Marketing for artificial intelligence (AI) tools aimed at medical practices is everywhere — and promising everything. But experts warn that you should not rush out and purchase something labeled AI, no matter how exciting the demo or urgent the problem it promises to fix.
The HHS Office of Inspector General (OIG) released the inspection report
“Medical Billing Software and Processes Used to Prepare Claims” in 2000. The agency reviewed “software literature and claim preparation processes to determine whether Medicare may be vulnerable to claims generated by electronic systems.” In the section on various types of software, OIG observed that interactive software “uses a form of artificial intelligence to ‘learn’ from past claims activity which services will be paid or denied.”
Fast forward 25 years: Marketing for AI tools aimed at medical practices is everywhere — and promising everything. But experts warn that you should not rush out and purchase something labeled AI, no matter how exciting the demo or urgent the problem it promises to fix.
“The worst way to try to implement AI is to do it in a reactionary fashion,” says Donovan Campbell, CEO of Medbridge. “And you actually are starting to see this across the clinical landscape, where someone’s been told by their CEO that we need to go get an AI strategy and they scramble to come up with quote: ‘an AI strategy.’”
In fact, AI programs can create more problems. For example, physician-owners of a practice that is struggling with a coder shortage and that doesn’t have time to conduct internal reviews might think an AI system can fill in the gaps. But AI isn’t autonomous; practices need skilled coders and compliance professionals to monitor the system and educate users.
“If the practice doesn’t have any process to review documentation or coding accuracy, AI can quickly compound issues,” warned Erin Kreider, MSN, APRN, NP-C, CCDS, product lead for clinical documentation improvement (CDI) and utilization management (UM) at Ambience. “AI needs oversight — through audits, coder review, or provider education — to avoid over- or under-documentation or misalignment with CMS guidelines,” Kreider said.
An upcoming article in Part B News will show subscribers how they should approach AI adoption in their practice. The short answer is with forethought, planning and conversations with everyone who will be affected by the new system.
A good place to start is with definitions and reasonable expectations: What does “artificial intelligence” mean and what are AI systems able to do today?
The term AI has been around for a long time and can mean a lot of things, Campbell said. An early application of “AI” is a machine learning system that monitors manufacturer’s inventory and orders parts or supplies before that part or supply runs out, he said.
“When we say AI today, we don’t mean that. What we mean is generative AI … and generative AI can come up with things that are novel based on natural interaction with an interface,” he explained.
In other words, you don’t need a degree in computer or data science to submit a question or a prompt, and the generative AI system creates a novel response based on masses of data. It doesn’t simply give a response from a limited set of pre-programmed answers, Campbell said.
Because discussion about what AI tools can do might be based on speculation about the future, practices must also understand what AI can do for health care providers today, rather than what might be possible several years or even decades from now. Kreider provided a list of ways AI tools are being used in the medical field today:
- Surface codes.The AI identifies potential diagnoses or conditions based on information in the chart and makes suggestions that the provider rejects, modifies or confirms, Kreider said.
- Help with real-time clinical documentation.
- Transcribing visits.
- Organizing notes.
But Kreider cautioned against expecting too much from these tools, because they are not "a shortcut to documentation and billing,” she said. “The best systems are designed with compliance in mind: They cite clinical indicators, preserve provider judgment, and can be audited to show that documentation decisions were made appropriately,” she explained.