Fraud, Waste and Abuse (FWA) is an area of focus in the healthcare industry. In the US alone, even conservative estimates of losses due to fraud in healthcare peg it around $80 Billion. A few other sources suggest that it could be as high as $200 Billion. Unfortunately, less than 5% of this is being recovered. These trends are relevant to even countries like Norway, Sweden, Canada and United Kingdom that have adopted single-payer model where healthcare is paid for by a single public authority.
There may be less of fraud, but more of waste and abuse.
How is FWA detected today?
Companies are primarily using rule-based models to detect fraud. Rules are created when a pattern is unearthed. Patterns are identified when repeated misrepresentations occur. A human element exists today to spot repeated misrepresentations before that particular pattern can be fed into a rule engine. These efforts tend to be retrospective by design and can be treated as simple AI.
Detecting FWA may not appear to be complicated at the outset. However, when you consider the amount of data that needs to be processed and the transformation that must precede before people can identify trends, the complexity becomes evident. Another concern is that the analytical tools and rule engines today work on structured data. But, for identifying patterns, unstructured data hold the key.
Problems and available solutions?
There are 3 main factors at the base of the problems that industry faces today.
- Vast amounts of data, which is heterogeneous, in multiple formats, multi-spectral and unstructured in nature.
- Finding patterns in such data which is deep and wide, is akin to looking for a needle in a haystack with no unique identifiers. This is an issue of data consolidation.
- Healthcare data is many cases can be incomplete and comprise of imprecise observations derived from multiple sources using incongruent sampling.
It is possible to leverage a combination of emerging technologies to break-down the bigger challenge into smaller problems and tackle them in a manageable way. A combination of AI, ML (both are in nascent stage) along with big data technologies can solve many problems listed above, if not all.
What problems are we solving today?
Big data is for sure the game changer as it can manage unstructured data that is heterogeneous and in multiple formats. Machine learning can help us to learn non-human programmed insights and discover patterns. It can also create new rules by itself once a pattern is identified. By adding AI interfaces to these, we can go couple of steps ahead and save quite a lot of money for the industry.
The AI/ML advantage can help us process vast amounts of unstructured/heterogeneous data, apply human learning via machines & detect patterns pre-emptively all in one harmonized flow.
How to go about it?
The first step on the AI/ML roadmap is to build the capability to ingest and operate on unstructured, heterogeneous data. This can translate to significant savings in data pre-processing stage. Implementing ML will help one reduce effort in programming and building rule engines and it can act as an analyst in much more precise manner.
Unearthing an entirely new fraud scenario can still be a challenge. But moving ahead, the new technologies can help with a better handling of FWA. We need to move from a traditional retrospective approach to a preventive model by processing the patient, pharmacy and hospital data through AI/ML.
The benefits and ROI are clear, and it now needs the industry to wholeheartedly embrace new technologies that can greatly reduce the FWA issue that confronts it.
AUTHOR: Srinivas Iyengar, Vice President & BU Head – Healthcare & Insurance, EVRY India