Sleep Apnea and the Nervous System

Sleep apnea begins in the one stretch of airway with no rigid support

Air passes the nose and the voice box inside walls of bone and cartilage. Between them lies the pharynx, whose wall has neither, and John Remmers showed in 1978 that it shuts in the moment its muscle goes quiet.

The nose has bone in its walls. The voice box and the windpipe below it are ringed with cartilage, which is why you can feel hard bands under the skin of your throat. Along the stretch of pharynx that matters here there is no bone and no cartilage anywhere in the wall. It is a soft tube. It has to be, because the same passage carries food when you swallow and shapes sound when you speak, and a rigid pipe can do neither.

A soft tube stays open only if something holds it open. In the pharynx that something is muscle. The largest single contributor is the genioglossus, a fan of fibers running from the inside of the chin back into the body of the tongue. When it fires, the tongue is pulled forward and the space behind it widens. It is the reason there is a space behind your tongue at all.

The tongue muscle goes quiet as sleep deepens

In 1976 two neuroscientists in the United States wanted to know what that muscle actually does while a person sleeps. Nobody had listened. They placed fine wire electrodes into the genioglossus of sleeping volunteers and recorded it through the night. Awake, the muscle pulsed with every breath in, tightening a fraction before the diaphragm pulled. As sleep deepened the pulses shrank. In the dreaming stage the muscle went nearly silent.

That is normal. Every human airway narrows at sleep onset, and for most people the narrowing costs nothing.

Two years later the respiratory physiologist John Remmers asked the next question with his colleagues. Why does an ordinary narrowing become a closure in some people? They recorded the tongue muscle and the pressure inside the chest at the same moment, in patients whose breathing stopped at night. The throat shut when the tongue muscle was quiet while the diaphragm was still pulling hard. It opened again when the tongue muscle fired strongly.

Obstruction turned out to be a balance of forces. Suction pulling the tube shut against muscle holding it open, settled breath by breath, all night.

Tone here is literal. The airway is open because muscle tone holds it open, and it closes when that tone falls below what the suction demands.

Structural findings are real and they set the range

Some people have a genuinely small or crowded throat. A recessed jaw, large tonsils, a long soft palate, a thick neck and fat deposited in the tissues around the airway are real structural findings with real consequences. Structural causes are treated on their own terms and nothing here displaces them.

What the Unified Model of Tone claims sits alongside that. Every disease has a tonal expression, and many are initiated, maintained or amplified by failures of tonal regulation. Sleep apnea is that claim in its plainest instance. Anatomy sets the range everything below works inside. Regulation decides where inside that range a person lands.

Sleep apnea is common, mostly undiagnosed, and dangerous at the wheel

In a random sample of 602 employed adults in Wisconsin, 24 percent of the men and 9 percent of the women met the threshold for disordered breathing. Almost none of them knew.

That 1993 study recruited its participants at random rather than from a clinic, and every one of them got a full overnight recording. About 2 percent of women and 4 percent of men had the events together with the daytime sleepiness that defines the syndrome. That was one American state, employed adults, and 1993 scoring rules, so the exact figures have moved since.

The global estimate is larger and much softer. A 2019 analysis put roughly 936 million adults aged 30 to 69 at mild obstructive sleep apnea or worse, and 425 million at moderate to severe. Reliable primary data existed for only 16 countries. Everything else was matched by proxy, and the analysis was funded by a manufacturer of sleep apnea devices. Read it as an order of magnitude.

Untreated sleep apnea multiplies the odds of a highway crash

In Spain, researchers recruited 102 drivers treated in emergency departments after highway crashes. They added 152 age-matched and sex-matched controls from primary care and sent all of them for sleep testing. Drivers with ten or more events an hour had 6.3 times the odds of having crashed. The association held after the analysis controlled for alcohol, eyesight, body mass index, driving experience, previous crashes, sedating medication and sleep schedule. Among those who had apnea, drinking alcohol that day raised the risk further.

Falling asleep at the wheel, gasping or choking awake, or a partner who reports that you stop breathing are reasons to see a physician now, not after more reading.

The same goes for morning headaches with heavy daytime sleepiness, and for apnea alongside blood pressure that will not come down or a heart that is failing. Diagnosis is never made from a symptom list. It requires a sleep study a physician orders, in a laboratory or at home, and there is no substitute for it. Nothing on this page is a reason to reduce, stop or change any prescribed therapy.

The throat holds all day because the brain drives the muscle harder

Patients with sleep apnea ran the genioglossus at 40.6 percent of maximum while sitting awake, against 12.7 percent in matched controls. Put them on positive pressure and the extra effort switched off.

An obvious question sat unasked for years. If these patients have a narrow, floppy throat, why can they breathe perfectly well all day?

A group in Denver, working with the respiratory physiologist David White, went and measured it. They recorded tongue muscle activity during wakefulness in eleven patients with apnea and fourteen matched controls. Each recording was expressed as a percentage of that person's own maximum effort, so that a large muscle and a small one could be compared honestly. The patients ran at 40.6 percent of maximum and the controls at 12.7 percent.

The awake brain was driving the muscle more than three times harder to keep the same tube open. Holding the airway open mechanically abolished the extra drive, so it was a response to load rather than a habit.

Read through the model, the shape of the disease changes. By day these patients still carry the disorder. They are paying for it, continuously, with a regulator recruited to do a job the anatomy no longer does by itself. The diagnosis arrives on the night the payment stops.

Collapsibility can be measured in centimeters of water

Then the field learned to put a number on the floppiness itself. Any collapsible tube with air moving through it stays open until the pressure inside falls below some critical value, and then it shuts. Suck hard enough on a paper straw and it flattens against itself. The pressure at which that happens is a property of the straw.

In 1988 Alan Schwartz and his colleagues at Johns Hopkins applied gentle suction to the noses of seven sleeping volunteers and watched airflow fall in step with the pressure they applied. Below a certain value the throat closed completely and the sleeper woke. In healthy people that critical closing pressure sat around minus 13 centimeters of water. You would have to suck hard to shut a normal throat.

Three years later the same laboratory measured that value in three groups: people who merely snored, people with partial obstruction, and people with full apneas. The numbers ran minus 6.5, then minus 1.6, then plus 2.5 centimeters of water. A straight line.

Snoring and full apnea sit on the same axis, read at different points along it.

Once that critical pressure reaches zero or climbs above it, the throat wants to close at ordinary atmospheric pressure. Which is to say, all by itself, with no suction required. This is the constraint everything else sits on, and it is measurable in centimeters of water. What it does not tell you is which of the people sitting at any given value will actually stop breathing.

Sleep medicine already measures four traits behind sleep apnea

Danny Eckert measured all four in 75 men and women across three nights each, and roughly half of patients carried a substantial non-anatomical trait.

By 2013 the field could measure four separate things in a sleeping person. Nobody had measured all four in the same one. Eckert, a sleep physiologist, did it with his colleagues at Brigham and Women's Hospital in Boston, with each trait isolated and measured one at a time.

36 percent of patients had almost no genioglossus response during sleep. In another 37 percent the arousal threshold sat too low. High loop gain turned up in 36 percent. More than one non-anatomical feature was present in 28 percent. And 19 percent had an airway no more collapsible than many healthy people, in whom loop gain ran almost twice as high.

On a weighting scheme the authors proposed, non-anatomical features played an important part in 56 percent of patients. That is the figure usually quoted as roughly half, and it should be quoted with its provenance attached. It is an estimate from a proposed three-point scale rather than a directly measured quantity. The component percentages above are the measured ones, and they carry the same message without the scaffolding.

In more than half of patients with sleep apnea, something other than the shape of the throat is doing real work.

The same traits can be read out of an ordinary sleep study

Three nights of invasive measurement is useless outside a research laboratory. The sleep researcher Scott Sands and colleagues asked whether the same traits were already hiding inside an ordinary sleep study. They built an automated method that reads collapsibility and muscle compensation out of the breath-by-breath rise and fall of airflow.

In 29 patients the estimates tracked the invasive gold standard closely. Carry the limit with the result. Accuracy for sorting patients into high and low groups ran between 69 and 86 percent, so misclassification is common.

Same severity, different machinery underneath

The dials are set differently in different people. Ten younger and ten older patients, matched for body size and sex, had all four measured. The older group had a floppier airway. The younger group had a far more sensitive controller. Same diagnosis, same severity band, different machinery underneath, in twenty people.

Four traits, each of which can sit high or low independently of the others. Respiratory medicine arrived at them empirically and named them in its own vocabulary. Collapsibility is constraint. Twitchiness is gain. The arousal threshold is a set point. Muscle responsiveness is the reflex that ties the three together, and it gets a section of its own later on.

A breathing controller that overcorrects can close a throat

Magdy Younes amplified the breathing control loop in patients whose airways were already held open, and nine of twelve with severe sleep apnea broke into periodic breathing.

A thermostat measures the room, compares it with a target, and turns the heat on or off. A well-tuned one holds the room steady and you never think about it. A badly tuned one overshoots.

It blasts heat until the room is too warm, shuts off completely until the room is too cold, then blasts again. Nothing in it is broken. The sensor works and the heater works. The relationship between them is wrong, and the room now swings between extremes it would never have reached on its own.

Engineers call the strength of that relationship loop gain. Gain above one means the correction is larger than the disturbance that provoked it, and a system like that will oscillate on its own, indefinitely, with nothing further done to it.

Your breathing runs on exactly this architecture. Sensors read the carbon dioxide in your blood, compare it with a target, and set how hard you breathe. High gain means a small disturbance draws a large correction. You over-breathe, blow off more carbon dioxide than you should, and the drive to breathe then falls below the level that keeps breathing going at all. The muscles of the airway lose their command at the same instant, because the same respiratory rhythm drives them.

The throat then closes, at whatever floppiness it has, because the instruction to hold it open has momentarily stopped arriving.

Where in the population the controller decides the outcome

Younes, a respiratory physiologist, set out to separate the tube from the thermostat. He held the airways of 32 patients open with positive pressure, so that anything unstable that remained had to be coming from the controller.

Then he used a ventilator to amplify the loop step by step until periodic breathing appeared. Nine of twelve patients with severe apnea broke into cycles. Only six of twenty with milder apnea did, and that was despite being pushed harder. The controller itself was less stable in the severe group.

Andrew Wellman and his colleagues at Brigham then asked where in the population the controller actually decides anything. They sorted 25 patients by critical closing pressure and measured loop gain in each. In the middle group, whose closing pressure sat near atmospheric, loop gain and apnea severity tracked each other almost perfectly. At either extreme, nothing. If the airway is stiff enough, no controller can cause apnea. If the airway is hopeless, no controller can prevent it. In between, the controller decides.

That single experiment carries the composition of sleep apnea, and it rests on 25 people split three ways, so the correlation is held up by a handful of them. Constraint sets the range. Gain decides where inside that range a person lands.

Acetazolamide moves loop gain and leaves the other traits alone

Bradley Edwards, a sleep researcher, and colleagues gave thirteen patients acetazolamide, a drug that damps the chemical control of breathing, for a week, then remeasured all four traits. Loop gain fell from 3.4 to 2.0. Collapsibility did not change. Arousal threshold did not change. Muscle responsiveness did not change. Apnea severity roughly halved. Thirteen people, one week, and the correlation between the fall in gain and the fall in severity was modest.

Move one trait and the others stay where they were. These are separately perturbable axes of one organization rather than one dial wearing four labels. A drug aimed at the controller reaches the controller and nothing else, which is why it matters where each axis sits in a particular person.

Waking easily is part of the cause of sleep apnea

Magdy Younes lowered the pressure on 82 patients and watched the airway reopen before the awakening, or with no awakening at all. The deeper the sleep, the more often it opened quietly.

Everyone assumed the brief awakening at the end of an apnea was the rescue. The throat shuts, oxygen falls, the brain wakes you just enough to open the airway, and you drop back into sleep. The awakenings were the price of survival, unpleasant and necessary.

Younes tested the belief directly, breath by breath, watching what actually reopened the airway. It frequently opened before the awakening, or with no awakening at all. And when an arousal did occur, the following breath overshot far more, at 267 percent of the initial flow decline against 180 percent without one. That overshoot is precisely the over-breathing that drives the controller into its next oscillation.

His conclusion was that arousals are incidental. They are not required to open the airway, and they probably make the disorder worse by feeding ventilatory instability.

Events that end quietly are less likely to repeat

A natural experiment supports it. Twenty to thirty percent of respiratory events end with no visible awakening on the brain trace at all, which gives two kinds of ending to compare. Sixteen patients slept with electrodes in the tongue and palate muscles, in a study led by the sleep researcher Amy Jordan.

Events that ended with an arousal produced a bigger ventilatory overshoot, though that difference did not survive once the severity of the event itself was accounted for. Those events were also more often followed by another event. Events that ended quietly were less likely to repeat. Muscle activity after the quiet endings was not low, which the simplest version of this story would have predicted.

The tongue muscle fires fastest in the deepest sleep

Then the mechanism underneath. A Boston group recorded individual motor units inside the genioglossus across the transition into slow-wave sleep, the deepest stage of the night. Twenty-nine people were studied and usable units were isolated in seven of them. Those units fired faster once deep sleep arrived, at 22.7 hertz against 20.3, and they fired for longer. Almost no new units were recruited. The ones already working simply worked harder. Twenty-six motor units from seven sleepers is a thin base for a result this load-bearing.

The airway is most stable in the state a person with a low arousal threshold almost never reaches. Waking easily prevents the very deepening that would have made the waking unnecessary.

This is the set point in its plainest form. The arousal threshold is a value the nervous system defends, and in these patients it is defended too low. The system is succeeding, in the sense that the airway does reopen every single time.

It is succeeding through a strategy so expensive that the strategy consumes the sleep which would have made it unnecessary. The architecture of a night belongs to the sleep page. What belongs here is the cost of never getting far enough into one.

Raising the arousal threshold with a sedative lowers the event rate

Eszopiclone raised the threshold and cut events in seventeen patients, and a survey of three other sedatives found none of the airway effects the standard warning predicts.

A claim earns its keep by making a prediction that can go badly for it. If a low arousal threshold is part of the cause rather than the rescue, then raising the threshold should reduce events. That prediction runs straight into the standard warning that sedatives are dangerous in sleep apnea, on the reasoning that they relax the airway muscles and blunt the response to falling oxygen.

Danny Eckert and colleagues tested it in seventeen patients, three nights each, with a pressure catheter in the throat to measure the effort at which each person woke. On eszopiclone, the arousal threshold rose and the event rate fell. Events did not become longer and the oxygen dips did not become deeper. Among the eight patients who already had a low arousal threshold at baseline, the improvement was largest.

Then a broader physiological survey complicated it. A group at Neuroscience Research Australia in Sydney ran 84 sleep studies across 21 people on temazepam, zolpidem, zopiclone and placebo, measuring the traits each time. Two of the three drugs raised the arousal threshold. None of them reduced tongue muscle activity. None made the airway more collapsible. Zolpidem did the opposite of what the warning predicts and tripled muscle responsiveness during narrowing.

The feared mechanism did not appear, and the effect on the threshold was real but specific to the drug. Both studies are single-night physiological experiments in small groups. Both excluded or lacked patients with severe oxygen desaturation. Neither is remotely a license for anyone to take a sedative.

What the pair establishes is narrower and still worth having. The arousal threshold is a value that can be moved independently of the other three, and moving it moves the disease. That is a statement about machinery. It belongs in a section about machinery rather than in any conversation about prescriptions.

Years of obstruction injure the nerves that report on the airway

Palate biopsies from 21 habitual snorers carried the signature of nerve injury and repair, and the picture worsened with the hours each person spent obstructing.

So far the airway has been treated as a passive object with a certain floppiness. It is living tissue, vibrated and collapsed several hundred times a night, and it answers back.

A group at the Karolinska Institute, working with the ear, nose and throat surgeon Danielle Friberg, wondered whether years of snoring damage the throat itself. They took muscle biopsies from the palate of 21 habitual snorers, ten of whom had apnea, and from ten people who did not snore. Under the microscope the snorers' tissue showed the signature of nerve injury and repair.

Sensation inside the throat is measurably blunted

If the nerves supplying the throat are injured, the throat should feel less. A Montreal team tested exactly that. They measured two kinds of sensation inside the throat of 37 patients with apnea, twelve snorers and fifteen controls. One was the ability to feel a vibration.

The other was two-point discrimination, how far apart two touches have to be before you feel two rather than one. Each person's lip and hand served as comparison sites. Throat sensation was blunted in the snorers and the patients. Lip and hand were normal, so this was local rather than general. After six months on positive pressure the vibration thresholds partly recovered. Two-point discrimination did not.

Both studies are cross-sectional and small, and causation could run either way. The reading that makes sense of them is a loop rather than an arrow. Obstruction injures the nerves that report on the airway. The report degrades. The reflex depending on that report weakens, and the obstruction worsens. A disease that produces its own causes is what the model calls an attractor: a state the system keeps returning to because being in it makes it easier to return.

The overweight sleepers who should have apnea and do not

The strongest evidence that the reflex matters, and not only the tube, comes from people who ought to have sleep apnea and do not. Scott Sands and colleagues compared eighteen overweight people without apnea against 25 matched patients who had it and eleven normal-weight controls. The protected group had airways genuinely more collapsible than the lean controls. Their tongue muscles answered a narrowing three times more strongly than either other group.

The same anatomy, with a live reflex sitting on top of it, produces no disease at all.

That is 54 subjects across three groups, cross-sectional, so it cannot say whether the strong reflex came first or grew in response to the load. Either way the finding stands. Two people with the same throat can have opposite nights, and the quality of one reflex is what separates them. That is input quality deciding an outcome that anatomy appeared to have already settled.

Sympathetic traffic in sleep apnea never comes back down

Ten untreated patients had high sympathetic nerve traffic while sitting awake in the afternoon, and during sleep it did not fall at all. It surged to nearly 300 percent of the waking level at the end of every obstructed breath.

Everything so far has happened inside the throat. The consequences do not stay there.

To follow them you need one technique. Microneurography threads a needle-thin electrode through the skin of the leg and into a nerve, positioning it beside the fibers that command blood vessels to tighten. Once it sits in place you can hear the traffic on the accelerator branch of the nervous system live, burst by burst, in a person who is awake or asleep and otherwise undisturbed.

What a normal night does to that traffic belongs to the sleep page. It falls with depth and surges in dreaming. The cardiologist Virend Somers and colleagues at Iowa put numbers on it in eight healthy volunteers. Sympathetic traffic fell to 41 percent of the waking level in the deepest sleep, and blood pressure fell from 90 to 80 millimeters of mercury. In dreaming sleep the traffic more than doubled, to 215 percent.

Then the same technique in ten patients with obstructive sleep apnea. Their sympathetic traffic was already high while they sat awake in the afternoon. During sleep it did not fall at all. Mean blood pressure peaked at 116 in light sleep and 127 in dreaming sleep. Positive pressure brought the traffic down. Ten patients, and only four were restudied on treatment.

Whatever gets reset at three in the morning is still set that way at three in the afternoon.

The daytime traffic takes six months to fall

The Iowa group followed eleven patients on positive pressure for a year against nine untreated. The untreated group did not change. The treated group's nerve traffic fell, and the timing is the useful part. Not at one month. The change appeared at six months and at a year. Blood pressure and heart rate did not move in either group. Twenty subjects, and not randomized.

Read through the model, that timing is the finding. A value which took years to set does not unset in a night. The splint works on the timescale of a machine and the regulator recalibrates on the timescale of a nervous system, and those are not the same clock.

Only the oxygen sensor is turned up

One more result identifies which sensor carries the gain. Sixteen untreated patients and twelve matched controls were challenged three ways: with low oxygen, with high carbon dioxide, and with a hand held in ice water. Only the low-oxygen challenge separated them.

The patients answered a fall in oxygen with a much larger rise in breathing, heart rate and blood pressure. Carbon dioxide responses were normal. The cold response was normal. This is gain, selectively, at one sensor, and a potentiated oxygen sensor is exactly what raises loop gain.

Intermittent oxygen dips raise daytime blood pressure on their own

Twelve healthy volunteers spent fourteen nights breathing air with the oxygen dipped, and finished 8 millimeters higher systolic by day. No airway obstructed at any point.

Sleep apnea travels with obesity, and obesity raises blood pressure by itself, so for years nobody could say which was doing the damage. Two experiments removed the confound by building the apnea without the person.

A laboratory in Heidelberg pulsed nitrogen into a sleeping chamber every thirty seconds, seven hours a day. It reproduced the oxygen dips of apnea in rats with no airway obstruction anywhere. After five weeks the animals' daytime blood pressure had risen by 13.7 to 21 millimeters of mercury, depending on how it was measured.

Control animals sat in identical chambers breathing compressed air and did not change. It is an animal model, and the dips are far more regular than human apnea. Sleep itself was never recorded, so nothing rules out fragmentation as a contributor.

The human version came later, in Grenoble. After a single night of dipped oxygen, daytime blood pressure in those twelve volunteers was already up by about 3 millimeters. After two weeks it was up 8 systolic and 5 diastolic. Sympathetic traffic had increased, and the baroreflex that normally restrains it had weakened. Twelve people and fourteen nights is a small experiment. It did what no observational study can. It delivered the input and watched the regulator move.

The overnight dip that stops happening

What follows in patients belongs to the blood pressure page, which owns the Wisconsin cohort and its finding that the odds of developing hypertension rise with apnea severity over years.

Blood pressure is supposed to fall by at least a tenth overnight. The pattern is called dipping. In 328 adults followed for 7.2 years, those with more events at baseline were three to four times more likely to lose that dip. The limit is that this held for systolic non-dipping alone. The diastolic association was not statistically significant, and the confidence interval in the severe group is very wide.

Pressure that will not come down on three drugs

When a value refuses to move at all, look at what is holding it there. A university hypertension clinic in Toronto took 41 patients whose pressure would not come down on three or more drugs at full dose. Every one was sent for a sleep study. Obstructive sleep apnea turned up in 83 percent of them, and almost none knew. One referral clinic, no control group. That percentage will not generalize.

A Spanish multicenter trial later randomized 194 patients with resistant hypertension and apnea to positive pressure or to nothing extra, on top of an average of 3.8 drugs each. Twelve weeks later the treated group had dropped 3.1 millimeters of mercury on the 24-hour average.

The proportion displaying a nocturnal dip at twelve weeks was 35.9 percent on positive pressure against 21.6 percent on control. The size of the fall tracked hours of use. It was open-label, twelve weeks long, and the systolic change on its own did not reach significance.

Three millimeters is small next to what a drug does. The dip returning in roughly one patient in seven is a different kind of result. What came back was a rhythm.

Positive pressure abolishes obstruction, and two large trials missed their endpoints

Colin Sullivan abolished obstruction in five patients with 4.5 to 10 centimeters of water in 1981. SAVE later randomized 2,717 patients for 3.7 years and the cardiovascular endpoint did not move.

Sullivan, an Australian respiratory physician, reasoned from the physics. If the throat collapses because the pressure inside it falls too low, then raising the pressure inside it should hold it open. He built a mask, connected it to a blower, and tried it on five patients with severe apnea. The applied pressures abolished the obstruction completely and gave each of them a whole night of sleep.

His own paper calls the device a pneumatic splint, which is the plain description and the one used here. It does not change the airway. It holds the airway.

The sleepiness effect depends entirely on who receives it

It works, and it works most where the burden is largest. A meta-analysis of twelve randomized trials found sleepiness scores falling 2.94 points more than placebo across the 706 patients in whom they were measured. In the trials that recruited severe apnea together with real sleepiness, the fall was 4.75 points. In the rest it was 1.10.

The same treatment, four times the effect, depending entirely on who received it. Sex, age, body mass index and country did not explain the difference. The objective effect was much smaller. On a laboratory test of how fast a person nods off, treated patients took under a minute longer than placebo to fall asleep.

That is input meeting tone, measured inside a single meta-analysis. The therapy did not change between trials. The systems receiving it did.

Dose matters too, and there is no single number. Terri Weaver, a sleep researcher, and colleagues followed 149 patients with severe apnea for three months and fitted dose-response curves to hours of nightly use. Subjective sleepiness normalized at about four hours a night. Objective alertness needed six. Everyday functioning needed seven and a half. Different outcomes have different thresholds, which is worth holding on to for the next paragraph.

What SAVE, ISAACC and APPLES found

The trial meant to settle the cardiovascular question was called SAVE. It randomized 2,717 adults with moderate to severe apnea and established cardiovascular or cerebrovascular disease. One arm got positive pressure plus usual care and the other got usual care alone, and both were followed 3.7 years.

The therapy worked on the sleep study, taking events from 29.0 an hour down to 3.7. Snoring, sleepiness, quality of life and mood all improved significantly. The primary cardiovascular endpoint did not move, at 17.0 percent against 15.4 percent. The treated group's event rate ran at 1.10 times the control rate, and no component of the endpoint favored treatment.

Mean adherence was 3.3 hours a night. That sits below the four hours at which subjective sleepiness normalizes, and at less than half of what everyday functioning needs. Sleepy patients were also largely excluded on ethical grounds, so the trial tested an under-used therapy in people who were not sleepy.

Then the null replicated. ISAACC, in Spain, asked the same question in people who had just had an acute coronary syndrome. Of 2,551 recruited, 1,264 with apnea were randomized, alongside 603 apnea-free patients as a reference arm, and followed for a median of 3.35 years. Event rates were the same in all three arms. Having apnea did not raise risk in this population and treating it did not lower it. Mean adherence was 2.78 hours a night.

Cognition came out null as well. APPLES randomized 1,105 patients for six months, double-blind against a sham device, which is a genuine achievement in a field where the treatment is a mask on your face. Of those, 1,098 contributed to the primary analysis. Three neurocognitive outcomes were specified in advance. One differed at two months and the difference had gone by six. Sleepiness improved clearly, by questionnaire and by objective testing.

None of that is a reason for anyone to stop using a prescribed therapy, and the reading is narrower than the headlines went. Two large trials tested a splint that was in place for roughly a third of the night, in people selected for not being sleepy.

The literature on positive pressure and cardiovascular events is unresolved. What is not in doubt is the sleepiness, the snoring, the quality of life and the mood, every one of which improved, in every trial, including the ones that missed their primary endpoint.

The oral appliance loses on events and ties on outcomes

Positive pressure took events to 4.5 an hour against 11.1 for a jaw-advancing appliance. Sleepiness, driving simulator performance and disease-specific quality of life came out the same on both.

If the number of events were the only thing that mattered, ranking the treatments for sleep apnea would be simple. It is not the only thing that matters.

A sleep medicine group in Sydney had 126 patients use positive pressure for a month and a jaw-advancing dental appliance for a month, in random order. The appliance holds the lower jaw forward, which carries the base of the tongue with it and widens the space behind.

Positive pressure was clearly better at the number, 4.5 an hour against 11.1. The appliance was clearly better at being used, at 6.5 hours a night against 5.2. It also beat positive pressure on four general quality-of-life domains. Appliance use was self-reported while positive pressure use was machine-recorded, which favors the appliance. Neither treatment improved blood pressure in this trial.

The pattern holds at scale. A network meta-analysis pooled 51 randomized trials and 4,888 patients. Positive pressure lowered systolic pressure by 2.5 millimeters of mercury and the dental appliance by 2.1. There was no statistically significant difference between them, despite the appliance abolishing far fewer events. Each extra hour of nightly use bought about another 1.5 millimeters. Only six of the fifty-one trials contributed the appliance-against-inactive comparison, so that arm is thin.

Efficacy multiplied by acceptance is the quantity that actually reaches a patient, and the treatment that wins on the sleep study does not automatically win on that product.

Sleeping off the back is the cheapest doorway

More than half of all patients have sleep apnea substantially worse on their back, because the geometry is unfavorable, the lungs hold less air, and the dilator muscles do not make up the difference. Avoiding the supine position works. The source is a narrative review rather than primary data, and its authors are explicit that long-term adherence data and adequately powered trials are both lacking.

Through the model, these are different doorways into one system rather than rival claims about what the disease is. Each changes a different constraint on the same loop. Which one fits a given person is a clinical question, answered with a physician, and the answer depends on which of that person's traits is set wrong.

Hypoglossal stimulation supplies the reflex instead of splinting the tube

A pulse of 1.05 to 1.46 milliamps timed to each breath raised airflow from 215 to 509 milliliters per second in 30 patients with sleep apnea, and nobody woke.

Every treatment so far holds the tube open from outside. One does something categorically different, and it is the cleanest illustration of this model available anywhere in respiratory medicine.

A small implanted electrode is wrapped around the hypoglossal nerve, the nerve that drives the tongue forward. A sensor reads the breathing effort. With each inward breath the device delivers a pulse timed to that breath, so the tongue moves forward exactly when the suction is about to arrive. Nothing is splinted. The reflex that was supposed to fire, and did not, is supplied from outside.

A group at Johns Hopkins measured what it does. In 30 implanted patients they raised the current step by step during sleep and watched the airflow. Flow rose in a straight line with current, from 215 to 509 milliliters per second. Inspiratory flow limitation is the flattened breath that shows the tube already narrowing, and it was abolished entirely in 57 percent of them. All of this happened without waking anyone.

A milliamp delivered to the right nerve at the right millisecond does what no amount of pressure delivered to the wrong variable could. Specificity here is correspondence rather than force.

The withdrawal arm shows the device is holding the value

The clinical trial, called STAR, implanted 126 patients with sleep apnea who could not tolerate positive pressure. At twelve months events per hour had fallen 68 percent. Then the elegant part. Responders were randomized either to leave the device switched on or to have it switched off. In the switched-off group severity went straight back up, from 7.6 to 25.8 events an hour.

The main trial was an uncontrolled single-group cohort in a highly selected population of people who had already failed another therapy, and it was funded by the device manufacturer. Only the 46-patient withdrawal phase was randomized.

The withdrawal result also cuts both ways, and the model has to say so. It proves the device is doing the work, which is the strongest evidence a single-arm study can produce. It also shows the device is holding the value rather than resolving it. Switch it off and the disease is back within days. A precisely matched input delivered by a machine every breath is still an input the system has come to lean on.

Throat exercises halve the event rate in sleep apnea

Thirty minutes a day of tongue, palate and pharyngeal exercises took events from 22.4 an hour to 13.7 in a randomized trial with a sham arm, and nobody lost weight.

If a weak reflex is part of the problem, the obvious question is whether the muscle can be trained. It has been tried. The results are real and partial, which is the most useful kind of answer.

A group at the Heart Institute in Sao Paulo, working with the speech therapist Katia Guimaraes, took a simple idea to a trial. The exercises used to rehabilitate swallowing train the same muscles that hold the throat open. Could they be turned to this? Thirty-one patients with moderate apnea were randomized to sham therapy or to the exercises, for three months.

Weight did not change, so the obvious confounder is gone. Events per hour fell from 22.4 to 13.7. Snoring intensity fell from 3 to 1. Sleepiness fell from 14 to 8. Neck circumference came down by about a centimeter, and the change in neck size tracked the change in apnea severity.

Pooled, the effect is smaller and better characterized. A meta-analysis gathered nine adult studies totaling 120 patients and two pediatric studies totaling 25 children. In adults, events per hour fell from 24.5 to 12.3, which is about half. In children the reduction ran to about 62 percent. The lowest oxygen saturation of the night improved by about 4 percentage points. The authors position the therapy as an adjunct rather than a replacement. Half is real. Half is not enough on its own.

Didgeridoo playing is resistance training for the airway

The most charming study in the field is also a real randomized trial. A didgeridoo instructor named Alex Suarez noticed his students were sleeping better and took the observation to the Zurich epidemiologist Milo Puhan. They randomized 25 patients with moderate apnea either to lessons plus about 25 minutes of daily practice for four months, or to a waiting list. Sleepiness fell by 3 points. Events per hour fell by 6.2. Partners reported far less disturbance.

Circular breathing is what didgeridoo playing demands, and it amounts to resistance training for the upper airway. The control arm was a waiting list with no sham, so expectation cannot be separated out. Sleep quality itself did not improve, and the confidence interval on the apnea change reached almost to zero.

What these share is the direction of the input. They do not hold the airway open. They ask the muscle to work and let the system rebuild the response, a different aim with a smaller and slower effect. The model predicts precisely that shape. Restoration should be partial, slow, and it should persist in a way a splint does not, and that last part of the claim has never been properly tested here.

In children, enlarged tonsils are usually the driver

Adenotonsillectomy normalized the sleep study in 79 percent of the 464 children randomized in the CHAT trial. It also normalized in 46 percent of the children who simply waited.

In children the picture is different, and the discipline of this model is tested hardest here. The commonest driver of obstructive sleep apnea in a child is enlarged tonsils and adenoids. That is a structural obstruction, and calling it anything else would be false. What the model adds sits on top of that fact rather than in place of it.

The clearest demonstration of what it costs started from the wrong end. David Gozal, a pediatric sleep researcher, did not take children from a sleep clinic and ask how they were doing at school. He took 297 first-graders whose school performance sat in the lowest tenth of their class and screened them at home overnight.

Disordered breathing during sleep turned up in 18 percent of them. Of those, the 24 whose parents chose surgery improved their grades the following year, from 2.43 to 2.87. The 30 whose parents did not, did not, moving from 2.44 to 2.46. Parents chose, so this was not randomized, and the screening used oximetry rather than a full sleep study.

What the randomized trials moved, and what they did not

CHAT randomized 464 children aged 5 to 9 with obstructive sleep apnea to early adenotonsillectomy or to watchful waiting with support. Seven months later the primary outcome, a formal test of attention and executive function, showed no significant difference. Behavior, symptoms, quality of life and the sleep study itself were all significantly better after surgery. Nearly half the untreated children got better anyway, which sets a high bar for any intervention in this age group.

PATS asked the same question for children who snore but barely register events at all, in 459 children aged 3 to 12.9. Both primary outcomes were null. Executive function and attention did not differ at twelve months. Behavior, symptoms, sleepiness and quality of life improved. Blood pressure percentiles fell. Progression past the pediatric apnea threshold dropped from 13.2 percent to 1.3 percent. Serious surgical adverse events occurred in 2.7 percent.

The number normalizes, the behavior improves, and the cognitive measure does not move. Three explanations are live. The tests may be the wrong instrument for whatever changed. What the surgery restores may not be the thing those tests read. Or the window in which the load was carried may already have closed, which is the time course argument and it has not been settled here.

Central sleep apnea is the same controller failing at a different point

Men with heart failure and central apnea answered a rise in carbon dioxide more than twice as strongly, at 5.1 liters per minute per millimeter of mercury against 2.1. The stronger the response, the more apneas they had.

Some people with heart failure stop breathing at night with no obstruction at all. The throat is wide open. The brain simply stops sending the command.

Shahrokh Javaheri, a sleep and pulmonary physician, measured carbon dioxide sensitivity in 20 men with stable heart failure, ten of whom had central apnea and ten of whom did not. The apnea group answered more than twice as strongly.

That is loop gain again, arriving at a different point of the same loop. An over-responsive controller overshoots, blows off more carbon dioxide than it should, and then has nothing left to trigger the next breath. A failing heart compounds it by slowing the circulation, so information about carbon dioxide reaches the sensors late, and delay is what makes any control system oscillate. Cardiac disease itself belongs to the cardiovascular page. What belongs here is the controller.

CANPAP and SERVE-HF corrected the number and not the patient

CANPAP randomized 258 patients with heart failure and central apnea to positive pressure or control, for two years. Every physiological measure moved the right way. Events fell and night-time oxygen rose. Ejection fraction, the share of blood the heart pushes out with each beat, improved by 2.2 percent.

Norepinephrine fell and six-minute walk distance rose. Survival free of transplant was identical, at 32 events in each arm. The trial was stopped early and was underpowered as a result. The early survival curves favored the control group before crossing at 18 months, which no mechanism explains cleanly.

SERVE-HF went further. A cleverer machine was built, one that reads a patient's own breathing rhythm and supplies exactly the support needed to smooth it. In 1,325 patients with a weakened ejection fraction and predominantly central apnea it worked beautifully on the sleep study, driving events below 7 an hour. The primary composite endpoint did not improve. All-cause mortality was 28 percent higher in the treated group. Cardiovascular mortality was 34 percent higher.

In advanced heart failure, with that machine, correcting the number made the patients worse.

Where the SERVE-HF finding does and does not apply

Nothing here says the machine was badly built or that the physicians were wrong to try. The model states the principle directly. A protective pattern that is costly and still serving a purpose can leave the system worse off when it is removed before another strategy is in place.

Central apnea in advanced heart failure may be partly a compensation. Abolishing a compensation before the thing being compensated for has changed takes away a strategy the system was using. The mechanism of harm is still debated in the field.

Misreading the boundary would be dangerous. Nothing in SERVE-HF transfers to obstructive sleep apnea, and nothing in it applies to any other device. It is a specific finding in a specific population with a specific machine. A reader on positive pressure for obstructive apnea has learned nothing about their own therapy from it.

Sleep apnea is one organization read at four sites

Three signature aspects and the reflex that couples them shape one another night after night, which is why the same event count can be produced by different bodies.

Tone is the integrated organization of the body's interacting state, taken as one bound state rather than a list of parts. Vibration and tension are the intuitive entry point and they are incomplete. Tone is the chord, and oscillation is the carrier the chord is written on, which is why an unstable breathing rhythm reads the organization rather than being it.

Health on that definition is the range of states the airway can enter and leave: relaxing at sleep onset as everyone's does, deepening into slow-wave sleep, answering a narrowing when one arrives. A permanently tense airway would be as disordered as a collapsed one. Sleep apnea is the narrowing of that range until only the expensive cycle is left.

In the upper airway that organization has a specific composition, and each part of it is built above. Constraint is a tube with no rigid support, whose collapsibility reads out in centimeters of water. Gain is a breathing controller that answers a small disturbance with a large correction and starts to oscillate. Set point is an arousal threshold defended so low that sleep never deepens far enough to stabilize anything. The fourth measurement, muscle responsiveness, is the reflex that couples the other three.

How the traits feed one another

The coupling is the part a list of traits cannot show. A low arousal threshold prevents the deep sleep in which the tongue muscle fires fastest, so the set point degrades the reflex. Every arousal produces a ventilatory overshoot, so the set point feeds the gain. Every obstruction vibrates and injures the nerves reporting on the airway, so the constraint degrades the input the reflex depends on. And every night of oxygen dips potentiates the oxygen sensor, so the constraint feeds the gain in turn.

Three aspects, and the reflex that couples them, each shaping the others, producing a state the system returns to night after night. That is a chord, and the event count is only its loudest note.

Put back together, several things that looked like anomalies stop being anomalies. The awake patient running the tongue muscle at three times normal is compensated dysregulation, and the diagnosis names the night the compensation ran out. The overweight person with a collapsible airway and no disease has a live reflex holding a bad constraint. The older patient and the younger patient with identical severity have different machinery, because the same number can be produced by different organizations.

Same event count, different biology

The sharpest version comes from a study that sorted patients by their symptoms instead of their numbers. The sleep researcher Diego Mazzotti and colleagues took 1,207 people from a large American community cohort, every one of them with at least fifteen events an hour, and clustered them.

Four groups fell out: disturbed sleep, minimally symptomatic, excessively sleepy, and moderately sleepy. Only the excessively sleepy group carried raised cardiovascular risk, running events at 1.7 to 2.4 times the rate of the other groups and more than threefold the odds of prevalent heart failure.

Same event count. Different biology. The subtypes come from a statistical clustering method and the number of clusters is a modeling choice, and the study is observational, so hold it as a strong signal rather than a settled fact.

It also explains, with no special pleading required, why the two large cardiovascular trials found nothing. Both enrolled non-sleepy patients almost exclusively. An input averaged across systems that receive it differently produces a modest mean describing nobody, which is what the model predicts of any trial that does not stratify by the state of the receiver.

Masking holds the sleep apnea event count, restoring moves a driver

Withdraw positive pressure for two weeks and morning systolic pressure rises 8.5 millimeters of mercury. That is what holding a value looks like on the night the holding stops.

The model draws one line here, and the line runs across the treatments rather than between the people delivering them. An intervention that masks improves the experience and leaves the organization intact. An intervention that restores resolves the organization, and the body reorganizes around the change.

Both are legitimate. They are different achievements, and the first is routinely recorded as the second. The instrument does not decide which is which, because a machine can hold a value and a drug can move a driver. Aim decides.

What a withdrawal trial shows

The masking case has been measured with unusual clarity. A group at the University Hospital Zurich randomized 41 patients who were already doing well on positive pressure. Half continued. Half were switched, without their knowledge, to a pressure too low to work.

The apnea returned within days. Within two weeks, morning systolic pressure had risen 8.5 millimeters of mercury, diastolic 6.9 and heart rate 6.3. Endothelial function, how well the lining of a blood vessel widens on demand, had fallen by 3.2 percent. Epinephrine and norepinephrine measured in the urine had gone up. Inflammatory markers, insulin resistance and blood lipids were unchanged.

The splint holds the value while it is in place and not one night longer. That is the definition of masking, stated with no criticism attached, because a mask that holds a value every single night is worth having.

A pneumatic splint is a good intervention. It is the most reliable way anyone has found to abolish obstruction, and it improved sleepiness, mood and quality of life in every trial that measured them. The Unified Model of Tone is explicit that relief has real and honorable value, granted without qualification. It is doing a different thing from restoring the regulator, and knowing which one is happening is what lets a person and their physician ask a better next question.

Where a driver has actually been moved

Weight is the clearest case, and the evidence for moving it is randomized. Sleep AHEAD randomized 264 obese adults with type 2 diabetes to an intensive lifestyle program or to group education. The lifestyle arm lost 10.8 kilograms against 0.6. Events per hour fell 9.7 more than in the control arm.

Three times as many people reached complete remission, and the biggest losers had the biggest reductions. The limits are real. Obese adults with type 2 diabetes only, and most patients in the successful arm still had apnea after a year of intensive support. A real structural driver, really moved, with a partial result.

Newer evidence moves the same driver with a drug. Tirzepatide produces large weight loss, and two phase 3 trials followed adults with moderate to severe apnea and obesity for 52 weeks, one cohort using positive pressure and one not. Body weight fell 17.7 and 19.6 percent on the drug, against 1.6 and 2.3 percent on placebo.

Events per hour fell by about 25 to 29 with the drug against about 5 with placebo. Systolic pressure, C-reactive protein and patient-reported sleep all improved. So did hypoxic burden, the total area under each oxygen dip across a whole night. The trials were manufacturer-funded, restricted to obesity, and there is no cardiovascular outcome data yet.

A drug pair aimed squarely at the reflex

A group at Brigham and Women's Hospital gave 20 people two old drugs together for a single night, one raising norepinephrine and one blocking acetylcholine. Events per hour fell 63 percent. Of the 15 who had real apnea at baseline, every one improved by at least half. Tongue muscle responsiveness roughly tripled. Neither drug did anything on its own. Twenty people, one night, nothing about durability or outcomes.

None of these paragraphs is advice and none of them names a dose. What they establish is that the drivers of sleep apnea can be moved, through different doorways and at different magnitudes, and that moving a driver behaves differently over time from holding a value.

What would establish this reading of sleep apnea

Four results would establish it. The sharpest is bidirectional restoration, where an input brings a high arousal threshold down and a low one up.

A framework explaining this much owes its readers a list of the results that would establish it, stated before the data.

The four traits have to load on one factor

First, the four traits have to behave like readings of one organization. Acetazolamide already shows that one of them can be shifted on its own, so they are separately perturbable axes. The claim underneath is stronger.

Measure all four together in the same people and they should load on a common underlying factor and vary together from person to person. Loading on that common factor across a population confirms the one-organization reading in the previous section, and makes a chord of what would otherwise be only a list.

Second, the traits have to predict who responds to what. The model holds that correspondence rather than magnitude decides an outcome. So measure the traits before treatment, assign each person the therapy their own traits select, and compare against the same therapies assigned at random. Matched assignment outperforming random assignment confirms specificity as correspondence, and with it this reading of sleep apnea.

Bidirectional restoration is the sharp test

A splint pushes one way regardless of where a person started. A restored regulator should move a value toward the middle of its healthy range from either side. Take people whose arousal threshold sits above the healthy window and people whose threshold sits below it.

Apply an input selected by a measure of tone recorded before the outcome is known, against a sham arm matched for contact and attention. The model predicts the two groups converge, and converge further than the sham arm does. If both groups move the same direction regardless of where they began, this is masking rather than restoration.

A splint holds the number for exactly as long as it is worn. A restored regulator changes what the system does on the nights when nobody is holding anything.

Fourth, and softest. A measure of a person's regulation, recorded before treatment, should carry information about their outcome that the event count does not.

That last one has already begun to arrive. The sleep researcher Ali Azarbarzin and colleagues noticed that the standard severity score counts events per hour and throws away how deep and how long each one was. They measured hypoxic burden instead, in 2,743 older men and in 5,111 adults from a second cohort.

That measure predicted cardiovascular death, with a death rate up to 2.73 times higher in the top fifth. The event count did not. It is observational, the metric was developed and tested in the same data era, and it predicted all-cause mortality clearly in one cohort and not in the other.

What you choose to measure decides what you are able to see. A frequency count of a defended value tells you less than the burden the system actually carried.

What a restored airway would look like on the instruments

The instruments already exist. Critical closing pressure reads the constraint. Loop gain reads the controller. The arousal threshold reads the defended value. Genioglossus responsiveness reads the reflex, and automated methods are beginning to pull all four out of an ordinary sleep study. Overnight sympathetic recording reads the accelerator. The nocturnal blood pressure dip reads whether a rhythm has come back. Hypoxic burden reads the cost.

A nervous system regaining its range would show it on those instruments in a particular pattern. Collapsibility would matter less, because the reflex answering it would matter more. Loop gain would fall toward one from above. The arousal threshold would rise toward the middle from below, and in the person whose threshold sits too high it would fall.

Deep sleep would return, and with it the fastest firing the tongue muscle achieves all night. Events would fall because none of the traits was set to produce them, rather than because a machine was holding the tube open.

Nobody has run that study. How much obstructive sleep apnea can be resolved by restoring regulation rather than by holding an airway open has not been answered. It is the study the model calls for. What is already clear matters more to the person reading. The throat closes because a regulator stopped holding it open, and regulation is the one thing in a body that was always built to change.

How sleep apnea relates to the rest of the library

Sleep apnea is a condition, and the foundations of the Unified Model of Tone are properties expressed through it. The library keeps them separate because collapsing them is what makes explanations vague.

Four condition pages sit closest to this one.

• Sleep owns the architecture of a normal night, including the depth this airway never reaches.
• Insomnia is the other way an arousal threshold can be set wrong, and the two conditions occur together often enough to have their own name in the literature.
• Blood pressure owns the Wisconsin cohort and the dose-response between apnea severity and incident hypertension.
• Cardiovascular disease is where the nightly surges are eventually paid for, and where SERVE-HF drew its hard boundary.

Frequently asked