An outpatient medical insurance auditing method and system based on a deterministic path and a cascade amplification mechanism
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- WENZHOU LOGOS TECHNOLOGY CO LTD
- Filing Date
- 2026-06-04
- Publication Date
- 2026-08-07
AI Technical Summary
（1）审核规则僵化，无法适应复杂多变的临床诊疗场景，导致误判率高；
Smart Images
Figure CN122529903A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical insurance data processing technology, specifically to an outpatient medical insurance review method and system based on deterministic paths and cascading amplification mechanisms, applicable to scenarios such as the rationality review of outpatient medical insurance prescriptions, detection of drug sales rebate behavior, and risk prevention and control of medical insurance funds. Background Technology
[0002] With the expansion of basic medical insurance coverage, outpatient medical insurance costs have been rising year by year, increasing the pressure on medical insurance fund supervision. Current outpatient medical insurance audits mainly rely on post-event manual spot checks or automated audit systems based on static rules, which presents the following technical problems: (1) The review rules are rigid and cannot adapt to complex and ever-changing clinical diagnosis and treatment scenarios, resulting in a high misjudgment rate; (2) The lack of continuous modeling of patients’ visits across time and departments makes it difficult to detect hidden drug sales kickbacks and transfer of benefits. (3) The review process is independent of each other and lacks a hierarchical judgment logic, which leads to the omission of high-risk prescriptions; (4) The existing system is unable to quantify the core risk of “pharmacoeconomics influencing doctors’ prescribing tendencies”.
[0003] Therefore, there is an urgent need for an intelligent auditing method and system that can perform path modeling of outpatient prescriptions and automatically identify high-risk prescriptions and drug salesperson kickback behavior through a cascading amplification mechanism. Summary of the Invention
[0004] The purpose of this invention is to provide an outpatient medical insurance review method and system based on deterministic path and cascade amplification mechanism to solve the problems mentioned in the background art.
[0005] To achieve the above objectives, the present invention provides the following technical solution: An outpatient medical insurance review method based on deterministic paths and cascading amplification mechanisms includes the following steps: Step S1, Data Access Step: Obtain outpatient visit data from the hospital HIS system and medical insurance settlement system, and construct a patient visit dataset; the visit data includes at least: patient ID, visit time, department code, doctor ID, diagnosis code, prescription drug list, drug unit price, drug quantity, examination and test items and cost amount; Step S2, Deterministic Path Construction Step: Based on the medical visit dataset, construct a unique "deterministic path" for each outpatient prescription; the deterministic path is a linear sequence composed of several ordered nodes, and the node types include at least: registration node, diagnosis node, prescription node, examination node, and payment node; each node is cascaded in chronological order to form a complete behavioral link from the initiation of medical visit to the settlement of payment; Step S3, Cascade Amplification Coefficient Calculation Steps: Define the basic risk weight of each node, and introduce the cascade amplification coefficient based on the causal dependency between nodes; when the risk value of the current node is passed to the next node, an amplification operation is performed so that the comprehensive risk value at the end of the path can reflect the cumulative anomaly of the entire path. Step S4, CVI risk index calculation steps: Based on the risk values of each node after cascading amplification, calculate the CVI risk index of the prescription; the CVI risk index is a normalized value with a value range of [0,1], and the larger the value, the higher the risk of medical insurance violation; Step S5, Drug Rebate Detection Step: When the drugs appearing in the prescription match the preset "High-Risk Drug Reseller-Associated Drug List", the drug rebate detection sub-process is triggered; the drug rebate suspicion coefficient is calculated by combining the doctor's prescription frequency, the proportion of high-priced drugs selected in the same indication, and the deviation of the drug's use among similar doctors. Step S6, Review Decision Step: Input the CVI risk index and the suspected drug salesperson kickback coefficient into the review decision model and output the review results; the review results include three categories: automatic approval, automatic interception, and manual review.
[0006] Furthermore, in step S2, the method for constructing the deterministic path includes: The starting point T0 of the path is the registration time of the patient's current visit; The path termination point Tn is the time when the doctor issues the prescription; Extract all medical behavior events generated between T0 and Tn, sort them by timestamp, and generate an event sequence; The event sequence is denoised to remove events irrelevant to the square root decision, thus obtaining a deterministic path.
[0007] Furthermore, in step S3, the formula for calculating the cascade amplification factor is: A_i = A_{i-1} × (1 + α · w_i) Where A_i is the amplified risk value of the i-th node, A_0 is the basic risk value of the starting node of the path, α is the cascade amplification factor (ranging from 0.1 to 1.0, dynamically set by the system according to the risk level of the department), and w_i is the basic risk weight of the i-th node.
[0008] Furthermore, in step S4, the formula for calculating the CVI risk index is as follows: CVI = (1 / N) × Σ_{i=1}^{N} A_i × I(A_i>θ) Where N is the total number of nodes in the path, and I(·) is an indicator function that takes the value 1 when A_i>θ (risk threshold) and 0 otherwise; the closer the CVI value is to 1, the more high-risk nodes there are in the prescription path.
[0009] Furthermore, in step S5, the calculation of the suspected kickback coefficient for pharmaceutical sales representatives comprehensively considers the following three dimensions: (a) Drug dimension: Whether the drug is on the list of drugs frequently promoted by pharmaceutical sales representatives, and the clinical necessity score of the drug; (b) Physician dimension: the historical frequency of the physician prescribing the drug, and the probability deviation of the physician choosing the drug for the same indication; (c) Patient dimension: Does the patient exhibit an abnormal pattern of repeatedly visiting different doctors to obtain the same medication?
[0010] This invention also provides an outpatient medical insurance review system based on deterministic paths and cascading amplification mechanisms, comprising: The data access module is used to acquire outpatient visit data; The path building module is used to build a deterministic path for each prescription; The cascade amplification calculation module is used to perform cascade amplification calculations on the risks of nodes in the path; The CVI calculation module is used to calculate the CVI risk index for each prescription. The pharmaceutical kickback detection module is used to detect whether there is any suspicion of pharmaceutical kickbacks in prescriptions; The review decision module is used to output review results based on the CVI index and suspicion coefficient; The results output module is used to push the review results to the terminals of medical insurance reviewers and the hospital's HIS system.
[0011] The beneficial effects of this invention are: First, by constructing a "deterministic path," the outpatient process is transformed into a computable directed linear sequence, enabling the review system to understand the causal links of medical behavior rather than judging individual prescription items in isolation. Second, the introduction of a "cascading amplification mechanism" enables the abnormal risks of front-end nodes to be transmitted and amplified along the path, effectively solving the technical defects of "single-point judgment and ignoring association" in traditional rule review. Third, a dedicated sub-process for detecting drug sales kickbacks was designed, which combines the CVI risk index with the drug sales kickback suspicion coefficient to accurately identify hidden benefit transfer behaviors and protect the security of the medical insurance fund. Fourth, the CVI index output by the system is a continuous variable, which can provide medical insurance regulatory authorities with a quantitative basis for risk assessment, support graded review according to risk level, and greatly improve review efficiency. Attached Figure Description
[0012] Figure 1 This is an overall flowchart of the outpatient medical insurance review method described in this invention.
[0013] Figure 2 These are the five key characteristics of a deterministic path model.
[0014] Figure 3 The advantage of a deterministic path, or a unique path, over multiple paths.
[0015] Figure 4 This refers to the technical architecture of the entire system.
[0016] Figure 5 This is a schematic diagram illustrating the risk transmission of a cascading amplification mechanism. Detailed Implementation
[0017] The present invention will now be described in further detail with reference to the accompanying drawings.
[0018] This embodiment 1 discloses an outpatient medical insurance review method based on deterministic path and cascading amplification mechanism, which is applied to the intelligent review scenario of outpatient prescriptions in municipal medical insurance bureaus.
[0019] In the specific implementation of step S1, the system incrementally synchronizes outpatient data daily from the HIS systems of various hospitals within the jurisdiction via the standard HL7 interface, and simultaneously obtains the corresponding prescription medical insurance reimbursement data from the medical insurance settlement system. The data fields must include at least: patient ID number (anonymized hash value), date of visit, hospital code, department name, doctor's employee number, primary diagnostic ICD-10 code, generic name of the drug, drug dosage form, specifications, unit price, quantity, usage and dosage, and examination item code.
[0020] In step S2, a deterministic path is constructed based on a single complete outpatient visit for each patient. For example, patient Zhang San registers at the Internal Medicine Department of Hospital A at 8:30 AM on March 15, 2025, sees the doctor at 9:00 AM, is diagnosed with "upper respiratory tract infection," and is prescribed "Amoxicillin capsules 0.25g × 24 capsules × 2 boxes" and a "complete blood count test." Payment is made at 9:20 AM, and medication is picked up at 9:30 AM. After extracting the above events, the system constructs the following deterministic path: Path ID: P_20250315_A Hospital_Zhang San_001 Node 1 (T0=8:30): Registration node, Department = Internal Medicine Node 2 (T1=9:00): Diagnosis node, diagnosis = upper respiratory tract infection Node 3 (T2=9:05): Check node, item = complete blood count Node 4 (T3=9:10): Root extraction node, Drug = Amoxicillin Capsules Node 5 (T4=9:20): Payment node, total fee = 186.5 yuan In step S3, the cascade amplification factor is calculated using the following specific parameters: The basic risk value A0 of the starting node is set to 0.1 (default low risk); the basic risk weight w_i of each node is set according to the node type: registration node w=0.05, diagnosis node w=0.10, examination node w=0.15, prescription node w=0.30, payment node w=0.20; the cascade amplification factor α is set according to the department's risk level: internal medicine α=0.3, orthopedics α=0.6, oncology α=0.8. Taking the above path as an example, the prescription node contains "amoxicillin," a first-line antibiotic, which is a reasonable prescription; therefore, this node has no additional risk score, A4 = 0.1×(1+0.3×0.30) = 0.109.
[0021] In step S4, the threshold for calculating the CVI risk index is θ = 0.5. If all Ai in the path are less than 0.5, then CVI ≈ 0, the system automatically determines it as a low-risk prescription, and the approval is granted.
[0022] Example 2: This embodiment, based on Embodiment 1, further describes the specific implementation of the drug sales rebate detection steps.
[0023] The system has a built-in "List of Drugs Associated with High-Risk Pharmaceutical Representation". This list is maintained in the following ways: regularly screening cases involving pharmaceutical representation kickbacks from the medical insurance bureau's audit case database, extracting the drugs involved, and forming an initial list; and adding drugs with a sales expense ratio exceeding 30% and abnormally concentrated doctor prescriptions to the list based on drug sales data.
[0024] When a prescription contains a drug that matches this catalog, the system initiates a suspected drug sales rebate calculation. The specific calculation steps are as follows: (1) Obtain the prescription data of the drug in this hospital in the past 12 months, and count the frequency of each doctor prescribing the drug and the proportion of the doctor's total prescriptions; (2) Calculate the deviation D between the current rate of doctors prescribing this drug and the average rate of doctors in the same department throughout the hospital; (3) If D>2.0 (i.e., the doctor's prescription rate is more than twice the average level), then a first-level suspicion marker is triggered; (4) Further search whether the patient has been prescribed the same drug in multiple departments in the past 3 months. If so, trigger a level 2 suspicion flag. (5) Based on the combined markers at each level, output the drug sales kickback suspicion coefficient R, R∈{0,1,2}, which correspond to low suspicion, medium suspicion and high suspicion respectively.
[0025] In step S6, the review decision rules are as follows: If CVI < 0.3 and R = 0, the audit result is "automatic approval"; If CVI ≥ 0.3 or R ≥ 1, the review result is "manual review". The system will automatically push the prescription to the medical insurance reviewer's pending queue and highlight the CVI index and the suspected mark. If CVI ≥ 0.7 and R = 2, the audit result is "automatic interception". The system will suspend the medical insurance settlement of the prescription and generate an audit work order to be pushed to the medical insurance bureau's supervisory personnel.
[0026] Example 3: This embodiment discloses an outpatient medical insurance review system that implements the above method. The system is deployed on the medical insurance bureau's private cloud server and adopts a microservice architecture.
[0027] The data access module is developed based on the Spring Boot framework and supports HL7 and FHIR standard medical data interfaces as well as the medical insurance bureau's dedicated API interface; The path building module uses the graph database Neo4j to store deterministic paths, with patients as vertices and visit events as edges, supporting efficient path querying and traversal. The cascaded amplification calculation module and the CVI calculation module are deployed as Python services, using NumPy for vectorized calculations, with a single prescription review time of <50ms. The drug sales rebate detection module has a built-in drug catalog management submodule, which supports regular updates and version management of the catalog; The review decision module has a built-in rule engine, Drools, which supports dynamic configuration and hot updates of review rules. The results output module supports sending the audit results back to the hospital's HIS system in standard JSON format, thus achieving closed-loop notification of audit results.
[0028] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for outpatient medical insurance review based on deterministic paths and cascading amplification mechanisms, characterized in that, Includes the following steps: Data access steps: Obtain outpatient visit data; Deterministic path construction steps: Construct a deterministic path for each outpatient prescription, consisting of several ordered nodes, with the nodes arranged in a cascaded order according to time. Steps for calculating the cascade amplification factor: Define the basic risk weight of each node, and calculate the cascade amplification of risk based on the causal dependency between nodes; CVI Risk Index Calculation Steps: Calculate the CVI risk index based on the amplified node risk value; Steps for detecting drug sales rebates: When a prescription drug is included in the list of high-risk drugs associated with drug sales representatives, calculate the drug sales rebate suspicion coefficient; Review decision-making process: Based on the CVI risk index and suspicion coefficient, the review results are output as either automatically approved, automatically blocked, or manually reviewed.
2. The method according to claim 1, characterized in that, The node types of the deterministic path include at least: registration node, diagnosis node, prescription node, inspection node, and payment node.
3. The method according to claim 1, characterized in that, The formula for calculating the cascade amplification factor is: A_i = A_{i-1} × (1 + α · w_i), where α is the cascade amplification factor and w_i is the basic risk weight of the i-th node.
4. The method according to claim 1, characterized in that, The CVI risk index is calculated as follows: CVI = (1 / N) × Σ_{i=1}^{N} A_i × I(A_i > θ), where θ is a preset risk threshold.
5. The method according to claim 1, characterized in that, The calculation of the suspected drug sales rebate coefficient takes into account the characteristics of three dimensions: drug dimension, doctor dimension, and patient dimension.
6. An outpatient medical insurance review system based on deterministic paths and cascading amplification mechanisms, characterized in that, include: The system includes a data access module, a path construction module, a cascade amplification calculation module, a CVI calculation module, a pharmaceutical sales rebate detection module, an audit and decision-making module, and a result output module.
7. The system according to claim 6, characterized in that, The path construction module uses a graph database to store deterministic paths.
8. The system according to claim 6, characterized in that, The review decision module has a built-in rule engine that supports dynamic configuration of review rules.