1. Industry application background
With the development of technologies such as big data and artificial intelligence, health insurance has entered the 3.0 era, and services in the digital transformation of the insurance industry have gradually migrated to online, and the use of data compliance has brought unlimited possibilities for the optimization of insurance claims models.
Among them, commercial health insurance, as an important part of promoting the construction of a multi-level medical security system, is of great significance to national medical health. In January 2022, the Personal Insurance Department of the China Banking and Insurance Regulatory Commission issued the "Notice on Printing and Distributing Reports on Issues and Suggestions for the Development of Commercial Health Insurance" to all personal insurance companies across the country: "Strive to achieve sufficient information sharing with the information system of medical institutions and improve settlement services; On the basis of ensuring information security and personal privacy, strengthen the application of medical and health big data, promote the reform of medical payment methods, and better serve medical insurance policy formulation and medical expense management."
In the face of claims business upgrade needs and regulatory requirements, insurance companies need to give priority to data compliance in the process of service innovation, so privacy computing provides this neutral and credible technical support.
In order to solve the problems of claim settlement experience, cost and efficiency of inpatient medical insurance for tens of millions of insured users, the ant insurance technology team cooperated with the insurance company to build a "claim settlement brain" based on the claim settlement technology platform and **privacy computing framework "lingo"** " Intelligent claims system .
2. Overall introduction of the case
The system is based on machine learning of inpatient medical insurance claim voucher images, with the help of data advantages (hundreds of thousands of typical claim settlement cases) and supplemented by certain knowledge constraints, the multi-modal medical voucher recognition of visual recognition + text classification + text semantic understanding is realized The model (with an accuracy rate of more than 95% for 100+ types of medical claim settlement vouchers) breaks through the in-depth structure of medical vouchers and the "expert-level" high-confidence auxiliary claim decision-making ability that can be commercially applied on a large scale , helping insurance institutions to improve the efficiency of claim settlement by more than 70%.
The system's large-scale online data-based cooperative investigation capability based on the "lingo" framework further reduces the cost and time spent on offline investigations by insurance companies, and pushes the digital and intelligent application of medical fact investigations to a new level.
3. Application of privacy computing in this case
In order to effectively find positive leads and reduce the risk of wrong claims. It is necessary to use external medical data in compliance with regulations to give full play to its value. Existing MPC technology is suitable for multi-party joint modeling (such as joint risk control scenarios), but it is not suitable for policy-driven and strong rule calculation positive claim risk discovery scenarios.
In the whole system, health insurance customized multi-party data joint analysis solution is one of the core modules. Based on the MPC SQL multi-party joint analysis field-specific language provided by "Yiyu", this project builds a customized multi-party data joint analysis solution for health insurance, covering official legal source medical data in provinces that account for 50% of the country's cases. Standard use provides a typical case.
This solution is based on multi-party secure computing technology, using secure encryption algorithms to jointly analyze multi-party data. Help insurance companies and their external medical data ISVs to conduct joint analysis without leaving the original data locally and protecting the value of the data. On the basis of protecting user privacy to the greatest extent, it meets business needs.
Fourth, the multi-party joint analysis process
Based on the multi-party joint analysis capability of the "crypto" framework, Ant Insurance and the insurance company introduce external data sources to complete the joint analysis process under the premise of ensuring the data privacy of all participants in the joint project as follows:
- Node deployment
The user obtains the deployment package with one click, fills in the node ID & token information and executes the script, and the local node deployment can be completed in a lightweight manner within hours;
- Data preparation (for pre-security configuration)
Both parties load their respective sample data to their respective local analysis nodes, register the data table structure of the corresponding sample on the platform, and agree to authorize access to the multi-party security analysis project.
- rule development
Rich MPC SQL operator support based on argots, users can describe secure calculations based on multiple data sources in scripts, and complete the union by combining statements such as "SELECT FROM", "JOIN ON", and "GROUP BY" The statistical results of the analysis are generated; the intersection results can be exported to the local node through the "SELECT INTO" statement.
- rule deployment
Subsequently, users can use the data authorized by the ISV to complete online debugging and optimization rules on the platform through the online SCQLIDE provided by joint analysis, and the rules after debugging and verification can be used as standard rules for large-scale deployment in more data sources.
- calling rules
After the rules are deployed, users can initiate calls to the rules on the platform side, and the argot supports simple visualization of data analysis results, such as the distribution of doctor visits, frequency of visits and other analysis results.
- Data source expansion
Users can also continue to introduce external medical data to further enhance the ability to enrich the underlying data, further enhance their own claims settlement and risk control capabilities, and increase intelligent decision-making services for underwriting scenarios.
5. Breakthrough in privacy computing technology
- Pre-data security configuration Data resource classification
In the data preparation process, users can use the CCL pre-security configuration function of Lingyu, with the support of MPC-related technical capabilities, to classify data assets, and ensure the security of data with a high security level through pre-configuration. Private data is not leaked during calculation.
- Rich MPC SQL operators support writing scripts to describe secure calculations based on multiple data sources
Argument support: arithmetic calculation (+, -, *, /, %), comparison (>, <, >=, <=, =, <>, IN, NOT IN), logic calculation (AND, OR, NOT), Window aggregation (group by … min, max, avg, sum, count, median ), control (IF, CASE WHEN), sorting (RANK, ROW NUMBER, ORDER BY), date functions (DATE_DIFF, DATE_ADD), other functions (ceil , floor, round…) rich operators.
- Underlying data capabilities continue to enrich capabilities
As shown in the layered overview diagram of the lingo framework above, lingo is oriented to the business delivery team in the resource management layer, which can shield the differences in the underlying infrastructure of different organizations and reduce the deployment and maintenance costs of the business delivery team. On the other hand, the nodes, data, and member core resources in the joint project can be managed in a centralized manner, and an efficient and collaborative data collaboration network can be built.
- Combined Exploration of Privacy Computing and Multiple Technologies
In the entire intelligent claim settlement system, the lingo framework focuses on the medical data introduced into the ISV in compliance with regulations. In the process of data analysis and machine learning, it also involves technologies such as multi-modal classification and recognition of medical vouchers, medical text NLP deep learning engine, etc. It is a typical exploration of the comprehensive application of privacy computing and other technologies, and has a paradigm effect on the collaborative mining and utilization of more types of data value such as images and texts.
6. Case business results
The customized multi-party data joint analysis solution for health insurance based on the adage MPC SQL multi-party joint analysis language is conducive to improving the ability to identify positive cases and investigate path planning. It covers official medical data from legal sources in provinces where cases account for 50% of the country . The compliant use of medical data provides a typical example. It is beneficial to effectively find positive clues, reduce the risk of wrong claims, control the operating costs of claims through digital investigation and review, and is more conducive to expanding the service scope of inclusive medical care and improving the service efficiency of inclusive medical care.
After the "Claims Brain" intelligent claims system was launched, compared with traditional offline investigation operations, the average cost of investigation cases was reduced by 40%, and the compensation rate was controlled at a reasonable level, ensuring the sustainable and healthy development of the business.
7. Prospect of case promotion
The implementation of the customized multi-party data joint analysis solution for health insurance is not only conducive to the sound development of commercial health insurance in terms of cost reduction and efficiency increase, but also expands its application in cutting-edge technology cooperation, innovative drug research and development, high-end medical device research and development and application in the medical industry. Disease risk assessment, disease prevention, classified diagnosis and many other scenarios, connecting various types of medical and health data.
In my country, health and medical big data, as an important national strategic resource, has far-reaching influence in many aspects such as management decision-making, public health, clinical research, services for the benefit of the people, industry governance, and industrial development. The construction of the big data ecology of medical and health is conducive to balancing the differences in the geographical distribution of medical resources, promoting the rational distribution of social resources, and improving the overall level of national health services.
| This article is contributed by Ant Group
Lingyu official website:
https://www.secretflow.org.cn