Our client, the Municipal Compulsory Medical Insurance Fund, is a state agency formed to finance compulsory health insurance programs and control the rational use of funds.
Insurance fraud is a serious problem for the Fund. Fraudulent claims make up a significant part of all claims received by the Fund, which leads to severe financial losses every year.
To prevent such heavy losses, the Fund decided to take a more rigid approach to detecting and responding to fraud claims. Toward this goal, they were looking to hire specialists to develop an automated fraud detection solution, and that is how they turned to our partner company.
This partner had been working with us for several years and knew about our strong expertise in building complex algorithms and optimizing business processes. The partner invited us to participate in the project and take on some of the heavy tasks.
A typical medical fraud insurance scenario
In the case of our client, the most frequent violations were related to fraudulent deals. They could occur in the following way:
- Patients visit medical institutions and receive medical services.
- Once a month medical institutions send invoices to insurance companies. At this stage, fraudulent collusion is possible.
- Insurance companies submit the statistics on all the invoices received from medical institutions to the Municipal Compulsory Medical Insurance Fund.
Working together with our partner, we divided the functions as follows:
- The partner company was responsible for business analysis: their team developed specific rules and analyzed cases (see below).
- Our experts were responsible for development: we built machine learning algorithms, set up batch processes, and created a system that processed data under high-load conditions according to the developed rules.
To identify fraud committed by the insurance companies and medical providers subordinate to the Fund, we developed a comprehensive ML-based solution.
It covers the following steps:
- Creation of rules, a set of characteristics used to group invoices into cases.
- Algorithmization of rules.
- Application of rules to cases with their further analysis, where a case is a series of procedures followed by a medical center when treating a particular condition.
- Rule updates that result in further training of algorithms.
- Automated decision-making for detecting incidents, or suspected violations.
Getting down to insurance fraud detection software development
Adoption of business rules.
To manage business rules, we used the FICO Blaze Advisor system. One of the project challenges came from the fact that these business rules were applied for the first time at the client’s organization. Our partner’s analysts couldn’t yet define the sensitivity of the coefficients, so we suggested they adjusted them in the process by defining lower and upper threshold values and observing the indicators.
Fraud detection with an invoice grouping algorithm.
All the invoices, which were to be verified with automated algorithms, were coming from different channels and contained information in different formats. Thanks to the invoice grouping algorithm, RNDpoint’ engineers were able to group all new invoices into cases. Our team created the profiles of medical conditions (descriptions of typical treatment procedures), applied them to these cases, and checked the magnitude of the data discrepancy. In 72% of cases, we correctly determined the characteristics of the treatment and identified whether the case contained any violations.
RNDpoint’ engineers focused on making the system as user-friendly as possible. Our team made the user flows clear and coherent so that the Fund’s representatives could find and manage key indicators on their own using the business rules.
How the invoice grouping algorithm detects fraud: an example
- A patient consults a doctor regarding a certain condition.
- The doctor prescribes treatment, which includes physiotherapy sessions.
- The patient’s invoices are submitted to the Fund for verification.
- The algorithm groups the invoices into a particular case.
- The profile of the attributed medical condition is applied to the case.
- It turns out that the prescribed physiotherapy is unnecessary for the treatment of this condition.
- The doctor’s actions are analyzed and it turns out that the prescription of such treatments was unreasonable, with the only goal to make more profit for the medical organization.
The case is marked as fraudulent.