How medical AI can alleviate the pain and accelerate prior authorization

To take advantage of new accelerated prior authorization timelines, providers must step up their internal processes to ensure they can scale while maintaining quality control over prior auth requests and denial appeals.

Obtaining prior authorization from a health insurer for a medical test or procedure is a major burden on providers and can negatively impact patient outcomes by causing delays in treatment. In a 2022 survey by the American Medical Association, 89% of responding physicians said prior auths have “a somewhat or significant negative impact” on clinical outcomes, with 33% of physicians reporting that prior authorization has led to a serious adverse event for a patient in their care.

The cumbersome prior auth process also creates severe operational challenges for provider organizations: 88% of physicians in the AMA survey describe the prior auth burden as high or extremely high. On top of creating unnecessary administrative problems, prior authorization delays can seriously disrupt a provider organization’s revenue cycle because they have to wait for an approval before delivering care to the patient, and then submit a claim to the payer. This prolonged and unpredictable process introduces serious cash flow issues that can spell disaster for provider organizations operating on thin margins.

Preparing for new CMS rules

To help speed the prior auth process, CMS has finalized new rules that require insurers to make prior auth decisions within 72 hours for urgent requests and within seven days for nonurgent ones (half of the current 14 days for nonurgent requests). Further, payers must develop prior authorization application programming interfaces for providers, divulge their reason for denying a request, and report prior authorization metrics to CMS.

These new rules, which go into effect in 2026, should benefit providers and patients by shortening (and enabling greater transparency into) the process. Still, to take full advantage of the accelerated prior authorization timeline, providers must step up their internal processes to ensure they can scale while maintaining quality control over prior auth requests and denial appeals. Given the serious staffing shortages faced by provider organizations — which lead to backlogs, burnout, and high error rates — technology solutions are the best path forward to streamline the prior auth process while meeting their goal of controlling healthcare costs.

Providers typically assign staff to find, fill in, or correct patient data to complete a prior auth request or respond to a denial. This labor-intensive process takes time, is prone to error, requires oversight, costs money, and increases the risk to patients whose outcomes may depend on prompt treatment. It’s also ill-suited to handle the vast and rapidly growing volume of health care data being generated today.

Fortunately, advances in large language models (LLMs) combined with multiple artificial intelligence capabilities and automation are giving providers the digital tools to expedite the prior auth process, which is slowed by missing and incorrect patient data.

One of the challenges in retrieving information for a prior auth request or denial appeal is that 80% of EHR patient data is in unstructured form. Early versions of natural language processing software were unable to interpret unstructured medical data. The highly structured way in which early NLP software outputs data makes it difficult to use. Getting needed information, for example, typically requires navigating an SQL database, a skill many clinicians and support staff lack.

An entirely different problem arises when keyword searches are used. Searching for words like “hepatitis” or “cancer” might yield hundreds of returns, many or most of which aren’t relevant to the patient’s current condition. For clinicians at the point of care, such high-volume, low-quality search results waste time and make their jobs more difficult.

In contrast, medical AI using LLMs can understand, extract, and summarize unstructured EHR data through natural language prompting. Medical AI helps provide the context and nuance hidden in the unstructured data that traditional NLP can’t access. Critically, medical AI results include auditable, verifiable evidence from the source data. Automating this process eliminates the need for revenue cycle staff to hunt down and insert unstructured data into PA requests. LLM algorithms can generate data outputs in minutes or even seconds.

Keeping humans in the loop

Not only can medical AI extract clinical data, it can also uncover data regarding social determinants of health that may be used to inform prior auth decisions. Medical AI is able to analyze massive datasets from multiple sources, enabling clinicians to create personalized care plans based on an individual’s specific circumstances.

In a recent study at Mass General Brigham, researchers found that compared to “diagnostic codes entered as structured data, text-extracted data identified 91.8% more patients with an adverse SDoH.”

So, in a case where medical AI reveals that a patient with heart disease has chronic transportation issues, for example, clinicians can develop a care plan that includes remote monitoring and/or connected technologies, telehealth visits, home care, and, if needed, arrangements for rides to and from appointments. This ensures the patient continues to receive the appropriate care.

Similarly, medical AI also can be deployed to find the clinical or SDoH data needed to successfully prevent and appeal denials. Using medical AI and automation, providers can determine if a patient meets all criteria for prior auth approval – such as lab results or a letter of medical necessity – and that all necessary data is included far more quickly than if staffers were tasked with gathering this information.

No matter how advanced medical technologies become, however, they can never fully replace humans. And that is not medical AI’s intent. While medical AI and LLMs can augment provider organization staff workers by retrieving and organizing unimaginably large amounts of data, humans should always be in the clinical loop.

Conclusion

Prior authorization long has been the bane of providers. New CMS rules set to go live in 2026 will force payers to make much faster decisions regarding prior auths. To take advantage of the accelerated timetables, providers must have the right tools in place. Medical AI can be deployed in the revenue cycle to streamline the process and increase overall efficiency.