Medical insurance policy and clinical guide interpretation system based on natural language processing

Abstract

The invention discloses a medical insurance policy and clinical guideline interpretation system based on natural language processing, and relates to the technical field of medical insurance policy interpretation and clinical guideline application, and the system comprises a double-source knowledge graph module which collects and converts data, and constructs a dynamically updated integrated knowledge graph; the bidirectional conflict classification module detects conflicts, classifies the conflicts and outputs a report; the conflict self-healing scheme module generates compliance substitution and other schemes; the multi-scene reasoning adaptation module analyzes a user demand and generates a customized decision; the interpretable interaction optimization module visually displays the basis and collects and feeds back the optimization system; according to the method, the medical insurance and clinical DRG integrated knowledge graph is constructed, accurate detection is performed, medical insurance and clinical conflicts are classified, a self-healing scheme is given, various user scene requirements are analyzed, a customized decision result and a visual display basis are generated, dynamic updating and optimization are performed, experience and efficiency are improved, compliance and efficient resource utilization are promoted, and medical insurance payment reform is supported.

Description

Technical Field

[0001] This invention relates to the field of medical insurance policy interpretation and clinical guideline application technology, specifically a medical insurance policy and clinical guideline interpretation system based on natural language processing. Background Technology

[0002] The medical field is currently undergoing gradual reforms in medical insurance payment methods, with DRG and DIP payment models becoming the mainstream development direction. This reform trend places higher demands on the synergy between medical insurance policies and clinical guidelines. Treatment practices must simultaneously meet both clinical norms and medical insurance compliance standards to balance medical quality and efficient use of medical insurance resources. Medical insurance policies are mostly presented in structured data form, containing core content such as reimbursement rules, payment grouping norms, and deductible standards; clinical guidelines, on the other hand, are primarily unstructured text, covering professional information such as disease treatment pathways, drug indications and contraindications, and surgical indications. With the expanding application of artificial intelligence technologies such as natural language processing and knowledge graphs in the medical field, the industry's demand for automated analysis and integration of these two types of data is increasingly prominent. However, existing technologies have not yet formed an integrated system of medical insurance and clinical knowledge, making it difficult to achieve dynamic updates and deep correlations of data. This results in a lack of systematic technical support for the connection between medical insurance policies and clinical guidelines, failing to provide a comprehensive and accurate knowledge foundation for subsequent conflict detection and decision support.

[0003] In the traditional model, the comparison and conflict resolution between medical insurance policies and clinical guidelines mainly rely on manual work. Staff need to manually consult a large number of policy documents and guideline texts, which not only consumes a lot of time and manpower, but is also prone to incomplete and inaccurate conflict identification due to human judgment bias or information omissions. In the conflict classification and handling stage, the traditional method lacks scientific classification standards, making it impossible to accurately distinguish between conflicts of different natures, and thus difficult to develop targeted solutions. At the same time, traditional technology has not established an effective dynamic knowledge update mechanism. When medical insurance policies are adjusted, clinical guidelines are iterated, or payment standards change, the knowledge system cannot be updated in a timely manner. Decisions based on old information are prone to losing timeliness and accuracy. In addition, the traditional processing method cannot adapt to the needs of different users' scenarios. The output results are mostly general content, which is difficult to meet the personalized needs of different stakeholders such as specialists, patients, and hospital administrators. Moreover, it lacks the interpretability of the decision-making process. Users cannot clearly understand the basis of the results, nor can they optimize the processing process through feedback. The overall processing efficiency and practicality are difficult to meet the actual needs of current medical insurance payment reform and clinical diagnosis and treatment. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a medical insurance policy and clinical guideline interpretation system based on natural language processing. This system integrates structured medical insurance policy and unstructured clinical guideline data through a dual-source knowledge graph module to construct an integrated DRG knowledge graph. The system utilizes a bidirectional conflict classification module to detect and classify conflicts, generates compliant alternatives through a conflict self-healing solution module, and a multi-scenario reasoning adaptation module to customize decision results to meet the needs of different users. An interpretable interactive optimization module displays the decision-making basis in a visual form and collects feedback to continuously optimize the system, ensuring decision transparency and accuracy.

[0005] To address the aforementioned technical problems, this invention provides the following technical solution: a medical insurance policy and clinical guideline interpretation system based on natural language processing, comprising a dual-source knowledge graph module, a bidirectional conflict classification module, a conflict self-healing scheme module, a multi-scenario reasoning adaptation module, and an interpretable interactive optimization module. The dual-source knowledge graph module is used to collect structured medical insurance policy data and unstructured clinical guideline data, and completes data parsing and standardization conversion through natural language processing technology to construct an integrated medical insurance and clinical DRG knowledge graph and establish a dynamic knowledge update mechanism. The bidirectional conflict classification module is used to call the integrated knowledge graph, detect the conflict between medical insurance policies and clinical guidelines through bidirectional semantic verification technology, classify the conflict into three categories: absolute conflict, conditional conflict, and cost conflict, and output a structured conflict identification report. The conflict self-healing solution module is used to receive conflict identification reports, perform differentiated self-healing operations for different conflict types, and generate a list of compliant alternative solution exception application paths or cost-effective solutions. The multi-scenario reasoning and adaptation module is used to parse users' natural language needs, combine an integrated knowledge graph and a self-healing solution, and generate customized decision results that adapt to different user scenarios through rule reasoning and machine learning. The interpretable interactive optimization module is used to visualize the basis of the entire decision-making process, provide a two-way interactive interface, collect user feedback, and optimize the knowledge graph and reasoning model in reverse.

[0006] Furthermore, the structured medical insurance policy data collected by the dual-source knowledge graph module specifically includes the national unified medical insurance catalog, regionally differentiated reimbursement rules, deductible standards, DRG payment grouping specifications, DIP payment grouping specifications, and disease cost range quotas; the unstructured clinical guideline data specifically includes disease diagnosis and treatment pathways, drug indications, drug contraindications, surgical operation indications, and efficacy evaluation index systems.

[0007] Furthermore, the dual-source knowledge graph module specifically includes: defining three core entities—medical insurance, clinical, and DRG—and their diverse relationships, constructing an integrated knowledge graph with an entity-relationship-entity triple structure; and a dynamic update mechanism that captures updated information from authoritative data sources through daily scheduled inspections, using an incremental update method to update only changed nodes, with the node weight calculation formula during the update process as follows: ,in For the first The updated weights of each node. The initial weights of the nodes, For feedback adjustment coefficient, To provide user feedback on node satisfaction, different versions of data are marked with timestamps, supporting historical version backtracking queries, and updates are verified through knowledge verification.

[0008] Furthermore, the bidirectional conflict classification module specifically includes: establishing a dual-source knowledge comparison dataset containing dimensions of treatment compliance, population matching, and cost adaptability by calling an integrated knowledge graph; simultaneously detecting two types of conflicts using bidirectional semantic verification technology: those recommended by clinical guidelines but restricted by medical insurance policies, and those permitted by medical insurance policies but not recommended by clinical guidelines; and calculating the conflict risk level after conflict classification, using the following formula: ,in For conflict risk level, 3> For low risk, 7> ≥3 indicates medium risk, 10≥ ≥7 indicates high risk. For conflict type coefficients, For the influence range coefficient, For the degree of influence, , , The weighting coefficients are used to determine the output structured identification report, which includes conflict type, core nodes, related knowledge, and risk level.

[0009] Furthermore, the conflict self-healing solution module specifically includes: matching self-healing strategies based on conflict type, screening clinically equivalent and medical insurance compliant alternatives for absolute conflicts, generating exception application paths for conditional conflicts, and screening cost-effective solutions for cost-effectiveness conflicts; the formula for calculating the cost-effectiveness score of alternative solutions is: ,in Rate it for value for money. For clinical efficacy matching, The actual cost of the solution, V represents the maximum payment limit for DRG / DIP, and V represents the reimbursement rate for medical insurance. The weighting coefficients are used to output a list of candidate solutions, sorted by score, and each solution's core parameters and implementation recommendations are marked.

[0010] Furthermore, the conditional conflict self-healing solution module specifically includes extracting the patient's clinical information, such as disease classification, complication diagnosis results, previous treatment records, age, and physical condition assessment results, and matching them one by one with the conditions of the medical insurance exception clause. If there is a complete match, an application material list is generated, including a disease diagnosis certificate, a complication-specific examination report, a summary of previous treatment medical records, a doctor's recommendation letter, the patient's medical insurance participation certificate, and an exception application approval form, as well as approval process nodes for department head review, hospital medical insurance office filing, and local medical insurance bureau approval, with the estimated time for each node marked. If there is a partial match, the required supplementary clinical supporting materials are clearly marked. If there is no match, a medical insurance-compliant alternative treatment plan is recommended.

[0011] Furthermore, the cost conflict self-healing solution module specifically includes using efficacy similarity assessment technology to search for equivalent alternative treatment items based on the dimensions of consistency of diagnosis and treatment principles, matching degree of indications, overlap of efficacy indicators, and no difference in contraindications. After cost verification, it is confirmed that the total sum of the drug procurement price, consumable bid price, surgical operation fee standard, and examination item fee standard of the candidate solution is within the corresponding DRG or DIP payment range. The weighted sum score is calculated by weighting the clinical efficacy matching degree (40%), medical insurance reimbursement ratio (30%), hospital cost saving margin (20%), and patient co-payment amount (10%). The solutions are then sorted from high to low to form a list of high cost-effective solutions.

[0012] Furthermore, the user scenarios of the multi-scenario reasoning adaptation module specifically include compliance verification of specialist doctors' treatment plans, compliance consultation of general practitioners' cross-specialty treatment plans, medical insurance reimbursement inquiries for outpatients, medical insurance reimbursement planning for long-term treatment plans of inpatients, compliance spot checks of batch treatment items by hospital medical insurance management specialists, and optimization of profitable diseases for DRG or DIP by medical institution operation and management personnel. The customized decision results are as follows: the doctor's end output includes a compliance report and a multi-plan comparison table containing basic plan information, medical insurance compliance status judgment results, clinical guideline reference chapters, DRG or DIP cost adaptation analysis, conflict risk warnings, alternative plan suggestions, etc.; the patient's end outputs reimbursement calculation results and plan explanations in a simplified manner; and the hospital's end outputs a cost control analysis report containing treatment plan cost details, DRG or DIP payment range comparison, expected profit or loss amount, cost optimization space, and a list of compliance risk points. All results are linked to knowledge graph nodes to ensure traceability.

[0013] Furthermore, the visualization of the interpretable interactive optimization module specifically includes: displaying the associated links generated by patient information conflict identification and self-healing treatment decisions through a node-connected graph, with each node labeled with the entity name and type, and the connection labeled with the relationship type; displaying the conflict classification logic, self-healing operation steps, and reasoning decision-making path through a step-by-step flowchart, with key steps labeled with their sources; comparing the efficacy indicators, reimbursement ratios, and cost amounts of different plans through bar charts, and displaying the proportion of reimbursement amount and out-of-pocket amount through pie charts; and presenting the policy provisions and clinical guidelines through a combination of excerpts from the original text and popular interpretations.

[0014] Furthermore, the dynamic optimization mechanism of the explainable interactive optimization module specifically includes daily scheduled inspections of official release channels to capture medical insurance policy revision notices, clinical guideline version update announcements, and DRG or DIP payment standard adjustment documents. Incremental update technology is used to analyze only the changed parts and update the graph nodes. Data from different versions is marked with timestamps to support historical version backtracking queries. Monthly, doctors' adoption marks, non-adoption reason selections, and custom opinions are collected. Patients' satisfaction star ratings, opinions and suggestions, and frequently asked questions are collected. Hospitals' batch verification results are fed back and optimization suggestions are submitted. Based on the feedback data, missing knowledge in the graph is supplemented, and error information is corrected. The scenario weight parameters of the inference model are adjusted, and the solution recommendation logic for high complaint rate scenarios is optimized. The conflict classification rule base and self-healing strategy base are updated, and the application path template and alternative solution matching rules are improved.

[0015] Compared with existing technologies, this natural language processing-based system for interpreting medical insurance policies and clinical guidelines has the following advantages: I. This invention integrates structured medical insurance policy data with unstructured clinical guideline data to construct an integrated knowledge graph of medical insurance and clinical DRGs and establishes a dynamic update mechanism to ensure the timeliness and completeness of the knowledge system. It relies on bidirectional semantic verification technology to achieve comprehensive detection of two types of conflicts. Through scientific classification and risk level assessment, it accurately locates the mismatch between medical insurance and clinical practice, and provides differentiated self-healing solutions for different conflict types. It achieves intelligent generation of compliant alternatives, exception applications, or cost-effective solutions, breaking the limitations of the traditional model of fragmented medical insurance and clinical knowledge and inefficient conflict handling. This helps doctors quickly clarify the boundaries of medical insurance compliance, reduce the risk of violations, and provide patients with more suitable treatment and reimbursement solutions. It significantly improves the efficiency of medical insurance policy implementation and the scientific nature of clinical decision-making, and promotes the standardized development of medical insurance and clinical collaboration.

[0016] Second, this invention generates customized decision-making results by accurately analyzing the core scenario needs of various users, including specialists, patients, and hospital administrators. It adapts to diverse demands such as compliance verification, reimbursement inquiries, and cost optimization. Through visual presentation and easy-to-understand explanations, it clearly presents the basis for the entire decision-making process, lowering the barrier to understanding policies and guidelines. It achieves two-way interaction and feedback collection, capturing updates to official policies and guidelines through a dynamic update mechanism. Combined with user feedback, it continuously optimizes the knowledge graph and reasoning model, constantly improving conflict resolution rules and solution recommendation logic. This not only enhances the user experience and decision-making efficiency for all stakeholders but also promotes compliance in medical treatment and the efficient use of medical insurance resources. It achieves an organic balance between clinical efficacy assurance, medical insurance compliance requirements, and cost control goals, providing intelligent support for diagnosis and management under the background of medical insurance payment reform.

[0017] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0019] Figure 1 A flowchart of a medical insurance policy and clinical guideline interpretation system based on natural language processing; Figure 2 This is a schematic diagram of data transmission for each module of a medical insurance policy and clinical guideline interpretation system based on natural language processing. Detailed Implementation

[0020] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0021] Example 1: Outpatient medical insurance reimbursement inquiry scenario.

[0022] The dual-source knowledge graph module initiates a data processing flow, collecting structured medical insurance policy data such as the national unified medical insurance catalog, regionally differentiated reimbursement rules, deductible standards, DRG / DIP payment grouping specifications, and disease-specific cost range quotas, as well as unstructured clinical guideline data such as disease treatment pathways, drug indications, drug contraindications, surgical operation indications, and efficacy evaluation indicator systems. Natural language processing technology is used to complete data parsing and standardization, unifying the scattered data from medical insurance policies and clinical guidelines into a unified format, laying the foundation for subsequent cross-data type association analysis. Simultaneously, three core entities—medical insurance, clinical, and DRG—and their diverse relationships are defined, constructing an integrated medical insurance, clinical, and DRG knowledge graph with an entity-relationship-entity triple structure. This clearly presents the logical relationships between the three types of data, providing a clear knowledge framework for conflict detection and reasoning. Furthermore, a dynamic update mechanism is established, capturing updated information from authoritative data sources through daily scheduled inspections. Incremental updates are used, updating only changed nodes to reduce invalid data processing and improve update efficiency. The node weight calculation formula is used during the update process. The calculation formula is as follows: ,in For the first The updated weights of each node. The initial weights of the nodes, For feedback adjustment coefficient, Based on user feedback on node satisfaction, node importance is adjusted to better align the graph content with actual usage needs. Timestamps are used to mark different versions of data, supporting historical version lookups and facilitating users' access to past information. After updates, knowledge verification ensures data accuracy and compliance, preventing erroneous data from interfering with subsequent decision-making processes. Figure 1 As shown.

[0023] The bidirectional conflict classification module utilizes the pre-built integrated knowledge graph to establish a dual-source knowledge comparison dataset encompassing dimensions of treatment compliance, population matching, and cost suitability. This dataset covers conflict detection dimensions from three core levels: treatment rationality, applicable population matching, and cost suitability, ensuring no conflict is missed. Bidirectional semantic verification technology is employed to simultaneously detect two types of conflicts: those recommended by clinical guidelines but restricted by medical insurance policies, and those permitted by medical insurance policies but not recommended by clinical guidelines. This achieves comprehensive capture of the two-way contradiction between "recommendation and restriction" and "permission and non-recommendation." Conflicts are categorized into three types: absolute conflict, conditional conflict, and cost conflict, providing accurate matching criteria for subsequent differentiated self-healing solutions. Subsequently, the risk level of each type of conflict is calculated using a conflict risk level calculation formula, which is as follows: ,in According to the conflict risk level, For conflict type coefficients, For the influence range coefficient, For the degree of influence, , , Using weighted coefficients, the system clearly distinguishes between low risk (3 > R), medium risk (7 > R ≥ 3), and high risk (10 ≥ R ≥ 7) through quantitative values, helping users prioritize the handling of high-risk conflicts. The final output is a structured conflict identification report containing conflict type, core nodes, related knowledge, and risk level, allowing users to intuitively grasp the core information of the conflict and the priority of handling it, providing clear guidance for subsequent self-healing operations.

[0024] The conflict self-healing module receives structured conflict identification reports and matches corresponding self-healing strategies based on conflict type: For absolute conflicts, it screens clinically equivalent and medical insurance compliant alternatives, ensuring compliance with medical insurance policies while meeting clinical treatment objectives and avoiding illegal reimbursement; for conditional conflicts, it extracts disease classification, complication diagnosis results, previous treatment records, age, and physical condition assessment results from the patient's clinical information and matches them one by one with the conditions of medical insurance exception clauses. If a complete match is found, it generates an application material list including a disease diagnosis certificate, complication-specific examination report, previous treatment medical record summary, doctor's recommendation letter, patient's medical insurance participation certificate, and exception application approval form, as well as an approval flow involving department head review, hospital medical insurance office filing, and local medical insurance bureau approval. The system provides users with a complete application path and timeline. For partially matched cases, it clearly indicates the required supplementary clinical documentation, reducing the cost of repeated document preparation. For non-matched cases, it recommends compliant alternative treatment options to ensure uninterrupted and compliant treatment processes. To address cost conflicts, it uses efficacy similarity assessment technology to search for equivalent alternative treatments based on consistency of treatment principles, indication matching, overlap of efficacy indicators, and lack of difference in contraindications. Cost verification confirms that the total cost of candidate solutions—including drug procurement prices, consumable bid prices, surgical procedure fees, and examination fees—is within the corresponding DRG or DIP payment range, preventing cost overruns. Simultaneously, a cost-effectiveness scoring formula is used to comprehensively evaluate candidate solutions. The formula is as follows: ,in Rate it for value for money. For clinical efficacy matching, The actual cost of the solution, DRG / DIP payment cap For medical insurance reimbursement ratio, Using weighted coefficients, the system sorts the solutions from highest to lowest score to create a list of cost-effective options. It also marks the core parameters and implementation suggestions for each solution, helping users quickly select the optimal solution that balances efficacy, cost, and reimbursement.

[0025] The multi-scenario reasoning and adaptation module analyzes the natural language needs of outpatients, accurately capturing their core demands regarding reimbursement scope and amount calculation. Combining an integrated knowledge graph and a conflict self-healing solution, it generates customized decision-making results adapted to outpatient scenarios through rule-based reasoning and machine learning. Specifically, it outputs reimbursement calculation results and solution explanations in a simple and easy-to-understand manner, avoiding misunderstandings caused by technical jargon. All results are linked to knowledge graph nodes, allowing users to trace the corresponding policy clauses and guidelines, enhancing the credibility of the results.

[0026] The interpretable interactive optimization module visually presents the entire decision-making process: a node-connected graph shows the interconnected links between patient information, conflict identification, self-healing processing, and decision generation, with each node labeled with its entity name and type, and connections labeled with their relationship types, allowing users to clearly see the logical connections between each stage of the decision-making process; a step-by-step flowchart illustrates the conflict classification logic, self-healing operation steps, and reasoning decision-making paths, with key steps labeled with policy provisions and clinical guidelines to help users understand the rationality of the decision-making process; bar charts compare the efficacy indicators, reimbursement ratios, and costs of different alternatives, and pie charts show the ratio of reimbursement to out-of-pocket expenses, reducing the difficulty for users to compare options with intuitive charts; relevant medical insurance policies and clinical guidelines are presented through a combination of excerpts from the original text and simplified explanations, allowing users to easily understand the decision-making basis; and a two-way interactive interface is provided to collect patient satisfaction ratings, suggestions, and frequently asked questions, providing data support for subsequent knowledge graph improvement and reasoning model optimization, making the system more aligned with the needs of outpatients.

[0027] Example 2: A scenario where hospital medical insurance management specialists conduct random inspections to ensure the compliance of batch medical treatment items.

[0028] The dual-source knowledge graph module performs batch data collection and processing, acquiring structured medical insurance policy data such as the national unified medical insurance catalog, regionally differentiated reimbursement rules, DRG / DIP payment grouping specifications, and disease-specific cost range quotas, as well as unstructured clinical guideline data such as disease treatment pathways, surgical indications, drug indications and contraindications, and efficacy evaluation index systems. Natural language processing technology is used to perform data parsing and standardization, unifying the format of batch policy and guideline data to meet the data source needs of specialists for simultaneous sampling of a large number of treatment items. The module defines three core entities—medical insurance, clinical, and DRG—and their diverse relationships, constructing a medical insurance-clinical triplet structure. The integrated DRG knowledge graph clearly presents the relationships between various diagnostic and treatment items and medical insurance policies and clinical guidelines, providing knowledge support for rapid matching and conflict detection during batch sampling. Relying on a dynamic update mechanism, it regularly checks official release channels daily to capture updates such as medical insurance policy revision notices, clinical guideline version update announcements, and DRG / DIP payment standard adjustment documents, ensuring that the graph data is synchronized with the latest policies and guidelines and avoiding deviations in sampling results due to data lag. An incremental update method is used, updating only changed nodes, reducing the processing pressure of batch data updates and improving update efficiency. The node weight calculation formula is used during the update process. The node weight calculation formula during the update process is as follows: ,in For the first The updated weights of each node. The initial weights of the nodes, For feedback adjustment coefficient, To assess user satisfaction with each node, the importance of nodes is adjusted based on user feedback, making the graph more aligned with the actual business needs of hospitals for batch sampling. Timestamps are used to mark different versions of data, supporting historical version lookups and facilitating comparisons of compliance changes in medical procedures under different policies. After updates, knowledge validation ensures the accuracy and consistency of batch data, guaranteeing the reliability of subsequent batch sampling results. Figure 2 As shown.

[0029] The bidirectional conflict classification module utilizes an integrated knowledge graph to establish a dual-source knowledge comparison dataset for batch treatment projects to be inspected by specialists. This dataset includes dimensions of treatment compliance, population matching, and cost suitability. It comprehensively covers the key compliance detection points for batch projects from three levels: whether the treatment conforms to guidelines, whether it matches the applicable medical insurance population, and whether the cost is suitable for the payment standard. Bidirectional semantic verification technology is used to simultaneously detect two types of conflicts: those recommended by clinical guidelines but restricted by medical insurance policies, and those permitted by medical insurance policies but not recommended by clinical guidelines. This achieves efficient capture of bidirectional contradictions in batch projects, avoiding omissions caused by single-type conflict detection. Conflicts in each treatment project are classified into three categories: absolute conflict, conditional conflict, and cost conflict, providing a classification basis for the accurate matching of subsequent batch self-healing solutions. The conflict risk level calculation formula is used to calculate the conflict risk level of each treatment project. The formula is as follows: ,in For conflict risk level, 3> For low risk, 7> ≥3 indicates medium risk, 10≥ ≥7 indicates high risk. For conflict type coefficients, For the influence range coefficient, For the degree of influence, Using weighted coefficients, low, medium, and high risks are distinguished by quantitative values, helping specialists quickly filter out high-risk conflict projects from a batch of projects and prioritize compliance rectification; the final output is a structured conflict identification report containing the conflict type, core nodes, related knowledge, and risk level of each diagnosis and treatment project. Specialists can use the report to grasp the compliance status of each project in batches, greatly improving the efficiency of sampling inspection.

[0030] The conflict self-healing module receives batch structured conflict identification reports and performs differentiated self-healing operations for each treatment item according to the conflict type: For absolute conflict items, it screens clinically equivalent and medical insurance compliant alternative treatment plans to ensure that the items comply with medical insurance policies without affecting clinical treatment effects, providing a feasible direction for hospitals to adjust non-compliant items in batches; For conditional conflict items, it extracts the clinical information of the patient group corresponding to each item and matches it one by one with the conditions of medical insurance exception clauses. If a complete match is found, it generates an application material list for each item, including disease diagnosis certificates, special examination reports of complications, etc., as well as an approval process involving department head review, hospital medical insurance office filing, and local medical insurance bureau approval. The system uses nodes to facilitate batch processing of exception applications by specialists. For partially matched applications, it clearly marks the required supplementary clinical documentation to prevent specialists from overlooking crucial materials during batch processing. For non-matched applications, it recommends compliant alternatives to ensure project compliance. For cost-conflicting projects, it uses efficacy similarity assessment technology to search for equivalent alternative treatments for each project based on dimensions such as consistency of treatment principles and indication matching. Cost verification confirms that the total cost of each candidate solution falls within the corresponding DRG or DIP payment range, preventing cost overruns in batch projects. Simultaneously, it uses a cost-effectiveness scoring formula to comprehensively evaluate each candidate solution. The cost-effectiveness scoring formula for alternative solutions is as follows: ,in Rate it for value for money. For clinical efficacy matching, The actual cost of the solution, DRG / DIP payment cap For medical insurance reimbursement ratio, Using weighted coefficients, a list of cost-effective solutions is generated by sorting the scores from high to low, and the core parameters and implementation suggestions of each solution are marked. Specialists can use the list to select the best rectification solution for conflicting projects in batches, reducing the cost of repeated decision-making.

[0031] The multi-scenario reasoning and adaptation module analyzes the batch sampling needs of hospital medical insurance management specialists, accurately understanding their core demands for compliance assessment and cost analysis of batch projects. Combining an integrated knowledge graph and conflict self-healing solution, it generates customized decision results adapted to the scenario through rule reasoning and machine learning. The specific output includes a cost control analysis report containing detailed cost breakdowns for each treatment plan, a comparison of DRG / DIP payment ranges, expected profit or loss amounts, cost optimization potential, and a list of compliance risk points. All results are linked to knowledge graph nodes, allowing specialists to trace the policy basis and guidelines corresponding to each conclusion in the report, ensuring that the report data is verifiable and trustworthy, and providing data support for hospitals to adjust treatment projects in batches and control cost risks.

[0032] The interpretable interactive optimization module visualizes the entire decision-making process for batch treatment projects: It uses a node-connected graph to show the conflict identification, self-healing process, and decision generation links for each project. Nodes are labeled with entity names and types, and connections are labeled with relationship types. Specialists can view the logical connections between decision-making stages for each project in batches. A step-by-step flowchart illustrates the batch conflict classification logic, self-healing operation steps, and reasoning decision paths. Key steps are labeled with corresponding medical insurance policy clauses and clinical guidelines, helping specialists understand the compliance and rationality of the batch decision-making process. Bar charts compare the efficacy indicators and reimbursement ratios of different treatment plans. Cost figures are presented using pie charts to show the percentage of reimbursement and out-of-pocket expenses for each item, helping specialists quickly grasp the differences in plans and cost structures for batch projects in an intuitive graphical format. Relevant medical insurance policies and clinical guidelines are presented through a combination of excerpts from the original text and simplified explanations, reducing the difficulty for specialists to understand the professional terms. At the same time, a two-way interactive interface is provided to collect feedback on batch verification results and optimization suggestions from hospital medical insurance management specialists. This feedback data will be used to improve the conflict rules in the knowledge graph and adjust the scenario weight parameters of the inference model, making the system more accurate and efficient in subsequent batch sampling requests.

[0033] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A system for interpreting medical insurance policies and clinical guidelines based on natural language processing, characterized in that, It includes a dual-source knowledge graph module, a bidirectional conflict classification module, a conflict self-healing solution module, a multi-scenario reasoning adaptation module, and an interpretable interaction optimization module. The dual-source knowledge graph module is used to collect structured medical insurance policy data and unstructured clinical guideline data, and completes data parsing and standardization conversion through natural language processing technology to construct an integrated medical insurance and clinical DRG knowledge graph and establish a dynamic knowledge update mechanism. The bidirectional conflict classification module is used to call the integrated knowledge graph, detect the conflict between medical insurance policies and clinical guidelines through bidirectional semantic verification technology, classify the conflict into three categories: absolute conflict, conditional conflict, and cost conflict, and output a structured conflict identification report. The conflict self-healing solution module is used to receive conflict identification reports, perform differentiated self-healing operations for different conflict types, and generate a list of compliant alternative solution exception application paths or cost-effective solutions. The multi-scenario reasoning and adaptation module is used to parse users' natural language needs, combine an integrated knowledge graph and a self-healing solution, and generate customized decision results that adapt to different user scenarios through rule reasoning and machine learning. The interpretable interactive optimization module is used to visualize the basis of the entire decision-making process, provide a two-way interactive interface, collect user feedback, and optimize the knowledge graph and reasoning model in reverse.

2. The medical insurance policy and clinical guideline interpretation system based on natural language processing according to claim 1, characterized in that, The structured medical insurance policy data collected by the dual-source knowledge graph module specifically includes the national unified medical insurance catalog, regionally differentiated reimbursement rules, deductible standards, DRG payment grouping specifications, DIP payment grouping specifications, and disease cost range quotas; the unstructured clinical guideline data specifically includes disease diagnosis and treatment pathways, drug indications, drug contraindications, surgical operation indications, and efficacy evaluation index systems.

3. The medical insurance policy and clinical guideline interpretation system based on natural language processing according to claim 1, characterized in that, The dual-source knowledge graph module specifically includes: defining three core entities—medical insurance, clinical, and DRG—and their diverse relationships, constructing an integrated knowledge graph with an entity-relationship-entity triple structure; and a dynamic update mechanism that captures updated information from authoritative data sources through daily scheduled inspections, using an incremental update method to update only changed nodes. The node weight calculation formula during the update process is as follows: ,in For the first The updated weights of each node. The initial weights of the nodes, For feedback adjustment coefficient, To provide user feedback on node satisfaction, different versions of data are marked with timestamps, supporting historical version backtracking queries, and updates are verified through knowledge verification.

4. The medical insurance policy and clinical guideline interpretation system based on natural language processing according to claim 1, characterized in that, The bidirectional conflict classification module specifically includes: establishing a dual-source knowledge comparison dataset containing dimensions of treatment compliance, population matching, and cost suitability by calling an integrated knowledge graph; simultaneously detecting two types of conflicts using bidirectional semantic verification technology: those recommended by clinical guidelines but restricted by medical insurance policies, and those permitted by medical insurance policies but not recommended by clinical guidelines; and calculating the conflict risk level after conflict classification, using the following formula: ,in For conflict risk level, 3> For low risk, 7> ≥3 indicates medium risk, 10≥ ≥7 indicates high risk. For conflict type coefficients, For the influence range coefficient, For the degree of influence, , , The weighting coefficients are used to determine the output structured identification report, which includes conflict type, core nodes, related knowledge, and risk level.

5. The medical insurance policy and clinical guideline interpretation system based on natural language processing according to claim 1, characterized in that, The conflict self-healing solution module specifically includes: matching self-healing strategies based on conflict type, screening clinically equivalent and medical insurance compliant alternatives for absolute conflicts, generating exception application paths for conditional conflicts, and screening cost-effective solutions for cost-effectiveness conflicts; the formula for calculating the cost-effectiveness score of alternative solutions is as follows: ,in Rate it for value for money. For clinical efficacy matching, The actual cost of the solution, V represents the maximum payment limit for DRG / DIP, and V represents the reimbursement rate for medical insurance. The weighting coefficients are used to output a list of candidate solutions, sorted by score, and each solution's core parameters and implementation recommendations are marked.

6. The medical insurance policy and clinical guideline interpretation system based on natural language processing according to claim 1, characterized in that, The self-healing conflict resolution module specifically handles the following: extracting the patient's clinical information, including disease classification, complication diagnosis results, previous treatment records, age, and physical condition assessment results, and matching them one by one with the conditions of the medical insurance exception clause; a complete match generates a list of application materials including a disease diagnosis certificate, complication-specific examination report, previous treatment medical record summary, doctor's recommendation letter, patient's medical insurance participation certificate, and exception application approval form, as well as approval process nodes for department head review, hospital medical insurance office filing, and local medical insurance bureau approval, with the estimated time for each node marked; a partial match clearly marks the clinical supporting materials that need to be supplemented; and a mismatch recommends a medical insurance-compliant alternative treatment plan.

7. The medical insurance policy and clinical guideline interpretation system based on natural language processing according to claim 1, characterized in that, The cost conflict self-healing solution module specifically includes using efficacy similarity assessment technology to search for equivalent alternative treatment items based on the dimensions of consistency of diagnosis and treatment principles, matching degree of indications, overlap of efficacy indicators, and no difference in contraindications. After cost verification, it is confirmed that the total sum of the drug procurement price, consumable bid price, surgical operation fee standard, and examination item fee standard of the candidate solution is within the corresponding DRG or DIP payment range. The weighted sum score is calculated by weighting the clinical efficacy matching degree (40%), medical insurance reimbursement ratio (30%), hospital cost saving margin (20%), and patient co-payment amount (10%). The solutions are then sorted from high to low to form a list of cost-effective solutions.

8. The medical insurance policy and clinical guideline interpretation system based on natural language processing according to claim 1, characterized in that, The user scenarios of the multi-scenario reasoning adaptation module specifically include: compliance verification of specialist doctors' treatment plans, compliance consultation of general practitioners for cross-specialty treatment plans, medical insurance reimbursement inquiries for outpatients, medical insurance reimbursement planning for long-term treatment plans of inpatients, compliance spot checks of batch treatment items by hospital medical insurance management specialists, and optimization of profitable diseases for DRG or DIP by medical institution operation and management personnel. The customized decision results are as follows: on the doctor's end, a compliance report and a comparison table of multiple plans are output, including basic information of the plan, medical insurance compliance status judgment results, clinical guideline reference chapters, DRG or DIP cost adaptation analysis, conflict risk warnings, and alternative plan suggestions; on the patient's end, a reimbursement calculation result and plan explanation are presented in a popular way; and on the hospital's end, a cost control analysis report is output, including treatment plan cost details, comparison of DRG or DIP payment ranges, expected profit or loss amount, cost optimization space, and a list of compliance risk points. All results are linked to knowledge graph nodes to ensure traceability.

9. The medical insurance policy and clinical guideline interpretation system based on natural language processing according to claim 1, characterized in that, The visualization of the explainable interactive optimization module specifically includes: displaying the associated links generated by patient information conflict identification and self-healing treatment decisions through a node-connected graph, with each node labeled with the entity name and type, and the connection labeled with the relationship type; displaying the conflict classification logic, self-healing operation steps, and reasoning decision-making path through a step-by-step flowchart, with key steps labeled with their sources; comparing the efficacy indicators, reimbursement ratios, and cost amounts of different plans through bar charts, and displaying the proportion of reimbursement amount and out-of-pocket amount through pie charts; and presenting the policy provisions and clinical guidelines through a combination of excerpts from the original text and popular interpretations.

10. The medical insurance policy and clinical guideline interpretation system based on natural language processing according to claim 1, characterized in that, The dynamic optimization mechanism of the explainable interactive optimization module specifically includes daily scheduled inspections of official release channels to capture medical insurance policy revision notices, clinical guideline version update announcements, and DRG or DIP payment standard adjustment documents. Incremental update technology is used to analyze only the changed parts and update the graph nodes. Data from different versions is marked with timestamps to support historical version backtracking queries. Monthly, the module collects doctors' adoption marks, non-adoption reason selections, and custom opinions; patients' satisfaction star ratings, opinions and suggestions, and frequently asked questions; and hospitals' batch verification results feedback and optimization suggestions submission. Based on the feedback data, the module supplements missing knowledge in the graph, corrects errors, adjusts the scenario weight parameters of the inference model, optimizes the solution recommendation logic for high complaint rate scenarios, updates the conflict classification rule base and self-healing strategy base, and improves the application path template and alternative solution matching rules.

Images