The first oral drug for sleep apnoea has published its phase 3 data

The peer-reviewed results of the largest trial of a drug for obstructive sleep apnoea are now published, and they are more mixed than the headline "first pill for sleep apnoea" suggests. AD109 works on the airway. Whether it makes patients feel better is, on this evidence, unresolved.

SynAIRgy randomised 646 adults with mild-to-severe obstructive sleep apnoea who were intolerant of positive airway pressure therapy or had refused it, across 69 centres, to AD109 or placebo for 26 weeks [s1]. AD109 is a fixed-dose oral combination of aroxybutynin 2.5 mg and atomoxetine 75 mg, designed to target the neuromuscular dysfunction that lets the upper airway collapse during sleep rather than the anatomy of the airway itself [s1].

The population matters for interpreting what follows. Median age was 58 years, 49.3% were female, and median body mass index was 32.4 kg/m2 [s1]. Median baseline apnoea-hypopnoea index was 19.6 events per hour, with 35% of participants classed as mild, 42% moderate and 23% severe [s1]. This was not a severe-apnoea population.

What it did to the airway

At week 26, the mean treatment difference in apnoea-hypopnoea index was −4.0 events per hour (95% CI, −6.4 to −1.6; P = .001), which the investigators describe as a model-estimated 44.1% decrease from baseline against 17.6% on placebo (P < .0001) [s1]. AD109 also improved the oxygen desaturation index and hypoxic burden against placebo at week 26 [s1].

Those are real effects on breathing, and they are the reason the trial is being treated as a milestone. A drug that reduces obstructive events without a mask or an implant would fill a genuine gap, because the alternative for someone who cannot tolerate positive airway pressure has until now been surgical or mechanical.

The size of the effect deserves its own sentence. Four events an hour is a modest absolute change, and it is measured in a group whose median starting point was 19.6 events per hour [s1]. A 44.1% reduction from that baseline does not put most patients into a normal range.

What it did not do

The trial's key secondary endpoint list included the PROMIS-Fatigue T-score, and there was no statistically significant difference between AD109 and placebo on it [s1].

This is the finding that complicates the story, and it is the one most likely to be lost in summary. Patients do not present asking for a lower apnoea-hypopnoea index. They present tired. A treatment that improves the index while leaving fatigue statistically unchanged has demonstrated mechanism without yet demonstrating the benefit the patient came for.

There are defensible explanations. Fatigue is noisy, the trial was not powered as a fatigue trial, and a population with a median index under 20 events per hour may have had limited room to improve. None of those explanations is a result. On the published evidence, the fatigue endpoint did not separate.

Tolerability

Discontinuation because of adverse events ran at 21.2% on AD109 against 3.1% on placebo [s1]. The most common adverse events with AD109 were dry mouth, nausea, insomnia and urinary hesitation, and no serious treatment-related adverse events were reported [s1].

A roughly sevenfold difference in discontinuation is a substantial tolerability signal, and it is the number to hold alongside the efficacy result. It also has a specific irony worth noting: insomnia was among the common adverse events in a trial treating a sleep disorder [s1]. Adherence is the reason drug therapy is attractive for people who could not tolerate a mask. A drug that one in five people stops is not a solved adherence problem, only a differently shaped one.

Where it sits against the alternatives

The same period produced published randomised evidence for a device approach in a related population. OSPREY tested proximal hypoglossal nerve stimulation in 104 patients with moderate-to-severe apnoea across 23 US health centres, and at month 7, 58.2% (95% CI, 45.5% to 70.2%) of treated patients met a primary endpoint of greater than 50% improvement in apnoea-hypopnoea index and an index below 20 events per hour, against 13.5% (CI, 4.5% to 28.8%) of controls [s2].

The two are not directly comparable — different populations, different severities, different endpoints, and OSPREY required an implant in every participant including controls [s2]. But they mark the shape of the field for patients who cannot use positive airway pressure: an implant with a large responder rate in a small trial, or a pill with a modest average effect and a meaningful discontinuation rate in a large one.

What to watch

Whether a trial powered for symptoms, in a population with a higher baseline index, separates on fatigue or sleepiness. Whether the 21.2% discontinuation rate holds outside a trial setting, where it usually rises [s1]. And whether regulators treat a four-event-per-hour improvement in the index as sufficient without a corresponding patient-reported benefit.

This article describes published trial results. It is not medical advice, and decisions about apnoea treatment belong with a clinician who has seen the patient's sleep study.

Sources

• [s1] Aroxybutynin and atomoxetine (AD109) for obstructive sleep apnea: a randomized phase 3 trial (SynAIRgy), American Journal of Respiratory and Critical Care Medicine, 2026;212(7):1569–1584.
• [s2] Proximal Hypoglossal Nerve Stimulation for Obstructive Sleep Apnea in the OSPREY Study: A Randomized Controlled Trial, Annals of Internal Medicine, 2026;179(6):812–822.