From the editor
It is estimated that fraud, waste, and abuse (FWA) is equivalent to 10% of total healthcare spending per year, or more than $400 billion annually in the US alone. That is a staggering figure which becomes even more mind boggling when we begin to think about it on a global scale. Yet, despite the impact FWA, improper payments, and other errors have on health plans, honest providers, and patients, it has remained difficult to overcome. Why is it so challenging for health payor organizations to spot, and stop, fraudulent behavior?
In this edition of Shift Insurance Perspectives we explore the world of FWA and improper payments, and the approaches health payors can take to mitigate the financial and operational impact of fraud on their organizations. We examine the importance of data, and why understanding “who’s who” in your provider network is critical to anti-FWA initiatives. We address how third-party and external data can be used more effectively in the fight against FWA. We illustrate how artificial intelligence (AI) can bolster both postpay and prepay detection efforts. And, we take a look at how emerging technologies such as generative AI are driving new fraud trends, and being used to stop them.
As always, publishing Shift Insurance Perspectives is a collaborative effort. Thank you to the Shift Healthcare Team for their contributions in bringing this edition to life.
An identity crisis
In 2022, there were 53 recorded mergers and acquisitions in the US healthcare industry. In 2023, that number rose to 65
In 2022, there were 53 recorded mergers and acquisitions in the US healthcare industry. In 2023, the number rose to 65. For health payors, this level of M&A activity can make it more difficult to detect suspicious activities or errors simply because from a data perspective, they do not have a clear picture of with whom they are doing business. M&A activity also has the propensity to introduce new data silos and/or data formats, complicating detection and analytics efforts and making it difficult to gain a clear understanding of overall plan activity.
The situation can be further complicated when providers add new services or specialties, such as behavioral health, that may operate under a different address, provider name, or National Provider Identifier (NPI). It is significantly more difficult to find provider or provider network behavior patterns if it is not clear how those entities are connected. This increases the occurrence of false positives, wasted resources, and missed opportunities for savings or recoveries.
AI-powered entity resolution, the process of identifying, matching, and merging records corresponding to the same entity across various datasets, can be incredibly useful in creating a strong foundation supporting anti-FWA initiatives. Shift’s experience has shown that health plans can significantly consolidate the number of provider entities, in some cases by close to 50%, by identifying and eliminating overlap, resolving the same data captured in different formats, and/or breaking down data silos. Knowing that the multiple “Dr. Smiths” scattered throughout the data are in fact a single “Dr. Smith” makes it much easier to see when services attributed to this provider are either inline with industry norms or represent outliers that should be investigated.
As important, effective entity resolution produces clearly defined provider IDs which allow payors to accurately uncover relationships and much larger networks of suspicious activity. The ability to identify shared addresses, patients, collusive or other shared billing characteristics, among other connections enable plans to uncover exponentially higher value cases and prioritize them for investigation. If we reuse the “Dr. Smith’’ example, once it’s known that what had been identified as myriad individual “Dr. Smiths’’ are in fact a single “Dr. Smith’’ it becomes much less difficult to see associations between different provider entities which may indicate patterns of ongoing suspicious connections.
Corroborating evidence
Another interesting benefit of entity resolution for health payors is that it opens the door to more effectively use third-party data, including publicly available provider reviews, to bolster investigations. This can be particularly useful in situations where there is evidence that a provider may be submitting false claims reimbursements associated with certain codes. Because the health plan now knows with certainty who the provider is, they are well positioned to look for corroborating evidence from a variety of different sources.
the AI can employ natural language processing, OCR, and other techniques to not only gain an understanding of the services provided, but also examine the policy level data
And while it is not likely a health plan would ever begin an investigation based on a Google or Yelp review, these may provide relevant information and context pertaining to the suspicious behavior identified in the claims documentation. For example, online reviews associated with a provider may indicate that requested reimbursement does not match the patient experience, such as reports that the provider “was rushing my appointment” or the provider “didn’t even show up for my appointment and I was seen only by a nurse or an assistant.” This type of third-party data can help verify that a provider was not administering care in a contracted manner and adds to the preponderance of the evidence.
AI and the ability to identify new schemes
Healthcare fraud is always evolving. The ability for bad actors to quickly hatch new schemes is just one of the reasons why FWA has proven so difficult to fight. We have seen this firsthand with providers taking advantage of the ambiguity of Place of Service (POS) code 99, particularly in the area of Applied Behavioral Analysis (ABA) where certain services should only be legitimately billed in a home, clinic, or school setting. Using POS 99 may be an attempt to secure a higher billing rate for the services provided.
In this case, AI is able to self-learn and spot anomalous activity identified as an increase in POS code 99 billings. From there, the AI can employ natural language processing, OCR, and other techniques to not only gain an understanding of the services provided, but also examine the policy level data associated with the procedure described to determine if it can legitimately be coded POS 99 or if there must be a clinical location associated with the claim. For example, some States allow for POS 99 to be billed with some ABA services.
In addition, the emergence of generative AI is impacting FWA in a number of interesting ways. Not only does it represent new opportunities to commit fraud, of which health payors need to be aware, but also new ways to fight it. It has become incredibly easy to create hyper realistic images — such as x-rays — or other medical documents, using DALL E and other similar image generators. And while these faked images may be difficult to discern from the real thing under normal scrutiny, AI and generative AI are able to more easily spot the fakes by examining the associated metadata. AI can also be used to help determine if the same documents or images are associated with more than one claim.
Complex medical record review
Determining whether a claim is legitimate or not often relies on a physical review of the medical records associated with the patient/procedure. Despite its effectiveness, medical record review is an arduous, highly manual, and time consuming task. AI is ideally suited, specifically generative AI and Large Language Models (LLMs), to eliminate some of the most onerous aspects of this process. These solutions can be trained on medical records, claims data, and relevant external data. Expert, industry specific prompt engineering can be applied. The result of which helps health payors determine if the information contained in the medical record supports the claim billed. This is not to say that AI/generative AI should or would replace qualified human record review.
Applying AI to the challenge of complex medical record review is about augmenting the human experience rather than replacing it
Applying AI to the challenge of complex medical record review is about augmenting the human experience rather than replacing it. AI provides the focus and resulting information to help a professional make the best decision possible about the status of a claim.
Making postpay work for prepay
Most current approaches to fighting FWA prior to a claim being paid are tied to more traditional thinking around payment integrity. Payors are looking to see if the procedure is covered; if billing is aligned to industry norms; and if it is coded correctly, among other basic attributes to the claim. And while this does prevent some improper payments from being made, far too many avoid detection. Bad actors have become adept at manipulating the system so as not to be caught.
We do see in some cases that the results of postpay investigations are used to put either individual practitioners, or in some cases entire healthcare systems, on prepay review in an effort to stop fraud. And while this may help payors increase the amount of FWA and improper payments discovered, this approach is highly inefficient, costly, and can introduce significant levels of unnecessary provider abrasion.
This is yet another area in which AI can support a switch in how the industry thinks about prepay review and its role in FWA/improper payment detection. Instead of taking an “all or nothing view” when it comes to prepay review, AI can leverage the best practices that drive successful postpay claim review and investigation to more effectively examine claims prior to payment. Analyzing post-payment claims data empowers the development of more effective and even new claims edits. The outputs of post-payment investigations can help healthpayers effectively identify those individual claims that warrant further review and separate them from legitimate claims, even when the claims are submitted by the same provider.
