Here's a thing that's been sitting with us since the nausea episode.
The dopamine blocker one.
Right. We talked about metoclopramide and the other dopamine antagonists for vomiting, functional GI stuff, and we mentioned tardive dyskinesia in passing. Daniel wrote in with a follow-up, and it's a good one. He says he has real respect for people living with psychosis and schizophrenia, and he thinks people are far too quick to criticise antipsychotics when what came before them, and what would happen without them, was brutal. Fair. But he also says TD is a sad condition and a heavy price to pay, and he's always wondered how you can take a drug and end up with something irreversible. What's the mechanism? And how much has the newer generation of antipsychotics actually reduced the incidence over prolonged treatment?
So we're taking that seriously.
We're taking that seriously. The mechanism, the irreversibility, the numbers, and the honest trade-off. And I'll say up front, Daniel's framing is the right one. This isn't an episode about how antipsychotics are bad. It's an episode about a real cost attached to a drug that does real work.
Which is exactly why it's worth doing properly. TD is a potentially irreversible hyperkinetic movement disorder resulting from prolonged exposure to dopamine receptor blocking agents. That's the definition. Most commonly antipsychotics, but also antiemetics with central dopaminergic antagonism, which is the bridge back to the last episode. Schönecker first described it in 1957, about three years after chlorpromazine came into psychiatric use.
So it showed up almost immediately.
Almost immediately. And the classic presentation is orofacial. Repetitive lip smacking, tongue protrusion, lateral tongue movements, cheek puffing, rhythmic jaw movements. It can spread to the limbs as choreiform finger movements, the so-called piano-playing fingers, to the trunk as rocking or pelvic gyration, and rarely to the respiratory muscles, which is where it gets dangerous. Dysphagia, aspiration, in the worst cases asphyxia.
And the scale. Because I think people hear "rare side effect" and move on.
Mean prevalence among antipsychotic-treated patients is twenty to twenty-five percent. Some estimates run twenty to thirty. This is not rare. It's one of the more common serious adverse outcomes in all of psychopharmacology.
One in five.
One in five, roughly, across treated populations. And the antiemetic bridge matters here, because metoclopramide carries the same black box warning and the same mechanism. It's not a psychiatric problem. It's a general pharmacology lesson about what happens when you block D2 receptors for long enough.
So let's get into the mechanism. Why does the brain start doing this?
The dominant theory is dopamine D2 receptor supersensitivity. You chronically block D2 receptors in the striatum, and the brain responds the way it responds to any sustained blockade. It upregulates the postsynaptic receptors. It grows more of them, and they become more sensitive to whatever dopamine is still floating around. So you end up with a striatum that is hypersensitive to endogenous dopamine.
And the evidence for that is what?
Three things, and they line up nicely. One, if you raise the antipsychotic dose, you temporarily suppress the TD. Because you're blocking the receptors again. Two, first-generation drugs with higher D2 affinity cause more TD than lower-affinity drugs. Three, the VMAT2 inhibitors, which deplete presynaptic dopamine, treat it. All three point at the same receptor.
Okay. But that model has a problem, doesn't it.
It has a big problem. In animals, D2 supersensitivity develops quickly and resolves quickly after you stop the drug. Caroff called it an inexact model for TD in humans. Because in humans, the movement doesn't go away when the drug goes away. And the model also doesn't explain why TD doesn't hit every patient, why onset varies so much, or why it correlates so strongly with age.
So the receptor story explains why the movement starts. It doesn't explain why it stays.
And the leading explanation for the permanence is maladaptive synaptic plasticity. Teo, Edwards and Bhatia, in Movement Disorders in 2012. The idea is that dopamine receptor sensitization plus altered NMDA receptor function causes the striatum to encode an abnormal motor program. The brain doesn't just get chemically sensitized. It learns the movement. It stores it the way it stores any learned motor pattern.
The brain learns the tic.
Which is why removing the drug doesn't remove the movement. You're not dealing with a chemical state anymore. You're dealing with a stored program.
That's the part that makes the whole thing click for me. Because a chemical effect should wash out. A learned program doesn't.
And that's why the irreversibility is so stubborn. Caroff's summary is that the potential irreversibility appears to stem from selective damage to striatal interneurons, and it may implicate glutamate, acetylcholine, GABA, oxidative mechanisms, or some combination. The structural correlates back this up. Reduced basal ganglia and thalamic volumes on imaging. Neuronal loss and gliosis on post-mortem. Mitochondrial abnormalities. Elevated markers of oxidative damage.
So there's actual tissue change.
There's actual tissue change in at least a subset of patients. Which is why the word "side effect" undersells it so badly. A side effect is something that happens alongside the treatment and stops when the treatment stops. This is a second condition.
Two competing theories sit alongside the plasticity story, right?
Two main ones, and they're not mutually exclusive. The GABA hypothesis says you're damaging striatal GABAergic neurons, especially the parvalbumin fast-spiking interneurons, which unbalances the direct and indirect pathways through the basal ganglia. The reactive oxygen species hypothesis says dopamine metabolism itself generates hydrogen peroxide and free radicals, which drive lipid peroxidation and cell membrane damage. Both of those could be running at the same time as the plasticity story. Takeuchi and colleagues in 2022 were explicit that these aren't competing so much as complementary.
And then there's the clinical trap. The dose-masking thing.
The dose-masking paradox. If you raise the antipsychotic dose, you temporarily hide the TD. But you're also perpetuating the exact mechanism that's causing it. So the patient looks better on the surface while the underlying process keeps consolidating. Clinicians fell into this historically, and it's a nasty trap, because the short-term feedback is telling you the opposite of the truth.
And TD often only shows up after withdrawal.
In somewhere between five and sixty-seven percent of patients, depending on the study. You take the drug away, the masking comes off, and there it is.
Which means there's a window.
There appears to be a window of consolidation during which the mechanism is still reversible. Early detection matters enormously, and that's the practical argument for regular AIMS screening. If you catch it while the striatum is still in the process of encoding the program, you have options. If you catch it after, you mostly have symptom suppression.
Can I push on that, though? Because I want to make sure the window is real and not just a nice story we tell ourselves. What does the evidence actually say about catching it early?
It's a fair push. The evidence is indirect but consistent. The old literature reported remission rates of fifty to seventy-five percent when the causative drug was stopped early. The newer literature, where patients have typically been on treatment for years before anyone notices, reports remission rates closer to two percent. That gap is not explained by the drug being different. It's explained by when you catch it.
So the window is inferred from the difference between the early cohorts and the late cohorts.
Inferred, yes. Nobody has run the trial you'd want, where you randomise people to early detection versus usual care and follow them for a decade. That trial doesn't exist and probably never will. But the natural history data point the same direction, and the plasticity model gives you a mechanism for why the window would close. Once the program is consolidated, you're not reversing a chemical state. You're trying to unlearn something the brain has physically encoded.
And the AIMS exam is the tool for catching it early.
The Abnormal Involuntary Movement Scale. Twelve items, seven of them rate specific body regions, and it takes about ten minutes. It's not a sophisticated instrument, but it's the one we have, and the guidelines say to do it at baseline, then every six months, or every three months if there's a risk factor. The problem is adherence. In real-world settings, AIMS screening rates are notoriously low. Some audits put them under thirty percent.
So the window exists but we're often not looking through it.
That's the honest version. The mechanism gives you a window. The screening gives you a way to use it. And the practice frequently doesn't.
That's the mechanism. Now the numbers. How much have the newer drugs actually changed the risk?
The definitive figure is Carbon, Kane, Leucht and Correll in World Psychiatry in 2018. They meta-analyzed fifty-seven head-to-head randomized trials. Annualized TD incidence was six and a half percent on first-generation antipsychotics versus two and a half percent on second-generation antipsychotics. Risk ratio of zero point four seven. Annualized rate ratio of zero point three five. Number needed to treat of twenty.
So roughly a two and a half times reduction.
Roughly. And importantly, that advantage was not driven by high-dose FGA comparators. They controlled for that. The reduction is real and it's attributable to the drug class, not to a dosing artifact.
Within the second-generation drugs, is there variation?
There is. Olanzapine and aripiprazole had a significant advantage over the other non-clozapine second-generation drugs. Olanzapine beat non-clozapine SGAs at a rate ratio of zero point six six, number needed to treat of one hundred.
And in older adults the gap is even wider, isn't it?
Much wider. O'Brien in the International Journal of Geriatric Psychiatry in 2016. Probable TD incidence at one year was twenty-three percent with first-generation drugs versus seven percent with second-generation drugs. And persistent TD at one year on SGAs was only three percent. The FGA incidence climbed to forty-two percent at two years and fifty-seven percent at three years.
Fifty-seven percent.
In older adults on first-generation antipsychotics, at three years, more than half.
That's not a side effect. That's the coin landing the other way.
And yet the reduction is not elimination. Caroff's line is that newer antipsychotics are less likely to cause TD but the risk is still significant. Stegmayer and colleagues in CNS Drugs in 2018 estimated around twenty percent prevalence in patients on atypicals. So we've moved from something like one in four to something like one in five. Better. Not solved.
And there's a rate-versus-count distinction buried in there.
That's the thing I keep coming back to. Caroff warns that the absolute number of people at risk of TD may be growing even as the rate falls, because antipsychotics are prescribed more widely and off-label. So you can have a headline that says the risk per patient is down by half, and simultaneously have more total cases than ever, because the denominator exploded.
The rate went down. The count went up.
The rate went down. The count may well have gone up.
What about treatment? What do we actually have?
VMAT2 inhibitors are the only treatments with high evidence of efficacy. Valbenazine and deutetrabenazine. They deplete presynaptic dopamine, which opposes the hypersensitivity directly. They were the first drugs ever approved specifically for TD, both in 2017.
They suppress rather than reverse.
They suppress. They do not reverse the underlying mechanism. Symptoms return after withdrawal, though not worse than baseline. Roughly forty to sixty percent of patients show clinically meaningful improvement. Complete resolution remains uncommon. Number needed to treat for an AIMS response is about five for valbenazine, seven for deutetrabenazine.
That's a strong NNT, to be fair.
It's a strong NNT for symptom control. It's not a cure. Switching to clozapine has moderate evidence and is recommended by most guidelines, because clozapine's low D2 affinity may allow natural improvement over time.
The reversibility numbers, if you just stop the drug?
Poor. In one study of a hundred and eight patients who stopped the causative agent, only thirteen percent achieved complete resolution, and no clinical characteristic predicted who would recover. Caroff notes newer surveys suggest as few as two percent resolve after drug withdrawal, versus older reports of fifty to seventy-five percent remission if caught early.
The old numbers were optimistic because they were catching the reversible window.
That's the reading. The early catches resolved. The late catches didn't. And because the field got better at keeping people on treatment, more of what we see now is the late kind.
What about mortality?
TD is associated with higher mortality. A six-year Asian cohort showed mortality rising with TD severity, with pneumonia and sepsis as common causes. That's the respiratory and swallowing involvement showing up as an outcome. Plus worse quality of life, social withdrawal, cognitive decline.
Let me bring in the antiemetic thread, because this is where Daniel's original episode connects. Metoclopramide.
Metoclopramide is a dopamine D2 antagonist used as an antiemetic and a gastroprokinetic. The FDA issued a black box warning in 2009 for long-term or high-dose use because of TD risk. Older estimates put the risk at one to ten percent. Newer data are much lower. Al-Saffar and colleagues in 2019 estimated around zero point one percent per thousand patient-years. McCallum and colleagues this year found zero point three seven percent in treated gastroparesis patients between 2011 and 2020, against historical guideline estimates of one to fifteen percent.
The real-world risk is far lower than the label implies.
Far lower for typical use. But the high-risk groups are the ones you'd expect. Elderly women, diabetics, liver or kidney failure, and anyone on concurrent antipsychotics. And there's a signal in the FDA's adverse event database. Between 2013 and 2023, TD was the single most-reported adverse event for metoclopramide. Three hundred and ninety-three of one thousand and eighty-five reports. Thirty-six percent.
The most-reported event for the drug.
The most-reported event.
The everyday nausea drug and the antipsychotic are running the same playbook. Same receptor, same warning, same outcome.
Same mechanism. Different dose, different duration, different population, same destination.
Which brings us to the honest trade-off, and I want to be careful here, because Daniel was careful. Untreated schizophrenia carries enormous mortality and suffering. The pre-antipsychotic era was not a gentler time. It was warehouses and restraints.
It was. And the mortality data on untreated psychosis are grim. Excess deaths from suicide, from cardiovascular disease, from accidents, from untreated medical comorbidity. The drugs work. They work well enough that we accept a serious risk of a permanent movement disorder as the price.
But the price is real, and it's paid by specific people. That's the tension. It's not a debate where one side is wrong.
The honest framing is much better, not solved. The rate is down roughly two and a half times. The mechanism is still only partially understood. There is no cure. And the count may be rising even as the rate falls, because we're prescribing these drugs more widely.
That's the shape of it. Much better, not solved.
Hilbert: The one I remember was in her sixties. Night shift, state hospital, long-term ward. She'd been on the ward longer than most of the staff. And she had it, the whole thing, the lip smacking, the tongue, the jaw. We called it the chewing. That was the word on the floor. The chewing. You'd chart it, you'd note it, you'd move on.
The chewing.
Hilbert: That was the word. And I want to say one thing about how you two are talking about it, because I think it matters. You keep saying side effect. That's not what it is. A side effect is a rash. A side effect is something that happens alongside the thing and goes away when the thing goes away. What she had was a second condition. She came in with one illness and she left with two, and the second one was the one she couldn't put down. The first one, the drugs helped. That's real. I'm not arguing with you about the drugs. I'm saying the word is wrong. Side effect makes it sound small, and it made it invisible. We called it the chewing and we wrote it on a chart and nobody said the word for what it was.
You're right that the naming does work. Once you've got a nickname for it, it stops being a diagnosis and becomes a feature of the ward.
Hilbert: It becomes furniture. Anyway. You're right about the newer drugs. I've seen both eras. The old ones, you'd see it all the time. The newer ones, less. That part's true.
The ones who got it early, did they get better?
Hilbert: Some. The young ones sometimes. The ones who'd been on it for years, no. It just stayed. I've got to go, I need to pick something up from a place that closes at six and I'm not going to make it if I stand here.
The image of the chewing is going to stay with me. Let's land this properly, because the science isn't finished. The pathophysiology of the irreversibility is still described in the literature as poorly understood, remains elusive. Caroff, Takeuchi, the current StatPearls chapter, all of them say the same thing. The maladaptive synaptic plasticity hypothesis is explicitly a hypothesis. It's the best explanation we have and it is not established fact.
The long-term SGA advantage isn't fully settled either.
O'Brien notes that evidence is lacking on the longer-term risk of TD with second-generation drugs. The 2018 Carbon -analysis is the strongest evidence for a real reduction, but it's built on randomized trials that may under-capture the long-latency cases. The ones that show up at year eight, not year one.
No cure exists. VMAT2 inhibitors suppress, they don't reverse. There's no drug that undoes the encoded program.
There isn't. So the honest framing stays where we put it. Much better, not solved. The rate is down, the count may be up, and the mechanism is still a hypothesis.
If you found this useful, a review helps other people find the show. That's it for today. This has been My Weird Prompts, the human-AI collaboration podcast. Thanks as always to our producer, Hilbert Flumingtop.
We'll be back soon.
See you then.
