Unlocking The Potential Of AI In Prior Authorization

Editor’s note: This is part one of a series looking at ways insurers can transform utilization management. Upcoming articles will examine utilization management policies and processes.

They are two of the most dreaded words in healthcare: prior authorization. Physicians loathe the process, claiming it can negatively impact patient outcomes and results in undue administrative burden. Lawmakers are pressing regulators to take a closer look at the process as well. While there are legitimate concerns around its varied implementation, prior authorization, as well as the broader set of utilization management tools, can play a meaningful role in controlling high pharmaceutical spending, limiting unnecessary procedures, and directing patients to the most appropriate sites of care.

Those efforts are necessary to curb unsustainable growth in total healthcare spending, which are projected to exceed $5 trillion in 2024. The challenge is especially pronounced in Medicare Advantage where spending nearly tripled over a 10 year period, driven largely by growing enrollment. Prior authorizations in MA soared to 46 million in 2022, up from 37 million in 2019.

Advances in technology pose opportunities and challenges for insurers to overhaul their prior authorization process. There are three areas in particular that can have a sweeping impact: digitization, machine learning, and generative artificial intelligence. Insurers wanting to set themselves apart need to be nimble and creative in how they adopt these tools.

Managing the wave of technology changes

Here’s our hot take on how the three technology trends will impact prior authorization:

There’s wide variation nationwide in adoption of electronic prior authorization. In some markets across the southeast, we’ve seen uptake exceeding 90% of prior authorizations. Nationally, the estimate is that 31% were fully electronic in 2023, up slightly from 28% in 2022. But now, a growing number of states are starting to require it. And the Centers for Medicare and Medicaid Services mandated that payers accelerate digitization of prior authorization starting in 2026.

Digitizing prior authorization requests offers a host of benefits over the antiquated use of fax and phone. For starters, payers can convert unstructured data to structured data and speed up the time needed for such things as assessing medical necessity. Also, payers can more easily exchange data with providers and reduce the bothersome back-and-forth that currently occurs manually. This should contribute to reductions in administrative errors and duplication of tasks or medical procedures. By some estimates, electronic prior authorization can cut healthcare spending by $449 million annually and save a clinician more than 10 minutes per transaction.

Enhanced use of machine learning goes hand in glove with the transition to digitization. For starters, machine learning, which is subset of artificial intelligence, can process larges amounts of data quickly. One large national insurer reported that use of an AI tool made the prior authorization process 1,400 times faster.

Machine learning can also aide payers in tracking and analyzing trends. For instance, looking at what percentage of requests are approved or denied. That data can then be turned into rules engines that are used to streamline the approval process. AI can perform auto-approvals where medical necessity is clear-cut, accessing clinical evidence, claims history, and past patient interactions. This will hasten the time it takes for a treatment to be authorized and for the patient to get care. A large regional insurer reported that use of an AI tool cut the time to get immediate decisions by 10 days. CMS allows Medicare Advantage plans to use algorithms and artificial intelligence to assist in making coverage determinations, but, importantly, the technology cannot override standards related to medical necessity.

Once digitization and machine learning take hold, insurers can pivot to adopting more disruptive technologies like generative AI. With the volumes of documents required during a request, generative AI can comb through reems of guidelines and identify specific codes and other data points for clinicians. It can spin out a simple, easy to understand summary for insurers to use when talking with a patient or caregiver. And, for more complex cases, it can help recommend treatment options, providing alternatives that address access and affordability.

Rethinking the operating model

The role out new technologies necessitates that insurers do a thorough evaluation of their current operating model. A host of new vendors with sophisticated capabilities are emerging offering the opportunity to outsource some functions. But since utilization management impacts several functions and departments, it is important to be strategic about selecting workflows to keep in-house and those that will be outsourced. There’s a direct impact on human capital to consider. Importantly, technology infrastructures need to be equipped to work alongside more advanced applications like generative AI. It is also critical to put guardrails in place that ensure the quality of care is not minimized as AI tools get wider use. Protecting against AI hallucination — incorrect or misleading results — is essential. AI systems overall hallucinate between 2.5% and 22.4% of the time, depending on the platform.

Insurers also need to consider the impact two key external constituencies: members and providers. Improving the member experience and easing access to care should be paramount. Technology tools, whether in-house or outsourced, should streamline back-office processes and improve service delivery. Insurers need to be careful to not introduce complexities in care coordination and communication between the member, insurer, and provider. Providers, meanwhile, must be trained on new processes and become champions to spread the word with colleagues that the goal is to reduce administrative burden and improve the patient experience. Processes that aren’t fully integrated into workflows could limit the ability of insurers and providers to exchange data in real time, slowing critical functions like reviewing treatment plans, claims approval and denial, and more.

Optimizing for the future

Utilization management has come under fire in recent years. But, as we noted earlier, it can be an extremely useful tool in helping stakeholders manage costs and reduce unnecessary care. To ensure that patients are getting the right care, at the right time, in the right place, however, health plans must continuously assess their operating and staffing models, technology infrastructure, and processes. Insurers that do so will be able to identify areas more quickly for improvement and, ultimately, enhance the member and clinician experience.