Cross-industry health insurance data sharing and privacy protection method based on homomorphic encryption

Abstract

The invention discloses a homomorphic encryption-based cross-industry health insurance data sharing and privacy protection method, and relates to the technical field of health insurance data security, and the method specifically comprises the steps: standardizing medical and insurance industry core data, quantifying key indexes such as a disease clinical risk value and an occupational basic risk coefficient, achieving the format unification of cross-industry data, and achieving the privacy protection of the cross-industry data; further generating ownership, risk and business labels, carrying out encryption processing on data through homomorphic encryption on the risk labels based on three scenes of data sensitivity and business influence grading, business label binding pricing, underwriting and claim settlement prediction, adding an integrity check code to high-risk data, screening ciphertexts based on the labels, and then executing fusion calculation; the comprehensive risk adjustment coefficient ciphertext is generated, the data integrity is verified, the comprehensive risk adjustment coefficient is obtained after collaborative decryption, the comprehensive risk adjustment coefficient is applied to premium pricing, underwriting conclusion generation and claim risk pre-judgment, and the scientificity and accuracy of health insurance pricing and underwriting decision making are improved.

Description

Technical Field

[0001] This invention belongs to the field of health insurance data security technology, specifically involving a method for cross-industry health insurance data sharing and privacy protection based on homomorphic encryption. Background Technology

[0002] As the health insurance industry moves towards refinement and precision, the quality of decision-making in core business areas such as insurance product pricing, underwriting risk control, and claims prediction increasingly relies on deep collaboration between health data from the healthcare industry and business data from the insurance industry. The core data accumulated by healthcare providers, such as disease diagnoses and physical examination indicators, and the business data held by insurance providers, such as occupational risks and historical claims, belong to different industries and are stored in independent systems. Cross-industry data sharing has become a crucial prerequisite for improving the scientific nature of health insurance business.

[0003] However, existing cross-industry data sharing models face multiple technical bottlenecks, severely restricting the development of health insurance business:

[0004] Medical data contains a large amount of sensitive personal health information, while insurance data involves the privacy of policyholders and trade secrets. Existing sharing models mostly rely on data anonymization or plaintext transmission. The former is prone to data value loss, while the latter cannot resist the risk of privacy leakage during transmission and storage, and lacks a special integrity verification mechanism for highly sensitive data.

[0005] Existing risk assessments rely heavily on data from a single industry, failing to fully integrate multi-dimensional risk factors across industries. Furthermore, they lack the ability to directly compute in encrypted domains while ensuring privacy, thus hindering the accurate integration and quantification of risk factors.

[0006] In traditional encryption schemes, private keys are usually kept by a single institution, which poses a risk of data being illegally decrypted due to private key leakage. Meanwhile, multi-node collaborative decryption mechanisms lack reasonable threshold design and are difficult to balance privacy and security with operational efficiency.

[0007] Therefore, there is an urgent need for a cross-industry health insurance data sharing and privacy protection method based on homomorphic encryption to solve the above problems. Summary of the Invention

[0008] The purpose of this invention is to provide a method for cross-industry health insurance data sharing and privacy protection based on homomorphic encryption, in order to solve the technical problem of low accuracy in risk quantification and fusion calculation in the prior art.

[0009] To achieve the above objectives, the present invention adopts the following technical solution:

[0010] Methods for cross-industry health insurance data sharing and privacy protection based on homomorphic encryption include:

[0011] Step 1: Classify the clinical risk level of the disease and calculate the basic risk value corresponding to different risk levels. Clearly collect core data from the medical side, calculate the occupational basic risk coefficient and the correlation value between claim frequency and amount, and clearly collect core data from the insurer.

[0012] Step 2: For the data collected in Step 1, generate ownership tags based on hash algorithms combined with institutional identifiers and data indexes; generate risk tags based on data sensitivity and business impact classification; and generate business tags by dynamically binding pricing, underwriting, and claims prediction scenarios.

[0013] Step 3: Construct a privacy protection framework based on 3D labels for fusion computing of core business data across industries in an encrypted state;

[0014] Step 4: Regulatory nodes that meet the threshold conditions collaboratively decrypt the data and apply the decrypted comprehensive risk adjustment coefficient to scenarios such as premium pricing, underwriting conclusion generation, and claims risk prediction.

[0015] Furthermore, the clinical risk level of the disease is classified, and the specific method is as follows:

[0016] The claim incidence rate of the lowest risk level is set as the benchmark value A. The basic risk values ​​of other levels are the ratio of their claim incidence rate to the benchmark value A. The classification of multi-level risk is determined based on the multi-dimensional characteristics of the disease, such as prognosis, treatment complexity, long-term health impact, recurrence probability, and treatment costs. The core data of the medical side is processed in a standardized format. The occupational risk classification and claim frequency of the insurer are classified according to preset rules. The classification rules are determined based on the cross-industry data interaction adaptation needs and data characteristics.

[0017] Furthermore, the basic occupational risk coefficient is calculated using the following method:

[0018] The occupational base risk coefficient is determined by the ratio of the occupational category's experience payout rate to the benchmark category's experience payout rate. The benchmark category is selected from the occupations with the most standard or lowest risk, and its base risk coefficient is set as the benchmark value, denoted as Ba. The claim frequency correlation value is quantified by an exponential saturation growth function, and the amount correlation value is obtained by mapping the maximum relative severity of a single claim through a scaling function. The mapped value range is [C, D]. The values ​​of C and D are determined based on the distribution characteristics of historical claim data and the requirements for risk quantification accuracy.

[0019] Furthermore, ownership tags are generated based on a hash algorithm combined with the organization identifier and data index. The specific method is as follows:

[0020] The hash algorithm of the national cryptographic standard is used to generate the final tag. After XOR operation between the organization identifier and the unique index of the data, the first E bits of the hash result are taken as the final tag. The value of E is determined based on the dual principles of non-duplication of ownership tags generated from different data of the same organization and compliance of tag storage space with the data storage efficiency optimization requirements.

[0021] Furthermore, risk labels are generated based on data sensitivity and business impact, using the following method:

[0022] The sensitivity coefficient and business impact coefficient of the risk label both range from [F, G]. The risk level is determined by a non-linear mapping function, where the product of the two is greater than or equal to the threshold H, the threshold I is less than or equal to the product and less than the threshold H, the risk level is medium, and the product is less than the threshold I, the risk level is low. The values ​​of H and I are derived from the statistical results of historical cross-industry data breaches. Business label binding requires the calculation of keyword similarity through a semantic mapping algorithm. Binding can only be completed when the cosine similarity is greater than the threshold J. The value of J is determined based on the accuracy requirements of the matching between the business label and the data features.

[0023] Furthermore, the key management center generates public and private keys. The data provider uses the public key to encrypt data and filters ciphertext data that matches business tags. The specific method is as follows:

[0024] The encryption scheme uses fully homomorphic encryption or partially homomorphic encryption that supports addition and scalar multiplication. The private key is split into (t, n) threshold secret sharing scheme and jointly kept by multiple supervisory nodes. t is the minimum number of nodes required for decryption, and n is the total number of nodes keeping the private key fragments. The values ​​of t and n are determined based on the balance between privacy and security control level and collaborative operation efficiency.

[0025] High-risk data, after encryption, needs to have its ciphertext fingerprint calculated using a secure hash algorithm, and then an integrity check code generated by combining it with the ownership tag. Medium- and low-risk data can be directly uploaded as encrypted ciphertext, and the check code generation rules are determined based on the integrity assurance requirements for the transmission and storage of high-risk data.

[0026] Furthermore, a fusion calculation is performed to generate the encrypted comprehensive risk adjustment coefficient. The specific method is as follows:

[0027] Homomorphic encryption is performed on the insurer's core data using a public key (PK) to obtain ciphertext insurance factors, including ciphertext-formatted occupational basic risk coefficient [c], ciphertext-formatted historical claim frequency correlation value [q], and ciphertext-formatted historical amount correlation value [s], which are then associated with the corresponding business tag B and risk tag. The basic risk value [r] of the corresponding risk label after encryption is determined. By extracting business scenario keywords from the request, precise matching is performed with the business label B of the data in the encrypted storage pool. All encrypted medical data and insurance data that are consistent with the request scenario and related to the target insured are selected. Predefined fusion calculation logic is directly executed in the encrypted domain, using formulas. Generate encrypted comprehensive risk adjustment coefficients, in which Represents the plaintext weight coefficient, satisfying The specific value is determined by historical data analysis and business rules. The encrypted baseline risk offset is set to the lowest risk baseline value in the business scenario. When high-risk medical data ciphertext with integrity check code MAC is involved in the calculation process, the computing node needs to use the linear operation characteristics of homomorphic encryption to synchronously generate the aggregate check code.

[0028] Furthermore, an aggregated verification code is generated synchronously, specifically using the following method:

[0029] Synchronous generation of aggregated verification codes is achieved using the formula Received, among which This represents the encrypted underlying risk value of the disease. The corresponding ciphertext fingerprint, and other factor fingerprints refer to those that participate in the ciphertext fusion calculation, excluding... fingerprints corresponding to other high-risk encrypted data, To aggregate the ownership tags of high-risk data participating in the calculation, the local verification code and the data uploaded by the computing node are combined. If the comparison is consistent, the verification is successful; otherwise, the calculation process is deemed abnormal.

[0030] Furthermore, regulatory nodes that meet the threshold conditions collaboratively decrypt the data, and the resulting comprehensive risk adjustment coefficient is applied to premium pricing, underwriting conclusion generation, and claims risk prediction scenarios. The specific method is as follows:

[0031] Collaborative decryption requires the data user to submit compliant application materials, which include a business scenario description, the target policyholder's authorization document, aggregated verification code, and aggregated value of ownership tag, ensuring the relevance of the application and decryption requirements. The threshold condition is based on the minimum number of decryption nodes t, and the value of t is determined according to the balance between privacy and security control level and collaborative operation efficiency. The key management center will initiate the private key reassembly and decryption process only when at least t regulatory nodes report valid private key fragments.

[0032] The entire decryption process is logged, including information on the response of regulatory nodes, the time taken for private key reassembly, and the information of the person performing the decryption. The logging rules are determined based on the requirements for operational traceability and liability determination. The comprehensive risk adjustment coefficient obtained from decryption is adapted to the corresponding rules for each scenario, taking into account the specific characteristics of that scenario.

[0033] In the premium pricing scenario, the calculation logic of benchmark premium × comprehensive risk adjustment factor × additional risk factor is adopted. The value of additional risk factor is determined based on the scenario-related characteristics such as region, policy, and industry risk trends, so as to achieve accurate matching between risk and premium.

[0034] In underwriting scenarios, multiple judgment thresholds are set according to the comprehensive risk adjustment coefficient, which correspond to underwriting conclusions such as standard underwriting, underwriting with additional premiums, and rejection. The division of the thresholds is determined based on the insurer's risk tolerance, industry risk control standards, and business rules.

[0035] In claims prediction scenarios, the comprehensive risk adjustment coefficient is mapped to three levels of claims risk: low, medium, and high. The boundaries of the classification are determined based on the distribution characteristics of historical claims data and the accuracy requirements of risk prediction. A special tracking mechanism is established for policyholders with high risk levels, and the specific form of the tracking mechanism is adapted to the needs of optimizing the allocation of claims resources.

[0036] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0037] 1. This invention achieves direct fusion computation of ciphertext domains through homomorphic encryption, avoiding the risk of leakage from plaintext transmission and storage of sensitive data. High-risk data is equipped with an integrity verification mechanism to ensure that the data has not been tampered with. At the same time, it does not rely on data anonymization, fully preserving the original value of cross-industry data and achieving a balance between privacy protection and effective data utilization.

[0038] 2. This invention breaks down the collaboration barriers caused by data silos by unifying the data format standards of the medical and insurance industries. By quantifying multi-dimensional cross-industry risk factors and performing precise fusion calculations, the generated comprehensive risk adjustment coefficient can fully reflect the true risk level of the insured, and significantly improve the scientificity and accuracy of health insurance pricing and underwriting decisions.

[0039] 3. Dynamically bind to core health insurance business scenarios to provide targeted data support for pricing, underwriting, and claims prediction, avoiding data usage from being out of touch with business needs. The threshold collaborative decryption mechanism not only eliminates the risk of private key leakage at a single node, but also ensures that the decryption process is compliant and traceable, taking into account both business operation efficiency and security control requirements. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1The diagram illustrates the steps of the cross-industry health insurance data sharing and privacy protection method based on homomorphic encryption according to the present invention.

[0042] Figure 2 The figure shows the steps of the data tag generation method of the present invention;

[0043] Figure 3 The diagram illustrates the steps of generating the ciphertext of the comprehensive risk adjustment coefficient and verifying the aggregate check code according to the present invention. Detailed Implementation

[0044] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0045] Example 1, such as Figure 1 The cross-industry health insurance data sharing and privacy protection method based on homomorphic encryption, as shown, specifically includes the following steps:

[0046] Step 1: Classify the clinical risk level of the disease and calculate the basic risk value corresponding to different risk levels. Clearly collect core data from the medical side, calculate the occupational basic risk coefficient and the correlation value between claim frequency and amount, and clearly collect core data from the insurer.

[0047] The core data from the medical side includes critical illness diagnosis codes, key physical examination indicator values, and corresponding collection times. These data are sourced from the medical side's electronic medical record (EMR) system and the laboratory information system (LIS) of the physical examination center. In this embodiment, the critical illness diagnosis codes adopt the standardized disease codes established by the International Classification of Diseases, 10th Revision (ICD-10). The key physical examination indicator values ​​refer to quantitative data that reflect human physiological functions and health status obtained through medical testing methods. These data need to be standardized according to a unified format, such as fasting blood glucose levels and systolic / diastolic blood pressure.

[0048] By selecting different dimensions of the disease (such as prognosis, treatment complexity, long-term health impact, recurrence probability, and treatment cost), the disease code is divided into multiple clinical risk levels. The classification in this embodiment is as follows:

[0049] Category A (Very High Risk): Malignant tumors (specific advanced or highly aggressive types), acute myocardial infarction (with serious complications);

[0050] Category B (High Risk): Coronary artery bypass grafting, end-stage renal disease, benign brain tumor.

[0051] Category C (Medium Risk): Diabetes mellitus with specific complications, decompensated chronic liver failure.

[0052] Category D (Low Risk / Baseline Risk): Acute appendicitis, simple fracture, mild pneumonia.

[0053] Using historical claims databases accumulated by insurance companies or industry platforms, the data must contain a large number of insured individuals' disease diagnosis codes (ICDs) and their corresponding claims information (whether a claim was made, the claim amount, and the coverage period). During the observation period, the probability of claims (especially critical illness claims) occurring for the insured population under each clinical risk level is statistically analyzed, i.e., the claim incidence rate. The claim incidence rate of the lowest risk level in the clinical risk level classification is set as the benchmark value (the benchmark value is usually defined as 1.0). The ratio of the claim incidence rate of other levels to level D is calculated to determine the basic risk value of a certain risk level, i.e., the basic risk value of a certain risk level = the claim incidence rate of that risk level / the claim incidence rate of level D.

[0054] The insurer's core data includes the basic risk coefficients corresponding to occupational risk classifications, and the correlation values ​​of claim frequency and amount corresponding to historical claim records, which are derived from the insurer's customer relationship management system (CRM) and claims settlement system;

[0055] Occupational risk classification involves compiling historical policy data, applying national occupational classification standards, and incorporating risk assessment experience within the insurance industry. Occupations are initially categorized into several classes (e.g., classes 1-3 or 1-6). The experience loss ratio for each occupational class is calculated, which is the cumulative claims amount of all policies in that class divided by the cumulative premium income. A class considered to have the most standard or lowest risk is selected as the benchmark class (e.g., teacher), with its base risk coefficient set at 1.0. The base risk coefficients for other occupational classes are determined by the ratio of their experience loss ratios to the benchmark class's experience loss ratio. That is, the base risk coefficient for an occupational class = experience loss ratio of that occupational class / experience loss ratio of the benchmark class. For example, if the experience loss ratio for class 1 (office worker) is 0.5%, while the experience loss ratio for class 3 (construction worker) is 1.625%, then the base risk coefficient for class 3 = 1.625% / 0.5% = 3.25. This 3.25 means that, all other things being equal, the expected risk cost for class 3 is 3.25 times that of class 1.

[0056] The correlation between the frequency and amount of claims in historical claims records is used to quantify the dynamic impact of the insured's past claims behavior on their future risks;

[0057] Select a historical policy cohort, track the number of claims for each policy during the observation period (e.g., the past 3 years), and whether a claim will occur again in the subsequent prediction period, fitting an exponential saturated growth function. Used to quantify the correlation value of claim frequency This indicates the base risk level when there are zero claims (set based on historical data and actual circumstances). This represents the maximum increase in risk that can be achieved by increasing the number of claims from 0 to infinity (set based on historical data and actual circumstances). This represents the attenuation parameter, which controls the rate at which risk approaches its upper limit as the number of claims increases. It is derived by fitting historical data. This indicates the number of claims made during the waiting period.

[0058] Divide the amount of a single claim by the average amount of claims for that type to obtain the relative severity. Calculate the maximum relative severity of all claims for an insured during the observation period. Then, map this value to a risk correlation value range (e.g., 0.9 to 1.5) using a scaling function to obtain the amount correlation value. The higher the amount and the more severe the illness, the larger the amount correlation value.

[0059] For medical institutions, disease coding uniformly adopts the International Classification of Diseases, 10th Revision (ICD-10) standard coding, and physical examination indicator values ​​uniformly retain a specified number of decimal places (determined based on clinical testing accuracy); for insurance institutions, occupational risk classification is uniformly mapped to a three-level classification code, and the number of claims is categorized, such as 0 times / 1 time / ≥2 times, to achieve consistency in data format across industries.

[0060] Step 2: For the data collected in Step 1, ownership tags are generated based on hash algorithms combined with institutional identifiers and data indexes. Risk tags are generated based on the classification of data sensitivity and business impact. Business tags are generated by dynamically binding pricing, underwriting, and claims prediction scenarios.

[0061] Each piece of data accessed by the medical / insurance provider (such as a critical illness diagnosis record or a claim record) is assigned a globally unique data index by the interactive platform to ensure that different data indexes from the same institution do not duplicate each other. Ownership tags are generated using a hash algorithm based on the national cryptographic standard. ,in Indicates ownership label, Represents a hash algorithm. For the XOR operation, the unique identifier of the organization (e.g., a hospital-001, an insurance company-002) is mixed with the data index to avoid the leakage of single information. The first N bits of the hash result are taken as the final ownership label. The value of N is determined based on the balance between the uniqueness of the ownership identifier and storage efficiency. In this embodiment, the value of N is based on: different data generated by the same organization No repetition, and The storage usage shall not exceed 5% of the total storage of a single data entry.

[0062] Define a data sensitivity coefficient S (determined based on the degree of privacy damage caused by data leakage, with a value range of [0, 1]) and a business impact coefficient I (determined based on the degree of impact of data on health insurance business decisions, through statistical analysis of historical business data, with a value range of [0, 1]), and use a nonlinear mapping function. Output tiered risk labels (e.g., high / medium / low risk labels), with mapping rules derived from historical cross-industry data breach loss statistics. Embedded data fields, this embodiment uses a non-linear mapping function. Output three levels of risk labels: high, medium, and low. The mapping rules are derived based on the loss statistics of historical cross-industry data breach events. Because when S×I is in the interval [0.7, 1.0], the proportion of events defined as high risk exceeds 90%, and when it is in the interval [0.4, 0.7), the proportion of events defined as medium risk exceeds 85%, therefore... The criteria for judgment are:

[0063] High risk, S×I≥0.7 (e.g., critical illness code, blood glucose risk value);

[0064] Medium risk, 0.4≤S×I<0.7 (e.g., occupational risk level);

[0065] Low risk, S×I<0.4 (e.g., the number of years of insurance coverage).

[0066] Business tag B is generated and dynamically bound when the data user (insurer) initiates a business request. The business scenarios are divided into pricing scenario (B1), underwriting scenario (B2), and claims prediction scenario (B3). The binding logic is: extraction of business requirement keywords → data feature matching → tag assignment. The matching accuracy is determined based on the correlation analysis between the business scenario and the data features (by using a semantic mapping algorithm based on the business scenario ontology, the similarity between the business requirement keywords and the standard description of the data features is calculated. Binding is only performed when the cosine similarity is greater than the threshold θ (e.g., 0.9) to ensure high accuracy).

[0067] Step 3: Construct a privacy protection framework based on 3D labels for fusion computing of core business data across industries in an encrypted state;

[0068] The key management center generates public and private keys. The data provider uses the public key to encrypt the data, filters the ciphertext data that matches the business tags, performs fusion calculations without decryption, generates ciphertext of the comprehensive risk adjustment coefficient, and verifies the aggregate verification code.

[0069] Key generation is performed by a trusted interactive platform or an independent Key Management Center (KMC). Based on a selected fully homomorphic encryption (FHE) or partially homomorphic encryption scheme supporting addition and scalar multiplication (such as Paillier), a public-private key pair (PK, SK) is generated. PK is the public homomorphic encryption public key, and SK is the secret decryption private key. The public key PK is distributed to all data providers (medical providers, insurers). The decryption private key SK is split by the interactive platform using a (t, n) threshold secret sharing scheme and distributed to n independent regulatory nodes or participants for joint safekeeping, where t is the minimum number of nodes required for decryption, ensuring that no single participant can independently decrypt the original ciphertext. For a medical record data M (containing basic risk values ​​and key physical examination indicator values), the ciphertext [M] is obtained by encrypting it with the public key PK. Simultaneously, based on the risk label of this data... Different encryption post-processing strategies are adopted for different risk levels (high, medium, and low). For high-risk data, the ciphertext fingerprint FP of the ciphertext [M] is calculated using the SHA-256 (Secure Hash Algorithm 256-bit) algorithm. The first 128 bits are taken as a fixed-length digest, and then the digest is combined with the ownership tag. Combined, an integrity check code is generated, and the formula is used. Indicates an integrity check code. Upload along with the encrypted text; for medium / low risk data, upload the encrypted text [M] directly.

[0070] Homomorphic encryption is performed on the insurer's core data using a public key (PK) to obtain ciphertext insurance factors, including ciphertext-formatted occupational basic risk coefficient [c], ciphertext-formatted historical claim frequency correlation value [q], and ciphertext-formatted historical amount correlation value [s] (all factors are business-normalized, with values ​​mapped to [0, 1] to avoid absolute zero values), and associated with corresponding business tag B and risk tag. Determine the base risk value [r] of the corresponding risk label after encryption, and perform the calculation without decryption according to the following logic:

[0071] Extract business scenario keywords from the request and perform precise matching with the business tag B of the data in the encrypted storage pool. Filter out all encrypted medical and insurance data that match the request scenario and are related to the target policyholder. Without decrypting any data, directly execute predefined fusion calculation logic in the encrypted domain, using formulas... Generate encrypted comprehensive risk adjustment coefficients, in which Represents the plaintext weighting coefficient (satisfying) The specific value is determined by historical data analysis and business rules. The encrypted baseline risk offset (valued as the lowest risk baseline value in the business scenario, such as 0.1) is used. If the calculation process involves high-risk medical data ciphertext with an integrity check MAC, the computing node needs to utilize the linear operation characteristics of homomorphic encryption to synchronously generate an aggregate check MAC. ,in This represents the encrypted underlying risk value of the disease. The corresponding ciphertext fingerprint, and other factor fingerprints refer to those that participate in the ciphertext fusion calculation, excluding... fingerprints corresponding to other high-risk encrypted data, To aggregate the ownership tags of high-risk data participating in the calculation, the local verification code and the data uploaded by the computing node are combined. If the comparison is consistent, the verification is successful; otherwise, the calculation process is deemed abnormal.

[0072] Step 4: Regulatory nodes that meet the threshold conditions collaboratively decrypt the data and apply the decrypted comprehensive risk adjustment coefficient to scenarios such as premium pricing, underwriting conclusion generation, and claims risk prediction.

[0073] After the aggregated verification code is verified, the collaborative decryption process is triggered. t or more supervisory nodes holding the decryption private key SK (based on the (t, n) threshold scheme set in step three) execute a secure multi-party computation protocol to jointly decrypt the ciphertext. The decryption application is submitted by the data user (insurer) to the Key Management Center (KMC). The application materials must include a scenario description corresponding to business tag B, the target policyholder's authorization document, and the aggregated verification code generated by the computation nodes. and ownership tag aggregation value To ensure the compliance and relevance of the application, KMC first verifies the completeness of the materials upon receiving the application, then sends private key sharding call notifications to n regulatory nodes. Only when at least t nodes return valid private key shards is the threshold decryption algorithm initiated to reconstruct the private key SK, and the ciphertext with the comprehensive risk adjustment coefficient is processed. The overall risk adjustment coefficient for performing decryption and generating plaintext results. The entire decryption process is logged, including node response status, private key reassembly time, and decryption executor information, ensuring traceability. No single institution or node can complete the decryption independently, eliminating the risk of private key leakage.

[0074] In the pricing scenario (B1), As a core risk factor, it is substituted into the premium calculation formula: Premium = Base Premium × ×Additional risk coefficients (such as regional risk coefficients) to achieve "risk-premium matching"—for example, an insured person with a Class 3 occupation (basic risk coefficient 3.25) and suffering from a Grade B high-risk disease (basic risk value 2.0), their After integration and calculation, the figure is 2.8, which is 180% higher than the benchmark level. This ensures that the insurance company's risk is controllable and avoids over-pricing.

[0075] In underwriting scenario (B2), set underwriting thresholds (e.g.) <3.5 is the standard coverage, 3.5≤ <4.5 is for coverage with additional premiums. (≥4.5 indicates rejection) The underwriting conclusion is automatically generated based on the insured's health declaration information.

[0076] In the claims prediction scenario (B3), based on Classify claims risk levels (e.g.) <1.5 indicates low claim risk; 1.5≤ <3.0 is for claims risk, (≥3.0 indicates high claim risk). A special tracking mechanism will be established for high-risk policyholders to predict the probability of claims in advance and optimize the allocation of claim resources.

[0077] The above formulas are all dimensionless calculations, and the preset parameters in the formulas should be set by those skilled in the art according to the actual situation.

[0078] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

[0079] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to specific implementations. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for cross-industry health insurance data sharing and privacy protection based on homomorphic encryption, characterized in that, include: Step 1: Classify the clinical risk level of the disease and calculate the basic risk value corresponding to different risk levels. Clearly collect core data from the medical side, calculate the occupational basic risk coefficient and the correlation value between claim frequency and amount, and clearly collect core data from the insurer. Step 2: For the data collected in Step 1, generate ownership tags based on hash algorithms combined with institutional identifiers and data indexes; generate risk tags based on data sensitivity and business impact classification; and generate business tags by dynamically binding pricing, underwriting, and claims prediction scenarios. Step 3: Construct a privacy protection framework based on 3D labels for fusion computing of core business data across industries in an encrypted state; The key management center generates public and private keys. The data provider uses the public key to encrypt the data, filters the ciphertext data that matches the business tags, performs fusion calculations without decryption, generates ciphertext of comprehensive risk adjustment coefficients, and verifies the aggregate verification code. Step 4: Regulatory nodes that meet the threshold conditions collaboratively decrypt the data and apply the decrypted comprehensive risk adjustment coefficient to scenarios such as premium pricing, underwriting conclusion generation, and claims risk prediction.

2. The method for cross-industry health insurance data sharing and privacy protection based on homomorphic encryption according to claim 1, characterized in that, The specific method for classifying the clinical risk level of a disease is as follows: The claim incidence rate of the lowest risk level is set as the benchmark value A. The basic risk values ​​of other levels are the ratio of their claim incidence rate to the benchmark value A. The classification of multi-level risk is determined based on the multi-dimensional characteristics of the disease, such as prognosis, treatment complexity, long-term health impact, recurrence probability, and treatment costs. The core data of the medical side is processed in a standardized format. The occupational risk classification and claim frequency of the insurer are classified according to preset rules. The classification rules are determined based on the cross-industry data interaction adaptation needs and data characteristics.

3. The method for cross-industry health insurance data sharing and privacy protection based on homomorphic encryption according to claim 1, characterized in that, The specific method for calculating the basic occupational risk coefficient is as follows: The occupational base risk coefficient is determined by the ratio of the occupational category's experience payout rate to the benchmark category's experience payout rate. The benchmark category is selected from the occupations with the most standard or lowest risk, and its base risk coefficient is set as the benchmark value, denoted as Ba. The claim frequency correlation value is quantified by an exponential saturation growth function, and the amount correlation value is obtained by mapping the maximum relative severity of a single claim through a scaling function. The mapped value range is [C, D]. The values ​​of C and D are determined based on the distribution characteristics of historical claim data and the requirements for risk quantification accuracy.

4. The method for cross-industry health insurance data sharing and privacy protection based on homomorphic encryption according to claim 1, characterized in that, The ownership tag is generated based on a hash algorithm combined with the organization identifier and data index. The specific method is as follows: The hash algorithm of the national cryptographic standard is used to generate the final tag. After XOR operation between the organization identifier and the unique index of the data, the first E bits of the hash result are taken as the final tag. The value of E is determined based on the dual principles of non-duplication of ownership tags generated from different data of the same organization and compliance of tag storage space with the data storage efficiency optimization requirements.

5. The method for cross-industry health insurance data sharing and privacy protection based on homomorphic encryption according to claim 1, characterized in that, Risk labels are generated based on a tiered system that considers data sensitivity and business impact. The specific method is as follows: The sensitivity coefficient and business impact coefficient of the risk label both range from [F, G]. The risk level is determined by a non-linear mapping function, where the product of the two is greater than or equal to the threshold H, the threshold I is less than or equal to the product and less than the threshold H, the risk level is medium, and the product is less than the threshold I, the risk level is low. The values ​​of H and I are derived from the statistical results of historical cross-industry data breaches. Business label binding requires the calculation of keyword similarity through a semantic mapping algorithm. Binding can only be completed when the cosine similarity is greater than the threshold J. The value of J is determined based on the accuracy requirements of the matching between the business label and the data features.

6. The method for cross-industry health insurance data sharing and privacy protection based on homomorphic encryption according to claim 1, characterized in that, The key management center generates public and private keys. The data provider uses the public key to encrypt data and filters ciphertext data that matches business tags. The specific method is as follows: The encryption scheme uses fully homomorphic encryption or partially homomorphic encryption that supports addition and scalar multiplication. The private key is split into (t, n) threshold secret sharing scheme and jointly kept by multiple supervisory nodes. t is the minimum number of nodes required for decryption, and n is the total number of nodes keeping the private key fragments. The values ​​of t and n are determined based on the balance between privacy and security control level and collaborative operation efficiency. High-risk data, after encryption, needs to have its ciphertext fingerprint calculated using a secure hash algorithm, and then an integrity check code generated by combining it with the ownership tag. Medium- and low-risk data can be directly uploaded as encrypted ciphertext, and the check code generation rules are determined based on the integrity assurance requirements for the transmission and storage of high-risk data.

7. The method for cross-industry health insurance data sharing and privacy protection based on homomorphic encryption according to claim 1, characterized in that, Perform fusion calculations to generate encrypted comprehensive risk adjustment coefficients. The specific method is as follows: Homomorphic encryption is performed on the insurer's core data using a public key (PK) to obtain ciphertext insurance factors, including ciphertext-formatted occupational basic risk coefficient [c], ciphertext-formatted historical claim frequency correlation value [q], and ciphertext-formatted historical amount correlation value [s], which are then associated with the corresponding business tag B and risk tag. The basic risk value [r] of the corresponding risk label after encryption is determined. By extracting business scenario keywords from the request, precise matching is performed with the business label B of the data in the encrypted storage pool. All encrypted medical data and insurance data that are consistent with the request scenario and related to the target insured are selected. Predefined fusion calculation logic is directly executed in the encrypted domain, using formulas. Generate encrypted comprehensive risk adjustment coefficients, in which Represents the plaintext weight coefficient, satisfying The specific value is determined by historical data analysis and business rules. The encrypted baseline risk offset is set to the lowest risk baseline value in the business scenario. When high-risk medical data ciphertext with integrity check code MAC is involved in the calculation process, the computing node needs to use the linear operation characteristics of homomorphic encryption to synchronously generate the aggregate check code.

8. The method for cross-industry health insurance data sharing and privacy protection based on homomorphic encryption according to claim 7, characterized in that, The method for synchronously generating aggregated verification codes is as follows: Synchronous generation of aggregated verification codes is achieved using the formula Received, among which This represents the encrypted underlying risk value of the disease. The corresponding ciphertext fingerprint, and other factor fingerprints refer to those that participate in the ciphertext fusion calculation, excluding... fingerprints corresponding to other high-risk encrypted data, To aggregate the ownership tags of high-risk data participating in the calculation, the local verification code and the data uploaded by the computing node are combined. If the comparison is consistent, the verification is successful; otherwise, the calculation process is deemed abnormal.

9. The method for cross-industry health insurance data sharing and privacy protection based on homomorphic encryption according to claim 1, characterized in that, Regulatory nodes that meet the threshold conditions collaboratively decrypt the data, and then apply the decrypted comprehensive risk adjustment coefficient to scenarios such as premium pricing, underwriting conclusion generation, and claims risk prediction. The specific method is as follows: Collaborative decryption requires the data user to submit compliant application materials, which include a business scenario description, the target policyholder's authorization document, aggregated verification code, and aggregated value of ownership tag, ensuring the relevance of the application and decryption requirements. The threshold condition is based on the minimum number of decryption nodes t, and the value of t is determined according to the balance between privacy and security control level and collaborative operation efficiency. The key management center will initiate the private key reassembly and decryption process only when at least t regulatory nodes report valid private key fragments. The entire decryption process is logged, including information on the response of regulatory nodes, the time taken for private key reassembly, and the information of the person performing the decryption. The logging rules are determined based on the requirements for operational traceability and liability determination. The comprehensive risk adjustment coefficient obtained from decryption is adapted to the corresponding rules for each scenario, taking into account the specific characteristics of that scenario. In the premium pricing scenario, the calculation logic of benchmark premium × comprehensive risk adjustment factor × additional risk factor is adopted. The value of additional risk factor is determined based on the scenario-related characteristics such as region, policy, and industry risk trends, so as to achieve accurate matching between risk and premium. In underwriting scenarios, multiple judgment thresholds are set according to the comprehensive risk adjustment coefficient, which correspond to underwriting conclusions such as standard underwriting, underwriting with additional premiums, and rejection. The division of the thresholds is determined based on the insurer's risk tolerance, industry risk control standards, and business rules. In claims prediction scenarios, the comprehensive risk adjustment coefficient is mapped to three levels of claims risk: low, medium, and high. The boundaries of the classification are determined based on the distribution characteristics of historical claims data and the accuracy requirements of risk prediction. A special tracking mechanism is established for policyholders with high risk levels, and the specific form of the tracking mechanism is adapted to the needs of optimizing the allocation of claims resources.

Images