A medical insurance instant settlement method and system based on a claim case intelligent agent

Abstract

The application discloses a medical insurance instant settlement method and system based on a claim case intelligent agent, and mainly relates to the technical field of insurance technology. The method comprises the following steps: obtaining and preprocessing claim-related data, and converting unstructured medical text into structured data; performing case complexity evaluation on the preprocessed structured data, and dynamically routing the case to a corresponding intelligent agent for processing according to the evaluation result; the intelligent agent automatically matches the insurance liability according to an electronic insurance policy model, and outputs a liability determination result and an estimated compensation amount; automatically inputting medical expense bills and cross-verifying the bills with medical record information; triggering an instant settlement process according to the liability matching result and the bill verification result, and completing payment of the claim settlement. The application has the beneficial effect of realizing intelligent distribution and instant settlement of claim cases, solving the problems of low efficiency of manual review, slow compensation, and realizing risk identification and automatic verification.

Description

Technical Field

[0001] This invention relates to the field of insurance technology, specifically a method and system for real-time settlement of medical insurance based on intelligent agents for claims cases. Background Technology

[0002] Currently, medical insurance claims processing by insurance companies primarily relies on manual procedures: after an incident, the customer submits hospital records, medical expense invoices, and other documents to the insurance company. Claims personnel meticulously review each page of the medical records, extracting key information such as diagnosis, surgery, and medication. This extracted medical information is then manually compared with the policy's coverage details to verify whether it falls within the agreed-upon scope of liability or whether any exclusions have been triggered. Simultaneously, the invoice amounts are manually entered. After review, the finance department is notified to pay the claim. In this process, some insurance companies have introduced OCR (Optical Character Recognition) technology to assist with invoice entry or employ simple rule engines for initial screening. For example, Shanghai Meixin Health Technology Co., Ltd. has published "A Method and Device for Reviewing Insurance Claim Materials Based on a Large Language Model" (CN118350773A), proposing to automatically review claim materials using a large language model.

[0003] However, the aforementioned existing technologies still have the following shortcomings: First, manually reviewing a complete medical record typically takes 20 to 30 minutes, which is inefficient. The fundamental reason is that medical record text is unstructured data, lacking the technical means to accurately understand the content of the medical record and automatically associate it with the policy terms. Second, matching insurance liability relies on human experience, with an exclusion rate of approximately 8% to 10%. The fundamental reason is the lack of an automated mechanism for structured comparison and conflict detection between medical record diagnostic information and the policy's liability scope and exclusion clauses. Third, document entry still relies on manual verification, with an error rate of approximately 5%. The fundamental reason is that existing OCR systems only perform text recognition and lack the ability to cross-verify with medical record information. Fourth, customers typically have to wait T+1 to T+3 business days from submitting documents to receiving claims payments, making instant settlement impossible. The fundamental reason is that existing systems adopt a single path of "payment only after all reviews are completed," lacking a tiered processing mechanism and a prepayment mechanism.

[0004] In particular, existing AI solutions, exemplified by CN118350773A, use the same large language model to process all cases. However, standard cases account for more than 80% of the total claims, and using a large model for these cases results in a significant waste of computing power and processing delays. Complex cases lack more targeted deep reasoning capabilities. At the same time, existing AI solutions do not involve innovation in the settlement process; the review and payment are still sequential processes, failing to achieve flexible settlement with "payment first, review later".

[0005] Therefore, there is an urgent need for a medical insurance instant settlement method and system based on intelligent agents for claims cases to solve the above problems. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for real-time settlement of medical insurance based on intelligent agents for claims cases. It realizes intelligent diversion and real-time settlement of claims cases, solves the problems of low efficiency and slow payment of manual review, and also realizes risk identification and automatic verification.

[0007] To achieve the above objectives, the present invention employs the following technical solution: On the one hand, the present invention provides a method for real-time settlement of medical insurance based on a claims case intelligent agent, comprising the following steps: Step S1: Obtain claims-related data and preprocess it, converting unstructured medical text into structured data; Step S2: Evaluate the case complexity of the preprocessed structured data, and dynamically route the cases to the corresponding agents for processing based on the evaluation results; Step S3: The intelligent agent automatically matches insurance liabilities based on the electronic policy model and outputs the liability determination result and the estimated compensation amount; Step S4: Automatically enter medical expense receipts and cross-verify them with medical record information; Step S5: Based on the liability matching results and invoice verification results, trigger the instant settlement process to complete the payment of the claim.

[0008] Preferably, in step S1, the acquisition and preprocessing of claim-related data specifically includes: The medical insurance settlement data of customers can be obtained in real time through the medical insurance interface. The medical insurance settlement data includes total cost, medical insurance payment amount, personal co-payment amount and details of Category C items. Obtain the inpatient medical records, discharge summaries, medical expense invoices and detailed expense lists uploaded by customers through self-service terminals at the counter or mobile terminals; Perform OCR text recognition on the acquired unstructured medical text; A medical named entity recognition model is used to extract key fields from the recognition results. The key fields include disease diagnosis code, surgical procedure name, length of hospital stay, drug name and out-of-pocket payment identifier, and main examination and testing items. The extracted key fields are integrated into structured data.

[0009] Preferably, step S2 specifically includes: The case complexity assessment unit automatically determines the complexity of the current case based on the following assessment dimensions: Diagnostic type dimension: Common diseases and single-disease diagnoses are characterized by low complexity, while rare diseases and diagnoses of multiple diseases are characterized by high complexity. Treatment method dimension: Conservative treatment and simple surgery are characterized by low complexity, while complex surgery and multiple hospitalizations are characterized by high complexity. Cost amount dimension: features below the preset threshold are considered low-complexity features, and features above the preset threshold are considered high-complexity features; Past medical history dimension: No past medical history is a low complexity feature, while past medical history or chronic disease history is a high complexity feature; Medication usage dimension: All drugs are covered by medical insurance catalog, which is a low complexity feature; drugs that include self-paid drugs or imported drugs are a high complexity feature. Based on the evaluation results, if the case is determined to be a low-complexity standard case, it is dynamically routed to a fast intelligent agent for processing. The fast intelligent agent adopts a lightweight model with a response time of less than 1 second. If the case is determined to be a high-complexity case, it is dynamically routed to a deep intelligent agent for processing. The deep intelligent agent uses a large language model combined with a medical knowledge graph for semantic understanding and reasoning.

[0010] Preferably, step S3 specifically includes: The system reads the digital model of the customer's insurance policy, which is pre-set with parameters such as coverage scope, waiting period days, deductible amount, payout ratio, exclusion list, and maximum coverage amount. The disease diagnosis code extracted from the medical record is compared with the policy coverage to determine whether the current incident falls under the insurance liability stipulated in the policy. Automatically detect whether there are any exclusions, including whether the incident occurred within the waiting period, whether the diagnosis is on the list of excluded diseases, and whether the treatment is out-of-pocket or not covered by the insurance. Assuming liability is established, the estimated compensation amount is calculated and expressed as follows: ; in, To estimate the amount of compensation, This refers to the portion of the total medical expenses that falls within the scope of insurance coverage. For deductible, This refers to the compensation ratio; Output the liability determination result and the estimated compensation amount.

[0011] Preferably, in step S3, the automatic matching of insurance liabilities further includes a medical record-policy conflict detection step: Step S31: Extract chronic disease records from past medical history and present medical history from the structured data of the medical records. The chronic diseases include hypertension, diabetes and heart disease. Step S32: Compare the extracted medical history information with the health declaration information filled in by the customer in the insurance application form item by item; Step S33: If a potential inconsistency is detected, i.e., the medical record contains a specific medical history but the corresponding item on the insurance application is checked as "none", then a risk indicator is output and the case is transferred to the manual review queue first. If no inconsistency is detected, the subsequent processing procedure continues.

[0012] Preferably, step S4 specifically includes: Automatically identify the total amount, medical insurance payment amount, personal cash payment amount, and out-of-pocket item amount in medical expense invoices; Automatically identify each charge item and its corresponding amount in the detailed expense list; The treatment items in the invoice are compared with the treatment plan recorded in the medical record. If it is found that there are chargeable items in the invoice that are not recorded in the medical record, or if there is no corresponding chargeable record in the invoice for the treatment items recorded in the medical record, an abnormal verification mark is output to trigger manual review. Once the verification is approved, an electronic claims list will be automatically generated.

[0013] Preferably, step S5 specifically includes: The automatic full payment process is triggered when the following conditions are met simultaneously: liability is determined to be established, there is no risk indicator and the verification is normal, and the estimated compensation amount is lower than the preset threshold; the system calls the financial interface to pay the full amount of the claim to the customer's bank card. The temporary payment-verification-adjustment closed-loop settlement process is triggered when any of the following situations occur: the estimated compensation amount is higher than the preset threshold, the medical record-policy conflict detection outputs a risk indicator, or the invoice-medical record cross-verification outputs an abnormal indicator. The aforementioned temporary payment-verification-adjustment closed-loop settlement process specifically includes: The amount of temporary payment is calculated as follows: ; in, For temporary payment amount, To estimate the amount of compensation, This is a temporary payment ratio. The range of values ​​is ; Pay the customer in advance according to the temporary payment amount; The case will be sent to the back-end manual review queue for final review. Once the review is completed, if the final compensation amount is higher than the provisional payment amount, the difference will be automatically paid out; if the final compensation amount is lower than the provisional payment amount, a difference refund notice will be generated and the customer will be informed of the refund method through the channels reserved by the customer. The entire process of temporary payment, verification, and adjustment will be recorded and written to the system log.

[0014] On the other hand, the present invention also provides a medical insurance instant settlement system based on a claims case intelligent agent, for implementing the above-described medical insurance instant settlement method based on a claims case intelligent agent, including: The data acquisition module is used to acquire claims-related data and preprocess it, converting unstructured medical text into structured data. The case complexity assessment module is used to assess the case complexity of preprocessed structured data and dynamically route cases to the corresponding intelligent agents for processing based on the assessment results. The intelligent agent processing module includes a fast intelligent agent unit, a deep intelligent agent unit, and a dynamic routing unit. The dynamic routing unit routes low-complexity standard cases to the fast intelligent agent unit and high-complexity cases to the deep intelligent agent unit based on the case complexity assessment results. The liability matching module enables the intelligent agent to automatically match insurance liabilities based on the electronic policy model, and outputs the liability determination result and the estimated compensation amount; the liability matching module also includes a medical record-policy conflict detection submodule, which is used to automatically detect inconsistencies between medical records and insurance disclosure information; The invoice entry module is used to automatically enter medical expense invoices and cross-verify them with medical record information; the invoice entry module includes an invoice-medical record cross-verification subunit. The instant settlement module is used to trigger the instant settlement process based on the liability matching result and the invoice verification result to complete the payment of the claim. The instant settlement module includes a full payment unit and a temporary payment unit. The temporary payment unit is used to make an advance payment according to a preset ratio in high-value or questionable cases, and complete the difference adjustment settlement after background review.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention uses an automatic case complexity assessment and dynamic routing mechanism to divert the majority of standard cases to a fast intelligent agent for processing, achieving a response time in seconds. This fundamentally changes the inefficient traditional method of manually reading medical records page by page.

[0016] 2. This invention uses OCR recognition combined with medical named entity recognition technology, supplemented by a cross-validation mechanism between invoices and medical records, which greatly improves the accuracy of data entry and effectively solves the problems of insufficient accuracy of traditional manual data entry and single OCR recognition.

[0017] 3. This invention automatically determines the scope of liability and detects exclusions through an intelligent agent, and has a built-in medical record-policy conflict detection module that can automatically identify potential inconsistencies between medical records and insurance disclosure information, overcoming the shortcomings of manual review being prone to omissions and the lack of conflict detection mechanisms in single AI models.

[0018] 4. This invention employs a settlement method for high-value or questionable cases: advance payment based on a percentage, followed by refunds or additional payments upon back-end review. This allows customers to receive most of their compensation within a very short time after submitting their documents. This mechanism reduces the claim payment time from the traditional "days" to the "minutes," representing a groundbreaking breakthrough while ensuring risk control. Existing technologies do not offer this differentiated effect.

[0019] 5. This invention uses a two-stage dynamic routing approach, where only complex cases are processed by calling large language models and knowledge graphs for in-depth processing, while standard cases are processed by lightweight and fast intelligent agents. The overall computing power consumption is far lower than that of a single architecture that calls large models for all cases.

[0020] 6. The medical record-policy conflict detection function of this invention can automatically identify adverse selection risks such as failure to disclose pre-existing conditions, with a detection rate far higher than that of traditional manual methods, effectively reducing the probability of subsequent claims disputes. Attached Figure Description

[0021] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a flowchart of the medical record-policy conflict detection process of the present invention; Figure 3 This is a schematic diagram of the system structure of the present invention. Detailed Implementation

[0022] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope defined in this application.

[0023] In this invention, terms such as "upper," "lower," "left," "right," "front," "back," "vertical," "horizontal," "side," and "bottom" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are used only to facilitate the description of the structural relationships of the various components or elements of this invention and do not specifically refer to any component or element in this invention. They should not be construed as limiting the invention.

[0024] Example 1: like Figure 1 As shown, this embodiment provides a method for real-time settlement of medical insurance based on a claims intelligence agent, including the following steps: Step S1: Obtain claims-related data and preprocess it, converting unstructured medical text into structured data; Step S2: Evaluate the case complexity of the preprocessed structured data, and dynamically route the cases to the corresponding agents for processing based on the evaluation results; Step S3: The intelligent agent automatically matches insurance liabilities based on the electronic policy model and outputs the liability determination result and the estimated compensation amount; Step S4: Automatically enter medical expense receipts and cross-verify them with medical record information; Step S5: Based on the liability matching results and invoice verification results, trigger the instant settlement process to complete the payment of the claim.

[0025] This example illustrates the process of a customer applying for medical insurance reimbursement after completing hospitalization for acute appendicitis.

[0026] After discharge, the customer logs into the insurance company's claims service interface via a mobile app, and follows the prompts to take and upload photos of their inpatient medical records, discharge summary, medical expense invoices, and detailed expense lists. Simultaneously, the system obtains the customer's medical insurance settlement data for this visit in real time through the medical insurance interface, including total costs, medical insurance reimbursement amount, out-of-pocket expenses, and details of Category C items. Upon receiving the multi-source data, the data acquisition module first performs OCR text recognition on the uploaded medical records and discharge summary, converting the text in the images into an editable text format. Subsequently, the system uses a medical named entity recognition model to extract key fields from the recognition results. This model, built on a deep learning framework, can accurately identify and extract information such as disease diagnosis codes (ICD-10), surgical procedure names, length of hospital stay, drug names and out-of-pocket expense indicators, and major examination and testing items. For example, in this case of acute appendicitis, the system extracted the diagnostic code K35.9 (acute appendicitis, not specifically identified) from ICD-10, the surgical procedure was appendectomy, the length of hospital stay was 5 days, and the medication list included antibiotics such as cefuroxime sodium. After extraction, the system integrated the above-mentioned scattered key fields into structured data in a unified format, which served as input for subsequent intelligent agent processing steps.

[0027] After data preprocessing, the system enters the case complexity assessment and dynamic routing stage. The case complexity assessment unit receives the preprocessed structured data and automatically determines the complexity of the current case from multiple dimensions. In terms of diagnosis type, common diseases and single-disease diagnoses are classified as low complexity, while rare diseases and multiple diseases combined are classified as high complexity. In terms of treatment method, conservative treatment and simple surgery are classified as low complexity, while complex surgery and multiple hospitalizations are classified as high complexity. In terms of cost, a preset threshold (preferably 5000 yuan in this embodiment) is used as the boundary for judgment. In terms of medical history, the system judges whether the structured data contains past medical history or chronic disease history. In terms of medication, the system determines whether all drugs are covered by medical insurance. For this case of acute appendicitis, the diagnosis was a common disease, the treatment was a simple appendectomy, the estimated cost was less than 5,000 yuan, there was no prior medical history, and all medications were covered by medical insurance. All five dimensions met the low complexity criteria. Therefore, the case complexity assessment unit classified this case as a low-complexity standard case. Based on the assessment results, the dynamic routing unit routed the case to a fast agent for processing. The fast agent uses a lightweight Transformer model with a response time of less than 1 second.

[0028] After receiving the structured data of the case, the rapid intelligent agent begins the automatic matching step of insurance liability. First, it reads the digital model of the customer's policy. This electronic policy model pre-sets the coverage scope, waiting period (30 days for this policy), deductible (100 yuan for this policy), reimbursement ratio (80% for this policy), exclusion list, and maximum coverage parameters. The rapid intelligent agent compares the disease diagnosis code K35.9 extracted from the medical record with the policy's coverage scope, confirming that acute appendicitis falls under the policy's stipulated inpatient medical coverage. Subsequently, the rapid intelligent agent automatically checks for exclusions: comparing the date of the incident with the policy's effective date to confirm that the 30-day waiting period has exceeded; comparing the diagnosis code with the list of excluded diseases to confirm that it is not within the exclusion scope; and comparing the treatment items with the list of out-of-pocket expenses to confirm that appendectomy is a covered treatment item. After all exclusion checks are passed, the process proceeds to the reimbursement calculation step. The rapid intelligent agent calculates the reimbursement amount according to the formula... Calculate the estimated compensation amount, of which To estimate the amount of compensation, This refers to the amount of total medical expenses that falls within the scope of insurance coverage (in this case, it is 4230 yuan). The deductible is 100 yuan. The estimated payout amount is calculated based on the payout ratio (80%). Yuan.

[0029] During the automatic matching of insurance liabilities, a medical record-policy conflict detection step is also performed in parallel. For example... Figure 2 As shown, the medical record-policy conflict detection process specifically includes: extracting chronic disease records from the structured data of the medical records, including chronic diseases such as hypertension, diabetes, and heart disease that require long-term management. In the acute appendicitis case of this embodiment, no chronic disease history records were found in the medical records. The extracted medical history information was compared item by item with the health declaration information filled in by the customer on the insurance application. Both showed no history of chronic diseases, and no inconsistencies were detected. Therefore, this case has no risk indicator, and the subsequent processing procedure continues.

[0030] While the liability matching is completed, the invoice entry module automatically processes the medical expense invoices and detailed expense lists. The system first automatically identifies the total amount (4230 yuan), medical insurance payment amount (1680 yuan), personal cash payment amount (2550 yuan), and out-of-pocket item amount (0 yuan) in the invoice image. Then, it automatically identifies each charge item and its corresponding amount in the detailed expense list, including bed fees, surgery fees, medication fees, and examination fees. After completing the above identification, the invoice entry module performs invoice-medical record cross-validation: comparing the treatment items in the invoice (appendectomy and related medications) with the treatment plan recorded in the medical record, confirming a complete match. No charge items not recorded in the medical record were found in the invoice, nor were any treatment items recorded in the medical record found to have corresponding charge records in the invoice; the cross-validation is normal. After successful verification, the system automatically generates an electronic claim list, eliminating the need for manual entry.

[0031] At this point, both the liability matching result and the invoice verification result were normal, triggering the instant settlement process. In this embodiment, the three conditions were met simultaneously: liability determination was established, there was no risk indicator and the verification was normal, and the estimated compensation amount of 3304 yuan was lower than the preset threshold of 5000 yuan. Therefore, the system triggered the full automatic payment process. The system called the financial interface to transfer the 3304 yuan compensation to the customer's linked bank card account in real time and sent a compensation receipt notification SMS to the customer. From the customer submitting the documents to the compensation arriving in the account, the entire process was completed within 2 minutes.

[0032] Example 2: This example illustrates how the method in Example 1 can be used to handle highly complex cases, using the case of a customer who applied for medical insurance compensation after undergoing coronary artery stent implantation for coronary heart disease.

[0033] After discharge, the patient uploads their inpatient medical record, discharge summary, expense invoice, and detailed list via the self-service terminal at the counter. The system obtains medical insurance settlement data through the medical insurance interface. The data acquisition module performs OCR text recognition and medical NER extraction, extracting from the medical record the diagnosis code I25.1 (atherosclerotic heart disease) in ICD-10, the surgical name percutaneous coronary intervention (PCI) and stent implantation, the length of hospital stay of 8 days, and the medication list including antiplatelet drugs such as clopidogrel and the imported drug everolimus-eluting stent. The case complexity assessment unit evaluated the structured data as follows: the diagnosis was coronary heart disease, a chronic disease requiring long-term management, indicating high complexity; the treatment method was PCI and stent implantation, a complex interventional procedure, also indicating high complexity; the estimated cost was 68,000 yuan, exceeding the 5,000 yuan threshold, indicating high complexity; no other past medical history was found in the medical records, indicating low complexity in the past medical history dimension; the medication included imported drugs, indicating high complexity. Considering that four of the five dimensions exhibited high complexity, the case complexity assessment unit classified the case as high-complexity. Based on this, the dynamic routing unit routed the case to the deep agent for processing.

[0034] The deep agent employs a large language model (preferably ChatGLM in this embodiment) combined with a medical knowledge graph for in-depth semantic understanding and reasoning. The deep agent first reads the customer's electronic insurance policy model. This policy is a high-end medical insurance product with a 0-day waiting period (no waiting period), a deductible of 0 yuan, a reimbursement rate of 100%, and a maximum coverage of 1 million yuan. The deep agent compares the diagnosis code I25.1 with the policy's coverage scope, confirming that coronary heart disease is covered. In the exclusions check, the policy has no waiting period, the diagnosis is not on the exclusion list, and the treatment is covered. The deep agent then calculates the estimated payout amount. It costs 68,000 yuan. It is 0 yuan. 100%, Yuan.

[0035] In the medical record-policy conflict detection step, the deep agent extracts past medical history and present medical history information from the structured medical record data. It finds that the medical record records "8-year history of hypertension," while the customer has selected "no" for the "Do you have hypertension?" section in the health declaration of the insurance policy. The medical record-policy conflict detection submodule detects this potential inconsistency, outputs a risk indicator, and marks the case as a priority for transfer to the manual review queue.

[0036] The invoice entry module processes expense invoices and detailed lists, identifying a total amount of 68,000 yuan, medical insurance payment of 18,000 yuan, and personal cash payment of 50,000 yuan. During invoice-medical record cross-validation, the system found that the "everolimus-eluting stent" in the invoice was priced at 26,500 yuan. This treatment item is consistent with the surgical plan recorded in the medical record, so the verification passed, and no abnormality flag was output.

[0037] During the immediate settlement phase, although liability was established and invoice verification was normal, the estimated compensation amount of 68,000 yuan exceeded the preset threshold of 5,000 yuan, and a risk indicator was output by the medical record-policy conflict detection. Therefore, the system triggered a closed-loop settlement process of temporary payment-verification-adjustment. The system first followed the formula... Calculate the amount of temporary payment, of which In this embodiment, the proportion of temporary payment is... Take 80%, temporary payment amount The system initially paid the customer 54,400 yuan via the financial interface. Simultaneously, the system pushed the case to the back-end manual review queue for verification by professional auditors. The manual review confirmed that the hypertension history recorded in the medical record was true, indicating a failure to disclose it truthfully during the insurance application process. Based on a comprehensive assessment, the auditors determined that the hospitalization for coronary heart disease and the hypertension history did not constitute a direct causal relationship, and the final compensation amount remained at 68,000 yuan. Since the final compensation amount (68,000 yuan) exceeded the provisional payment (54,400 yuan), the system automatically supplemented the difference of 13,600 yuan to the customer's account. The entire process of provisional payment, verification, and adjustment was recorded in the system log, forming a complete settlement loop.

[0038] In this embodiment, the customer received a temporary payment of 54,400 yuan within 2 minutes of submitting the information, experiencing "instant" claims service. Meanwhile, the insurance company ensured the time for risk control verification through the temporary payment mechanism, achieving a balance between customer experience and risk control.

[0039] Example 3: This embodiment illustrates the specific implementation method of incremental learning of the agent in Embodiment 1.

[0040] Upon completion of each claim processing step, regardless of whether it is handled by a fast or deep agent, and regardless of whether it has undergone manual review, the system automatically uploads the case's feature data, agent processing results, and manually reviewed and corrected data back to the historical case database. For example, in the coronary heart disease case in Example 2, the system records complete data including the case's diagnostic features (PCI procedure), cost amount (68,000 yuan), agent processing results (estimated payout of 68,000 yuan, conflict detection risk), and manual review results (confirmation of conflict, maintenance of payout amount). All data is anonymized before being stored in the database. The system periodically (preferably weekly in this example) retrains or fine-tunes the fast and deep agents using newly added data from the historical case database. For the fast agent, incremental learning is used to update the parameters of the lightweight Transformer model; for the deep agent, a low-rank adaptation method is used to fine-tune the large language model, while simultaneously updating the relationships in the medical knowledge graph. As the number of processed cases increases, the accuracy of the agent's judgments continues to improve.

[0041] Example 4: like Figure 3 As shown, this embodiment provides a medical insurance instant settlement system based on a claims case intelligent agent. This system is used to implement the above-described method embodiment. The system specifically includes the following modules: The data acquisition module is used to acquire and preprocess claims-related data, transforming unstructured medical text into structured data. This module includes a medical insurance interface connection unit, an OCR recognition unit, and a medical NER extraction unit. The medical insurance interface connection unit establishes a secure, encrypted connection with local medical insurance systems via standardized protocols to acquire customer medical insurance settlement data in real time; the OCR recognition unit performs text recognition on images of medical records and invoices uploaded by customers; and the medical NER extraction unit extracts key fields such as disease diagnosis codes, surgery names, and drug names from the recognized text.

[0042] The case complexity assessment module evaluates the complexity of preprocessed structured data and dynamically routes cases to corresponding agents for processing based on the assessment results. This module includes a diagnosis type analysis subunit, a cost amount judgment subunit, and a medical history detection subunit. The diagnosis type analysis subunit determines whether a disease is common or rare, and whether it is a single disease or a combination of multiple diseases. The cost amount judgment subunit compares the estimated cost with a preset threshold. The medical history detection subunit determines whether the medical records contain any past medical history or chronic disease history.

[0043] The agent processing module comprises a fast agent unit, a deep agent unit, and a dynamic routing unit. The dynamic routing unit routes low-complexity standard cases to the fast agent unit and high-complexity cases to the deep agent unit based on the output of the case complexity assessment module. The fast agent unit is built using a lightweight Transformer model or a decision tree ensemble model, with a response time of less than 1 second. The deep agent unit is built using a large language model combined with a medical knowledge graph, possessing stronger semantic understanding and reasoning capabilities.

[0044] The liability matching module enables the agent to automatically match insurance liabilities based on the electronic policy model, outputting the liability determination result and estimated compensation amount. This module includes a policy information reading subunit, a liability scope determination subunit, an exclusion liability detection subunit, and a compensation amount calculation subunit. The liability matching module also includes a medical record-policy conflict detection subunit, used to automatically detect inconsistencies between past medical history information recorded in the medical record and the health disclosure information in the insurance application.

[0045] The invoice entry module is used to automatically enter medical expense invoices and cross-verify them with medical record information. This module includes an invoice information recognition subunit and an invoice-medical record cross-verification subunit. The invoice information recognition subunit is used to extract expense items such as the total amount and medical insurance payment amount from the invoice image; the invoice-medical record cross-verification subunit is used to compare the treatment items in the invoice with the treatment plan recorded in the medical record to detect any anomalies.

[0046] The instant settlement module triggers the instant settlement process based on liability matching and invoice verification results to complete the payment of claims. This module includes a full payment unit and a temporary payment unit. The full payment unit triggers automatic full payment for standard cases where liability is established, there is no risk indicator, and the amount is below a preset threshold. The temporary payment unit makes an initial payment according to a preset ratio (preferably 80%) for high-value or questionable cases, and then makes up the difference or refunds it after manual review in the background, forming a complete settlement loop.

[0047] The above is a detailed description of the preferred embodiments of the present invention, but the present invention is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention, and these equivalent modifications or substitutions are all included within the scope defined by the claims of this application.

Claims

1. A method for real-time settlement of medical insurance based on intelligent agents for claims cases, characterized in that, Includes the following steps: Step S1: Obtain claims-related data and preprocess it, converting unstructured medical text into structured data; Step S2: Evaluate the case complexity of the preprocessed structured data, and dynamically route the cases to the corresponding agents for processing based on the evaluation results; Step S3: The intelligent agent automatically matches insurance liabilities based on the electronic policy model and outputs the liability determination result and the estimated compensation amount; Step S4: Automatically enter medical expense receipts and cross-verify them with medical record information; Step S5: Based on the liability matching results and invoice verification results, trigger the instant settlement process to complete the payment of the claim.

2. The method for real-time settlement of medical insurance based on a smart agent for claims cases according to claim 1, characterized in that, In step S1, the acquisition and preprocessing of claim-related data specifically includes: The medical insurance settlement data of customers can be obtained in real time through the medical insurance interface. The medical insurance settlement data includes total cost, medical insurance payment amount, personal co-payment amount and details of Category C items. Obtain the inpatient medical records, discharge summaries, medical expense invoices and detailed expense lists uploaded by customers through self-service terminals at the counter or mobile terminals; Perform OCR text recognition on the acquired unstructured medical text; A medical named entity recognition model is used to extract key fields from the recognition results. The key fields include disease diagnosis code, surgical procedure name, length of hospital stay, drug name and out-of-pocket payment identifier, and main examination and testing items. The extracted key fields are integrated into structured data.

3. The method for real-time settlement of medical insurance based on a smart agent for claims cases according to claim 1, characterized in that, Step S2 is as follows: The case complexity assessment unit automatically determines the complexity of the current case based on the following assessment dimensions: Diagnostic type dimension: Common diseases and single-disease diagnoses are characterized by low complexity, while rare diseases and diagnoses of multiple diseases are characterized by high complexity. Treatment method dimension: Conservative treatment and simple surgery are characterized by low complexity, while complex surgery and multiple hospitalizations are characterized by high complexity. Cost amount dimension: features below the preset threshold are considered low-complexity features, and features above the preset threshold are considered high-complexity features; Past medical history dimension: No past medical history is a low complexity feature, while past medical history or chronic disease history is a high complexity feature; Medication usage dimension: All drugs are covered by medical insurance catalog, which is a low complexity feature; drugs that include self-paid drugs or imported drugs are a high complexity feature. Based on the evaluation results, if the case is determined to be a low-complexity standard case, it is dynamically routed to a fast intelligent agent for processing. The fast intelligent agent adopts a lightweight model with a response time of less than 1 second. If the case is determined to be a high-complexity case, it is dynamically routed to a deep intelligent agent for processing. The deep intelligent agent uses a large language model combined with a medical knowledge graph for semantic understanding and reasoning.

4. The method for real-time settlement of medical insurance based on a smart agent for claims cases according to claim 1, characterized in that, Step S3 is as follows: The system reads the digital model of the customer's insurance policy, which is pre-set with parameters such as coverage scope, waiting period days, deductible amount, payout ratio, exclusion list, and maximum coverage amount. The disease diagnosis code extracted from the medical record is compared with the policy coverage to determine whether the current incident falls under the insurance liability stipulated in the policy. Automatically detect whether there are any exclusions, including whether the incident occurred within the waiting period, whether the diagnosis is on the list of excluded diseases, and whether the treatment is out-of-pocket or not covered by the insurance. Assuming liability is established, the estimated compensation amount is calculated and expressed as follows: ； in, To estimate the amount of compensation, This refers to the portion of the total medical expenses that falls within the scope of insurance coverage. For deductible, This refers to the compensation ratio; Output the liability determination result and the estimated compensation amount.

5. A method for real-time settlement of medical insurance based on a smart agent for claims cases, as described in claim 4, is characterized in that... In step S3, the automatic matching of insurance liabilities also includes a medical record-policy conflict detection step: Step S31: Extract chronic disease records from past medical history and present medical history from the structured data of the medical records. The chronic diseases include hypertension, diabetes and heart disease. Step S32: Compare the extracted medical history information with the health declaration information filled in by the customer in the insurance application form item by item; Step S33: If a potential inconsistency is detected, i.e., the medical record contains a specific medical history but the corresponding item on the insurance application is checked as "none", then a risk indicator is output and the case is transferred to the manual review queue first. If no inconsistency is detected, the subsequent processing procedure continues.

6. The method for real-time settlement of medical insurance based on a smart agent for claims cases according to claim 1, characterized in that, Step S4 is as follows: Automatically identify the total amount, medical insurance payment amount, personal cash payment amount, and out-of-pocket item amount in medical expense invoices; Automatically identify each charge item and its corresponding amount in the detailed expense list; The treatment items in the invoice are compared with the treatment plan recorded in the medical record. If it is found that there are chargeable items in the invoice that are not recorded in the medical record, or if there is no corresponding chargeable record in the invoice for the treatment items recorded in the medical record, an abnormal verification mark is output to trigger manual review. Once the verification is approved, an electronic claims list will be automatically generated.

7. The method for real-time settlement of medical insurance based on a smart agent for claims cases according to claim 1, characterized in that, Step S5 is as follows: The automatic full payment process is triggered when the following conditions are met simultaneously: liability is determined to be established, there is no risk indicator and the verification is normal, and the estimated compensation amount is lower than the preset threshold; the system calls the financial interface to pay the full amount of the claim to the customer's bank card. The temporary payment-verification-adjustment closed-loop settlement process is triggered when any of the following situations occur: the estimated compensation amount is higher than the preset threshold, the medical record-policy conflict detection outputs a risk indicator, or the invoice-medical record cross-verification outputs an abnormal indicator. The aforementioned temporary payment-verification-adjustment closed-loop settlement process specifically includes: The amount of temporary payment is calculated as follows: ； in, For temporary payment amount, To estimate the amount of compensation, This is a temporary payment ratio. The range of values ​​is ; Pay the customer in advance according to the temporary payment amount; The case will be sent to the back-end manual review queue for final review. Once the review is completed, if the final compensation amount is higher than the provisional payment amount, the difference will be automatically paid out; if the final compensation amount is lower than the provisional payment amount, a difference refund notice will be generated and the customer will be informed of the refund method through the channels reserved by the customer. The entire process of temporary payment, verification, and adjustment will be recorded and written to the system log.

8. A medical insurance instant settlement system based on a claims intelligent agent, used to implement the medical insurance instant settlement method based on a claims intelligent agent as described in any one of claims 1-7, characterized in that, include: The data acquisition module is used to acquire claims-related data and preprocess it, converting unstructured medical text into structured data. The case complexity assessment module is used to assess the case complexity of preprocessed structured data and dynamically route cases to the corresponding intelligent agents for processing based on the assessment results. The intelligent agent processing module includes a fast intelligent agent unit, a deep intelligent agent unit, and a dynamic routing unit. The dynamic routing unit routes low-complexity standard cases to the fast intelligent agent unit and high-complexity cases to the deep intelligent agent unit based on the case complexity assessment results. The liability matching module enables the intelligent agent to automatically match insurance liabilities based on the electronic policy model, and outputs the liability determination result and the estimated compensation amount; the liability matching module also includes a medical record-policy conflict detection submodule, which is used to automatically detect inconsistencies between medical records and insurance disclosure information; The invoice entry module is used to automatically enter medical expense invoices and cross-verify them with medical record information; the invoice entry module includes an invoice-medical record cross-verification subunit. The instant settlement module is used to trigger the instant settlement process based on the liability matching result and the invoice verification result to complete the payment of the claim. The instant settlement module includes a full payment unit and a temporary payment unit. The temporary payment unit is used to make an advance payment according to a preset ratio in high-value or questionable cases, and complete the difference adjustment settlement after background review.

Images