How can your health plan tell if a fraud solution’s claims are valid?
Just in the last year, dramatic leaps forward have occurred in advanced technology, and new healthcare fraud solutions have emerged to take full advantage of these improvements. According to futurist Peter Diamandis, these advances are made possible by one thing, but its impact is far-reaching: “Computation is the foundation. Be it classical or quantum computing, as it becomes faster and cheaper lots of technologies that use it also become more capable. For example, communication networks, sensors, robotics, augmented and virtual reality, blockchain, and AI are all exponentially improving.”
As applications of Artificial Intelligence (AI) evolve beyond buzzwords to usefulness in the real world, some legacy healthcare fraud solutions have adjusted their messaging to put a new shine on old technology. If you’re a data scientist, you may easily intuit the difference. For the rest of us, it requires a bit of investigating.
Maybe the fraud solution you are evaluating actually is an application of AI as the technology vendor claims. Maybe it isn’t. But, how can you tell? Here are three questions you should be asking when you seek out a modern healthcare fraud solution.
1. What do you mean by “AI”?
Too many technology vendors – in all industries – throw around the term “AI” because it’s largely and too easily misunderstood. Layperson understanding of the term implies AI is “computers doing stuff for you.” Academically speaking, AI refers to building algorithms that produce outcomes indistinguishable from human cognition. They key word here is “artificial” not “intelligence.”
True AI applications should not additionally tax your already-constrained data science resources, but rather free them up to focus on higher-value activities. In a recent survey, industry leaders said they see AI as a pragmatic solution today for “a variety of administrative challenges such as automating pre-authorizations, managing electronic health records, and detecting fraud, waste or abuse in reimbursement.”
In fact, one innovative regional health plan that serves 4.5 million members has reported significant progress towards detecting fraud schemes pre-pay after adding AI methods to their multi-pronged approach to mitigating FWA.
Red flag: Vendor can’t provide a thoughtful answer to this question.
2. What methods support your AI approach?
Because AI is an application of methods, it pays to question your potential vendor partner about these details. Especially if your “sixth sense” kicked in after their answer to the first question, this question is key to digging deeper into those fuzzy details. But first, let’s define a couple of terms you may hear during this line of questioning:
Neural networks: Seeks to simulate human brain processing, which is facilitated by networks of neurons. At its simplest, a neural network processes information in three layers: 1. Input layer where data enters the system, 2. Hidden layer where data is processed, and 3. Output layer where the system decides what to do with the information.
Deep learning: Allows for increasing numbers of layers through which data passes, where each layer of nodes trains on a distinct set of features based on the previous layer’s output. The further you advance through the layers, the more complex the features your nodes can recognize, since they aggregate and recombine features from the previous layer.
Though these terms are often used interchangeably, neural networks and deep learning are related but different. The way they differ lies primarily in how they process information. Neural networks require a great deal of structured historical data in order to train their learning and decision making. As a follow-up question for vendors claiming this method, ask them how they train the neural nets, which require historical data. Training on just SIU data, for instance, would limit the neural net’s effectiveness because it isn’t fully representative of broader claims data.
On the other hand, deep learning is able to learn from data that is both unstructured and unlabeled. A boon for healthcare where at least 80% of data – images, medical records, etc. – is unstructured. A recent article explains the advantage of this model. “Unlike the resource-intensive and largely static nature of traditional payment integrity processes, intelligent algorithms continually learn and evolve with each claim.”
Red flag: Vendor can’t articulate what type of underlying method supports their AI approach. Or, they refer to SQL queries, clustering, etc. that are remnants of older, less robust technology.
3. How is domain knowledge and expertise being incorporated into analytics?
No matter how reality-based the AI application is or how sophisticated the underlying methods are, the models will be insufficient to the task of detecting healthcare fraud unless they thoughtfully incorporate domain knowledge.
On one end of the spectrum are rules-based engines, which are static and therefore surface only known schemes. Legacy fraud solutions often fall into this category; still relevant but not flexible enough for the modern environment.
On the other end of the spectrum are naive data analytics, which detect plenty of outliers but don’t distinguish between good/bad data and therefore result in too many false positives. Think of credit card fraud detection systems, which can’t account for the complexity inherent to healthcare fraud scenarios.
What’s the happy medium between these two extremes? For SIU divisions to prioritize cases appropriately, they need to be able to understand the motivators for flagging potential fraud. By combining the best parts of each approach, a modern FWA solution should be able to detect aberrancies from a data perspective and use domain knowledge to impute meaning into findings.
Red flag: Vendor can’t speak to false positive rate or can’t articulate how domain knowledge is incorporated into analytic models.
You can have it all:
Modern technology has finally caught up to the complex scenarios inherent to healthcare FWA. Instead of persisting with outdated solutions that simply check fraud compliance off the list, your SIU department can find previously unknown schemes, reduce false positives and improve time to resolution.
ClarisHealth is continuing to develop Pareo Fraud Detection using deep learning methods to build nuanced algorithms and multi-tiered provider scoring on multiple models, weighted in importance, in a payer-specific hierarchical nature, so investigators can understand exactly why a claim is flagged for further investigation. This structure also ensures that the issues relevant to the individual payer are taken into consideration.
The models are being developed in partnership with the University of Illinois Chicago Center for Research in Information Management. Its leading-edge academic research, ties to the UIC College of Medicine, and analytic methods and technology is headed by a data scientist with over 20 years of experience leading Payment Integrity and FWA analytics for payer organizations are a boon for health plans adopting Pareo Fraud.
- Robotic Process Automation (RPA) allows your most valuable resources to concentrate on high-value activities instead of tedious administrative tasks. Get to exactly what you need in fewer clicks.
- Modern analytics based on deep learning methods push the leads to you, rather than manual queries to pull leads.
- Automated reporting for state Medicaid and non-government entities keeps you in compliance.
- Seamless integration of detection – analysis – case management to Audit, COB, Data Mining and other areas of Payment Integrity minimizes provider abrasion and administrative burden.
Now’s the time for total payment integrity
See the ClarisHealth 360-degree solution for total payment integrity in action.