Machine Learning Applied to Detecting Fraud in Healthcare
Fraud is a major contributing factor to escalating healthcare costs: fraud costs the U.S. about $60 billion annually and accounts for up to 10 percent of total healthcare spending.
Is there a way to prevent or preempt healthcare-related fraud, or at least make a dent in these numbers? Discovering fraudulent activities requires finding patterns in the data—integrating data from silos across the organization—including patient data, hospital and physician data. In other words, using advanced analytics to discover hidden connections that maps the data across the datasets.
The Cure is in the Data
An important tool in the fight against fraud is artificial intelligence. We can generate graph-based machine learning features for a low risk provider (“good doctor”) and a high risk provider (“bad doctor”) and use these features to train the artificial intelligence to look for these profiles within huge healthcare datasets.
Graph-based analytics can identify low risk providers (“good doctors”) and high risk providers (“bad doctors”)
For example, we can first drill down on a feature called “stable group for routine ICD codes.” The stable group includes ICD (International Classification of Diseases) code groups that are billed frequently in claims for the provider over a period of time. The “good doctor” has a stable group with one or more code groups. If a particular provider’s specialty is, for example, emergency medicine they are likely to see members with various healthcare conditions. This results in multiple ICD codes – but this is justified, as their specialty isn’t associated with a specific code group.
If a doctor, however, is routinely generating claims for multiple ICD codes that are unrelated to their specialty (such as a podiatrist generating claims for respiratory system and nervous system code groups), they may be a “bad doctor.”
The second feature called “cost of care” compares the cost of prescribed medications, tests and procedures by the prescriber and their referral network with the average cost for treating the healthcare condition for similar members. TigerGraph traverses from the patient to claims, doctors, pharmacies and treatment centers to map out the journey easily, to calculate the cost of care for each provider.
Next, the solution finds similar patients and calculates the average cost of care for opioid addiction treatment for each patient population. A “good doctor” has an average cost while a “bad doctor” will have a higher than average cost for the prescribed medications, tests and procedures.
The third feature called “potential undeclared prescriber-facility relationships” digs into undisclosed connections among providers and pharmacies as well as substance abuse facilities. The “good doctor” does not have such connections while a “bad doctor” has undisclosed connections.
An ability to look across multiple data connections can help find a doctor who is referring a large number of patients to a specific opioid addiction treatment center. This deep link analysis could uncover that one of the previous addresses for the administrator of the substance abuse treatment center is the same as the address for the doctor referring patients to that treatment center.
Detecting “Bad Doctors” At Scale
The size of the datasets that need to be examined in order to find high-risk providers can be a deterrent to even the most committed data scientist or business user. Artificial intelligence can be trained to identify the patterns associated with “bad doctors” and power this exploration at scale.
Machine learning feature generation and explainable AI power the risk assessment for providers
The graph-based machine learning features for “good doctor” and “bad doctor” are generated for each prescriber and provider and are fed into the machine learning solution as training data. They are also used for explainable AI. When a non-technical user wants to know why a prescriber was flagged as “bad doctor”, the graph-based features can provide a clear and digestible explanation.
Don’t be patient. Act now
There’s never been a better time to upgrade to graph. A Fortune 10 healthcare company is already using TigerGraph to gain an advantage by uncovering valuable connections hidden in its data.
You can read reviews by people like you who are using our technology here. You can start free with TigerGraph Cloud with multiple starter kits for healthcare as well as fraud detection or consider becoming a Certified TigerGraph Associate.