An Arm and a Leg: Attack of the Medicare machines

Note: "An Arm and a Leg" uses speech-recognition software to generate transcripts, which may contain errors. Please use the transcript as a tool but check the corresponding audio before quoting the podcast.

Dan: Hey there–

So this is kind of a horror story. But it’s not quite the kind of story it might sound like at first.

Because at first, it might sound like a horror story about machines taking over, making all the decisions– and making terrible, horrifying choices. Very age-of-Artificial Intelligence.

But this is really a story about decisions made by people. For money.

It’s also kind of a twofer sequel– like those movies that pit two characters from earlier stories against each other. Like Godzilla vs King Kong, or Alien vs Predator.

Although in this case, I’ve gotta admit, the two monsters are not necessarily fighting each other.

Let’s get reacquainted with them. 

On one side, coming back from our very last episode, we’ve got Medicare Advantage: This is the version of Medicare that’s run by private insurance companies. 

It’s got a bright and appealing side, compared to the traditional Medicare program run by the federal government, because: It can cost a lot less, month to month — saving people money on premiums. And it often comes with extra benefits, like dental coverage, which traditional Medicare doesn’t offer. [I know.]

But Medicare Advantage can have a dark side, which is basically: Well, you end up dealing with private insurance companies for the rest of your life. You need something — a test, a procedure, whatever — they might decide not to cover it.

Which can be scary. 

Our other returning monster — am I really calling them a monster? — well, last time we talked about them, in 2023, we had an expert calling them a behemoth. That's United HealthGroup. You might remember, they’re not only one of the biggest insurance companies 

— and maybe not-coincidentally the very biggest provider of Medicare Advantage plans —

they’ve also got a whole other business– under the umbrella name Optum. And Optum has spent the last bunch of years buying up a gazillion other health care companies of every kind. 

That includes medical practices — they employ more doctors than anyone else, by a huge margin. It includes surgery centers, and home-health companies, and every kind of middleman company you can imagine that works behind the scenes — and have their hands in a huge percentage of doctor bills and pharmacy visits. 

A few years ago, United bought a company called NaviHealth, which provides services to insurance companies that run Medicare Advantage plans. 

NaviHealth's job is to decide how long someone needs to stay in a nursing home, like if you’re discharged from a hospital after surgery, but you’re not ready to go home yet.

And the horror story– the stories, as dug up by reporters — starts after United bought NaviHealth.

And according to their reports, it involves people getting kicked out of those nursing homes who aren’t ready to go home. 

People getting sent home who can’t walk up the stairs in their house. Who can’t walk at all. Who are on feeding tubes. People who NaviHealth's own employees are saying, “Wait. This person isn’t ready to go home.”

But their new bosses have told them: You're not really making these decisions anymore. 

This is where machines do enter the picture.

NaviHealth's distinctive offering has always been its proprietary algorithm– an algorithm that makes predictions about how long any given patient might need to stay. 

Before United bought the company, that algorithm was used as a guide, a first-guess. Humans weighed in with their own judgment about what patients needed.

After United bought the company, people inside have told reporters, that changed: The new owners basically told their employees, If the algorithm says someone can go home after x days, that's when we're cutting them off.

 Like pretty much any horror movie, this story’s got people running around trying to tell everyone: HEY, WATCH OUT! THERE’S SOMETHING BIG AND DANGEROUS HAPPENING HERE.

And in this case, they’ve actually gotten the attention of some people who might have the power to do something about it. Now, what those people will do? We don’t know yet. 

And, by the way: Yes, I said at the end of our last episode that we’d be talking about Medicaid this time around. That’s coming! But for now, strap in for this one. 

This is An Arm and a Leg, a show about why health care costs so freaking much, and what we can maybe do about it. I’m Dan Weissmann. I’m a reporter, and I like a challenge. So our job on this show is to take one of the most enraging, terrifying, depressing parts of American life, and bring you something entertaining, empowering, and useful.

So. I said that, like every horror movie, this one has people who are seeing what’s going on and are trying to warn everybody?

Like those movies, we’re gonna follow one of those people, watch them discover the problem, see how deep it goes, and start ringing alarm bells. Let’s meet our guy.

Bob Herman: My name is Bob Herman. I’m a reporter at STAT News

Dan: Stat is an amazing medical news publication. Bob covers the business of medicine there. Bob started working on this story in November 2022, after talking to a source who runs nursing homes. Bob’s source was complaining about Medicare Advantage. 

Bob Herman: There were a lot of payment denials. They just weren’t able to get paid. And just offhandedly, the source mentioned like, um, you know, and they’re attributing everything to this algorithm. This algorithm said, You know, only 17 days for our patients and then time’s up and I went running to Casey Ross 

Dan: Casey is a reporter at Stat who focuses on tech and AI in healthcare. Bob said, hey, what do you think of this? Wanna team up?

Bob Herman: And he was hooked.

Dan: They started talking to people who worked at nursing homes, talking to experts, and talking to families. And it was clear: They were onto something. 

Bob Herman: It took so many families by surprise to be like, what do you mean we’re going home? The, you know, my husband, my wife, my grandma, my grandpa, they can’t go to the bathroom on their own. Like, what do you? It was just, it was so confusing to people. It seemed like such a, a cold calculation,​

Dan: One person they ended up talking with was Gloria Bent. Her husband Gary was sent to a nursing home for rehab after brain surgery for cancer. He was weak. He couldn’t walk. And he had something called “left neglect”: His brain didn’t register that there was a left side of his body. Here’s Gloria testifying before a Senate committee about how — when Gary arrived at the nursing home — the first thing he got was a discharge date. That is…

Gloria Bent: Before the staff of the facility could even evaluate my husband or develop a plan of care, I was contacted by someone who identified themselves as my Navi Health Care Coordinator

Dan: Gloria says when she told the nursing home staff she’d heard from NaviHealth, they groaned. And told her what to expect. 

Gloria Bent: I was told that I had just entered a battlefield, that I could expect a series of notices of denial of Medicare payment accompanied by a discharge date that would be two days after I got that notice.

Dan: Yeah, they said she’d get two days notice. Gloria says the nursing home staff told her she’d have 24 hours to appeal each of those, but even if she won, the denials would keep coming. In fact, they said,

Gloria Bent: If we won a couple of appeals, then we could expect that the frequency with which these denials were going to come would increase.

Dan: All of which happened. NaviHealth started issuing denials July 15, 2022, after Gary had been at the home for a month.

Gloria appealed. She told senators what the doctor who evaluated the appeal found: Gary couldn’t walk. He couldn't even move — like from bed to a chair — without help from two people.. That reviewer took Gloria’s side.

Her husband’s next denial came a week after the first. Gloria won that appeal too. She says the reviewer noted that Gary needed maximum assistance with activities of daily living. 

The third denial came four days later, and this time Gloria lost. 

Gary came home in an ambulance: As Gloria testified, he couldn’t get into or out of a car without assistance from someone with special training. 

And when he got into the ambulance, he had a fever. The next morning, he wound up in another ambulance — headed to a hospital with meningitis. He lost a lot of the functioning he’d picked up at the nursing home. 

He died at home a few months later. When Gloria testified in the Senate, all of it was still fresh. She told them that as awful as Gary's illness and decline had been, the fights with insurance were an added trauma.

Gloria Bent: This should not be happening to families and patients. It’s cruel. Our family continues to struggle with the question that I hear you asking today. Why are people who are looking at patients only on paper or through the lens of an algorithm

making decisions that deny the services judged necessary by health care providers who know their patients.

Dan: Bob Herman calls Gloria’s story heartbreaking, like so many others he's seen. 

And his attention goes to one part of Gloria's story beyond denial-by-algorithm. 

Because: It’s not just one denial. It’s that series of denials. You can appeal, but as Gloria testified, the denials speed up. And you have to win every single time. The company only has to win once. 

I mean, unless you're ready to get a lawyer and take your chances in court– which, in addition to being a major undertaking, also means racking up nursing home bills and legal bills you may never get reimbursed for, while the court process plays out. 

Bob Herman: This appeal system is designed in such a way that people will give up. If you have a job, you know, even if you don’t, and you’re, and you’re also trying to take care of a family member, um, it’s a rigorous monotonous process that will chew people up and spit them out and then the people are inevitably going to give up. And I think in some ways insurers know that.

Dan: Going out on a limb to say: I think so too. So Bob and Casey’s first story on NaviHealth came out in March of 2023. They were the characters in the movie who go, “HEY, I THINK THERE’S SOMETHING REALLY BAD HAPPENING HERE.”

And people started paying attention. Like the U.S. Senate. which held that hearing where Gloria Bent told her story. 

And like the federal agency that runs Medicare — the Centers for Medicare and Medicaid Services, CMS. 

CMS finalized a rule that told insurers: You can’t deny care to people just from using an algorithm. 

And something else happened too: Bob and Casey started suddenly getting a lot MORE information. 

Bob Herman: We received so many responses from people and it just opened the floodgates for former employees, just patients and family members, just everyone across the board.

Dan: And not just former employees. Current employees. And what they learned was: There was absolutely a strategy at work in how this algorithm was being used. It was strategy some people on the inside didn’t feel good about. 

And this strategy got developed after United HealthGroup — and its subsidiary, Optum– bought NaviHealth in 2020. And here's what NaviHealth employees started telling Casey and Bob about that strategy.

Bob Herman: For some of us, it’s creating this moral crisis. Like we know that we are having to listen to an algorithm to essentially kick someone out of a nursing home, even though we know that they can barely walk 20 feet.

Dan: What Bob and Casey learned from insiders– and how it connects to United’s role as a health care behemoth– that’s next.

This episode of An Arm and a Leg is produced in partnership with KFF Health News. That’s a nonprofit newsroom covering healthcare in America. Their reporters do amazing work, and I’m honored to work with them. We’ll have a little more about KFF Health News at the end of this episode.

So, NaviHealth — the company with the algorithm — got started in 2015.. And the idea behind it was to use data to get people home faster from nursing homes if they didn’t actually need to be there. 

Because there was a lot of evidence that some people were being kept longer than they needed. 

Bob Herman: There is some validity to the idea that there’s, there’s wasteful care in Medicare, like, you know, there’s been cases in the past proving that people stay in a nursing home for way longer than is necessary. And obviously there’s financial incentives for nursing homes to keep people as long as possible. 

Dan: Traditional Medicare does have limits on nursing home care — but if you need "post-acute care" — help getting back on your feet after leaving a hospital traditional Medicare pays in full for 20 days– pretty much no questions asked. One of the selling points of Medicare Advantage — like selling points to policy nerds and politicians — was that it could cut waste, by asking those kinds of questions. NaviHealth and its algorithm were designed to help Medicare Advantage plans ask those questions in a smart way. 

Bob Herman: There were… a lot of believers within NaviHealth that were like, okay, I think we’re doing the right thing. We’re trying to make sure people get home sooner because who doesn’t want to be at home. 

Dan: And as those employees told Bob and Casey: Before United and Optum came in, the algorithm had been there as a guide — a kind of first guess — but not the final word. 

NaviHealth has staff people who interact directly with patients. And back in the day, the pre-United day, Bob and Casey learned that those staff could make their own judgments. 

Which made sense, because the algorithm doesn’t know everything about any individual case. It’s just making predictions based on the data it has.

Bob Herman: And there was just, just this noticeable change after United and OptiMentor that it felt more rigid. There’s no more variation. 

Dan: If the algorithm says you go, you are pretty much going.

Bob Herman: United has said, no, that’s not the case, but obviously these documents and other communications that we’ve gotten kind of say otherwise.

Dan: Because these employees weren’t just talking. They were sharing. Internal memos. Emails. Training materials. All making clear: The company wanted people shipped out on the algorithm’s timetable. 

Bob Herman: Documents came in showing that like this was a pretty explicit strategy. You know, UnitedHealth was telling its employees. Listen, we have this algorithm. We think it’s really good. So when it tells you how many, how many days someone should be in a nursing home, stick to it.

Dan: Stick to it or maybe be fired. Bob and Casey got documents — employee performance goals– saying: How close you stick to the algorithm’s recommendations? That’s part of how we’re evaluating your job performance. 

Bob Herman: It’s okay. Algorithm said 17 days, you better not really go outside of that because your job is on the line. 

Dan: Here’s how closely people were expected to stick to it. In 2022, employee performance goals shared with STAT showed that workers were expected to keep actual time in nursing homes to within three percent of what the algorithm said it should be. Across the board.

So, say you had 10 patients, and the algorithm said they each should get 10 days. That’s 100 days. Your job was to make sure that the total actual days for those patients didn’t go past 103 days.

Then, in 2023, the expectations got more stringent: Stay within one percent of the algorithm’s predictions. 10 patients, the algorithm says 100 days total? Don’t let it get past a hundred and one. 

Bob Herman: Like that is, almost nothing. Like what, what, your hands are tied. If you’re that employee, what are you going to do? Are you going to get fired? Are you going to do what you’re told?

Dan: And one person who ended up talking, to did get fired. 

Bob Herman: Correct. Yes. Uh, Amber Lynch did get fired And what she said was what we had also heard just more broadly was it, it created this internal conflict, like, Oh my God, what I’m doing doesn’t feel right. 

Dan: Amber Lynch was a case manager. She told Bob and Casey about onepatient who couldn’t climb the stairs in his home after knee surgery. But the algorithm said he was ready. Amber’s supervisor said, “Have you asked the nursing home staff if they’ve tried to teach him butt bumping?” Amber grit her teeth and made the suggestion to the rehab director.

Amber Lynch: And she looked at me like I had two heads. She’s like, he is 78 years old. He’s not going to do that. He’s not safe to climb the stairs yet. He’s not doing it. We’re not going to have it butt bump the stairs.

Dan: Amber told Bob and Casey that when she got fired, it was partly for failing to hit the one percent target and partly for being late with paperwork– which she told Bob and Casey she fell behind because her caseload was so heavy.

She wasn't the only one with that complaint. 

Bob and Casey's story shows another NaviHealth case manager– not named in the story because they're still on the job — in their home office, struggling to keep up. 

That week, they were supposed to work with 27 patients and their families. Gather documents, hold meetings. Another week, shortly before, they’d had 40 patients. 

"Do you think I was able to process everything correctly and call everyone correctly the way I was supposed to?" the case manager asked. "No. It's impossible. No one can be that fast and that effective and capture all of the information that's needed."

Bob and Casey watched this case manager fill out a digital form, feeding the algorithm the information it asked for on a man in his 80s with heart failure, kidney disease, diabetes and trouble swallowing, who was recovering from a broken shoulder. 

A few minutes later, the computer spat out a number: 17 days. 

The case manager didn’t have a lot of time or leeway to argue, but they were skeptical that the algorithm could get that number exactly right based on only the data it had. 

And what data is the algorithm working with? What’s it comparing the data on any given patient TO? Bob Herman says that’s a big question.

Bob Herman: It’s something that for sure, like Casey and I, it’s been bothering us. Like, what, how is this whole system? Like, what is it based on? And we were never really given straight answers on that. NaviHealth and Optum and United have said it’s based on millions of patient records over time. The sources of that, it’s, it’s a little unclear, where all that’s coming from. 

Dan: Bob and Casey talked with an expert named Ziad Obermeyer, a professor at the University of California Berkeley School of Public Health, who is not anti-algorithm. He actually builds algorithmic tools for decision making in public health. 

AND he’s done research showing that some widely-used algorithms just scale up and automate things like racial bias.

He told Bob and Casey: Using an algorithm based on how long other, earlier patients have stayed in a nursing home — that’s not a great idea.

Because people get forced out of nursing homes, in his words, “because they can’t pay or because their insurance sucks.” He said, “So the algorightm is basically learning all the inequalities of our current system.”

And leaving aside that kind of bias, it seems unlikely to Bob that any algorithm could predict exactly what every single patient will need every single time. 

No matter how much data it’s got, it’s predicting from averages.

Bob Herman: It reminds me of, like, a basketball game where let’s say someone averages 27 points per game. They don’t have 27 points every single, the game they go out there. It just varies from time to time.

Dan: But the NaviHealth algorithm doesn’t have to be right every time for United to make money using it. 

Using it to make decisions can allow United to boost profits coming and going.

Bob Herman: United health and the other insurance companies that use Navi health. Are using this technology to more or less kick people out of nursing homes before they’re ready. And that is the claims denial side where it’s like, okay, let’s save as much money as we can instead of having to pay it to a nursing home.

Dan: And that’s just one side of it. The insurance side. Claims denial. But United isn’t just in the insurance business. 

United’s Optum side is in every other part of health care. 

Including — in the years since United took over NaviHealth — home health services. The kind of services you’re likely to need when you leave a nursing home.

In 2022, Optum bought one top home health company in what one trade publication called a “monster, jaw-dropping mega-deal” — more than 5 billion dollars. In 2023, Optum made a deal to buy a second mega-provider. 

Bob and Casey's story says NaviHealth's shortening nursing home stays is integral to United's strategy for these acquisitions. It does seem to open up new opportunities.

Bob Herman: You’re out of the nursing home because our algorithm said so. Now we’re going to send you to a home health agency or we’re going to send some home health aides into your home. And by the way, we own them. 

Dan: Oh, right, because: If you’re in a Medicare Advantage plan, your insurer can tell you which providers are covered. 

Bob Herman: So the real question becomes, how much is United potentially paying itself?

Dan: That is: How much might United end up taking money out of one pocket — the health insurance side — and paying itself into another pocket, Optum’s home-health services?

We don’t know the answer to how much United is paying itself in this way, or hoping to. And United has said its insurance arm doesn’t favor its in-house businesses.

But it seems like a reasonable question to ask. Actually, it’s a question the feds seem to be asking.

Optum hasn’t wrapped up its purchase of that second home-health company yet, and in February 2024, the Wall Street Journal and other outlets reported that the U.S. Department of Justice had opened an anti-trust investigation. 

And you don’t have to be in a Medicare Advantage plan run by United to get kicked out of a nursing home on an algorithm's say-so. 

Bob Herman says NaviHealth sells its algorithm-driven services to other big insurance companies 

He says, put together, the companies that use NaviHealth cover as many as 15 million people — about half of everybody in Medicare Advantage.

Bob Herman: Odds are, if you’re in a Medicare Advantage plan, there’s a, there’s a really good shot that your coverage policies, if you get really sick and need nursing home care, for example, or any kind of post acute care, an algorithm could be at play at some point.

Dan: This is the dark side of Medicare Advantage. 

Bob Herman: Everyone loves their Medicare Advantage plan when they first sign up, right? Because it’s offering all these bells and whistles. It’s, here’s a gym membership. It’s got dental and vision, which regular Medicare doesn’t have. And it’s also just, it’s, it’s cheaper. Like, if it’s just from a financial point of view, if, if you’re a low income senior, How do you turn it down? There’s, there’s so many plans that offer like free, there’s no monthly premiums in addition to all the bells and whistles. But Nobody understands the trade offs , When you’re signing up for Medicare and Medicare Advantage, you’re on the healthier side of, of being a senior, right?

Dan: And none of us can count on staying healthy forever. When you sign up for Medicare you’re signing up your future self — whether that’s ten or twenty or more years out. That future you, might really need good medical care. 

And at that point, as we explained in our last episode, if Medicare Advantage isn't working for you, you may not be able to get out of it.

Bob Herman: You could potentially not fully get the care that you need. We shouldn’t assume that, that this couldn’t happen to us because it can. 

Dan: So, yeah. Kind of a horror story. But: Unlike some horror movies, when Bob and Casey started publishing their stories, they started getting people’s attention.

We mentioned the new rules from the feds and the senate hearings after Bob and Casey’s first story in March 2023

Later in the year, when Bob and Casey published their story with documents and stories from inside NaviHealth, a class-action lawsuit got filed.

Since then, CMS has said it will step up audits under its new rules. 

Bob Herman: There was a memo that CMS sent out to Medicare advantage plans that said, Hey, listen, we’re telling you again, do not deny care solely on any AI or algorithms. Like just don’t do it. 

Dan: And in February 2024, the Senate held another hearing. 

Here’s Senator Elizabeth Warren at that hearing, saying these CMS rules aren’t enough. We need stronger guardrails.

Elizabeth Warren: Until CMS can verify that AI algorithms reliably adhere to Medicare coverage standards by law, then my view on this is CMS should prohibit insurance companies from using them in their MA plans for coverage decisions. They’ve got to prove they work before they put them in place.

Dan: So people — people with at least some power– are paying some attention. 

Bob Herman: I don’t think this is necessarily going to escape. Political scrutiny for a while. 

Dan: So, basically, the story isn’t over. 

This isn’t one of those horror movies where the monster’s been safely defeated at the end, and everybody just starts cleaning up the mess. And it’s not one where the monster is just on the loose, unleashing the apocalypse. 

Because it’s not a movie. There’s no ending. There’s just all of us trying to figure out what’s going on, and what we can maybe do about it.

One last thing: I got a lot of emails after our last episode, where we laid out a lot of information about Medicare Advantage and traditional Medicare. Most of it was along the lines of, Thank you! That was really helpful! Which made me feel really good.

And we got a couple notes about things we could have done better. Especially this: We said Traditional Medicare leaves you on the hook for 20 percent of everything, without an out of pocket limit. 

Which is true — but only for Medicare Part B: Doctor visits, outpatient surgeries and tests. Which can add up, for sure.

Medicare Part A — if you’re actually hospitalized — covers most services at 100 percent, after you meet the deductible. In 2024 that’s one thousand, six hundred thirty-two dollars. 

Thanks to Clarke Lancina for pointing that out. 

There have been a bunch of other, amazing notes in my inbox recently, and I want to say: Please keep them coming. 

If you go to arm and a leg show dot com, slash, contact, whatever you type there goes straight to my inbox. You can attach stuff too: documents… voice memos. 

Please let me hear from you. That’s arm and a leg show dot com, slash contact.

I’ll catch you in a few weeks. 

Till then, take care of yourself.

This episode of an arm and a leg was produced by me, Dan Weissmann, with help from Emily Pisacreta, and edited by Ellen Weiss. 

Adam Raymonda is our audio wizard. Our music is by Dave Weiner and blue dot sessions. Extra music in this episode from Epidemic Sound.

Gabrielle Healy is our managing editor for audience. She edits the first aid kit newsletter. 

Bea Bosco is our consulting director of operations. Sarah Ballama is our operations manager. 

And Arm and a Leg is produced in partnership with KFF Health News. That’s a national newsroom producing in depth journalism about healthcare in America and a core program at KFF, an independent source of health policy research, polling and journalism. 

Zach Dyer is senior audio producer at KFF Health News. He’s editorial liaison to this show. 

And thanks to the Institute for Nonprofit News for serving as our fiscal sponsor, allowing us to accept tax exempt donations. You can learn more about INN at INN. org. 

Finally, thanks to everybody who supports this show financially– you can join in any time at arm and a leg show dot com, slash, support — and thanks for listening.