Jordan Miller is a post-baccalaureate student, double majoring in statistics and data science and mathematics at UT San Antonio.
Robert Miller is a specialist in poison information with the South Texas Poison Center at UT San Antonio.
Episode Description
Research suggests that more than half of Americans could benefit from taking a GLP-1 RA medication. The drugs were originally designed to help manage type two diabetes, but in 2024, the Food and Drug Administration approved one GLP-1, semaglutide, for weight loss. With that approval came a spike in prescriptions for the drugs, and a spike in calls related to GLP-1s to poison control centers. That's the focus of this episode of Stats and Stories with guests Jordan and Robert Miller.
+Full Transcript
Rosemary Pennington
Research suggests that more than half of Americans could benefit from taking a GLP-1 RA medication. The drugs were originally designed to help manage type 2 diabetes, but in 2024, the Food and Drug Administration approved one GLP-1, semaglutide, for weight loss. With that approval came a spike in prescriptions for the drugs and a spike in calls related to GLP-1s to poison control centers.
That's the focus of this episode of Stats and Stories, where we explore the statistics behind the stories and the stories behind the statistics. I'm Rosemary Pennington. Stats and Stories is a production of the American Statistical Association in partnership with Miami University's Departments of Statistics and Media, Journalism, and Film.
Joining me today is regular panelist John Bailer, emeritus professor of statistics at Miami University. We have two guests joining us on the show today. Jordan Miller is a post-baccalaureate student, double majoring in statistics and data science and mathematics at UT San Antonio, and Robert Miller is a specialist in poison information with the South Texas Poison Center at UT San Antonio. They're two of the co-authors of an article in Significance about a spike in GLP-1-related calls to poison control hotlines.
Thank you so much for being here today.
Jordan Miller
Thank you. Thank you for having us. Thank you.
Rosemary Pennington
I think, yeah, I think to get started, it might be nice to remind our listeners what exactly a GLP-1 is before we get into a conversation about why there are these spikes and what this means.
Robert Miller
Yeah, so the history of the GLP-1 drugs is interesting in that I think for most people, it's a relatively new phenomenon that they're familiar with since the 2020s. But realistically, these drugs have been around for a pretty long time.
You know, there was the kind of prescription drug boom in the 1990s, this concept of the blockbuster drug, and that only increased in the 2000s, specifically with the passage of Medicare Part D. Right now, the government's going to foot a certain portion of the bill for prescription drugs. That tended to kind of slow the rate of increase in the concept of the blockbuster drug, kind of slowed down through the 2010s.
And so these GLP-1 drugs kind of emerging again for the purposes of weight loss—it's definitely kind of a new... It's kind of a "what's old is new again" phenomenon.
Specifically, these were primarily prescribed for diabetes, particularly type 2 diabetes, starting around the year 2004, and they've kind of existed as, for lack of a better word, kind of an "also drug." So they didn't really change what the mainline therapy for diabetes was, but this was another kind of what we would call adjunctive therapy that was available for people who had a lot of difficulty achieving their lab values that they wanted to, and just their glycemic control.
And they were kind of just always there. And then the realization was, when you look at the clinical studies, you did see this weight loss effect, and that was enough to really motivate some people to start looking at things closer.
You started getting clinical trials for weight loss in 2014, and there were some limitations, though. Some of the GLP-1 drugs they were looking at were once daily. Some were even twice daily. And the idea of doing a subcutaneous injection in your abdomen two times a day—you know, for diabetics, that's no big deal, of course. But as a weight loss drug, you have to think about the pragmatics of what you're actually bringing to market, and that was always a major limiting factor.
Things really changed with semaglutide because that was your once-weekly, and suddenly that was kind of the Goldilocks zone. People are willing to do a once-weekly subcutaneous injection for weight loss, and so when those clinical trials dropped, things just really, really exploded.
In addition to that, I think what makes the semaglutide kind of explosion, so to speak, unique is that we didn't really... So we've obviously had Instagram and TikTok and this short-form social media experience for a while now, but this was the first time where you had that Venn diagram of a blockbuster drug is coming out, plus we have this social media concept. And so it was just very, very different compared to, say, like when Lipitor came out. This was a blockbuster drug as well, but people were not super excited about their cholesterol falling in the same way they are with the GLP-1 drugs.
John Bailer
So let me ask: How did the two of you get involved in this project? I mean, what was it that brought you in and said, "Gee, I want to know more"?
Jordan Miller
Well, partially that, as you may or may not know, Robert and I are actually a married couple, and so Robert works for the Poison Center, and they certainly subjectively experienced the huge increase in calls about these drugs.
At the same time, again, going through my undergraduate statistics program, I had a class with Dr. David Hahn, who just kind of put out an open call for undergraduates who wanted to do research he was willing to assist. I thought he might have a project in mind. He said, "No, whatever you want."
And so, just talking about it at home with my husband, you know, he sort of had this idea of, "Hey, there's this suspicious increase in this particular drug. I bet, you know, with the statistical methods that I was learning at school, I could maybe provide a better look than..." Not to disparage my colleagues in the health professions, but they didn't quite have the tools to look at it.
John Bailer
So, it seems like the poison control centers are the key link in this story, and that's the source of information. Can you tell us a little just background information about poison control centers and how they function and what triggers contact with them?
Robert Miller
Yeah, so poison control centers have existed since the 1950s, and it started as a very interesting, just kind of out of necessity.
Basically, it was a pediatrician who realized he was just getting all these calls regarding kids getting into stuff, and it was just kind of like, okay, but do we treat that or do we not treat that? What's a problem? What's not a problem?
And he basically just started writing down feverishly things on index cards. And it was kind of funny because the original Poison Center, which was based out of Chicago, it literally was just a red—this, like, cartoon red phone on the wall in one of the emergency departments, and they would have... You would just be working in the ER, and you would go answer that phone when it rings, and you had your card catalog of information.
Starting in the year 2000, we dramatically modernized. Of course, now we're a 1-800 number. We have drug database resources, and generally, what's interesting is we get a little bit of everything.
So, in the sense that, you know, ostensibly we're for toxicology. So this idea that a kid gets into something, we need to make the decision: Is this safe to watch at home? Do we need to go to the emergency room?
But then, in addition to that, we also kind of do more what we would call consulting, in the sense that we also get calls from emergency departments saying, "Hey, we've got this..." The technical word we would use is like a toxic ingestion, which we tend to use even when it's not an ingestion, just one of those healthcare things. And they were asking, like, "How do we treat this?"
And at first it was kind of like, well, you're the ER doc. But then, needless to say, we kind of rose to that challenge pretty quickly.
And now it's pretty much 50-50. Half of our job is pediatrics; half of our job is adults in the emergency room. Even balance between therapeutic errors, chronic toxicity, but also obviously intentional self-harm.
And what's interesting, too, is it wasn't intentional that it worked out this way, right? Just we're answering these calls, we're generating data, and then starting around 2000—or really since the 1990s, but more in a modern sense in the 2000s—we started sending this to a database.
And so the National Poison Data System is pretty close to real time. So if a patient or, you know, someone calls and I walk them through... and I make a medical chart, that chart is actually uploaded to this database in around eight to 10 minutes, typically.
So it's, on one hand, like the job of this database is not epidemiology per se. So I'm not getting, like, every single vital sign, every single lab value, et cetera. My job is to help the caller. But as just an unintended side effect, we ended up with this kind of surveillance network, basically, where subtle signals—you have to have a lot of finesse. But at the same time, like, if a bunch of people start getting really sick, we actually can find out really quickly.
Jordan Miller
I was also going to add that, you know, in the '90s and early 2000s, many people did not have the internet as a resource at their fingertips, and that's sort of how poison control was born. And it would be, "Hey, my kid ate soap. What do I do?"
You could have these really common, benign situations, but now that everyone has a smartphone, there's the internet, and now also AI. People are turning to that.
I mean, I guess we don't have a statistical proof of this, but, you know, we kind of reason that a lot of people turn to the internet and whatever searches to just say, "Hey, you know, is this a problem?" And they get a quick, satisfactory answer.
And so the role in poison control really seems to be shifting toward these new drugs that the internet doesn't necessarily have a lot of information about, but also possibly higher-acuity or unusual exposures.
Whereas I think people know by now that if your kid eats soap, it's weird, but it's not a problem.
Rosemary Pennington
You're listening to Stats and Stories, and we're talking about GLP-1-related calls to poison control with Jordan Miller and Robert Miller.
So, Jordan was talking about how, you know, she had this product she wanted to work on, sort of coming out of a stats class. And Robert, you're working in poison control. I wonder what it was you were seeing that made you feel like, "Ah, this is something that we need to research," and sort of if there was a point in time where it really became clear to you that there was really something interesting happening.
Robert Miller
Yeah. So, from my perspective, there are two ways to kind of approach thinking about toxicology, right? One is from just the clinical syndromes themselves and the pathophysiology.
And on that note, honestly, it wasn't too surprising. Once again, these drugs didn't emerge out of nowhere. We had a lot of data from back in the kind of—what I guess I'll call the diabetes era—of what happens when these patients overdose, and we knew it was pretty substantial nausea, vomiting, abdominal cramping, almost like a food-poisoning-esque picture.
And this also wasn't too surprising, right? These are glucagon-like peptides. For anyone that works in the emergency room, you might be familiar with glucagon, which also causes a very similar kind of toxic presentation when it's either inadvertently overdosed on, you know, just through a therapeutic error as well.
So that part wasn't too surprising.
So what mostly jumped out at me was the other side of that equation, which is resource utilization.
So, like, in the U.S., we have a big problem where people don't have access to primary care in the way that they would like to. And so the idea is that, well, we can just go to the emergency room. And using the emergency room as your primary care source—not the greatest thing in the world, right?
So just when you think about how expensive that is, and the sort of opportunity cost of, well, they need to take care of stroke and heart attack patients predominantly.
And so it really was a situation where I wasn't too worried about the clinical syndrome per se. It was just I went from getting—you know, this was a call I would get every now and then with someone saying, taking a bit too much of their Victoza, versus now I'm getting calls like every single day, sometimes multiple calls per day.
And it was kind of like... So we also think a lot about disposition, right? How long do we expect these symptoms to last? How long is the patient going to be in the emergency room?
And so that's what kind of motivated me to take a look under the hood, so to speak.
Rosemary Pennington
Yeah. So, in looking at some of your figures and some of the discussion in your paper, you're tracking the number of calls over time, and you notice a couple of things.
One is you notice that the types of calls you were getting for these drugs was changing compared to when they were exclusively being used for diabetes or other purposes, and you also notice something about the trajectory.
Can you talk about those two results?
Robert Miller
Yeah, I can definitely talk first about the patients. I'll leave the trend to the expert.
So, with the patient, yeah, it was kind of like... So you would get calls on diabetic patients who were, "Oh, I accidentally did, like, a double dose," or, you know, something of that nature, and generally it wasn't too bad.
And part of that is, I think, these patients were primarily being managed by their endocrinologist, and those endocrinologists were very, very knowledgeable about, okay, I need to slowly titrate this drug up because that's what prevents the nausea and vomiting.
And so once a patient is stable, kind of at what we would call pharmacokinetically sort of a steady state, you know, if you take a double dose in that circumstance, it's a... it kind of comes out in the wash, so to speak.
What I noticed with these patients that were taking it for weight loss is, one, the idea that it's being managed by an endocrinologist kind of goes out the window.
Like, definitely there are endocrinologists involved, but now, like, family doctors are prescribing this. Obviously, there's a huge topic to be discussed regarding the emergence of these, like, Instagram companies, right, where you can just get on, like, a subscription service to get GLP-1s directly to your door.
And so that fundamentally changed the patient.
And then what I also noticed was that titration was not always happening. So some patients were—and you can... Part of this, I think, is healthcare system errors and, like, the prescribers not being knowledgeable, but also the patients themselves.
There's this understandable temptation that if a little bit is good, more is better. And so a lot of patients would just basically start taking one-milligram dosages, which you typically wouldn't start until, like, your third month, right away.
And then they get this nausea, vomiting, abdominal cramping, and that just contrasted heavily with what I was kind of used to in these patients.
Jordan Miller
So, as far as the trend, I guess one thing I've sort of learned in my statistics education is nothing is discrete, but also nothing is continuous.
And what I mean by that is, you know, we chose to partition this data into a pre-approval and post-approval period, with the line of demarcation being the FDA kind of first approving semaglutide for chronic weight management, because that's sort of the event that opened the floodgates to expand the usage of these drugs to a new patient population.
And so we have these two, you know, sort of discrete periods. As we know, in the real world, you know, everything is continuous and nothing can be neatly discretized. But, you know, our measurements are always inherently discrete.
I guess that's what I mean. We live in a continuous world and we only have discrete measurements available to us.
But so, as an undergraduate student, of course, the first thing I wanted to do is, can we, you know, build a regression model that will both reproduce the frequency of calls that poison control received and also, you know, perhaps predict—which I think is to say, I took the last couple of months off the timeline and asked the model to accurately predict the next two—for the data that we did have.
But, so, you know, I tried kind of a battery of models at first: a linear, a quadratic, you know, polynomial, negative binomial. And of course, to no statistician's surprise, the best regression model was the Poisson regression, because calls to a poison center are a Poisson process.
You know, it's measuring time between calls. So the time between calls basically decreased because as this drug becomes more popular, people are calling the poison center about this drug more often.
So the Poisson regression, you know, was the most accurate model based on our AIC and R-squared statistics.
But the main factor in that regression was the interaction of time and whether the drug was semaglutide, so that, you know, it was cool to show off in a fancy statistical way that semaglutide, among all GLP-1 RAs, was really the driver of this increase, especially, of course, in the post-approval period.
But there are many types of drugs that are used to treat both type 2 diabetes and for chronic weight management. But semaglutide just far and away led the pack here.
John Bailer
So basically, in this analysis, you say you're looking—you expect more calls with more use of a drug, so that's kind of what's in play.
But then there's this question: Are you seeing something that outstrips what you would expect if it was just tracking the drug being used more?
And that's kind of what you were doing with that, as you were describing that interaction term—that deviation from a pattern you would expect if it was following what you might expect just from more use.
Is that a fair summary?
Jordan Miller
Yeah, I'd say that's a fair summary.
Now, one limitation we have here is we don't actually have access to the number of prescriptions. This is clearly—or, sorry, simply—through calls to poison center.
So, right, we don't have that validation to make, nor is it a, you know, as Robert mentioned earlier, we're not... Well, I say "we" as if I'm a poison center, but I'm not. But we're not an epidemiological center.
And so there's a lot of other questions that we had that we wouldn't be able to answer because that data just doesn't get collected on the calls.
So our hats off to the great statisticians working in clinical trials. Famously difficult subject, but those are the people to go to for those kinds of questions.
Robert Miller
Yeah, and coming from this as a pharmacist, you know, my background is purely biostatistics. So much to my displeasure, I exist in a kind of frequentist space versus more like Bayesian thinking.
So when I first pitched this project, I went to Jordan and I was like, "Well, I don't get it, right? Just do pre-post medians."
And of course, she looked at me like I was a very adorable small animal and quickly corrected me, like, "No, the time-segmented analysis. Do that one."
Rosemary Pennington
Well, that's all the time we have for this episode of Stats and Stories. Jordan and Robert, thank you so much for joining us today.
Jordan Miller
Thank you so much. Pleasure.
Rosemary Pennington
Stats and Stories is a partnership between the American Statistical Association and Miami University's Departments of Statistics and Media, Journalism, and Film.
You can follow us on Spotify, Apple Podcasts, or other places where you find podcasts. If you'd like to share your thoughts on the program, send your email to statstories@amstat.org, or check us out at statsandstories.net.
And be sure to listen for future editions of Stats and Stories, where we discuss the statistics behind the stories and the stories behind the statistics.