Using AI
The paper’s authors have previously estimated that almost one-third of the costs to execute PAs could be saved in the next three years by using existing technology, especially AI.3,4
AI could relieve burden on providers in multiple ways, Mr. Sahni suggested. For example, AI may be used to extract the relevant information from multiple health organization data systems and pre-populate a PA form for review by a clinician. AI may catch some errors, such as identifying ZIP codes that are missing zeros because they were copied from spreadsheets. AI may also help staff determine individual insurers that require prior authorization for specific drugs.
The survey included questions to gauge support for using AI to lower the PA burden. Eleven percent of providers and 65% of private payer respondents reported they are considering incorporating AI in the next three to five years. The top provider concerns were lack of budget and lack of trust in AI technology and tools. Top payer concerns were cybersecurity and lack of technical infrastructure.
Overall, providers saw PA as a greater burden than patients did. “There is something about perception that’s tied to how often you engage, and patients just don’t happen to engage with prior authorizations, vs. a nurse who is submitting them daily,” said Mr. Sahni.
He added that with continued nursing and physician shortages, trying to free up time to do more clinical visits “is absolutely critical to address some of our patient-access issues in this country as well.”
Reactions
Although data about the perceptions of different stakeholders in the PA process are useful, Christopher Phillips, MD, a community rheumatologist in Paducah, Ky., questions “the idea that the whole prior authorization infrastructure is a cost saver to the healthcare system.” That’s because most PAs are ultimately approved. In his experience, “for those that are [initially] denied, especially for expensive specialty medicines, [PAs] are still usually approved, but for a different expensive specialty medicine.”
Dr. Phillips, who chairs the ACR’s Committee on Rheumatologic Care, would like to see advocacy and research on how to reduce the number of services subject to PA. He points to several AI considerations beyond those noted in the paper, most of which are detailed in a November 2023 ACR position statement on the role of AI in rheumatology. AI is a powerful tool, but it does not replace the role of the clinical judgment of rheumatology professionals, the statement says. AI can incorporate sources’ biases, and language processing programs such as ChatGPT present potential for inaccurate statements. Any healthcare data used in AI models must be kept secure, and AI programs used in healthcare should be thoroughly tested, regulated, monitored and verified for clinical use, the statement adds.