Oncology Experts and AI Advocates Explain the Evolving Role of AI in Insurance Appeals

With insurers now using artificial intelligence (AI) to issue denials for oncology treatments, embracing similar technology to automate and strengthen prior authorization submissions could help level the playing field and speed access to care. However, skepticism remains regarding the oncology field’s larger-scale ability to influence payer policies.

In interviews with OncLive®, the following experts shared their thoughts on the future of AI in this arena:

• Douglas Flora, MD, LSSBB, executive medical director of the Yung Family Cancer Center at St. Elizabeth Healthcare in Edgewood, Kentucky
• Lalan Wilfong, MD, a medical oncologist/hematologist with Texas Oncology
• Ruemu E. Birhiray, MD, a medical oncologist and hematologist at Ascension St Vincent in Indianapolis, Indiana
• Jim Chen, MD, an oncologist and hematologist at Genesis Cancer and Blood Institute in Little Rock, Arkansas
• Beau Hilton, MD, an oncologist and hematologist at Genesis Cancer and Blood Institute in Hot Springs, Arkansas

Flora: Prior authorizations and denials are the big [challenge] in oncology right now. We as doctors understand the evidence, we make a treatment recommendation, and now the insurance companies can say no—using AI, I should point out—and I feel like we’re losing that arms race a bit. The prior authorization letter denials can be adjusted quickly if we adopt these tools to, say, scrape the entire medical record and get everything that the insurance company says they need before you submit that first approval.

There are tons of commercially available tools that will do that. You’ll stop losing those reimbursements or approvals because you left out something like the ECOG performance status, because the tools understand that that’s important and [that the absence of the ECOG performance status] is something that insurance companies can [use to deny payment] for [a certain] drug. That’s a technical mishap, because the people who are submitting for authorization are not technically trained in oncology. However, if we can deliver a document that has [a patient’s] stage, grade, scan results, pathology results, ECOG performance status, exam [results, and notes about] the physician’s clinical decision-making, the insurance company has everything they need to know. Most of these tools [include] references to document and support the decision made by the doctor without us having to spend 45 minutes on a peer-to-peer review, submit denial approval letters or fight with the team at the insurance company when we should be working together to remove the work on both sides.

Wilfong: One of the biggest challenges with insurance authorizations right now is the manual labor that has to go into getting them done. Typically, the physician does their documentation in their notes, then somebody in the office who often doesn’t have a medical background is taking that information and putting it on the insurer’s portal to make sure it meets their criteria. A vast number of denials I’ve seen are for misinformation. There’s some mistake that happens along the way: either the physician hasn’t documented things exactly correctly, or the staff is not inputting it correctly.

Where AI has the biggest benefit, there is translating and taking all that information, especially in unstructured fields, and converting it into the payer portals. It’ll automate [the process], it’ll speed things up, and it should make it more accurate. That’s one thing AI can potentially do well to help us there.

One of the challenges that we will have, at least now, is when we disagree with what the payers’ policies are. A much smaller part of the whole problem we have with utilization management is disagreement with payer policies. [We are concerned] more with the difficulty of getting things done, which we could solve with technology.

Birhiray: The bane of most oncologists’ existence is the fact that they have to get prior authorization. Could you imagine an AI tool that could gather data from the patient’s chart, generate a letter, and generate the evidence and the argument, and the supporting recommendations that you should use a particular treatment model or particular test in a particular patient, and then get your insurance approval sooner rather than later? That would be amazing.

Chen: I believe appeals are going to be easier to write [with the use of AI], but at the end of the day, the holdup is going to be insurance companies not wanting to pay. They want to hold on to as much money as possible for as long as possible so they can invest. I don’t know if AI is the solution to change the philosophy of the for-profit business model that is the insurance company.

Hilton: One can hope we all have the same access to these tools, and that we’re going to escalate them. [However], if we can have an automated, well-reasoned appeal, there’s no reason why we can’t [receive] an automated, well-reasoned denial. It’s still going to be difficult. At the end of the day, though, typically right now, if we have basis for an appeal, we tend to get what we want, it just takes a lot of work. I hope the balance will swing toward making things easier, better, and having less delays in care.

Working on explainable AI is a huge interest of mine. One of the most powerful [opportunities] for explainable AI in health care is you [can receive results about] not just this is what the algorithm said you should do, but why. You have reasoning built into the tool that you can see as a clinician, an insurance company, or another interested party. There are technological approaches [to improving the insurance coverage process], but at the end of the day, it’s still a human problem. I hope that the human product gets easier as technology gets better.