Applications for artificial intelligence to improve revenue cycle management in healthcare show promise, but executives are concerned about accuracy and reliability of the technology.
These were among the results of an Inovalon survey of more than 400 revenue cycle and financial executives and managers, 84% of whom said they are optimistic about AI-enabled RCM in hospitals.
However, a third of respondents said they were concerned or skeptical about using AI in RCM, with worries about accuracy and reliability (31%), lack of familiarity/understanding (17%), and AI being too new/untested (15%) being the chief sticking points.
Humans better than AI
Twenty percent of respondents said they were convinced human performance – at least at this point – is superior to that of AI.
Julie Lambert, president and general manager of provider at Inovalon, told Healthcare IT News that is applicable across RCM, but there are definitely areas that can benefit in greater ways from AI.
“Rather than think of this as an either/or scenario, I’d challenge us to think about this more as expertise is a critical underpinning to creating AI/ML models that perform and are continually refined,” she said. “When technology and expertise are combined, the potential for the best outcomes is present.”
From her perspective, the areas where AI can be the most impactful in RCM are the areas that cause the most amount of pain and are the most manual to providers today.
Among these areas, denials, prior authorization and eligibility likely rank somewhere near the top for all providers, and she said it isn’t by accident that all of these items are related.
“Errors upfront in the registration process create denials on the back-end,” Lambert said.
What’s causing denials?
Knowing which scenarios are causing denials and how to catch or predict those denials before they happen is the perfect opportunity to use AI with expertise and claim results data to build and train a model.
“There are opportunities both within these processes themselves and in the overarching, connectiveness to use ML and AI for the betterment of providers,” she said.
Lambert added an important factor to consider is that AI is not static, and it should never be treated as such.
“Designing a model that is continuously learning is a core tenant of AI – models will continuously learn from the data and the feedback loop that comes in naturally from the results,” she said.
External factors
It is also critical that the knowledge of external factors that can impact a model are known and accounted for. This could mean regulatory changes that impact the structure of the data, the data elements in the responses, or other factors that could cause anomalies in the data.
“Make sure there is awareness about any changes that impact the model so that interpretation of the results isn’t drawing false assumptions or correlations,” Lambert advised.
She added it is important to make sure that people understand AI isn’t only for the C-suite or only for data scientists – it’s for everyone to be a part of, and that is what is going to make AI successful.
“AI needs the inputs of those with feet on the ground who are managing the data, performing the operations, and managing the workflow to help in making models,” she said.
Nathan Eddy is a healthcare and technology freelancer based in Berlin.
Email the writer: nathaneddy@gmail.com
Twitter: @dropdeaded209