Can AI rescue us from the mess of prior auth?

From the article:

Prior auth is the system by which the payer (generally an insurance plan, but sometimes a healthcare delivery organization accountable for the cost of care) sets and enforces standards to be met before agreeing to pay for a medication, test, or procedure. Prior auth requirements exploded with the growing popularity of Medicare Advantage (MA), the federal program in which patients’ Medicare benefits are turned over to a health plan, creating an arrangement known as capitation in which the health plan receives a fixed sum (adjusted for patient age and preexisting illnesses) to care for a patient, usually over a full year. Such arrangements give health plans a powerful incentive to minimize their expenditures. (The plans, of course, would say their incentive is to encourage the delivery of appropriate, evidence-based care.) Today, more than half of Americans over 65 are enrolled in Medicare Advantage, up from 17 percent a decade ago.

...

Physicians complain bitterly about prior auth, for good reason. For every physician in an office practice, the doctor and staff spend an average of twelve hours a week submitting prior auth requests. Physicians consider the process soul-crushing and often harmful to their patients, by either delaying needed care or sometimes blocking it altogether. For years, doctors have been desperate for a tool to help them do battle with the insurance companies.

AI entered the prior auth wars soon after the public release of ChatGPT in November 2022. Within weeks, Doximity, which describes itself as LinkedIn for physicians, rolled out a prior auth request generator. “All you had to do was type the letter ‘O’ and it automatically created a prior auth for Ozempic [the weight loss drug], addressed to UnitedHealthcare,” Doximity CEO Jeff Tangney told me, with a mix of amusement and awe. Today, tens of thousands of physicians use Doximity’s prior auth tool, noting that it not only saves time but, by pulling in key patient data from the electronic health record, markedly cuts their denial rate.

Of course, the insurance companies responded in kind, deploying their own AI to review – and often reject – doctors’ AI-generated prior auth requests. We quickly found ourselves in a ludicrous prior auth arms race, with AI serving as the primary weapon on both sides.

...

Dealing with the bureaucratic miasma of prior auth sometimes involves more than the generation of a written authorization request. One start-up, Infinitus Systems, built an AI program to automate the process of calls to the insurance company for authorization. Basically, the bot twiddles its digital thumbs during the on-hold period, then signals the clinic staff when a human from the insurance company picks up the phone. Just take that in for a second: A solution that mostly serves as an automated on-hold Task-Rabbit has raised more than $100 million in venture funding. That fact alone should give you a sense of how desperate health systems and physicians are to slice through the Gordian knot of prior auth.

...

For clinicians and patients alike, a world without prior auth seems blissful, but it’s not in the cards. There are simply too many examples of patients being subjected to expensive tests and treatments completely unsupported by evidence. Moreover, there are appalling (and to most physicians, embarrassing) examples of fraud and abuse. Just read this scandalous story about skin substitutes or my wife Katie Hafner’s exposé of Mohs surgery abuses in private equity-owned dermatology practices and try to defend the notion that there should be no oversight or accountability of any physician’s decision regarding any test or procedure, particularly when the program in question – Medicare – is funded by tax dollars and on the brink of insolvency.

...

To reimagine prior auth, we need to think about automating the entire process, starting with connecting the payer to the provider’s electronic health record. The ability of large language models to review clinical notes (i.e., to read unstructured data) means that such EHR-centered reviews can now serve as the core of prior auth decisions, replacing today’s system of ping-ponging faxes.

What would this look like? In response to a physician’s order for a limited number of prior-auth-requiring medications, tests, or procedures, the insurer would perform an AI-enabled review of the chart to see if the patient meets the criteria. If the answer is no, such a system could smooth – and perhaps automate – the process by which the clinician provides evidence supporting her choice.

...

Some health systems and payers are beginning to implement these types of automated, EHR-based, AI-enabled solutions. Louisiana-based Ochsner Health has connected its EHR to the computer systems of its largest insurers and now receives instant approvals for about half the requests on a select group of procedures. Even when an approval isn’t instantaneous, the link has sped up the process – often shortening the time from appeal to decision from days to hours.