Mismatch Medicine

How a diabetes drug and an old idea in evolutionary biology reframe why we get sick, and why I won’t call it a cure

By Loren Rauch, MD, MPH

For about twelve months in 2020 and 2021, I lived and slept in a small room outside my own house.

I’m an emergency physician. In the pre-vaccine stretch of the pandemic, I was reasonably sure the virus would eventually catch me, and genuinely afraid I would bring it home and kill someone I loved. So I moved into a tiny night-shift sleep room at the edge of the property. My kids podded with friends who were fairly wary of any household with an ER doctor. And in the hours between shifts I did what a lot of us did that year: I stayed in my room and trolled the literature for anything that looked like it might help. A treatment, a prophylaxis, an edge. Anything.

I want to tell you about a number I would have cheered in triumph for back then, which arrived about four years too late.

In 2024, a group of researchers revisited the SELECT trial, a large study of semaglutide (the drug in Ozempic and Wegovy), and examined what had happened to its participants during the pandemic. The trial was never designed to study COVID. The pandemic simply arrived in the middle of it, and the investigators had the sense to amend the protocol and start counting.

Semaglutide did not reduce anyone’s chance of catching COVID. Infection rates were identical in both groups. But among the people who got infected, the ones on semaglutide died of it far less often: 43 deaths versus 65, a hazard ratio of 0.66. Roughly a third fewer COVID deaths if you caught it. The same analysis found about a fifth fewer deaths from anything at all over the trial’s three-plus years.

I read that and sat back from my computer. A drug developed to control blood sugar, cutting COVID mortality by a third, not by stopping the infection but by changing whether it killed you. If someone had handed me that result in 2020, I would already have been on it.

That gap, between what this drug was built to do and what it keeps turning out to do, is what this essay is about.

The accidental experiment

Endocrinologists weren’t trying to cure obesity, or addiction, or heart disease. They were trying to build a better diabetes drug: the next insulin, but for type 2.

What they built instead was, to me, the most interesting uncontrolled experiment in modern medicine.

Tens of millions of people are now bathed continuously in supraphysiological concentrations of a gut-signaling molecule called GLP-1, a peptide most of us had never heard of and whose role we had badly underestimated.

Normally your gut releases a little GLP-1 after a meal, and it fades within minutes. These drugs hold it high, all the time, with a kind of armored persistence the body never evolved to experience. We pushed a small, ancient button with a sledgehammer.

Then patients started reporting things, and clinicians and researchers, to their credit, listened. The effects came in almost in sequence, each one a surprise:

• weight loss
• quieted cravings for food
• then for alcohol and nicotine and gambling
• lower inflammation
• protection for the heart and kidneys and liver
• fewer strokes
• signals in anxiety and depression
• early hints around dementia, aging, even cancer

A drug built for one job keeps doing a dozen. That breadth is the tell. It is not what you would expect from a narrow lever pulling on a single pathway. It is what you would expect if we had accidentally reached something central and old.

So the question this essay holds is simple:

Is GLP-1 a narrow lever, or have we stumbled upon something nearer to the root of how we get sick?

My answer, stated once and not oversold: GLP-1 looks like the most powerful and most accessible lever we currently have for a system whose set point was written elsewhere.

That is a claim about leverage, not authorship. Hold the distinction: I spend the end of this essay defending it against myself.

The evolutionary toolkit

Modern medicine is brilliant at one question: How does the body break?

It is nearly silent on a second: Why is the body built so that it breaks this way in the first place?

That second question is the whole of a small field called evolutionary medicine. It rests on a distinction the ethologist Niko Tinbergen drew for all of biology. Every trait has:

• a proximate explanation: the machinery operating right now in this body
• an ultimate explanation: the reason selection shaped it that way at all

A cough has a proximate cause (irritated airways, a reflex) and an ultimate one (expelling a pathogen was worth its cost over deep time). Most of medicine lives entirely in the proximate. The ultimate question is not decoration. It changes what you think the disease actually is.

Its single most useful idea is mismatch.

A large share of modern chronic disease is not a malfunction at all. It is a well-built adaptation running in an environment it was never designed for. The body isn’t broken. It is optimized for a world that no longer exists.

Obesity is the cleanest case. A body that aggressively sought calories, stored them efficiently, and defended that store survived famine. It is performing today exactly as designed. The design is simply colliding with cheap, engineered, always-available food. The disease is the collision, not a defect in either party.

To keep the rest straight, hold one picture: an orchestra with three roles.

• The composer wrote the score: in our story, the early-life programming that set the system’s sensitivities long before you had any say.
• The conductor is the integrating hub where many signals converge and get arbitrated: the body’s central energy-balance circuitry.
• The first chair is the loudest, most legible player, the one we can actually direct. That is GLP-1.

One discipline governs everything below, and I’ll state it once so you can hold me to it:

An idea's elegance is a reason to explore it, not a reason to believe it.

The pleasure of a clean, unifying story is exactly the feeling that has launched a thousand wrong theories. (It is the pattern I chase across everything I work on, from the order in which civilization and religion arose to how the human brain was shaped, and the pattern is seductive precisely because it feels true before it is shown true.)

So everything below is graded. Some rests on large randomized trials I would stake a great deal on. Some is plausible mechanism. Some is frank speculation, flagged as such.

The composer: willpower was never the variable

Start with the score.

An evolutionary principle called antagonistic pleiotropy holds that selection will keep a trait that helps you when young even if it harms you when old, because the early payoff dominates. Sensitive reward circuitry, efficient fat storage, a hair-trigger appetite are not flaws. In a world of scarcity and danger, they were advantages, and the machinery now failing a fifty-year-old is the same machinery that helped a twenty-year-old survive a lean winter.

There is a reason evolution never engineered a defense against the diseases that follow. Almost all of them strike after reproductive age, in what biologists call the selection shadow. Once reproduction is past, selection’s grip weakens, and there was little pressure to build a guard against late-life metabolic collapse. The body has no protection here because it never needed one. That is not a flaw in the argument. It is why the mismatch exists.

A more contested idea sits atop this. The developing body, sometimes in the womb, may read its early environment and make a bet about the world it is about to enter, pre-committing to a phenotype.

I’ll flag, honestly, that the predictive version of this is debated among those who study it. But the intuition is hard to shake in the clinic. For some patients, the metabolic thermostat was set to survival mode before they had any agency over it. It was a prediction made on their behalf, and for many of us, it didn’t arrive. We were tuned for famine and handed abundance.

This is why I no longer treat obesity as a question of character. Not because that’s unkind, though it is, but because it’s inaccurate given the wiring.

We don’t tell a thyroid it failed through inattention. Willpower was never the variable.

Why the alarm misfires now

Two more ideas explain the day-to-day experience.

1. The smoke-detector principle

The first is the smoke-detector principle, from the evolutionary psychiatrist Randolph Nesse. Alarms are tuned by the cost of being wrong, not by accuracy.

A smoke detector that shrieks at burnt toast is correctly designed, because one missed fire is catastrophic and a thousand false alarms are merely annoying.

Hunger and craving work the same way. In a world where a missed meal could be fatal, a constant low hum of seek food was the right setting.

What people on these drugs describe, the food noise going quiet, isn’t a broken signal being silenced. It is an alarm whose threshold finally got reset to a world it was never calibrated for.

2. The supernormal stimulus

The second is the supernormal stimulus, an exaggerated version of a natural cue that triggers a stronger response than the real thing ever could.

Hyperpalatable food, nicotine, slot machines, the infinite scroll: all engineered to overstimulate reward circuitry built for a tamer world.

This, I think, is why a single drug quiets so many different cravings at once. It isn’t treating a dozen separate addictions. It is turning down the gain on the one ancient seeking system they all hijack.

I’ll add my own data point, with the caveat that it is a single person. Within a few weeks of starting (titrated semaglutide, later dose-split tirzepatide, now orforglipron), I was comfortable self-managing only because I’m a physician who can read his own labs; the constant background hum I hadn’t known was there went quiet. The food noise, the low-grade impulse to buy the thing, the reaching.

The startling part wasn’t the quiet. It was realizing how loud it had always been.

An ancient signal, and the brain question

Here is where it gets genuinely strange, and where the evolutionary reading earns its keep.

GLP-1 is not a modern curiosity. The system it belongs to is ancient, conserved across the vertebrate lineage, with the same incretin, proglucagon machinery turning up in fish, frogs, and egg-laying mammals. It began as a gut signal coordinating digestion, plugged into the gut–brain axis, one of the oldest regulatory circuits animals have.

That deep conservation is almost certainly why a humble digestion hormone reaches all the way up into the brain’s reward centers and memory structures. The wiring is old enough to be wired into nearly everything.

The clearest clue is usually told as trivia. One of the first drugs in this family came from the venom of the Gila monster, a desert lizard. The lizard makes a peptide, exendin-4, that is only about half identical to our own GLP-1, but similar enough to be a potent key for the human receptor, and sturdier than our version, which is why it became a drug.

A reptile evolved a near-copy of our gut hormone, put it in its venom, because it works broadly, including in mammals like us. That isn’t a footnote about drug discovery: It is evidence of how old and how shared the underlying lock is.

Which raises the question that genuinely stops doctors when I ask it:

What is a GLP-1 receptor doing in your brain?

Because it is there, and the brain makes its own GLP-1 too. People assume these drugs are large peptides that can’t cross into the brain, and mostly they can’t do so freely. But the brain has built-in receiving stations, small windows where the barrier is naturally porous, and a sledgehammer-level, always-on dose gets enough signal through to matter.

The point isn’t that we are forcing it in. It is that the receiving equipment was already installed. The brain was built to listen for this.

This is also the answer to the cultural anxiety about these drugs, the worry that we are hacking the body, overriding natural desire, imposing chemical control. That worry assumes we are introducing something foreign.

The deep-time view suggests almost the opposite. We are boosting the volume of an ancient signal that modern abundance has drowned out, restoring a message that evolution has relied on for longer than there have been mammals. For anyone inclined to call these drugs unnatural, the phylogeny is the rebuttal. The signal could hardly be more native.

The newest drugs only sharpen the point. Tirzepatide hits not just GLP-1 but a second gut hormone, GIP, and the field is already moving toward triple agonists that add a third.

A lesser view would treat that as a complication. I read it as confirmation. We grabbed the gut end of an enormous, ancient, body-and-brain-spanning system and called it a digestion hormone. The more receptors we find that matter, the bigger the system we have stumbled into looks.

We are the blind men describing the elephant by holding its tail.

The shared channel: inflammation

Let me make it personal again, because it’s the part I can’t stop thinking about. I have the metabolic-syndrome starter kit: a tendency toward high blood pressure, high cholesterol, the early edge of type 2 diabetes. For years, the medical answer to people like me has been medication whack-a-mole.

A pill for the pressure, a pill for the cholesterol, a pill for the sugar, each knocking down its own number.

And here is what has always bothered me: Even when you win at whack-a-mole, every number controlled, people like me still have more heart attacks, strokes, and cancers. Cardiologists have a name for it: residual risk. You can normalize every readout we use and the disease keeps climbing out of some other hole.

Inflammation turned out to be a large part of what whack-a-mole leaves behind.

The proof that it is a cause and not just a marker came from a trial called CANTOS, which used an anti-inflammatory antibody that does nothing to weight, sugar, or cholesterol, and still cut cardiovascular events by about 15 percent. Inflammation down, events down, nothing else changed.

Now look back at the GLP-1 data. These drugs lower inflammation too. The heart benefit in SELECT shows up before the weight comes off, and the receptors keep turning up where they have no business being for a digestion hormone: on immune cells, in the brain, woven into the inflammatory machinery itself.

The COVID number I opened with fits exactly. The drug didn’t stop people catching the virus. It stopped them dying of it, a severity effect, the fingerprint of calming an overreaction rather than fighting a pathogen.

So inflammation looks less like another item on the list and more like a shared channel, one of the main lines through which this lever reaches the heart, the metabolism, and maybe the mind at once.

This is exactly where I have to discipline my own excitement. CANTOS itself is the caution. It cut heart events but did not lower overall death, and it raised deaths from infection, because when you turn an alarm down you turn it down for real fires too. The smoke detector cuts both ways.

The bet I would make

Here is the idea I find most generative, offered as a hypothesis rather than a result, and it is the one I would most like someone to test.

We have assumed these drugs override a hungry, overactive system. But what if, for people shaped by early adversity, the body’s own GLP-1 signal was quietly turned down to begin with? Fewer gut signals, a brain tuned to ignore them, so that the drug is less an override than a replacement for a signal that was suppressed before you could vote on it.

That bet makes three predictions you could actually test:

1. The drugs should work better in people who carry more early-life adversity.

2. Those people should show a blunted natural signal to begin with.

3. At least some of the benefit might outlast the drug, if the early marks turn out to be reversible.

There is a candidate mechanism: early stress reshaping the gut’s microbial ecology and, through it, the production of the signal, alongside changes in the brain circuits that make and read it.

The full chain is unproven. It is the experiment I would run, not the finding I’m reporting. I’ll say the uncomfortable part plainly, because it is the discipline the whole frame depends on. If the benefits don’t scale with early adversity, the story is wrong, and I would rather find that out than protect it.

The testable version of this needs people who can run the bench and the cohort. If that’s you, I want to hear from you. That is not a rhetorical flourish.

Where this could be wrong

The sharpest objection deserves a fair hearing. Maybe all of this (the cravings, the mood, the inflammation ) is just downstream of losing weight and feeling a little nauseated, and the grand evolutionary story is dressing up a simple appetite suppressant.

It’s the right question, and there are real answers.

• The heart benefit in SELECT, a 20 percent drop in cardiovascular death, heart attack, and stroke across more than 17,000 people, showed up before meaningful weight loss, and a later analysis found only about a third of it traced to reduced waist size.
• The COVID survival benefit cannot be a weight effect; losing weight during the months you have COVID does not save you.
• CANTOS cut heart events with no weight loss at all.

And there is a tell I find hard to wave away. In a Swedish national cohort of roughly 95,000 people, four drugs that hit the same receptor diverged sharply.

• Semaglutide, the most brain-penetrant of them, was tied to a lower risk of worsening depression, anxiety, and substance use, with reductions in the range of 40 to 50 percent.
• Liraglutide showed a weaker, narrower effect.
• Exenatide and dulaglutide showed essentially none.

If this were just weight loss, the chemistry of which drug should not matter so much.

I’ll be honest about the other side of the ledger. Some large observational studies report the opposite, links to worse depression or suicidality. I weight those lightly, and not reflexively. A big enough dataset with enough analytic freedom can produce a statistically significant association in almost any direction you’re inclined to find. That is a feature of the method, not a finding.

What makes the cohort above more persuasive isn’t its size but its design: each person served as their own control, which strips out most of the confounding the scarier studies can’t. None of which makes these drugs free.

Real costs and limits

The most important cost is one the scale hides. A quarter to nearly half of the weight lost can come from muscle, silently, and it has to be defended against deliberately with protein and resistance training; preserving lean mass is now an active drug-development target in its own right.

The same quieting that frees some people from craving can, pushed too far, flatten them, the Ozempic apathy that takes the color out of music, appetite, libido.

And the access problem is real and ugly. A drug that recalibrates a system set by early hardship is, at current prices, available mostly to people who didn’t face that hardship. I don’t have an answer to that, and I’m suspicious of anyone who pretends the personal protocol scales to a population.

There is one more hazard, the one that should worry an evolutionary storyteller most. The oldest charge against this kind of reasoning is that you can narrate anything as an adaptation after the fact, in an untestable way. It’s a fair charge. The only defense is the one I tried to meet above: convert the story into predictions that could fail. The replacement hypothesis is worth telling only because it is tethered to a test it could lose.

The oxytocin lesson

Now the warning I’ve been pointing at, and the reason I keep insisting on leverage, not authorship. Remember oxytocin, the love hormone, the bonding molecule. It was easy to measure and manipulate, so it was crowned the driver of human attachment.

A decade of careful work then walked most of that back. Many of the headline effects did not replicate, and oxytocin turned out to be one player in a redundant, context-dependent network, not its conductor. It was miscast, not unreal.

GLP-1 sits in a similar seat, and the same bias is at work. We credit the node we can see and drug.

But the cautionary analogy needs one correction, and the correction is the whole point. Oxytocin’s effects were often fragile and didn’t replicate. GLP-1’s effects are large, dose-dependent, and reproduced across major trials. So GLP-1 is not making oxytocin’s empirical mistake. The risk is that it makes oxytocin’s attribution mistake: crowning the loud, druggable instrument as the author of the music.

Back to the orchestra.

GLP-1 is the first chair: real, reliable, the player we can actually direct, and worth every bit of the attention.

But the composer is the early-life programming that set the score, and the conductor is the central system this drug feeds into alongside many others.

A thermostat dial has enormous leverage over a room without being the building's architect.

Keeping GLP-1 as first chair rather than composer is the version of this story I won’t have to apologize for in ten years.

What it opens, and the close

If even part of this holds, the picture it offers is one in which metabolic disease, addiction, and a meaningful share of mood disorders aren’t separate epidemics, but one over-responsive, mismatch-stressed system seen from different angles, which is why a single intervention keeps turning up in all of them.

Metabolic health is mental health stops being a slogan and starts looking like a mechanism.

In the clinic it changes the conversation. The frame isn’t add a drug so much as, potentially, stop playing whack-a-mole.

As the upstream physiology improves, you can sometimes peel away the blood-pressure pill, the sugar pill, the rest. Replace three drugs, don’t add a fourth.

And the thing I most want a patient to hear isn’t it’s not your fault. It’s something truer and stranger: Your body did exactly what it evolved to do. It prepared you for a harder world than the one you got. This doesn’t fix a flaw. It updates an old prediction.

That’s the personal part. People like me (and I mean my actual family, not a generic patient ) may not have to spend the rest of our lives white-knuckling our biology and quietly ashamed that willpower wasn’t enough.

The willpower was never the variable. There is a lever now. It is a powerful lever, not a cure, and it comes with costs, unknowns, and an access problem I can’t wave away. But it’s real.

I think often about the version of me, living in that sleep room in 2020, scrolling the literature at three in the morning, afraid. If someone had shown him the number that a diabetes drug would, four years later, turn out to cut COVID deaths by a third, he would already have tried it. He wouldn’t have known why it worked. He would have been holding the elephant’s tail.

We are all still holding the tail. We have found a remarkable instrument. We should be awed and grateful, and slow to call it the composer.

.

Note on confidence. The cardiovascular, COVID-mortality, agent-specificity, and substance-use findings rest on large randomized or national cohort data, and the figures are drawn from published trials. The mood and addiction signals are real but still emerging. The aging, cancer, and early-life-programming pieces are frontier: the questions I would chase, not claims I would defend. Where the evidence is thin, I’ve tried to say so plainly rather than let a good story carry more weight than it has earned.