Roughly a billion adults stop breathing repeatedly while they sleep. A literature-based analysis covering the 30-to-69 age band estimated 936 million people worldwide with mild-to-severe obstructive sleep apnea and 425 million with moderate-to-severe disease, with prevalence above 50 percent in some countries [1]. In the United States, modelling from the Wisconsin Sleep Cohort put moderate-to-severe sleep-disordered breathing at 17 percent of men aged 50 to 70 and 9 percent of women in the same band, increases of 14 to 55 percent over two decades driven largely by rising body weight [2]. A clinical review summarising US data gives 34 percent of men and 17 percent of women affected at the broader diagnostic threshold [3].
Most of those people do not know. That is the first thing worth understanding about this disease, and it shapes everything that follows.
Start here: what obstructive sleep apnea is
During sleep the muscles that hold the throat open relax. In some people the upper airway narrows or closes entirely; breathing effort continues against a shut airway, blood oxygen falls, carbon dioxide rises, and the brain surfaces briefly toward wakefulness to restore muscle tone and reopen the airway. Then sleep resumes and the cycle repeats — in severe disease, dozens of times an hour, all night, for years.
Two consequences follow. The first is fragmented sleep, and with it the daytime sleepiness that is the classic presenting symptom. The second is a nightly physiological stress: each event ends in a surge of sympathetic nervous activity. Direct nerve recordings in patients with obstructive sleep apnea found sympathetic activity elevated even during wakefulness, blood pressure that failed to fall during any stage of sleep, and sympathetic bursts at the end of apneic events reaching 299 percent of waking levels during stage II sleep — all attenuated by positive airway pressure [4]. Beat-to-beat cardiovascular variability is correspondingly disturbed [5], and the repeated cycles of oxygen desaturation and reoxygenation have been proposed as an oxidative-stress mechanism linking the disorder to vascular disease [6]. Broader reviews of the pathophysiology gather these threads [7] [8].
Obesity is the dominant risk factor, acting partly through altered respiratory mechanics — reduced lung volumes and increased work of breathing [9] — and partly through fat deposition around the airway. But it is not the whole story. A careful physiological study of 75 people measured both the anatomical contribution (the pressure at which the passive upper airway collapses) and three non-anatomical traits: how responsive the tongue muscle is during sleep, how easily an arousal is triggered, and the stability of respiratory control. All varied substantially between individuals. Thirty-six percent had minimal genioglossus responsiveness during sleep, 37 percent a low arousal threshold, and 36 percent unstable respiratory control; 19 percent had an airway no more collapsible than many controls. The authors concluded that non-anatomical features play an important role in 56 percent of patients [10]. Obstructive sleep apnea is not one disease with one cause, and later work has pushed toward an integrative model on that basis [11].
Pillar 1: measurement and diagnosis
The apnea-hypopnea index, and what it does and does not capture
Severity is conventionally reduced to a single number: the apnea-hypopnea index, the count of complete and partial breathing cessations per hour of sleep. Five or more per hour with symptoms, or fifteen or more regardless, is the usual diagnostic line.
The index earned its place. A prospective analysis of the Wisconsin cohort found a graded relationship between baseline apnea-hypopnea index and hypertension four years later — odds ratios of 1.42, 2.03 and 2.89 across rising severity bands, after adjustment for baseline blood pressure, body-mass index, neck and waist circumference, age, sex, alcohol and smoking [12]. A large community-based cross-sectional study found the same association [13], as did earlier population work [14]. An observational cohort of 1,022 patients followed for stroke or death from any cause found a hazard ratio of 1.97 for those with the syndrome after adjustment for age, sex, race, smoking, alcohol, body-mass index, diabetes, hyperlipidemia, atrial fibrillation and hypertension, with a significant trend across severity [15]. Prospective cohort studies of sleep-disordered breathing and mortality point the same way [16] [17].
The index is nonetheless a crude summary. It counts events without weighting how long they last or how far oxygen falls, and a review of severity metrics argues for supplementing it with measures such as hypoxic burden, arousal intensity, the odds ratio product and cardiopulmonary coupling — while conceding that the apnea-hypopnea index remains the best-studied metric available, imperfect as it is [18]. This matters practically: two people with identical indices can have very different physiology and very different symptoms.
Getting tested
Attended overnight polysomnography in a laboratory is the reference standard, and practice parameters set out its indications [19]. It is also expensive, capacity-limited and unlike a normal night's sleep. Home sleep apnea testing has therefore become routine for many patients, with a sensitivity of approximately 80 percent [3] — good enough to confirm suspected moderate-to-severe disease, not good enough to rule it out in someone with a convincing story.
Symptom questionnaires do the triage. The Epworth Sleepiness Scale asks how likely a person is to doze in eight everyday situations [20] and has been examined for its factor structure and reliability [21]. The STOP-Bang questionnaire combines snoring, tiredness, observed apneas, blood pressure, body-mass index, age, neck circumference and sex into an eight-item score [22]; high scores indicate high probability of the disorder [23], and it has been validated across several populations [24]. None of these is diagnostic on its own.
The gap between symptom and disease is the recurring problem. Excessive sleepiness, the most common presenting complaint, is reported by as few as 15 to 50 percent of people with the disorder in the general population [3]. Screening based on sleepiness alone therefore misses most cases — and, as the next section shows, treating people found by any other route turns out to be harder to justify than one might expect.
Centerpiece: a simple simulatable model of adherence and benefit
The central practical fact about obstructive sleep apnea is that the standard treatment works well and is frequently not used. Continuous positive airway pressure splints the airway open with a stream of pressurised air delivered through a mask; it reliably abolishes breathing events. Whether it helps a given patient depends almost entirely on how many hours a night they actually wear it — a problem recognised as the central challenge of the therapy for decades [25] and reviewed systematically across age groups [26].
A multicenter study of 149 patients with severe disease, tested before treatment and again after three months, fitted dose-response functions of nightly duration against the chance of returning to normal levels of sleepiness and daily functioning. It identified thresholds above which further improvement became less likely, and they were not the same number for every outcome: 4 hours for the Epworth Sleepiness Scale, 6 hours for the Multiple Sleep Latency Test, and 7.5 hours for the Functional Outcomes of Sleep Questionnaire. Below those thresholds the relationship was linear in nightly hours [27].
That is the model. The left panel below draws it directly from the source's own description — a straight rise in nightly hours to a plateau at each outcome's published threshold. The vertical scale is normalised to the share of the attainable response rather than an absolute probability, because the source reports the thresholds and the linearity beneath them rather than full probit coefficients; the thresholds themselves are the published values, and are the only fitted parameters.
The right panel simply evaluates that function at nightly adherence levels actually measured in published trials. Nothing new is assumed.
Two things fall out of the arithmetic.
First, adequate use is not one number. At 5.2 hours a night — the mean recorded in the pressure-therapy arm of a crossover trial [28] — a patient has captured the entire attainable improvement in subjective sleepiness but only 69 percent of the improvement in daily functioning under the published thresholds [27]. The conventional four-hour definition of adherence is calibrated to the easiest of the three outcomes.
Second, and more consequentially: the largest cardiovascular-outcomes trial of the therapy ran at a mean nightly use of 3.3 hours [29] — below every one of the three thresholds. Under the model that trial's participants stood to receive about 83 percent of the available sleepiness benefit and 44 percent of the available functional benefit. That trial, which randomised 2,717 adults with moderate-to-severe disease and established cardiovascular disease, found no reduction in its primary composite endpoint (hazard ratio 1.10, 95 percent confidence interval 0.91 to 1.32) while significantly reducing snoring and daytime sleepiness and improving quality of life and mood [29]. A separate trial in non-sleepy patients with coronary artery disease also found no cardiovascular benefit [30].
Whether the null result reflects a genuine absence of cardiovascular benefit or an inadequate dose of a therapy that works is unresolved, and it is the single most important open question in the field. It is worth being precise about what the model does and does not say: it was fitted to sleepiness and functioning, not to heart attacks, and extrapolating it to cardiovascular endpoints is an assumption, not a finding.
The figure's third lesson is quieter. In a randomised crossover trial, an oral appliance controlled fewer breathing events per night than positive airway pressure (residual index 11.1 versus 4.5 per hour) but was worn longer (6.5 versus 5.2 hours), and the two treatments produced similar improvements in sleepiness, driving-simulator performance and disease-specific quality of life [28]. Efficacy per night and effectiveness over a year are different quantities, and adherence is the exchange rate between them.
Pillar 2: treatments
Positive airway pressure
It remains first-line for symptomatic moderate-to-severe disease, and its symptomatic benefits are not in doubt. Its effect on blood pressure is real but modest and concentrated in the patients with the most to gain: in a randomised trial of 194 patients with resistant hypertension and an apnea-hypopnea index of 15 or above, twelve weeks of treatment lowered 24-hour mean blood pressure by 3.1 mmHg (95 percent confidence interval 0.6 to 5.6) and restored a normal nocturnal dipping pattern in more patients (35.9 versus 21.6 percent). That trial also found a significant positive correlation between hours of use and the size of the blood-pressure fall (r = 0.29) — the dose-response again, measured directly [31]. Earlier controlled work had established the blood-pressure effect [32].
In patients with a high event count but no daytime sleepiness, a randomised sham-controlled trial found no change in quality of life, objective sleepiness, cognitive function or arterial blood pressure after six weeks, and concluded the treatment was not indicated in that group [33]. A clinical review reaches a similar bottom line: managing asymptomatic disease to reduce cardiovascular events is not currently supported by high-quality evidence [3]. Supplemental oxygen is not a substitute; a trial comparing the two found they were not equivalent [34].
Oral appliances, position, and the non-device options
Mandibular advancement devices hold the lower jaw forward during sleep, enlarging the upper airway [35]. Practice parameters position them for snoring and for mild-to-moderate disease, and for patients who cannot tolerate pressure therapy [36]; as noted above, their lower per-night efficacy can be offset by better adherence [28]. Some patients have strongly position-dependent disease, and positional therapy has an underused role there [37]. The broader non-pressure landscape has been reviewed [38] and codified in a European Respiratory Society guideline [39].
Stimulating the airway open
If the problem is a tongue muscle that fails to activate during sleep, one answer is to activate it electrically. The idea was tested early [40]; a later single-arm trial of an implanted hypoglossal nerve stimulator reported use on 89 percent of nights for 5.8 hours per night, with the apnea-hypopnea index falling from 43.1 to 19.5 [41]. The pivotal trial of upper-airway stimulation enrolled 126 patients who could not accept or adhere to pressure therapy and reported a 68 percent median reduction in the event index at twelve months, from 29.3 to 9.0 per hour, with under 2 percent serious procedure-related adverse events; a randomised withdrawal phase confirmed the effect was the device's doing, as events returned in the withdrawal group [42]. It is effective in selected patients, generally those with a body-mass index under 32 [3].
Weight, and the first drug
Because obesity is the principal modifiable cause, weight loss is causal therapy rather than adjunct. A trial randomising 181 patients to pressure therapy, a weight-loss intervention, or both found that reductions in C-reactive protein, insulin resistance and triglycerides appeared in the weight-loss arms and not with pressure therapy alone, while blood pressure fell in all three groups [43] — a useful reminder that the two treatments address different parts of the problem.
In 2024 two phase 3 trials established the first drug therapy. Adults with moderate-to-severe disease and obesity, enrolled either off or on pressure therapy, received tirzepatide or placebo for 52 weeks. From a baseline event index around 50 per hour, the treatment difference versus placebo was 20.0 events per hour in those not on pressure therapy and 23.8 in those on it, with concurrent improvements in body weight, hypoxic burden, high-sensitivity C-reactive protein, systolic blood pressure and patient-reported sleep outcomes; adverse events were predominantly gastrointestinal [44]. This is a genuine change: as recently as 2020 the field's summary position was that no effective pharmacological therapy existed [3].
The honest framing is that this is a drug for the obesity that causes the apnea, not a drug for the apnea. That is not a criticism — treating the cause is usually the better move — but it means the benefit is unlikely to extend to the substantial minority of patients whose disease is driven by non-anatomical traits [10].
Pillar 3: what is unresolved
The cardiovascular question
The association between sleep-disordered breathing and hypertension, stroke and death is consistent, graded and adjusted-for-confounders across multiple cohorts [12] [15] [16]. The randomised trials of treatment have not shown a reduction in cardiovascular events [29] [30]. Reconciling these is the field's central problem, and the candidate explanations are not mutually exclusive: adherence too low to deliver a dose, trials enrolling patients selected for the absence of sleepiness (and therefore perhaps for the absence of benefit), follow-up too short, or an association that is confounded by obesity more thoroughly than adjustment can fix.
Endotypes and matching treatment to cause
If a third of patients have a low arousal threshold and a third unstable respiratory control [10], then treatments aimed at those traits should exist and should be targeted by measurement rather than by trial and error. Phenotype-based frameworks have been proposed for clinical and perioperative use [45] [11]. Turning research-grade physiological measurements into something orderable in a clinic remains the obstacle.
Finding the people who have it
Given that most cases are undiagnosed and that sleepiness identifies only a fraction of them [3], the case-finding problem is unsolved. It is sharpened by the settings where undiagnosed disease is dangerous — surgery and anesthesia in particular, for which specific guidelines exist [46] — and by paediatric disease, which has a distinct presentation and management [47].
What better sleep is worth
Sleep disruption has consequences beyond the cardiovascular [48], and there is intriguing experimental work connecting slow-wave sleep disruption to cerebrospinal-fluid amyloid-beta levels [49]. Whether treating sleep apnea alters long-term cognitive outcomes is not established.
Dig deeper in lmmol
Obstructive sleep apnea sits at the centre of a causal triangle that lmmol covers from the other two corners. Obesity is its principal cause and the target of its first drug therapy — see obesity. Hypertension is its best-established consequence, with the dose-response between event count and later blood pressure among the most replicated findings in the field — see hypertension [12]. The relationship runs onward into cardiac disease: see heart failure, where positive airway pressure has been studied for central as well as obstructive events [50]. The stroke association [15] connects to stroke, and one review treats sleep apnea, hypertension and kidney disease as a single spectrum of shared pathology [51] — see chronic kidney disease. Metabolic overlap links it to type 2 diabetes and glycemic control, and the contrast with fixed airflow obstruction is instructive against COPD and asthma. The sleep-and-amyloid thread [49] touches Alzheimer's disease. The full collection is at health.