Most tools used to detect potential fraud in health
insurance claims are rules-based. They analyze at
the level of a single claim, in isolation, and identify
forms of abuse that are either pre-determined or
already known.
In 2005, a Canadian Life Insurance company
provided a database of its dental insurance claims
to a company called Claim Analytics, to determine
if their methodologies could offer fraud-detection
benefits beyond those of rules-based methods.
Claim Analytics uses predictive modeling
techniques to offer an approach to insurance
analytics that takes full advantage of modern
computing powers.
For the project, Claim Analytics used the
techniques of Principal Components Analysis and
Clustering Analysis to analyze over 1,500 dentists,
and approximately 200,000 dental claims. The
approach considered each dentist’s entire claims
portfolio over a specified period, allowing the
identification of types of abuse that are otherwise
difficult to detect, e.g. practitioners who add claims
for minor unperformed services to each bill on a
regular basis.
The analysis proved to be very effective in
identifying dentists whose portfolio of claims
differed significantly from the norm. As
demonstration of its effectiveness, the study
includes an analysis of a wide-ranging sample of 14
dentists who were identified as having atypical
claims activity.,
As an exploration of how predictive modeling can
be used in fraud detection, this study delivered both
specific examples and a strong overall indication of
the increased capabilities of this approach.