CN-121998779-A - Intelligent medical insurance compliance examination method and system based on graph retrieval enhancement generation

CN121998779ACN 121998779 ACN121998779 ACN 121998779ACN-121998779-A

Abstract

The application provides an intelligent medical insurance compliance examination method and system based on graph retrieval enhancement generation, relates to the field of medical informatization and artificial intelligence, and solves the technical problem that the prior art cannot effectively identify implicit medical insurance violations with context dependency and structural combinability. The method comprises the steps of constructing a compliance knowledge graph, extracting clinical entities from diagnosis and treatment records, mapping the clinical entities to corresponding nodes in the compliance knowledge graph, performing multi-jump graph retrieval in the compliance knowledge graph based on the mapped clinical entities as query starting points to obtain policy rule subgraphs related to current diagnosis and treatment behaviors, generating a compliance examination result through structural reasoning, constructing an illegal risk propagation model based on historical auditing data, performing illegal risk assessment on the diagnosis and treatment behaviors to obtain risk early warning grades, and generating an auditing priority queue for a plurality of diagnosis and treatment records to be examined based on the illegal confidence and the risk early warning grades. The method is used in the medical insurance compliance examination process.

Inventors

LI YOUTAO
LIU JIE
WANG HAOJIE
Guo Husong

Assignees

安徽晶奇网络科技股份有限公司

Dates

Publication Date: 20260508
Application Date: 20260206

Claims (10)

1. An intelligent medical insurance compliance examination method based on graph retrieval enhancement generation is characterized by comprising the following steps: Constructing a compliance knowledge graph containing medical insurance policy rules and logic constraint relations thereof; extracting a clinical entity from the diagnosis and treatment record, mapping the clinical entity to a corresponding node in the compliance knowledge graph, and marking the clinical entity as a target mapping node, wherein the clinical entity comprises a charging project name, a surgical operation name or a medical consumable name; based on the mapped clinical entity as a query starting point, performing multi-jump map retrieval in the compliance knowledge map to obtain a policy rule subgraph related to the current diagnosis and treatment behavior; Inputting the diagnosis and treatment record and the policy rule subgraph into a generated inference model, and generating a compliance examination result through structural inference, wherein the compliance examination result comprises violation judgment, policy basis, correction suggestion and violation confidence; constructing an illegal risk propagation model based on historical auditing data, and carrying out illegal risk assessment on the diagnosis and treatment behaviors by utilizing the legal knowledge graph to obtain a risk early warning grade; and generating an audit priority queue for a plurality of to-be-inspected diagnosis and treatment records based on the violation confidence and the risk early warning level.
2. The intelligent medical insurance compliance review method based on graph retrieval enhancement generation according to claim 1, wherein the method for constructing the compliance knowledge graph comprises the following steps: Performing multi-granularity segmentation on the medical insurance policy document to generate a hierarchical text unit covering the total description, the chapters, the clauses and the sub-items; Based on the hierarchical text units, constructing corresponding violation scene nodes, and identifying logic constraint relations among the nodes through a medical insurance semantic pattern library, wherein the logic constraint relations comprise one or more of item inclusion relations, exclusionary content exclusion relations and charge exclusion relations which cannot be simultaneously carried out; And constructing and obtaining an initial compliance knowledge graph based on the violation scene nodes and the logic constraint relation thereof.
3. The method for determining the target mapping node of the extracted clinical entity according to claim 1, wherein the method for determining the target mapping node of the extracted clinical entity comprises the following steps: Generating a plurality of candidate knowledge map nodes for the clinical entity based on a pre-constructed multi-source coding mapping index library, wherein the multi-source coding mapping index library integrates medical insurance directory coding, ICD surgical coding, UDI consumable coding and local alias dictionary; Acquiring context information of the clinical entity in a diagnosis and treatment record, wherein the context information comprises patient diagnosis, an execution department, an associated charging item and operation time; Carrying out semantic compatibility evaluation according to the context information and policy applicable conditions associated with each candidate knowledge graph node to obtain compatibility scores of each candidate node, wherein the policy applicable conditions comprise one or more of medical institution level restriction, necessary accompanying diagnosis, mutex charging items or limited use scenes; And selecting a target mapping node based on the compatibility score, and outputting structured mapping information containing a mapping result and a confidence level.
4. The method for intelligent medical insurance compliance review based on graph retrieval enhancement generation according to claim 3, wherein the method for analyzing the compatibility score of each candidate node comprises the steps of: encoding the context information into a context feature vector; resolving policy applicable conditions associated with each candidate knowledge graph node into a structured constraint set; distributing preset weights for different types of policy applicable conditions, wherein the weight of the forbidden or restrictive conditions is higher than that of the recommended conditions; and calculating a compatibility score through a weighted semantic matching function based on the context feature vector and the structured constraint set, wherein the weighted semantic matching function applies a significant negative penalty when any hard constraint is not met, and accumulates positive contributions to the met constraint according to the weight of the weighted semantic matching function.
5. The method for intelligently inspecting medical insurance compliance generated based on graph retrieval enhancement according to claim 1, wherein the method for acquiring the policy rule subgraph related to the current diagnosis and treatment behavior comprises the following steps: generating query intention expression by taking the mapped clinical entity as an initial query node and combining the current diagnosis and treatment context; In each jump graph traversal, dynamically determining a path expansion priority according to semantic relativity of neighbor nodes and query intention representations and medical insurance rule effectiveness grades corresponding to connecting edges, wherein the medical insurance rule effectiveness grades comprise forbidden rules, limiting rules and explanatory rules, and edges corresponding to the forbidden rules share the highest priority in path expansion; and performing multi-hop retrieval based on the path expansion priority to obtain an initial policy rule subgraph, performing semantic redundancy clipping on the initial policy rule subgraph, and removing node branches with similarity lower than a preset threshold value with the query intention representation to obtain the policy rule subgraph.
6. The intelligent medical insurance compliance review method based on graph retrieval enhancement generation of claim 1, wherein the generation of compliance review results includes: constructing a chain type thinking reasoning template comprising five stages of fact restoration, rule matching, logic verification, rule violation judgment and correction suggestion; converting the policy rule subgraph into a structured rule statement set with unique identification, and embedding the structured rule statement set as a rule anchor point into a prompt word context of a generated reasoning model; The generated reasoning model sequentially fills the contents of each stage according to the chain thinking reasoning template, and each reasoning step must refer to at least one unique identifier of the rule anchor point and output a compliance examination result.
7. The method for intelligently inspecting medical insurance compliance based on graph retrieval enhancement generation according to claim 1, wherein the construction of the violation risk propagation model comprises the following steps: Acquiring historical medical insurance audit data, wherein the historical medical insurance audit data comprises historical audit labels for all charging projects, and the historical audit labels represent violation conditions of the projects; Taking a knowledge graph node mapped by a charging item with a corresponding history audit label as a positive sample and taking a node with a corresponding history audit label as a non-illegal sample as a negative sample; Constructing a graph neural network on the compliance knowledge graph, wherein the graph neural network is configured to output a violation risk score to each node, and dynamically distributes propagation weights according to the medical insurance rule types of the connecting edges in the message transmission process, wherein the edges corresponding to the forbidden rules are given higher propagation weights than the edges of the limiting rules and the explanatory rules; and performing supervision training on the graph neural network based on the positive sample and the negative sample to ensure that the violation risk score is consistent with the history audit label, and training to obtain a violation risk propagation model.
8. The method for inspecting compliance with intelligent medical insurance based on graph retrieval enhancement generation according to claim 1, wherein the performing the risk assessment of the diagnosis and treatment behavior includes: Inputting the mapped clinical entity as an initial activation node to a trained violation risk propagation model, and executing multi-jump risk signal diffusion on the compliance knowledge graph to obtain a violation risk score of each relevant node; Constructing a risk activation subgraph based on the violation risk score, clustering connected components of the risk activation subgraph, and identifying high-risk item combinations; Calculating the combined risk intensity of each high-risk item combination, wherein the combined risk intensity is obtained by weighting and fusing the average offending risk score of the nodes in the combination and the density of the sub-graph structure; mapping the combined risk intensity to a preset threshold interval, and determining a risk early warning level.
9. The method for intelligently inspecting medical insurance compliance based on graph retrieval enhancement generation according to claim 1, wherein the generating an audit priority queue for a plurality of diagnosis and treatment records to be inspected comprises: Obtaining the violation confidence and risk early warning levels respectively corresponding to a plurality of diagnosis records to be examined; mapping the risk early-warning level into a preset value, and carrying out weighted fusion on the violation confidence coefficient and the mapped risk early-warning level based on a dynamic weight to obtain an audit priority score of each diagnosis and treatment record; Sequencing the plurality of diagnosis and treatment records according to the auditing priority score, and generating an auditing priority queue by combining with the preset auditing resource capacity, wherein the auditing priority queue is pushed to a medical insurance auditing management system to guide manual review.
10. An intelligent medical insurance compliance inspection system based on graph retrieval enhancement generation, which is operated based on the intelligent medical insurance compliance inspection method based on graph retrieval enhancement generation according to any one of claims 1 to 9, and is characterized by comprising a compliance knowledge graph construction module, an entity mapping module, a policy rule subgraph retrieval module, a compliance inspection module, a risk assessment module and a priority scheduling module; The compliance knowledge graph construction module is used for constructing a compliance knowledge graph containing medical insurance policy rules and logic constraint relations thereof; The entity mapping module is used for extracting clinical entities from the diagnosis and treatment record, mapping the clinical entities to corresponding nodes in the compliance knowledge graph and marking the clinical entities as target mapping nodes; The policy rule subgraph retrieval module is used for executing multi-jump graph retrieval in the compliance knowledge graph based on the mapped clinical entity as a query starting point to obtain a policy rule subgraph related to the current diagnosis and treatment behavior; the compliance examination module is used for inputting the diagnosis and treatment record and the policy rule subgraph into a generated reasoning model and generating a compliance examination result through structural reasoning; the risk assessment module is used for constructing an illegal risk propagation model based on historical auditing data, and carrying out illegal risk assessment on the diagnosis and treatment behaviors by utilizing the compliance knowledge graph to obtain a risk early warning grade; And the priority scheduling module is used for generating an audit priority queue for the plurality of to-be-examined diagnosis and treatment records based on the violation confidence and the risk early warning level.

Description

Intelligent medical insurance compliance examination method and system based on graph retrieval enhancement generation Technical Field The application relates to the field of medical informatization and artificial intelligence, in particular to an intelligent medical insurance compliance examination method and system based on graph retrieval enhancement generation. Background With the continuous perfection of medical security systems and the continuous expansion of the scale of medical insurance funds, medical insurance fund safety supervision faces an increasingly serious challenge. In recent years, medical insurance audits identify abnormal behavior through big data analysis or machine learning models. However, the prior art suffers from a significant drawback in that implicit violations with context dependencies and structural combinability cannot be effectively identified. In particular, many medical insurance violations do not result from obvious errors of a single item, but rather items that are multiple surface compliance violate restrictive or forbidden logical constraints in policies under certain medical scenarios. Because the existing method lacks structural modeling of logical relations among policy rules, and the generated AI model is not deeply coupled with authoritative policy knowledge during reasoning, complex violations are not detected due to rule fragmentation, or a non-basis examination conclusion is generated due to 'illusion', and severe requirements of accuracy, interpretability and compliance of medical insurance audit check are difficult to meet. The defect becomes a core technical bottleneck for restricting the improvement of the supervision efficiency of the intelligent medical insurance, and a new paradigm capable of deeply fusing the structured policy knowledge and the generative reasoning capability is needed to solve, so that the application provides an intelligent medical insurance compliance examination method and system based on graph retrieval enhancement generation. Disclosure of Invention The application provides an intelligent medical insurance compliance examination method and system based on graph retrieval enhancement generation, which solve the technical problem that the prior art cannot effectively identify implicit medical insurance violation behaviors with context dependency and structural combinability. In order to achieve the above purpose, the application adopts the following technical scheme: in a first aspect, an intelligent medical insurance compliance review method based on graph retrieval enhancement generation is provided, including: Constructing a compliance knowledge graph containing medical insurance policy rules and logic constraint relations thereof; extracting a clinical entity from the diagnosis and treatment record, mapping the clinical entity to a corresponding node in the compliance knowledge graph, and marking the clinical entity as a target mapping node, wherein the clinical entity comprises a charging project name, a surgical operation name or a medical consumable name; based on the mapped clinical entity as a query starting point, performing multi-jump map retrieval in the compliance knowledge map to obtain a policy rule subgraph related to the current diagnosis and treatment behavior; Inputting the diagnosis and treatment record and the policy rule subgraph into a generated inference model, and generating a compliance examination result through structural inference, wherein the compliance examination result comprises violation judgment, policy basis, correction suggestion and violation confidence; constructing an illegal risk propagation model based on historical auditing data, and carrying out illegal risk assessment on the diagnosis and treatment behaviors by utilizing the legal knowledge graph to obtain a risk early warning grade; and generating an audit priority queue for a plurality of to-be-inspected diagnosis and treatment records based on the violation confidence and the risk early warning level. According to the technical scheme, in the intelligent medical insurance compliance examination method based on graph retrieval enhancement generation, a logically consistent medical insurance compliance knowledge graph is constructed, policy regulations are converted into structured rule nodes and constraint edges with efficacy grades, clinical entities in diagnosis and treatment records are precisely mapped, multi-jump and intention-aware graph retrieval is carried out by taking the clinical entities as starting points, policy subgraphs related to current behaviors are dynamically obtained, the subgraphs are used as rule anchor points and embedded into generated model prompt words, a driving model carries out structured reasoning according to a five-stage chain template, specific rule IDs are forcefully referenced, traceability and auditability are ensured, meanwhile, a graph neural network risk propagation model based on historical audi