Insurance fraud is a grave problem in America, costing insurance companies and their customers more than $40 billion a year, according to the FBI. This is excluding medical insurance fraud, which is estimated to cost tens of billions more. Typically, insurers utilize medical peer review to determine whether the treatment provided to a claimant was medically necessary and whether the claim is a legitimate one. The peer-to-peer (P2P) process involves a conversation between the claimant’s attending physician or consulting physician and the Insurance company’s clinical case manager or medical director. The attending/consulting physician may or may not convince the insurer’s physician that the medical services were appropriate for the patient. The claim could be genuine, in which case the claimant receives the reimbursement that is legally due to him/her, or it could be fraudulent. In the latter case it would unnecessarily increase healthcare costs.
Healthcare insurance fraud is a staggering reality with hospitals and other providers billing patients and their health insurers for lab tests they never performed, or by providing a test that is a less expensive one than what the hospital billed for. It is unfortunate that healthcare that is directed at helping people stay well or recover from injuries or illnesses, is subject to different types of fraud that causes financial harm and inflates costs in the healthcare system. Typically, human fraud examiners are assigned the task of collecting and analyzing huge amounts of data from multiple sources to detect fraud. Since these methods are not 100% successful, and fraud, waste and abuse are quite rampant in the industry, healthcare insurers are turning to artificial intelligence (AI) to better fight insurance fraud.
AI and machine learning tools imitate a human who is reviewing a claim, provider and member/beneficiary risk for fraud, abuse and waste. AI has the advantage of always being reliable, with keen insight, and providing consistent results. It is programmed to identify suspicious behavior and anomalies. Large-volume data that is usually handled by humans is now being fed into machine learning systems for scrutiny. Though such initiatives are only starting to be undertaken, some insurers such as Aetna are applying it to everyday business. The company uses a flexible, intuitive, and intelligent system powered by machine learning models to detect fraud more efficiently. Here are the steps the company took to implement the AI system, according to Aleksandar Lazarevic, a senior director at Aetna’s analytics organization and the person heading their machine learning fraud program.
- A data pipeline was built by collecting and aggregating all relevant data.
- A series of models were built, with the initial ones being based on the familiar signs of fraud – providers submitting unusually large number of claims for specific procedures, or seeing an unusually large number of patients per day.
- The next step was deploying the supervised machine learning models (second type of models) to identify variations of known fraud patterns, i.e., new types of fraud that are similar but not identical to existing patterns. This is something traditional analytics dashboards may find difficult to detect.
- The third type of models are anomaly detection models designed to detect new or emerging types of fraud. An anomaly detection model works on a kind of AI-powered intuition or instinct. Though it may not be clear why a specific pattern is relevant, it flags the data as suspicious. In healthcare fraud this pattern could be a combination of medical practice type, billing data and other clinical data that the AI-powered system thinks requires further investigation.
- As the final step, the new processes were incorporated into the fraud investigation team’s workflow. They were able to develop a tool that automatically sends all their findings to the fraud investigation unit.
The individual anomalies or patterns sent to the fraud investigation unit would be assigned to an individual investigator. The findings would also include a reason code that signifies why it was flagged by the AI system. This would help investigators who are following up and performing a complete investigation. In other words, the AI system provides a lead to investigators as regards why a particular provider is suspicious and thereby providing a starting point to work from.
Some fraud scenarios and how AI could assist are as follows.
- Providers billing for services patients did not receive: AI can review all the data available to see if there is documentation to prove that the patient actually received the billed service.
- Billing a simple procedure as a much more complex one (Upcoding): AI uses anomaly detection to understand the typical treatment for a condition and identify any deviations made by a provider. This anomaly is flagged as potential fraud.
- Corruption and kickbacks: AI and machine learning tools can analyze behavioral and transactional data to find out corruption. It would consider disparate data sets such as the history of referrals made by a physician and the travel expenses they had covered by a person/company benefiting from those referrals.
That said, how can insurers use AI and machine learning to the best effect? Here are some considerations.
- Ensure that you are ready for the adoption of AI: Ideally, insurers should base their adoption of this innovative system on a payment integrity strategy that clearly reflects applicable industry best practices as well as the specific requirements of their organization. It should also identify quantifiable performance gaps that they plan to resolve with artificial intelligence and machine learning solutions.
- Stay vigilant and ready to meet new challenges: Fraudsters may soon adapt and use new strategies. The AI and machine learning system insurers implement should be scalable and automatically identify/respond to any changes in claiming behavior.
- Ensure all stakeholders are on board: It is important to involve all stakeholders so that you can get their support and maximize the payment integrity results you achieve with the new AI system.
- Stay away from propagandists: When choosing a vendor, insurers must be aware of hype. Ensure that you are clear about your requirements and understand clearly what the vendor is offering to meet your needs.
A medical peer review company assisting insurance fraud lawyers with medical chart reviews understands how useful AI and machine learning tools can be for the insurance industry in their battle against fraud. When it comes to the success of AI tools to detect fraud, it is vital that all data is available at one place so that a comprehensive view can be provided into human behavior and processes from which AI models can be trained. The EHR transition can help immensely in this regard. Another important thing to note is that human review would remain a key element in determining whether someone actually intended to be deceitful. AI has limitations in that as yet it cannot always determine on its own whether some wrongdoing was intentional. For example, it cannot determine whether a wrong billing code was intentionally applied or whether it was a typo in the health record. So, until AI can be trained in this regard also, human contribution will continue to be relevant.