Nine point four millimeters. That's the gap between what a palatal expander claims to move and what it actually moves in adult basal bone.
Roughly half. Skeletal expansion is about forty-nine percent of the total in the Zeng meta-analysis — the rest is alveolar and dental, and some of it relapses.
And that gap is why Daniel's asking. He wants the field update on Emet Schneiderman's world — not the man again, we've done him — but what's moved since: MARPE in adults, whether anyone is testing it for sleep apnea, what tirzepatide did to the oral appliance business, and whether AI can read a suture off a CBCT yet. Briefing for a researcher, he says.
So the whole picture is a field being pulled two ways at once. On one side, more aggressive skeletal intervention — getting miniscrews into the palate to force the suture open in a thirty-year-old. On the other, a drug that shrinks the patient instead of widening the jaw. And both landed in the last few years, and they're aimed at the same night-time problem.
One moves the bone. One moves the person.
That's the frame, and dental sleep medicine now sits across all three of its old disciplines at once — craniofacial anatomy, sleep physiology, and pharmacology. The evidence is uneven across every thread, though, and that's the useful part. Solid in one place, a trial registration in another, and a vendor claim in the third. Worth being strict about which is which.
So where does MARPE actually stand?
Efficacy is solid. Evidence quality is not. The Zeng -analysis from twenty-three found a mean success rate of ninety-three point eight seven percent across twelve studies — and graded the whole body of evidence "very low." Only one of the twelve was prospective, and seven were at serious risk of bias.
Ninety-four percent sounds like a treatment. "Very low" sounds like a hunch with good numbers.
Both are true, and that's the tension. Nobody disputes the suture opens. What's weak is the long-term follow-up, the consistency of how success is defined, and the fact that most of the data come from retrospective cohorts of patients who did well enough to stay in the study.
And the skeletal share is the part people skip. Forty-nine percent basal bone — the rest is teeth tipping and alveolar bending. That's not nothing, but it is not what the brochure says.
Nor is it what the relapse data say you keep. There's some give-back over time, and the dental component is the part that gives back first. Which is exactly the question Schneiderman has been asking for years — how much of expansion is actually skeletal — and the answer keeps coming back lower than the field would like.
Age matters more than anything else here. Where's the line?
The twenty-four scoping review, seventy-five studies, was clear: highly effective up to age twenty-five, and it used the phrase "should be considered a standard procedure." Below that, you can generally get the suture to open. After thirty, success drops off in a way you can see in the data — the review from this year confirms it. Between twenty-five and thirty is the grey zone, and that's where imaging to stage the suture starts doing real work.
So it's a procedure with a shelf life.
In the biological sense, yes. And that's why the variants keep appearing. MISMARPE — minimally invasive surgical assistance alongside the miniscrews — reported ninety-six percent success in twenty-five cases. MASPE, the slow expansion protocol, got eighty-six percent skeletal success. Both are small numbers, both promising, and both are really saying the same thing: the field is still tuning how hard and how fast to push.
Twenty-five cases is a poster, not a proof.
It's a case series, and it reads like one. The complication story is the part I'd put in front of a cautious reviewer. A twenty-five case series in Progress in Orthodontics documented craniofacial fractures — zygomatic, orbital, frontonasal — in patients undergoing MARPE and MASPE. Not teeth. Not the palate. The bones of the midface.
That stops a conversation.
It should. Those are fractures you get from a high-velocity impact, and here they're associated with expansion activation. It's rare, the numbers are small, but the pattern points at rapid activation protocols as the culprit. If you're pushing the suture hard enough to open it in a resistant adult, you are also pushing at the buttresses holding the face together.
So the caution is: go slower, and accept that the success rate drops.
Or go slower and use the surgical assist. That's the trade-off the field is working out right now, and it hasn't settled.
Now the sleep apnea question — is anyone actually testing it?
This is the exciting part. The prior evidence has been decent for a while: a twenty-one multi-center prospective controlled trial showed MARPE improves adult OSA and quality of life. A twenty-five randomized trial from Aras and colleagues found MARPE produced significantly greater nasal airflow and reduced nasal resistance compared to conventional RPE — wider nasal, maxillary, and alveolar gains. That's physiological mechanism, and it's real.
But it's not an OSA outcome trial.
No. And the field's own systematic reviews call the OSA evidence low-quality — mostly retrospective, mostly small, mostly looking at surrogate measures like nasal resistance rather than apnea-hypopnea index. That's why the trial registration I've been watching matters.
The Malaya one.
NCT07734688. University of Malaya, twenty-four participants, ages sixteen to sixty, MARPE versus standard OSA care, with home sleep test outcomes. Registered late July, start date August this year, completion next February. It's the first trial specifically designed to test MARPE as a sleep apnea treatment.
Twenty-four patients is small.
Enormously small. But it's the right question asked the right way — a proper comparator, actual sleep outcomes, and a registered protocol. If it shows anything, it will change the conversation. If it doesn't, at least we'll have an answer instead of an inference.
I'd like to see the sample size after the first interim analysis.
I'd like to see the retention. Twenty-four people through six months of palatal expansion and home sleep testing is not a guarantee. But that's the trial to watch — it's the one that turns MARPE from an orthodontic procedure with a side effect to a therapeutic option, if the data come through.
And if they don't, we go back to where we were.
Which is where most of the evidence has been sitting for a while. Now — the other end of the treatment spectrum. A drug that just got approved for sleep apnea.
Tirzepatide. Zepbound.
December twenty-four, FDA approved it as the first drug for moderate-to-severe OSA in adults with obesity. That's a genuine paradigm shift — for the first time, there's a pharmacological treatment for a disease that has been entirely mechanical and device-based. CPAP, oral appliances, surgery, positional therapy. Now a weekly injection.
What did SURMOUNT-OSA actually show?
The numbers are striking. In the group not on PAP, Zepbound cut breathing disruptions by twenty-five an hour versus five on placebo. Twenty-five to five. On top of PAP, twenty-nine versus six. Forty-two percent on Zepbound and fifty percent on Zepbound-plus-PAP reached either remission or mild non-symptomatic OSA — versus sixteen and fourteen percent on placebo.
Forty-two percent off all treatment entirely.
That's the number Lilly led with, and it's not wrong. Weight loss averaged forty-five pounds — about eighteen percent of body weight — and fifty pounds on PAP, versus two percent on placebo. So it's a big effect and it's real. But it's also a weight-loss drug, and that's where the dental story starts.
Because the dental profession doesn't prescribe it.
Right. A piece in the Journal of the American Dental Association last year framed this exactly: dentists don't prescribe tirzepatide, but they're the ones doing the screening, making the referral, and fitting the oral appliances. The drug changes what happens at the next visit. If a patient comes in for a mandibular advancement device fitting and they're down thirty pounds, the calculus changes — maybe they need the appliance less, maybe they need it refit, maybe they need it entirely.
What's the hypothesis about appliances and GLP-1s — complementary or competitive?
A twenty-five review argued GLP-1 receptor agonists may enhance traditional OSA treatments rather than replace them — the phrase used was "a paradigm shift toward more comprehensive management." That's the nice version. The honest version is nobody knows, because there's no head-to-head trial. Tirzepatide versus mandibular advancement device, with the same endpoint, has not been run.
The dental field is reading the tea leaves.
It is. And the practitioners I've spoken to on this say the same thing: the patients who do well on a MAD are the ones who wear it, and the patients who do well on tirzepatide are the ones who tolerate it and keep injecting. It's the same adherence problem dressed differently.
The adherence axis is where the competition actually lives. A weekly shot versus a device you have to wear every night for the rest of your life.
That's exactly the fight. Now — what does the guidance actually say about MADs?
The AASM/AADSM joint guideline. Same one as always.
The twenty-fifteen document is still the foundational piece: oral appliance therapy is recommended — standard recommendation — for OSA patients intolerant of CPAP or who prefer an alternative, and for primary snoring. Custom titratable devices preferred over non-custom. Follow-up sleep testing suggested. AADSM updated their broader standards in twenty-five, but the joint guideline is still the reference.
And the evidence since then?
A twenty-six narrative review reaffirmed the finding that MADs consistently reduce AHI relative to placebo and produce symptom relief comparable to CPAP in mild-to-moderate OSA — and crucially, "largely due to superior adherence." That's the paradox: on the raw apnea-hypopnea index, CPAP is better. On real-world outcomes — how people actually feel and function — the oral appliance often wins, because it's in their mouth and CPAP is in the closet.
The better treatment is the one that gets used.
It's the oldest finding in sleep medicine and it still surprises people. And then there's the quality control issue — a twenty-five paper raised the concern that most appliances marketed as "custom" don't actually meet the guideline's definition of custom. Which is a problem, because the guideline is recommending custom and the market is supplying something else.
It's a word that has been commercialized harder than it has been defined. Now — the last thread, and this is the one that's still early. AI and the CBCT.
Suture staging first.
A twenty-five paper in Scientific Reports described an end-to-end deep-learning framework for classifying midpalatal suture maturation stages from CBCT — the Angelieri classification, running A through E. A twenty-six unified framework in the same journal tried to stage CVM, SOS, and midpalatal suture together. Single-phase accuracy: sixty-four percent. Two-phase, where it has both stages to work from: about seventy-eight.
Seventy-eight percent is a lot better than sixty-four. But neither is a clinical number.
That's the point. Seventy-eight percent for a decision that affects whether you put miniscrews in a child's palate is not where you want to be unsupervised. It's useful as a second read, useful as a triage tool — "this one needs a specialist look" — but not as a primary determinant.
The cephalogram work is more interesting to me. Predicting suture maturity from a plain X-ray instead of a CBCT is a radiation win if you can trust it.
A twenty-six paper in Progress in Orthodontics did exactly that — used deep learning to estimate CBCT-defined midpalatal suture maturity from a standard cephalogram, and reported AUC between point-nine-one and point-nine-seven. That's strong. That is close to a usable number.
Close.
AUC of point-nine is good discrimination, but there's a difference between discrimination and calibration, and there's a difference between the retrospective validation they ran and a prospective clinical test. I'd want to see the latter before I let anyone skip the CBCT. But the direction is right — the whole field wants to reduce imaging, and this is a way to reduce it.
Airway segmentation is further along.
Further, yes. Multiple convolutional and U-Net models are now segmenting the pharyngeal airway from CBCT — Diagnocat published one, there's a nnU-Net paper in the Journal of Dentistry this year. A systematic review in PeerJ last year covered the whole space of AI for CBCT airway assessment in sleep-disordered breathing. The segmentation is getting good — the accuracy of the outlines is high.
But segmentation is measurement, not diagnosis.
Telling me the volume of the pharyngeal airway in cubic centimeters is a solved-ish problem. Telling me what that volume means for the patient's apnea-hypopnea index, their arousal threshold, their response to a given therapy — that's a different model entirely, and it's barely started. The segmentation gets you a number. The number is not the diagnosis.
So where's the evidence strong and where's it early?
MARPE efficacy: strong, though the evidence quality is weak. Tirzepatide's OSA effect: strong, large randomized controlled data. MADs reducing AHI: strong, consistent. MARPE as an OSA therapy: early. AI suture staging: early. AI airway measurement: measurement is solid, clinical translation is early.
And the open questions for someone like Emet.
The Malaya trial, when it reports. Whether GLP-1-induced weight loss reduces appliance need or complements it. Whether AI can get suture staging to the point where it's reliable enough for a clinical call. And the real-world complication rate for MARPE — because what's published is what gets published, and the fractures in that case series are not a thing anyone wants to undercount.
So it's a field with two very good drugs and procedures, and an evidence base that's playing catch-up.
That's the shape of it. The interventions are ahead of the trials, and the imaging tools are ahead of the validation. Which is not a criticism — it's what a fast-moving field looks like.
Attention, residents of the tri-county area. The Department of Public Health reminds you that the midpalatal suture closes once, and it closes on its own schedule. Do not rely on folklore. Do not rely on a neighbor's experience. If your transverse dimension is compromised, seek evaluation from a licensed professional before the window closes. The Midline Skeletal Advisory Board urges all adults over the age of thirty to consult a craniofacial specialist, because time is not on your side, and neither is gravity. Only one product is endorsed for responsible palatal maintenance in this region: the Palatal Guardian Pro, a device worn nightly, endorsed by no one in particular but available at participating retailers. Ask about the warranty.
That started as a public service announcement.
The success rate claim is not right — MARPE success drops after thirty, it doesn't become urgent before then. That's backwards.
Herman. Don't.
Right. Back to the trial.
So if a researcher is going to spend the next year on one of these, which one?
The MARPE-for-OSA question, if I had to pick. It's the one with the biggest clinical implication, the one where a positive result would open a new treatment category, and the one where we don't know. The drug story is largely written — tirzepatide works and the guidelines will catch up. The AI story will take years to mature. But the MARPE question is answerable in a couple of trials and could change practice quickly.
The thing about palatal expanders is the screw. That's the whole issue. Four turns a day, a quarter millimeter each, and everyone in the trade argues about the rate. My orthodontist — a man called Freely, at the Arcade Dental Group, forty-one years at the same address — he turned it twice a day. Said the bone knows. He's the only one in that plaza still using a manual key. The others all went to the hydraulics. Hydraulics are a mistake. I told them so. They don't listen.
Twice a day is on the slow end of the protocols.
The MASPE data would actually support slow activation — eighty-six percent skeletal with slow expansion. He might have been ahead of the literature.
Freely doesn't read the literature. He reads the warranty cards that come with the hardware. He's got a filing cabinet. The expanders themselves — he uses a brand, I don't name brands on air — but it's got a screw head shaped like a flat blade, not a hex. Everyone else switched to hex in twenty-fourteen. Freely kept the blade. Says the hex strips. He's right about that, incidentally. I've seen the hexes strip.
What did you see him use it for that the data wouldn't have covered?
A man came in with a midpalatal suture that was, in Freely's words, "like rebar in concrete." Turned the expander for nine months. Nine months, Herman. The man's wife left him over it. Not over the expander — over the snoring that came after. He was sleeping with his mouth open. Anyway, Freely documented the whole case, took CBCTs every six weeks. This was before anyone was using AI to read them — he was using a man called Vespucci who sat in the back room and pointed at things. Vespucci could call the suture stage on a scan in about four seconds. No model I've seen matches that. He was eighty percent accurate by his own admission and he was right more often than that.
Four seconds for a staging call is impressive but I wouldn't put it in a validation study.
Vespucci wouldn't either. He did it for cash. Anyway — the business with Freely. He's still turning screws. He called me last Tuesday about a warranty dispute. The expander's manufacturer — I'm not naming them — they changed the screw thread pitch in twenty-twenty and stopped honoring the old replacement parts. Freely has forty-one devices in a drawer that are now, technically, unrepairable. He wants me to represent him. I've agreed. We filed. The hearing is in March.
You're litigating over palatal expander parts.
It's a real dispute. The manufacturer claims the thread pitch change was a safety improvement. Freely says it was to force the clinics to buy the new model. He's probably right. He's always right about hardware. He's wrong about the two turns a day, though — that's too slow, and he knows it, but he'll never admit it. Man's seventy-eight and he's still taking new patients three days a week. The other two days he does the repairs. He's got a lathe. He re-cuts the screws himself when they strip.
That's a whole repair ecosystem.
It's what happens when the manufacturer stops supporting the older parts — someone in the field picks it up. That's a story in every device-dependent specialty.
The thing is, I asked him about the AI. He has a CBCT in the back — a twenty-sixteen model, one of the ones with the rotating arm that squeaks. I asked him whether he'd use a model to stage the suture if one worked. He said he'd use it to check Vespucci. That's all he'd use it for. He doesn't trust anything he can't argue with.
That's not a bad instinct for a clinical call.
Second reader, not first. The cephalogram work supports that — the AUC is high enough to be useful as a check.
He also said something about tirzepatide that I'll pass on since you mentioned it. Half his patients are on it now. He said the appliance fittings are all wrong because the jaws are moving — not the palate, the fat pads. He's refitting more devices in a year than he used to in five. He thinks the whole thing is going to end in a recall. He said it with some satisfaction.
The mandibular advancement device fit changing with major weight loss is a real clinical question — there's not much written on it. Weight loss of eighteen percent changes pharyngeal tissue, changes mandibular posture, and the device that fit last year may not hold the jaw where it should. It's a fair concern.
The manufacturer is going to love hearing that the drug they didn't make is breaking the device they did.
Freely's already written to them. I told him not to expect an answer. He sends the letters anyway. He's got a stack. He reads them at the counter while people wait.
So there's a dental practitioner out there tracking appliance refits as a proxy for how many of his patients are on GLP-1s. That's a data set nobody's collecting.
Freely's collecting it. He's got it on index cards. He won't digitize them. He says the cards don't crash.
Index cards as a data store. There's a paper in that if someone could get at it.
A four-second suture read and a shoebox of refit records. The most valuable database in dental sleep medicine is in the back of a shop in a plaza.
He'd deny it was valuable. He'd say he just writes down what happens. Which is also true.
The question for the field is what the refit rate looks like across practices, not just one.
And the answer would tell you something real about the tirzepatide-appliance interaction that no trial has measured yet.
Freely says the answer is forty percent. He says it like it's obvious. I'd want to see the denominator before I quoted that number.
Cutting-room floor item. A twenty-fourteen CBCT study from the primate literature — rhesus macaques, midpalatal suture fusion timing — showed fusion happening in a pattern that tracked with dental eruption rather than with chronological age. It's the reason Schneiderman went from monkey jaws to human sleep medicine in the first place, and it's the reason age is a weak proxy for suture maturity. The AI work is chasing the same signal from images instead of dissection.
That's the thread that connects the monkey work to the imaging models. Same question, different tool.
Daniel's watching the Malaya trial, and so should anyone with an interest in whether this field treats bone or treats air. When that reports, we'll know a lot more than we do now.
My weird prompts dot com, or the Telegram channel, or the review tab in your podcast app. Take your pick.
To Hilbert Flumingtop, our producer. If you want more of this, try episode nine seventy-six, From Monkey Jaws to Sleep Apnea; episode four, If Your Voice Ages, Does Your Fine-Tune Become Useless; and episode twenty-nine oh six, How Much Bone Do You Actually Get From Palatal Expansion. This has been My Weird Prompts. See you tomorrow.
Tomorrow.
