A nerve-stimulation implant for sleep apnoea reports its randomised results

Most evidence for implanted devices in obstructive sleep apnoea comes from single-arm studies: implant everyone, switch everyone on, measure the change. That design cannot separate the device from the surgery, the follow-up, or the regression to the mean that any measured population shows over time.

OSPREY did something harder. Every participant was implanted, then randomly assigned in a 2:1 ratio to have therapy begin at month 1 or at month 7 [s1]. The control group carried the same hardware and the same operation, and simply waited. Whatever the difference at month 7 is, it is not the surgery.

The design and the population

The trial was a 7-month randomised controlled trial followed by a 6-month open-label extension, across 23 US health centres, in adults aged 22 or older with moderate-to-severe obstructive sleep apnoea [s1].

The 104 randomly assigned patients had a mean age of 55.6 years (SD, 9.0), body mass index of 30.6 kg/m2 (SD, 3.0), a preimplantation apnoea-hypopnoea index of 35.7 events per hour (SD, 12.8), and a preimplantation oxygen desaturation index of 36.7 events per hour (SD, 13.4) [s1].

That baseline index of 35.7 events per hour puts this squarely in moderate-to-severe territory, which is worth holding in mind when comparing against trials run in milder populations [s1].

The intervention was proximal hypoglossal nerve stimulation, which delivers multicontact stimulation to proximal portions of the nerve and, the authors note, allows easier electrode implantation than distal nerve stimulation [s1].

What it found

The primary endpoint combined two thresholds: greater than 50% improvement from baseline in the apnoea-hypopnoea index, and an index below 20 events per hour at month 7.

At month 7, 58.2% (95% CI, 45.5% to 70.2%) of patients assigned to treatment (n = 67) met that endpoint, against 13.5% (CI, 4.5% to 28.8%) of those assigned to control (n = 37) [s1].

Oxygen desaturation index was reduced by at least 25% in 68.7% (CI, 56.2% to 79.4%) of the treatment group against 37.8% (CI, 22.5% to 55.2%) of controls [s1].

On symptoms, median Epworth Sleepiness Scale score improved from baseline to month 7 in the treatment group, from 10.0 (IQR, 7.0 to 14.0) to 6.0 (IQR, 5.0 to 9.0), and did not improve in the control group, going from 9.0 (IQR, 7.0 to 11.0) to 9.0 (IQR, 6.0 to 11.0) [s1].

No serious procedure-related adverse events were reported [s1].

The control group is the interesting number

Look again at the controls: 13.5% met a primary endpoint requiring a halving of their apnoea-hypopnoea index, having received nothing but an implanted, inactive device [s1].

That is the value of this design. In a single-arm study, those responders would have been counted as treatment successes. Here they are visible as what happens anyway — measurement variability, night-to-night variation in the index, weight change, and whatever behavioural effect follows being enrolled in a sleep trial. Any implant study without a control arm is quietly absorbing that 13.5% into its response rate.

The same point applies to the sleepiness result, and here the contrast is sharper. Controls did not improve on the Epworth scale at all, from a median of 9.0 to 9.0 [s1]. The treated group moved from 10.0 to 6.0 [s1]. Symptom scores are the endpoint most vulnerable to expectation, and controls knew they were waiting.

The limits the authors state

The paper lists them: lack of blinding, small sample size, and short follow-up [s1].

Lack of blinding is the substantive one. Patients knew whether their device had been switched on, which bears directly on a self-reported sleepiness scale. The apnoea-hypopnoea index and oxygen desaturation index are measured rather than reported, so they are more robust to this, but the Epworth result is not immune.

The primary funding source was LivaNova PLC, the device manufacturer, and two authors are identified in the paper as affiliated with the company [s1]. That is disclosed, and it is ordinary for device trials; it is also a reason to want independent replication.

Where it sits

For patients who cannot tolerate positive airway pressure, 2026 has produced two randomised options with published data pointing in different directions on effect size and burden.

AD109, an oral combination of aroxybutynin 2.5 mg and atomoxetine 75 mg, produced a mean apnoea-hypopnoea index treatment difference of −4.0 events per hour against placebo over 26 weeks in 646 patients, with 21.2% discontinuing for adverse events against 3.1% on placebo, and no significant difference on the PROMIS-Fatigue score [s2].

An implant with a 58.2% responder rate and an operation, or a pill with a modest average effect, a fifth of users stopping, and no demonstrated fatigue benefit [s1][s2]. The trials enrolled different populations at different severities and measured different endpoints, so this is a description of the landscape rather than a comparison.

What to watch

Whether the 6-month open-label extension shows the month-7 responder rate holding once controls cross over [s1]. Whether an independently funded trial reproduces the effect. And whether longer follow-up in a larger sample narrows those wide confidence intervals — 45.5% to 70.2% on the primary endpoint leaves a lot of room [s1].

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] 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.
• [s2] 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.