How AI can accelerate prior and concurrent authorization processes

The potential of artificial intelligence (AI) tools to transform health care is demonstrably improving patient care and operational efficiency. Some administrative burdens ― automating various EHR tasks, completing repetitive patient forms ― are obvious areas where AI scripts can automate processes that lie outside the realm of clinical care. And more recently, the possibilities of generative AI have taken over headlines, but are still being trained and validated with respect to many operational use cases, including within prior authorization. Not all AI is generative AI, however, and not all automated tasks are mundane.

The accelerated pace of AI adoption in health care has highlighted several key areas to ensure the industry is deploying AI technologies responsibly – ensuring data standardization and quality, ensuring inclusiveness for population-representative data, ensuring transparent and real-time access to data sources to avoid biases or the notion of proprietary models that do not adequately reflect a patient’s real-time clinical picture. And of course, any AI use case that prioritizes the judgment of a digital script over a clinician’s judgment will continue to be met with scrutiny. That’s not to say that AI is unprepared to meet the challenges of 2024 when it comes to claims decisions. On the contrary, the technology capable of accelerating prior and concurrent authorization decisions already exists. The challenge for payers and providers in 2024 is to apply them in a nonproprietary and transparent manner, in partnership, elevating the emphasis on member or patient care.

At the core of this challenge is something that holds true for every industry grappling with how to apply AI to their existing workflows: a transparent, intentional focus on how AI models are taught. For payers and providers, Medicare provides specific lists of procedures that require prior authorizations, and a list of criteria for said procedures. When both parties can come together to train AI models to apply a set of standards (like Medicare’s) and understand how and when the standards will be applied, their path forward will be less fraught.

For example, as soon as any patient’s electronic health record is updated to include a recommended procedure that requires a prior authorization, AI can instantly run a review if a prior authorization is needed, speeding up decision-making on appropriate cases based on historical data and agreed-upon standards between the provider and payer. This streamlines the process, eliminating some of the work that must be performed by humans. Importantly, in this use case, AI doesn't replace human oversight. Instead, it efficiently pre-filters claims, highlighting those clearly necessary based on pre-defined criteria, those clearly unnecessary, and the gray areas requiring human expertise. For claims that have more clear-cut outcomes and fall into the black and white zones, both parties will already have insight into the “why” and “how” — saving time, hassle, and accelerating alignment. Similarly, for those cases that did not meet criteria, AI will provide insight into the “why not.” By prioritizing the gray zone, or most critical cases with supporting data, AI also elevates clinicians’ work, moving beyond a criteria-based mindset to pinpoint where their specialized, critical thinking skills can be put to the best use. With over a quarter trillion dollars identified as unnecessary administrative waste, such operational use cases of AI in health care are a no-brainer when they not only provide more consistency, standardization and focus on the most revenue-sensitive cases, but also have the potential to reduce payer-provider friction.

These processes lend shared efficiencies to both stakeholder groups, but are they actually more accurate in deeming medical necessity? In short, they should be. Consider the burden of reviewing 10 or more years of case histories involving similar approval or denial decisions — the foundational work of any claims process. The time-consuming process of culling the relevant data, then considering the merits of each side, is rife for an AI data-scraping script. While we still have data integrity and interoperability challenges, it must be noted that payers and providers have an overabundance of patient data sets to draw from, and increasingly less time to analyze it all. The same burden faced by the most thorough human claims reviewer is no different than the burden facing an AI script. The difference? One is growing more capable of consuming large amounts of data every year.

This is especially relevant for claims decisions involving prior authorization, which can directly affect patient health outcomes in a given moment. Yet delaying the delivery of expensive specialty drugs (like for CAR-T Cell Therapy, which can be lifesaving for cancer patients that failed previous treatment) is not uncommon, especially as the cost of these drugs rises and the level of scrutiny by payers increases. Even a four-week delay with cancer treatment has been linked to increased mortality for cancer patients, as cancer patients are generally more vulnerable to poorer outcomes due to delays with prior authorization. More broadly, 95% of physicians report care delays due to prior authorization, and 80% of physicians say it could even lead to abandoning treatment, per a survey by the American Medical Association. Focusing solely on the breakdowns that occur during prior authorization, however, would be short-sighted as it could exacerbate additional bottlenecks in care delivery. Expanding AI-driven efficiencies to also include concurrent review offers more comprehensive improvements that impact the patient experience as well as the bottom line. By continuously evaluating the medical necessity of ongoing treatments, concurrent review minimizes delays for patients as they are already receiving care, creating more seamless transitions of care. This ensures timely access to vital treatments while still allowing payers to manage costs responsibly. While it may not directly impact the immediate cost to the patient, it streamlines the process for providers, allowing them to focus on delivering quality care.

When debating the future of AI in health care, it’s always worth remembering the question that motivated us to enter the field in the first place: how can we navigate and improve patient care? Authorization processes are very tedious, with significant challenges, burdens and delays impacting patient care directly. Leveraging AI technologies to improve those processes, creating a more frictionless health care system, can help patients get care at the appropriate time without delays – making sure medical procedures that should be approved are, and what should not be approved are not.

Ankita Patel received her M.D. from American University of Antigua College of Medicine, her M.A. in bioethics from New York University, and her MBA in healthcare administration from Plymouth State University. She is currently a senior physician advisor at Xsolis, the AI-driven health technology company with a human-centered approach, is a per diem hospitalist, and is pursuing a two-year Integrative Medicine Fellowship at the Andrew Weil Center for Integrative Medicine.