Buried at the end of the obstructive sleep apnea section of the current tirzepatide label is a two-sentence paragraph headed Sleep-Related Impairment. It says that treated patients showed improvement compared with placebo, and that the measurement was made with the PROMIS Short Form Sleep-Related Impairment 8a. That is the whole of it. No point estimate, no interval, no p value.[1] Four inches above it, the same document reports the sleep apnea-specific hypoxic burden falling 70.1 %·min/hour further than placebo (95% CI, −90.9 to −49.3) and body weight falling 16.1% further (95% CI, −18.0 to −14.2).[1] The breathing got a confidence interval. The sleeping got an adjective.
That asymmetry is the subject here. The event count itself is set out in the sleep apnea article and the trial machinery in the SURMOUNT-OSA breakdown. What follows is narrower and, for anyone buying a drug because they are tired, more useful: which sleep instruments were carried into these trials, which were scored, and what the published rankings say about where this drug class actually sits on the outcome a person notices.
The instrument that was collected and never reported as a result
The Epworth Sleepiness Scale is an eight-item self-administered questionnaire asking how likely a person is to doze in ordinary situations. It is the standard measure of daytime sleepiness in sleep medicine, and it is a reliable one: in its validation work, 87 healthy medical students retested five months apart produced paired scores correlated at r = 0.82, with internal consistency of 0.88 by Cronbach’s alpha.[2]
The tirzepatide apnea trials collected it. It appears in the label twice. Once as a baseline characteristic — mean total score 10.5 (SD 5.2) in the first trial and 10.0 (SD 4.6) in the second. Once as a threshold inside the composite remission definition, where controlled disease means an event index below 5, or between 5 and 14 with an Epworth score of 10 or less.[1]
So the score was used to decide who counted as successfully treated, and was never itself reported as changing. A prospective buyer reading the prescribing information can learn the baseline sleepiness of the trial population and the arithmetic role the score plays in a composite, and cannot learn whether it moved. That is a choice about what to print, not a gap in what was measured.
A cohort with severe breathing and borderline sleepiness
The baseline number is worth sitting with. Those trials enrolled people whose breathing was badly disordered: 63.1% of the first trial and 68.2% of the second had severe disease by the label’s own definition of at least 30 events an hour, against 35.2% and 30.9% with moderate disease.[1] Yet mean Epworth scores of 10.5 and 10.0 sit essentially on the conventional line for excessive daytime sleepiness, which means roughly half the enrolled population was not, by that instrument, unusually sleepy at all.
That matters because in the instrument’s own validation data sleepiness and the event count travel together. Epworth scores distinguished people with simple snoring from people with apnea, rose with apnea severity, and in regression were more closely related to the frequency of apneas than to the degree of oxygen desaturation.[3] A cohort near 50 events an hour would, on that relationship, be expected to score higher than 10.5. Somebody with severe disease and a normal Epworth is common enough to fill half a trial, and that person is precisely the reader most likely to be told a drug will fix their tiredness.
Where this drug class ranks when every treatment is lined up
Two evidence syntheses published since have put the whole field on one table, and they reverse the intuition. An umbrella review screened 5,571 meta-analyses and retained 34, covering 230 randomized trials and 36,353 participants, 72.3% of them male.[4]
For lowering the event count, positive airway pressure ranked first at a mean difference of −30.7 events per hour (standardized mean difference −1.65; 95% CI, −1.87 to −1.43), and tirzepatide ranked second at −21.86 events per hour (SMD −0.84; 95% CI, −1.01 to −0.68), ahead of mandibular advancement devices at −11.91.[4] The certainty grading inverts the ordering: the review rated the airway-pressure estimate low certainty and the tirzepatide estimate moderate. The runner-up has the better-evidenced number.
Then the daytime outcome. The same review reports that airway pressure, wake-promoting stimulants, hypoglossal nerve stimulation and myofunctional therapy significantly reduced Epworth scores, at standardized mean differences of −0.80 to −0.88. Incretin therapy is not among them.[4] Quality of life produced a third ordering again, with physical activity giving the largest improvement (SMD 1.3; 95% CI, 0.58 to 2.02) and airway pressure a modest one (SMD 0.16; 95% CI, 0.11 to 0.21).[4]
A network meta-analysis of 34 randomized trials in 3,964 participants found the same split from a different direction. Airway pressure both cut the event index (mean difference −22.17 events per hour; 95% CI, −38.01 to −6.33) and improved the Epworth score (−2.75; 95% CI, −3.71 to −1.79). Liraglutide’s significant effects were metabolic: body-mass index −1.60 kg/m² (95% CI, −2.04 to −1.16) and glycated hemoglobin −0.19% (95% CI, −0.25 to −0.13), with the combination of liraglutide and airway pressure producing the largest body-mass reduction at −2.00 kg/m² (95% CI, −3.49 to −0.51).[5] The machine carried the sleepiness result; the drug carried the weight and the glucose.
Neither synthesis says incretin therapy makes sleepiness worse, and neither was powered as a head-to-head trial. What they establish is narrower and still useful: across the published randomized literature, the interventions with a demonstrated effect on how sleepy people feel are a different set from the interventions with the largest effect on how they breathe, and the drug sits firmly in the second group.
What the other five labels say about sleep, which is nothing
Among the six current prescribing documents for this class, the word insomnia appears exactly once: in the liraglutide 3 mg adverse-reaction table, reported by 2.4% of treated adults against 1.7% on placebo, sitting between viral gastroenteritis and dry mouth.[6] It appears in none of the other five.
The terms somnolence, sleep disorder, nightmare, abnormal dreams, sleep quality and restless legs appear zero times across all six documents. The same six contain the word nausea between 10 and 20 times each and diarrhea between 8 and 14 times each, so the zeros describe the labels rather than a failure to look. The word sleep does occur twice in each of the semaglutide, oral semaglutide, liraglutide and dual-agonist diabetes labels — both times inside the phrase “deep sleepiness (deep sedation),” which is the patient-language gloss on anesthesia in the aspiration warning, and has nothing to do with sleeping at night.[7] Only the label carrying the apnea indication discusses sleep as sleep.
Nobody has staged a night of sleep on this drug
Sleep quality in a laboratory means architecture: how much of the night is spent in deep non-REM sleep, how much in REM, how often a person surfaces. Those trials ran full overnight polysomnography three times per participant, which is the recording that produces exactly those numbers. None of it has been published as a staging result.
The indexed literature is the same shape. Combining the three molecule names with sleep-architecture terms and restricting to human studies returns three records, none of them a sleep-staging trial in people taking these drugs. The same architecture terms combined with positive airway pressure return 300, and the same three molecule names combined with weight loss return 2,412, so the vocabulary on both sides of the intersection retrieves plentifully and the emptiness is in the middle.
The one place electrodes have gone on is a rodent. In a study of adult male rats implanted with EEG and EMG electrodes, subcutaneous liraglutide dose-dependently decreased wakefulness and increased non-REM sleep at every dose except the lowest, shortening waking bouts and lengthening non-REM bouts, with the change concentrated in the animals’ dark phase.[8] That is a real measurement of sleep architecture on this drug class, and it was made in rats at doses chosen for their effect on opioid seeking. It is a hypothesis about people, not a finding in them.
The adjacent sleep disorders are thinner still. The published record on restless legs syndrome and this drug class is a single case observation of one patient with Parkinson’s disease whose refractory symptoms improved after starting semaglutide.[9] One patient with another neurological condition is a reason to look, not a reason to expect anything.
What this changes about the purchase
Someone whose breathing is measurably disordered has a well-evidenced reason to treat it, and this drug class is second in the field for that, on better-graded evidence than the therapy ranked above it. Someone whose complaint is that they sleep badly and wake unrefreshed, with no apnea diagnosis, is buying into a different question, and the published rankings put four other interventions ahead of the drug on that outcome. Daytime tiredness has its own separate treatment course in the fatigue article, and the broader sleep literature is collected in the sleep article. Schedules that fragment sleep by design are a further complication, covered in the shift-work article.
Every figure above belongs to branded, FDA-approved product at labeled doses, and the apnea indication belongs to one brand of tirzepatide. Compounded preparations are not FDA-approved, and the FDA does not review them for safety, efficacy or quality before they are dispensed; no compounded product holds an approved indication for anything, sleep apnea included. Most sellers on the tirzepatide board dispense compounded vials, and that distinction is set out in the compounded-versus-brand article. No sleep measurement of any kind — questionnaire, actigraphy or polysomnogram — has been published on a compounded preparation.
Apnea is diagnosed by a sleep study. An intake form cannot produce an event index, and a person who suspects their nights are the problem is better served by a recording than by a prescription written against a symptom nobody in these trials reported a number for.