Results of a 6-month pilot with AI (Artificial Intelligence) technology found TGH $1 million in uncaptured revenue, but was it worth the potential risk of improper coding?
Anthony Escobio, the Vice President of Revenue Cycle at Tampa General Hospital (TGH), pondered this question as he reviewed the findings from the Palantir 6-month pilot, which assessed 13,000 infusion patients in the Emergency Department. Palantir’s new AIP (Artificial Intelligence Platform) solution, which had a code review function to detect and learn coding rules, could help identify potential missed coding opportunities using Palantir’s Large Language Models (LLMs) to scan through unstructured clinical documentation notes.
Escobio’s decision of whether to implement this technology solution relied on several unanswered questions. First, there was the question of how accurate AI could be in this new application. Second, there was the question of scope and the ratio of AI to human involvement in the process. Were they ready to take humans out of the coding process altogether, and rely completely on Palantir’s solution? Or would they still need the 36 TGH coders involved, causing them to continue to pay for those resources, in addition to the Palantir AIP solution? According to the Centers for Medicare and Medicaid Services (CMS), the fiscal year (FY) 2022 Medicare FFS estimated improper payment rate was 7.46% (American, 2021). The TGH goal was to be 95% accurate.
If Escobio decided to move forward on the Palantir AIP solution, it would, without a doubt, bring in a significant volume of missed revenue for the organization. In addition to increasing revenue, the additional workload that could potentially be removed from the human coders, overworked physicians, and remaining Revenue Cycle team would have an invaluable impact on the workforce’s morale. Though the positive gain was obvious, Escobio had to evaluate if this was the right decision for the organization based on the potential risks of implementing a new and industry-disrupting technology. Was he confident in the technology? There was still much to be known about Artificial Intelligence – including if it was safe to use in this complicated healthcare business where so much was at stake.
Balancing these concerns with the knowledge that if he were to approve this solution, TGH could stand to gain a significant revenue increase. However, if he chose not to approve this solution, they would digress back to identify the best solution to review all the medical coding in the approximate 187,000 encounters per year, in hopes of finding similar recaptured revenue from missed codes. No, he thought to himself, this was not going to be an easy decision to make.
Authors: Stacey Brandt, Jessica Cox, Kraig Kalby, Jason Robb
Link: https://doi.org/10.28945/5392
Cite As: Brandt, S., Cox, J., Kalby, K. & Robb, J. (2024). Relying on artificial intelligence in medical coding review. Muma Case Review 9(2). 1-24. https://doi.org/10.28945/5392