Tinker, tailor, doctor, fraudster. Middle East Insurance Review, February 2019.
“Healthcare insurance fraud is a reality that the industry has to grapple with. Insurers need to embrace technology to defend themselves against waste, abuse and fraud”, says Mr Kiril Milev of Netcetera Middle East.
Adults consulting specialist pediatricians. Nephrologists conducting unrelated laboratory tests such as chest X-rays and EKG. Endless investigations ordered without consultations. These are just some of the cases of waste, abuse and fraud (WAF) insurers in the UAE failed to detect despite teams of claim handlers and medical experts.
In 2017, a team of experts from Netcetera decided it was time to address this issue and embarked on a mission to develop a solution which could potentially transform the healthcare industry in the UAE for good. With more than 10 years of experience in the local healthcare insurance industry, the team was well-equipped to take on the challenge that, so far, no one else had.
The problem is widely recognised, but insurers had been unsuccessful in dealing with it using old-fashioned rule-based systems. On the other hand, regulators had only taken a few enforcement actions and offered no official recourse. As a result, premiums kept increasing because payers had to account for losses incurred due to WAF.
Putting together the A-team
Fortunately, advances in machine learning (ML) and artificial intelligence (Al) proved to be a real game changer, providing the insurance industry with the perfect solution.
Netcetera’s office in Dubai brought together a group of data scientists, academics, Al experts and software engineers with a single goal in mind: to harness Al for detecting doctors and clinics making fraudulent, wasteful and abusive healthcare insurance claims in the UAE. By February 2018, teams from Dubai, Zurich, Skopje and Liechtenstein started working on a new solution and began building RiSIC, a fraud detection system.
The RiSIC team grew as claim handlers and medical experts from a leading UAE insurer joined the project to bring in depth of expertise and provide access to hundreds of thousands of real healthcare insurance claims needed to develop the software properly.
Of man and machine
The starting hypotheses was that perhaps 8-10% of e-claims could be WAF claims. The RiSIC development team started by creating a number of complex algorithms to match the most frequent and known WAF patterns such as upcoding, unbundling and overbilling. Based on these algorithms, the first version of the software was built using Al to score the risk of each claim indicating the likelihood of WAF. At the same time, the insurer’s claim management and medical team was also scoring a random sample of claims from the same batch of claims. This single-blind scoring exercise enabled the team to re-train the Al software algorithms to be even more accurate based on insights provided by the evaluators and medical experts. This was a prime example of man and machine working together.
In August 2018, RiSIC was finally fine-tuned to best possible detect accuracy and approved by the lead data scientist Gjorgji Madjarov (PhD) from the Faculty of Information Science and Computer Engineering in Skopje, Macedonia. A significant batch of claims that were kept completely separate from claims used to train the RiSIC Al models could now be processed for ‘live’ risk scoring. The Netcetera team uploaded over 370,000 UAE healthcare insurance e-claims into the RiSIC server. The complex set of neural networks started to check each claim against WAF patterns and scoring the claim based on the response. Only 18 seconds later, each one of these 370,000 claims had been processed accurately and given a risk score.
Stamping out fraud
Once the results were in, it was instantly evident that insurers are facing a systemic and endemic problem. The World Health Organization estimates that 3-10% of healthcare costs are WAF, a widely recognised benchmark. What RiSIC found out was that 15.7% of the 370,000 claims were
likely WAF. These claims should have been denied, but yet a majority of them had escaped scrutiny, and insurers had paid them.
Further analyses showed that, while a single claim might not appear as being a WAF claim on its own, but when such a claim was grouped with thousands of similar claims and subjected to Al to detect patterns undetectable even to the trained human eye, this same claim can turn out to be a WAF claim. In other words, the insurer and claims managers were far inferior to Al when it came to the ability to detect WAF claims in terms of speed and accuracy.
Further results were even more encouraging. An Al-based approach would have increased a denial rate considerably and thereby provided significant cost savings. With a little extra statistical analysis, the team was able to identify both doctors and clinics that consistently appeared to engage in abusive activity. Over 4,000 doctors had the highest risk scores pertaining to WAF claims. The team also found out that 27 doctors had filed claims worth over AED2.2m, flagged by the Al as WAF claims. Investigations performed by the claims and medical teams fully concurred that the Al findings were accurate.
The results are a testament to the incredible power of today’s Al and ML techniques. Furthermore, the RiSIC team had proven that it was possible to fight against WAF in the UAE healthcare insurance sector. However, challenges remain.
You can read a full article here (subscription is needed).
About Middle East Insurance Review
Middle East Insurance Review (MEIR), a monthly publication launched on 1 September 2006, aims to meet the information needs of insurance practitioners in the Middle East and North Africa (MENA) region and the global takaful industry.