Hot AI on AI Medical Reimbursement Action (AI Deep Dissection, Part 2)

UHC and Other Private Insurance Companies are Already Using AI for Evil.

Feb 24, 2025

AI generated image of a health-insurance AI bot. I’ll admit it made itself look pretty cool. — AI generated image of a health-insurance AI bot. It made itself look pretty cool.

Having cultivated a mostly sturdy foundation of understanding of what AI is, I wanted to know what it’s supposed to do to make my life and my patient’s lives better. Fortunately for me, the New England Journal of Medicine (NEJM) was wondering the same thing back in July of 2023.

This is when a review article on this general topic was published, written by Sahni, Carrus, et. al. entitled “Artificial Intelligence in U.S. Health Care Delivery.” Not a spicy title if I’m being honest. The authors did present an overview of the promises of AI in the future of medicine and I’ll say it appears bright for those of us eager to embrace this new technology with open hearts and minds. This seems a good time to point out that these authors are MBAs who work for McKinsey as AI industry consultants and can now claim a published credit in one of the world’s most prestigious medical journals on their LinkedIn profiles. I don’t suspect I’m reading from an unbiased source, here, but will try to give them the benefit of the doubt—for now.

This article is from the middle of 2023, which feels like two lifetimes ago at this point when we had the semblance of a semi-functioning government and the wet, hot “Barbenheimer” summer was all anyone cared about.

I’m sure much has changed since this article was published, but it seems like a good jumping off point for the bill of goods being sold to physicians—and by extension, patients—about what AI could do for them in the U.S. healthcare system. The article presents several areas in which AI was purported to change medicine for the better. Did the reality of AI in medicine hold up since this article? I guess I’ll look into that too along the way.

These authors addressed three main areas of AI use in medicine: reimbursement, clinical operations, and quality and safety, presenting them in that very order for discussion. I might have led with the reducing-harm-to patients-part of this pitch, but I’m not a Harvard trained MBA, nor have I published in the New England Journal of Medicine (yet).

Reimbursement

The professional AI business consultants wanted to lead off with the promise of cost savings and improvement to the bottom line. I don’t suppose I should be surprised. They pointed out that the predominant uses and most eager adoption of AI at the time of this article was in this domain, helping grease the wheels between the payors (insurance companies) and payees (patients).

American Style Medicine:

The way it generally works here is that when a patient fortunate enough to have a job that offers health insurance benefits has to go to a hospital, they are provided care, and ultimately receive an astronomical bill many weeks later in what seems to be a not-so-subtle effort to dissuade them from ever seeing a doctor again.

Behind the scenes, the hospital says to the insurance company, “hey, this person came to the hospital and here’s what you owe us.” The bill sent to the insurance company is vaguely generated based on services provided with what seem to be dice roll based mutlipliers and extra zeroes at the end of the amount requested, just for good measure.

Citing internal criteria that shift like the creative whims of horroscope readings, the insurance company rolls its eyes and sets a team of humans upon the hospital bill to try and figure out how little it should actually pay for these services, relying upon an age old practice of denying claims that the hospital makes, sometimes offering a pittance if anyting at all. Other imes the insurance company might try blaming the patient for being ill or existing and will try to stick them with the bill.

If all of that sounds horrifying, congratulations! I suspect you probably live anywhere but the only wealthy industrialized country without universal healthcare. If that doesn’t sound awful, I’d encourage you to look into emigrating to a country that does have universal healthcare. The outcomes are better.

All of this denying of claims at the insurance company costs time and money because a human has to look at the paperwork and engage in labyrinthine reasoning to justify not paying for medical care. The judgement is handed down by way of an overstuffed envelope that arrives at a patient’s mailbox notifying them that they owe an exhorbinant sum of money to the hospital. It would be nice if these notifications came with a promo code to a bankruptcy lawyer. This is the American way of medicine.

Algorithms For The People and Against the People

Our pro-AI consultant-authors suggest there must be a better way. Why don’t we have an impressively advanced AI algorithm take on the side of the hospital and patient to try and prevent those denied claims in the first place. They wrote, “With the use of this system, more than 10% of claims are denied or delayed because of eligibility issues and missing data, but up to 85% of those denied claims could have been avoided.” The computer could make the hospital billing side of things spiffy and denial-resistant, like a troublesome strain of pseudomonas. This would decrease administrative costs on the payee side and one could imagine the savings would altruistically be transferred onto the patient. Sounds like a win, I guess, if you believe corporate interests are capable of benefecence. I do not.

Of course, what if the insurance companies wanted to use their own AI products to fight the hospital/patient sided AI systems? Well, the authors contend, they certainly could do that.

Within the walls of the insurance company—which I imagine are wallpapered with hudred dollar bills using a glue made from the misery and tears of their enrolees—their AI computing models could do the jobs of the humans they employ to reduce the burden of adjudicating (fighting) claims. According to our authors, the "percentage of complex claims that were processed without denial increased from less than 80% to more than 90%.” In other words, if we take “compex claims” to mean those that were flagged for possible denial, less than 10% of claims were bogged down by administrative needs to try and deny care with the help of AI. This “reduced associated administrative spending by 30%,” a cost savings that could theoretically be passed along to the health insurance consumer in the form of cheaper insurance premiums. In reality, if we’ve learned anyting from noting ever increasing health insurance CEO bonuses during times of health crises, it’s that any profits they make with this technology will more likely be funneled back to their shareholders and superyacht fuel payments.

Maybe the bright side is that with all the cuts to administrative costs (i.e. human beings with jobs) there will be a new glut of unemployed hospital and health insurance billers hitting the rolls of people desperately seeking gainful employment trying to avoid being bankrupted by U.S. health care. The bleary-eyed former health insurance staff would see the plight most of our patients have known. Formerly employed hospital accounting department staffers, now replaced by a cold, unfeeling, and relentless computerized battle for health insurance dollars, might understand why the for-profit systems we’ve built to haunt our nightmares should be dismantled. If these authors can ignore reality, so can I.

Robot Reimbursement Wars: The Present Day

AI generated image of the battle between pro-insurance AI and pro-human beings AI in the style of a sci-fi, fantasy book cover. I kinda like it.

Since this article was published, a patient-based vs insurer-based AI vs. AI arms-race has developed. Predictably, AI models are battling one another. The computerized algorithms trying to avoid insurance denials are pitted against those trying to deny insurance claims, burning through electrons in order to decide a winner. One side is in the lead.

In November 2023, four months after the NEJM article was published, a Newsweek article pointed out UnitedHealthcare (UHC) had already deployed an AI tool (called nH Predict) developed by one of its subsidiary companies. This tool was aimed at denying care to elderly patients on Medicare Advantage (MA) plans. Insurance claim denial rates were found to be “more than 20 percent for private, commercial claims and nearly 56 percent for MA claims,” so noted the American Hospital Association in September, 2024. The same report cited that AI tools were directly responsible for MA claims’ denials to be greater than twice as prevalent as those from private commercial insurers. Companies like UHC adopted a “disadvantage the old and sick” kind of business plan, one to which other insurance companies can and do aspire.

Fast forward to January, 2025 and it seems the insurance company AI bots are still stronger than the patient and hospital favored AI bots. Impressively, the fact that the UHC nH Predict algorithm had a 90% error rate was a feature, not a bug. The UHC computer was wrong a vast majority of the time when it decided to deny a claim. The error rate meant that, if appealed, 90% of those claims would be overturned.

According to an article in The Guardian, the reality was that a tiny amount of patients (0.2%) would actually make the effort to appeal an in-nework claims denial. If UHC’s AI algorithm could generate multitudes more denials than the company’s human employees, the profits could be reaped as patient’s didn’t realize how flimsy those AI-based denials would be. Patients with previous experience trying to appeal a denial probably found the process to be so onerous, that they’d rather attempt a self-administered colonoscopy, someting likely to be a more successful and fulfilling use of their time.

UHC wasn’t alone in these tactics. Standing proudly alongside its fellow health insurance co-conspirators (Humana and Cigna), UHC et. al. have been named in numerous class action lawsuits citing the companies’ reliance upon computers to try and deny payments for life-saving care.

Hard to imagine why this industry would claim to be shocked at the public’s sardonic response to the gunning down of its CEO last year.

Thus far, I’m left feeling skeptical about the benefits of AI in medicine—especially as it’s delivered here in the U.S. of A. Maybe that’s because I started by focusing on the reimbursement bits. Next, I think I’ll examine the ways these AI enthusastic authors claim the algorithms will help my day-to-day work as a physician and, at the same time, protect my patients from potential medical pitfalls.

Hot AI on AI Medical Reimbursement Action (AI Deep Dissection, Part 2)

UHC and Other Private Insurance Companies are Already Using AI for Evil.

Discussion about this post