Healthcare insurance fraud represents a growing threat, as a weakened economy creates incentives for criminals. Larry Jacobson describes a technological approach to fighting back.
This leaves the door open for emerging threats that capitalise on the fact that rules-based systems don’t yet know what they’re looking for.
Many insurers, for example, have a rule that a claim filed less than 30 days after a policy is written should be flagged and investigated.
This seems sensible enough, but in reality it is precisely the kind of rule that sophisticated criminals will anticipate and outsmart.
In fact, FICO has seen evidence of policy holders that will pay the first couple of premiums and avoid claims for a period in order to masquerade as a ‘good’ account before making a significant claim to maximise the fraud.
This phenomenon also observed in the credit industry, where it is known as ‘bust out’ fraud.
The rule may end up penalising more good customers with genuine claims than it does fraudsters.
At the same time, abnormal patterns can go undetected. In one US case, a provider ordered repeated hearing tests on children, walking away with $3.5m in excess charges over the course of five years. Rules-based detection would have been challenging.
How would a rule be written so as not to exclude children who legitimately required multiple hearing tests? Yet claims that seem legitimate in isolation may be part of a larger pattern of fraud, waste and abuse.
Predictive analytics could flag those claims from that provider as unusual or aberrant, forwarding them to analysts for additional review, highlighting a systematic pattern of abuse and catching an expensive problem long before rules-based claims analysis identifies an issue.
Predictive analytics use algorithms to look for aberrant patterns in the data, that don’t make sense.
This allows unusual or problematic claims to be flagged and reviewed before the abuse becomes widespread.
Using complex, multi-dimensional analysis, predictive models score each claim. Scores indicating high risk of fraud can direct analysts to only look at the most aberrant claims and providers.
The score is accompanied by one or more explanations, highlighting the reasons why the claim or provider appears to be suspicious.
Analysts can use this information to quickly determine whether a claim is legitimate, fraudulent, or worthy of further investigation.
Used in pre-payment mode, predictive analytics can identify problems like adjudication errors, upcoding (a fraudulent practice in which provider services are billed for higher procedure codes than were actually performed), OCR scanning issues, unit inflation and payment policy weaknesses, thereby decreasing the volume of payments made on fraudulent, erroneous or abusive claims.
Most claims can be analysed in less than three minutes, ensuring that there are no processing delays that would risk problems with prompt payment legislation.
This gives insurers the ability to avoid ‘pay-and-chase’ treatment of claims. A fraud system can also address post-payment analysis, examining large data sets – often several terabytes of data – in complex, nonlinear ways to identify aberrant patterns in patients, providers or procedures that aren’t evident in smaller batches.
This increases detection accuracy while improving investigator productivity and recovery success.
