How the Marriage of AI and H.I. Impacts Healthcare Costs
This AIThority guest post is co-authored by Evan Pollack, MD, Chief Medical Officer, Medical Audit and Review Solutions.
Identifying healthcare fraud, waste and abuse is a highly evolved practice that is best done with a marriage of artificial intelligence (AI) and human intelligence (HI) capabilities. As the losses attributed to fraud continues to grow, unfortunately, we all share the responsibility of paying for it. The National Health Care Anti-Fraud Association (NHCAA) estimates that the financial losses due to health care fraud are in the tens of billions of dollars each year.1 The payment integrity review process of analyzing a healthcare claim can be strengthened by implementing a hybrid approach of both HI and AI. However, it’s important to understand the benefits and limitations of each to avoid pitfalls that can arise.
On July 20, 2022, the Department of Justice announced criminal charges against 36 defendants in 13 federal districts across the United States for more than $1.2 billion in alleged fraudulent telemedicine, cardiovascular and cancer genetic testing, and durable medical equipment (DME) schemes. While not involved in this litigation, it demonstrates the scope of healthcare fraud and how schemes are growing increasingly sophisticated.
Marrying AI and HI, we see complementary analysis. For example, AI algorithms are set up to compute what is in the data, which can include frequency by provider, against all time-based CPT codes, hours spent by frequency (daily) and “normal” patterns as established through statistical analysis or clinician definition. Adding on a layer of HI analysis, which looks at what’s on the claim, we’ll see the provider’s name appearing on a large volume of non-contact claims, CPT codes allowing time spent analysis and billing over 24 hours per day/7 days a week/365 days a year.
Steps to HI and AI integration include:
Identifying an issue
HI approach: Individuals identify issues in claims. Gather data with analyst team.
AI approach: Claims are identified as anomalous by statistical significance against assumed population characteristics.
Developing a defensible concept
HI approach: Physician or medical coder reviews information and builds argument/dismisses.
AI approach: Model interpretation algorithm provides statistical test measurements back. If labeled data, will have prediction of fraud, waste and abuse.
Putting it in production
HI approach: Provider is excluded from network. Ops team trained to spot this issue. Refer case to Special Investigations Unit (SIU).
AI approach: Every claim run through system and labeled in terms of level anomaly and % likelihood of FWA (if available).
Updating and improving as needed
HI approach: Anomalousness largely remains same. If provider appeals, perhaps re-addition.
AI approach: When behavior shifts and new norms emerge, anomalousness updates.
Recommended: AI in Advertising and Marketing – Hype, Disillusionment, Back to Hype Again?
Weighing the Pros and Cons in Healthcare Costs
We see pros and cons of implementing the hybrid HI and AI approach. Pros of human intelligence include high defensibility, quantifiable actionability and identification of nuanced issues. With AI, we find scalability and ability to handle large claim volume, as well as speed and the ability to identify nuanced patterns and recognize changing conditions. However, on their own, each has limitations. HI doesn’t scale well due to the cost, can be hard to staff and has difficulty recognizing changing conditions. On the other hand, AI doesn’t understand context, has lower defensibility, actionability is not understood if unsupervised and has a higher likelihood of provider abrasion.
Blending them into a hybrid approach helps with anomaly detection. Key issues/features can be identified through use cases and explanation of Special Investigations Unit (SIU) methods. Because model interpretability is built into processes, it provides an anomaly score, and an explanation. We see rough definitions of “expected behavior” and have the ability to train both analysts and physician teams.
The hybrid approach also helps to identify an issue and a defensible concept. Anomaly packets as well as human-identified concerns are sent to the analyst team, who reviews and identifies common patterns/trends. Humans define the defensible concept and, in the case of gray areas, may recommend multiple concepts to be applied in tandem. Expected behavior is defined in terms that relate to the available data.
Next steps are to test it in production and ideate on improvements or needed updates. Algorithms flag all claims requiring audit by an SIU team. Active monitoring is set up to understand if environmental variables have changed. Lastly, new anomalous areas continue to be identified by anomaly packets for review. Given the change in environment, thresholds are adjusted automatically and new standards in industry can lead to manual override of thresholds.
Top AI ML Blogs: How Agencies Can Use AI Image Generation to Build Better and More Creative Ads
A real-life example of the successful deployment of the HI/AI approach illustrates how it helps to recover costs and drive successful payment integrity outcomes. An investigation beginning with a lab claim for an uncommon genetic immunodeficiency disorder led to larger scale identification of testing abuse. Another began with a few additional claims than expected for a provider and ended in the realization that the physician was billing 30 hours in a day, 7 days a week.
Valuable and actionable findings like these can come from innocuous origins like simple claims edits but are identified and pursued because of the nuanced contextual understanding of medical experts. An unfortunate downside of this approach is that it is costly and manual.
On the flip side, automated identification of suspicious claims is possible at lightning speed, but historically have lacked the ability to understand the subtleties and contexts, which can limit the value of the data. Given the complementary nature of these two approaches, ‘man and machine’ solutions rather than man vs. machine bake-offs are showcasing increased identification of issues at faster rates.
It’s important to frame out and develop a feature-rich space for analysis, which is one of the big advantages of a merged strategy. The domain experts tell us where to look in data and what they think about, which helps to point the algorithms in the right direction.
Humans and algorithms each have distinct advantages and disadvantages and the marriage of the two is most successful when you play to both of their strengths. Computers look across many records, and trim the concepts to get the most value, while subject matter experts define what they see and provide invaluable context.
AI ML News: NLP Leader Huma.AI Launches Industry’s First Generative AI…
Comments are closed.