treatment process, directly impact medical safety, the reliability of clinical research, and the accuracy of medical insurance reimbursement. Therefore, medical record quality control (hereinafter referred to as "quality control") is an indispensable key link in the quality management of medical institutions. However, the quality control methods commonly used in the industry currently have significant limitations and cannot meet the needs of modern medicine for comprehensive, efficient, and precise control of medical record quality. Currently, mainstream quality control methods mainly include manual sampling review, traditional automated rule systems, and emerging pure large language model applications. Among these, manual sampling review, as a traditional method, relies on clinical experts or dedicated quality control personnel periodically sampling a small number of medical records for manual review. This method is not only inefficient and costly, but also has extremely low coverage due to manpower limitations, typically only able to check a small number of medical records, resulting in a large number of systemic quality problems being hidden within. Furthermore, the review results heavily depend on personal experience and subjective judgment, making it difficult for different reviewers to maintain consistency and objectivity. To improve efficiency, some institutions have introduced traditional automated quality control software based on preset rules. These systems can efficiently handle "formal quality control" issues such as missing fields and format errors through keyword matching, regular expressions, and simple logical judgments. However, their shortcomings lie in their inability to understand the semantics of medical context. They are completely powerless against "intrinsic quality control," which requires deep medical logical reasoning, such as whether the diagnosis and medical history description are contradictory or whether the treatment plan is reasonable. In addition, rule-based systems are rigid, have high maintenance costs, and are difficult to adapt to the changing clinical practices and document standards of different departments and institutions. In recent years, with the rise of large language model technology, there has been exploration of using it directly for medical record quality control. However, pure large language model solutions suffer from the "black box" problem of poor interpretability. Their judgment process is not transparent and it is difficult to provide convincing and traceable quality control reasons, which is a fatal flaw in a rigorous clinical environment. At the same time, the inherent "illusion" phenomenon of the model may lead to the generation of content that does not conform to the facts, and its judgment results are easily affected by prompts or context, making it difficult to guarantee stability and consistency, and unable to meet the stringent requirements for accuracy and reliability in medical scenarios. In summary, existing technical solutions each have shortcomings in terms of efficiency, depth, cost, and reliability. In particular, they lack an intelligent medical record quality control solution that can balance the determinism and high efficiency of rule engines with the deep semantic understanding capabilities of large language models, and is applicable to regionalized, multi-institutional medical networks. Summary of the Invention
[0003] To balance the determinism and high efficiency of rule engines with the deep semantic understanding capabilities of large language models, and to make it applicable to intelligent medical record quality control solutions in regionalized, multi-institutional medical networks, this invention proposes an intelligent medical record quality control method based on a hybrid engine of rules and large language models, comprising the following steps: S1: Obtain raw electronic medical record data from the medical information system; S2: Perform post-structuring processing on the original electronic medical record data, converting the unstructured or semi-structured text in the original electronic medical record data into a unified, machine-readable structured data object. S3: Traverse multiple quality control points in the preset quality control rule base, and for each quality control point, adaptively select and execute the corresponding quality control path based on the transformed structured data object; S4: Based on the quality control results of all quality control points, calculate the score and generate a quality control report; The quality control paths include pure rule-based quality control paths, pure large language model quality control paths, and hybrid quality control paths with rules at the forefront.
[0004] Further, in step S2, the post-structuring process includes: S21: Load the corresponding parsing module based on the text type in the original electronic medical record data; S22: Use natural language processing technology to divide the raw electronic medical record data into multiple logical modules; S23: Extract and semantically normalize the key fields in each logical module to generate a structured data object containing key-value pairs.
[0005] Furthermore, the semantic normalization process includes identifying negative statements in the original electronic medical record data of the current logic module and converting the negative statements into specific identifiers.
[0006] Furthermore, the pure rule-based quality control path directly performs logical judgments on structured data objects based on preset deterministic rules. The deterministic rules include at least one of the following: field integrity verification, field format and validity verification, logical consistency verification between fields, and entry timeliness verification, as well as fast content filtering based on keywords or regular expressions.
[0007] Furthermore, the pure large language model quality control path performs in-depth semantic analysis and logical reasoning-based quality control on structured data objects by calling the large language model.
[0008] Furthermore, the rule-based hybrid quality control path includes the following steps: The structured data object is initially assessed for quality control based on pre-defined deterministic rules. If the preliminary quality control assessment result is that the quality control has passed, then the conclusion that the quality control has passed will be output directly and the process at the current quality control point will be terminated. If the initial quality control assessment results in a failure, a large language model is invoked to perform deep semantic analysis on the structured data object, and the final quality control conclusion is obtained based on the output of the large language model.
[0009] Furthermore, when calling the large language model in the execution of the pure large language model quality control path or the rule-preceding hybrid quality control path, a dynamic prompt word construction method is adopted, specifically including the following steps: Based on the current quality control points, extract the corresponding prompt word templates from the quality control rule base; Fill the relevant fields from the structured data into the prompt word template to generate dynamic prompt words specifically for the task; Dynamic prompts are sent to a large language model for analysis.
[0010] Furthermore, the prompt word template includes the following information: relevant field data extracted from the structured data object, analysis instruction parameters of the current quality control point obtained from the quality control rule base, and response template used to constrain the output format of the large language model.
[0011] Furthermore, in step S3, the adaptive selection specifically involves selecting a quality control path based on the quality control mode parameters pre-configured for each quality control point in the quality control rule base.
[0012] Furthermore, in step S4, the scoring calculation includes: calculating deductions for failed quality control points based on the pre-configured weights or deduction values of the quality control points, and determining the medical record quality level based on the total score after deductions.
[0013] Compared with the prior art, the present invention has at least the following beneficial effects: (1) The present invention proposes an intelligent medical record quality control method based on a hybrid engine of rules and large language models, which can realize automated and in-depth quality control of the entire medical record. It not only covers formal inspection, but also performs internal logic analysis. Its modular and configurable design also gives the system excellent flexibility and regional expansion capability, providing reliable technical support for improving the overall level of medical quality management. (2) Through post-structuring processing technology, the original medical record texts from diverse sources and in different formats are transformed into a unified and standardized structured data object. This preprocessing method provides a high-quality, machine-readable input basis for all subsequent quality control links, fundamentally ensuring the accuracy and consistency of quality control analysis. (3) A hybrid quality control engine with intelligent scheduling logic is used to adaptively select pure rules, pure large language models or rule-based hybrid quality control paths for different quality control points. Among them, the rule-based hybrid path is particularly critical. It performs rapid initial screening through pre-set deterministic rules and only starts the high-cost large language model for deep semantic analysis when the rule judgment fails. This layered strategy ensures that the inherent quality control problems that traditional systems cannot identify are discovered, while greatly optimizing the allocation of computing resources and achieving a balance between quality control depth and operating efficiency. (4) When calling the large language model, the system adopts the dynamic prompt word construction method to programmatically integrate structured data, quality control instructions and output templates, which effectively constrains the output range of the large language model, significantly improves the accuracy, consistency and interpretability of the analysis results, and overcomes the "black box" and "illusion" risks of the pure large language model scheme. Attached Figure Description
[0014] Figure 1 A flowchart illustrating the steps of an intelligent medical record quality control method based on a hybrid engine of rules and large language models; Figure 2 This is a flowchart of the complete quality control method. Detailed Implementation
[0015] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to its technical solutions. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. This invention can be deployed and run on electronic devices with computing capabilities, such as regional medical data centers or hospital local servers. Its hardware environment includes a processor, memory, and necessary network interfaces, while the software environment includes an operating system, a database management system, and a programming language runtime environment. Figure 1 As shown, the present invention proposes an intelligent medical record quality control method based on a hybrid engine of rules and large language models, comprising the following steps: S1: Obtain raw electronic medical record data from the medical information system; S2: Perform post-structuring processing on the original electronic medical record data, converting the unstructured or semi-structured text in the original electronic medical record data into a unified, machine-readable structured data object. S3: Traverse multiple quality control points in the preset quality control rule base, and for each quality control point, adaptively select and execute the corresponding quality control path based on the transformed structured data object; S4: Based on the quality control results of all quality control points, calculate the score and generate a quality control report; The quality control paths include pure rule-based quality control paths, pure large language model quality control paths, and hybrid quality control paths with rules at the forefront.
[0016] Specifically, such as Figure 2 The overall quality control method flowchart of this invention first executes the data acquisition step. This step obtains raw electronic medical record data from a regional medical data platform or hospital information system. In actual implementation, this step needs to process heterogeneous data sources from different medical institutions and systems. This data includes structured data such as patient basic information and test results, as well as a large amount of unstructured natural language text such as chief complaints, present illness, past medical history, and progress notes. To ensure data integrity and accuracy, this method adopts a unified data interface specification, which can automatically identify and parse data files of different formats. In particular, for electronic medical record systems from different vendors, this method standardizes various fields through preset data mapping rules, laying the foundation for subsequent processing. The key to this step is to establish a stable and reliable data channel to ensure that the latest medical record data can be obtained in real time or periodically, while ensuring the security and privacy protection during data transmission.
[0017] After obtaining the raw data, the next step is post-processing of the medical records into a structured data object. This step uses natural language processing technology and configurable rules to transform the messy raw medical record text into a unified and standardized structured data object. In practice, the first step is to load the corresponding parsing configuration template based on the document type and department of origin of the medical record. These templates are predefined based on the analysis of a large number of medical documents and can accurately identify the characteristics of different departments and different medical record types. For example, internal medicine medical records and surgical medical records require different parsing strategies due to their different focuses and recording habits. Next, keyword recognition and regular expressions are used to segment the long text into logical modules such as "chief complaint," "present illness," "past medical history," and "physical examination." This process needs to handle common abbreviations, synonyms, and expression variations in medical text to ensure the accuracy of segmentation.
[0018] Within each module, key information fields are further extracted using a medical thesaurus and refined rules, followed by semantic normalization. This process involves complex natural language processing techniques, including entity recognition and relation extraction. For example, when extracting "past medical history" information, the system needs to accurately identify key elements such as disease name, onset time, and treatment details, and map them to standardized medical terminology. Crucially, the system can recognize negative expressions in medical text, such as parsing "deny of surgical history" as a Boolean value of "surgery: false," rather than simply extracting the keyword "surgery." This ability to recognize negation is vital for subsequent quality control assessments, preventing misjudgments.
[0019] To handle complex expressions in medical texts, this method also incorporates a contextual understanding mechanism. For example, when a medical record contains a statement like "The patient underwent an appendectomy three years ago and is currently recovering well," the system needs to accurately extract multiple information points, such as the surgery name, surgery time, and current status. Simultaneously, for complex information such as physical examinations, the system supports multi-level parsing, for instance, further breaking down "T 37.8℃, P 110 bpm, BP 140 / 90 mmHg" into specific indicators like temperature, pulse, and blood pressure. Through this series of processing steps, the original medical record is transformed into machine-readable JSON-formatted structured data. This structured representation greatly facilitates subsequent automated quality control.
[0020] After data structuring is completed, the core hybrid intelligent quality control judgment step begins. This step iterates through all applicable quality control points in a pre-built quality control rule base and adaptively selects and executes the corresponding quality control path based on the preset pattern of each quality control point. The quality control rule base is built based on medical quality control standards and clinical practice experience, and contains detailed definitions of hundreds of quality control points. Each quality control point clearly specifies its applicable medical record type, examination content, judgment criteria, and corresponding quality control path selection.
[0021] When performing quality control assessments, the system first needs to determine the set of quality control points applicable to the current medical record. This determination is based on basic information about the medical record, such as the record type, department, and patient age. For example, for surgical medical records, the focus will be on surgery-related quality control points; for pediatric medical records, the focus will be on specific indicators such as growth and development. After determining the set of quality control points, the system executes the checks on each quality control point sequentially according to a preset priority order.
[0022] In the execution of the pure rule-based quality control path, logical judgments are directly performed on structured data objects based on pre-defined deterministic rules. These rules cover various types, including field integrity verification, format validity verification, logical consistency verification, and data entry timeliness verification. For example, in field integrity verification, the system checks whether required fields such as patient name and hospital number are missing or have length limits; in format validity verification, it verifies whether fields such as ID number, address, and mobile phone number conform to the standard format; in logical consistency verification, it checks whether key information such as name and gender in the admission record matches the medical record cover page; and in data entry timeliness verification, it confirms whether the admission record was completed within 24 hours of the patient's admission. This type of judgment is characterized by fast execution speed and high certainty of results, and is suitable for all formal quality control points that do not require deep semantic understanding. To ensure the accuracy of the rules, the system also establishes a rule verification mechanism to regularly test and optimize the effectiveness of the rules.
[0023] For quality control points requiring deep semantic understanding, a pure large language model quality control path is adopted. In this path, the first step is to extract the corresponding prompt word template from the quality control rule base based on the current quality control point. The design of the prompt word template is a crucial step, requiring clear definition of task requirements, input data format, and output specifications. For example, for the quality control point of "logical consistency between diagnosis and medical history," the system will integrate fields such as chief complaint, present medical history, and diagnostic description into the prompt word, forming a specific task instruction requiring the large language model to analyze logical correlation. The construction of the prompt word needs to follow specific principles: providing sufficient contextual information while avoiding information overload; clearly defining task requirements while retaining a certain degree of flexibility to adapt to different situations.
[0024] Then, the relevant fields from the structured data are filled into a template to generate dynamic prompts specific to the task. This process requires careful data filtering and organization, retaining only information relevant to the current quality control point and avoiding irrelevant information from interfering with the model's judgment. Finally, the prompts are sent to a large language model for analysis, and the returned structured results are parsed. To ensure consistency, the system strictly regulates the model's output format, requiring the judgment results and reasons to be returned in a specific JSON format.
[0025] The implementation of a hybrid quality control path begins with a preliminary assessment of structured data using pre-defined deterministic rules. For example, when checking whether a surgical history exists and further confirming that the description must include key information such as the surgical time, surgical name, and postoperative condition, two quality control points are established: 1. Determining whether a surgical history exists; 2. Determining whether the surgical history includes key information such as the surgical time, surgical name, and postoperative condition. Specifically: 1) First, use rules to quickly determine if a surgical history description exists: By scanning the past medical history field, search for keywords such as "surgery" and "resection," and combine this with the results of negative word recognition. If the rule clearly indicates that no surgical history exists, directly output the conclusion that quality control fails. The rules used in this preliminary judgment stage are carefully designed to ensure both high recall and execution efficiency. The rule design is based on the analysis of a large amount of real medical record data and can cover common expression methods; 2) When the above rule passes, i.e. the rule detects the existence of a surgical history description, then further extract key information such as surgical time, surgical name and postoperative condition. If the extraction is successful, it is recorded as "passed"; if the key information cannot be extracted, then proceed to the next step of calling the large language model to analyze whether the surgical description contains necessary elements such as surgical time, specific surgical procedure and postoperative condition.
[0026] This layered strategy significantly optimizes the use of computing resources while ensuring the depth of quality control. In practical applications, this hybrid path can filter out most simple cases, only calling the large language model when necessary, thus ensuring both quality control effectiveness and controlling operating costs.
[0027] In quality control processes that require the invocation of large language models, the implementation of dynamic prompt word construction methods is particularly crucial. This method procedurally integrates information from three parts: relevant field content extracted from structured data, specific analysis requirements for the current quality control point obtained from the quality control rule base, and a preset output format template. This standardized prompt word construction method ensures that the large language model operates within clear task boundaries and analysis frameworks each time, thereby significantly improving the accuracy and consistency of judgments. The prompt word template also includes constraints, such as requiring the model to make judgments based on given textual evidence, avoiding arbitrary reasoning, which helps reduce the model's "illusion" phenomenon.
[0028] After all quality control points have been processed, the scoring and report generation process begins. This step first summarizes the "pass / fail" results for all quality control points, and then calculates the score based on the pre-configured weight or deduction value for each point. The weighting is determined based on the importance and risk level of the quality control point; critical quality control points, such as diagnostic accuracy, are assigned higher weights, while general formatting issues are assigned lower weights. The scoring process starts with a maximum of 100 points and deducts from the final score. Next, according to the preset scoring grading standards, the total score is mapped to the corresponding medical record quality level, such as Grade A, Grade B, or Grade C. The grading standards are set with reference to relevant medical quality control standards and expert opinions to ensure the scientific rigor and authority of the evaluation results.
[0029] Finally, a detailed quality control report is generated, including the overall score and grade, a list of all identified defects, and the specific cause and location information for each defect. During report generation, the system categorizes and organizes defects, grouping them according to severity and type for easy identification of key issues. For each defect, the report clearly indicates its location and provides improvement suggestions. These reports are presented in a clear and easy-to-understand format to clinicians and quality control personnel, providing direct evidence for quality improvement. The reports also support multiple export formats for easy transfer and use across different systems.
[0030] The entire implementation process also includes comprehensive data persistence. All original medical record data, structured transformation results, quality control judgment process records, and final scoring reports are properly stored. In particular, for calls to large language models, the system fully saves the input prompts and the original responses returned by the model for each call, ensuring that the entire quality control process is completely traceable and auditable. This feature is crucial for establishing clinical trust and meeting regulatory requirements. Data storage employs a hierarchical design, with hot data stored on high-speed storage devices for real-time querying, while cold data is archived in a lower-cost storage system.
[0031] To ensure the continued effectiveness of the method, a comprehensive update and maintenance mechanism has been established. The quality control rule base is regularly updated according to the latest medical guidelines and clinical practices, and the prompt word templates are continuously optimized based on usage results. Simultaneously, the system collects feedback during use to improve algorithms and processes. This continuous improvement mechanism ensures that the method can adapt to the ever-evolving medical practices.
[0032] The implementation of this method also takes into account the specific needs of different application scenarios. For primary healthcare institutions, the focus can be on checking basic quality control points; for tertiary hospitals, more stringent and detailed quality control standards can be used. Furthermore, the method supports specialty customization, allowing different departments to use specific sets of quality control points based on their own characteristics. This flexibility enables the method to adapt to diverse application environments.
[0033] Through the synergistic effect of the above steps, the method of this invention achieves comprehensive and multi-level intelligent inspection of medical record quality. In a pilot application at a tertiary hospital, testing was conducted on 10,000 inpatient medical records (see Table 1). For quality control points applicable to this path, the pre-processing rules successfully filtered out 75% of scenarios, avoiding unnecessary large model calls and reducing overall computational resource consumption by approximately 60%, while ensuring the accuracy of quality control. This demonstrates that the present invention achieves high-quality quality control while also possessing excellent operational efficiency and economy. It is evident that this method achieves automatic quality control coverage of all hospital medical records, improving quality control efficiency several times compared to traditional manual methods, while maintaining a high level of accuracy. In particular, through the design of a hybrid path, while ensuring the depth of content-based quality control, the number of calls to large language models is controlled within a reasonable range, achieving a balance between quality control effectiveness and cost.
[0034] Table 1: Comparison of the effects of different medical record quality control methods Furthermore, this method exhibits excellent scalability. As medical quality control requirements continue to increase and new quality control needs emerge, new requirements can be quickly supported simply by adding the corresponding quality control point definitions and judgment rules to the quality control rule base, without requiring large-scale modifications to the core system. This design significantly enhances the method's adaptability and lifespan.
[0035] In terms of security, this method strictly adheres to medical data security standards, and all data processing is conducted in a controlled environment to ensure full protection of patient privacy. Simultaneously, the system has established a comprehensive access control system, ensuring that users with different roles can only access functions and data within their authorized scope.
[0036] In summary, the intelligent medical record quality control method proposed in this invention, based on a hybrid engine of rules and large language models, firstly transforms the diverse and formatted original medical record texts into standardized structured data objects through post-structuring processing technology. This preprocessing method provides a high-quality, machine-readable input foundation for all subsequent quality control stages, fundamentally ensuring the accuracy and consistency of quality control analysis.
[0037] Building upon this foundation, the innovative hybrid quality control engine, through its intelligent scheduling logic, adaptively selects pure rule-based, pure large language model-based, or rule-pre-based hybrid quality control paths for different quality control points. Among these, the rule-pre-based hybrid path is particularly crucial. It performs rapid initial screening using pre-set deterministic rules, and only activates the more costly large language model for deep semantic analysis when the rule judgment fails. This layered strategy ensures that inherent quality control problems that traditional systems cannot identify are discovered, while greatly optimizing the allocation of computing resources and achieving a balance between quality control depth and operational efficiency.
[0038] In addition, when calling the large language model, the system adopts a dynamic prompt word construction method to programmatically integrate structured data, quality control instructions and output templates. This technique effectively constrains the output range of the large language model, significantly improves the accuracy, consistency and interpretability of the analysis results, and overcomes the "black box" and "illusion" risks of pure large language model solutions.
[0039] Ultimately, the synergistic effect of these technologies enables the present invention to achieve automated and in-depth quality control of the entire medical record, covering not only formal inspections but also performing intrinsic logical analysis. Its modular and configurable design also endows the system with excellent flexibility and regional expansion capabilities, providing reliable technical support for improving the overall level of medical quality management.
[0040] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.
[0041] Furthermore, in this invention, descriptions involving terms such as "first," "second," and "a" are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.
[0042] In this invention, unless otherwise explicitly specified and limited, the terms "connection," "fixed," etc., should be interpreted broadly. For example, "fixed" can mean a fixed connection, a detachable connection, or an integral part; it can mean a mechanical connection or an electrical connection; it can mean a direct connection or an indirect connection through an intermediate medium; it can mean the internal communication of two components or the interaction between two components, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.
[0043] Furthermore, the technical solutions of the various embodiments of the present invention can be combined with each other, but only if they are feasible for those skilled in the art. If the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.
Claims
1. An intelligent medical record quality control method based on a hybrid engine of rules and large language models, characterized in that, Includes the following steps: S1: Obtain raw electronic medical record data from the medical information system; S2: Perform post-structuring processing on the original electronic medical record data, converting the unstructured or semi-structured text in the original electronic medical record data into a unified, machine-readable structured data object. S3: Traverse multiple quality control points in the preset quality control rule base, and for each quality control point, adaptively select and execute the corresponding quality control path based on the transformed structured data object; S4: Based on the quality control results of all quality control points, calculate the score and generate a quality control report; The quality control paths include pure rule-based quality control paths, pure large language model quality control paths, and hybrid quality control paths with rules at the forefront.
2. The intelligent medical record quality control method based on a hybrid engine of rules and large language models as described in claim 1, characterized in that, In step S2, the post-structuring process includes: S21: Load the corresponding parsing module based on the text type in the original electronic medical record data; S22: Use natural language processing technology to divide the raw electronic medical record data into multiple logical modules; S23: Extract and semantically normalize the key fields in each logical module to generate a structured data object containing key-value pairs.
3. The intelligent medical record quality control method based on a hybrid engine of rules and large language models as described in claim 2, characterized in that, The semantic normalization process includes identifying negative statements in the original electronic medical record data of the current logic module and converting the negative statements into specific identifiers.
4. The intelligent medical record quality control method based on a rule-based and large language model hybrid engine as described in claim 1, characterized in that, The pure rule-based quality control path directly performs logical judgments on structured data objects based on preset deterministic rules. The deterministic rules include at least one of the following: field integrity verification, field format and validity verification, logical consistency verification between fields, and entry timeliness verification, as well as fast content filtering based on keywords or regular expressions.
5. The intelligent medical record quality control method based on a hybrid engine of rules and large language models as described in claim 4, characterized in that, The pure large language model quality control path performs in-depth semantic analysis and logical reasoning-based quality control on structured data objects by calling the large language model.
6. The intelligent medical record quality control method based on a hybrid engine of rules and large language models as described in claim 5, characterized in that, The rule-based hybrid quality control path includes the following steps: The structured data object is initially assessed for quality control based on pre-defined deterministic rules. If the preliminary quality control assessment result is that the quality control has passed, then the conclusion that the quality control has passed will be output directly and the process at the current quality control point will be terminated. If the initial quality control assessment results in a failure, a large language model is invoked to perform deep semantic analysis on the structured data object, and the final quality control conclusion is obtained based on the output of the large language model.
7. The intelligent medical record quality control method based on a hybrid engine of rules and large language models as described in claim 6, characterized in that, When calling the large language model in the execution of the pure large language model quality control path or the rule-preceding hybrid quality control path, a dynamic prompt word construction method is adopted, which specifically includes the following steps: Based on the current quality control points, extract the corresponding prompt word templates from the quality control rule base; Fill the relevant fields from the structured data into the prompt word template to generate dynamic prompt words specifically for the task; Dynamic prompts are sent to a large language model for analysis.
8. The intelligent medical record quality control method based on a rule-based and large language model hybrid engine as described in claim 7, characterized in that, The prompt word template includes the following information: relevant field data extracted from the structured data object, analysis instruction parameters of the current quality control point obtained from the quality control rule base, and response template used to constrain the output format of the large language model.
9. The intelligent medical record quality control method based on a hybrid engine of rules and large language models as described in claim 1, characterized in that, In step S3, the adaptive selection specifically involves selecting a quality control path based on the quality control mode parameters pre-configured for each quality control point in the quality control rule base.
10. The intelligent medical record quality control method based on a hybrid engine of rules and large language models as described in claim 1, characterized in that, In step S4, the scoring calculation includes: calculating deductions for failed quality control points based on the pre-configured weights or deduction values of the quality control points, and determining the medical record quality level based on the total score after deductions.
Citation Information
Patent Citations
Medical record content quality control method and related device
CN120412870A
Electronic medical record content quality control method and system based on large language model
CN120496867A
Medical record quality control method based on medical record segmentation and scene quality control rule matching, electronic equipment and storage medium
CN120544770A
Big model-based receipt identification method, system and equipment
CN120564217A
Electronic medical record intelligent evaluation method based on complex quality control indexes
CN120564931A
Cited By
Disease quality control marking system based on heterogeneous mixed mode
CN121938531A
Heterogeneous mixed mode-based disease quality control labeling system
CN121938531B
Electronic medical record examination method, system, equipment and medium
CN121983216A
