← Back to Evidence MedBug insights
24 Oct 2026 · Breathing control
Loop gain

Loop gain, without a wire in sight

A sensor under the mattress feels your breathing, an AI reads it, and a physiology model turns it into loop gain: how steady your breathing control is, night after night.

Duc Phuc Nguyen & the Flinders Sleep Health team
Flinders Health & Medical Research Institute · Flinders University
0.86
Correlation with the standard lab method (81 studies)
72%
Same low, moderate or high group as the lab method
91%
Home sleep periods with a usable result
6.5%
False alarms, 7-night average (23% for one night)

The problem with one night in a lab

Loop gain sounds abstract, but it explains something simple: why breathing pauses in sleep apnoea keep coming back, night after night. It describes how strongly your brain’s breathing control reacts when your breathing changes, and until now, measuring it meant a night in a lab with sensors on your face, chest and body.

Think of a thermostat that overshoots. After a pause, an over-sensitive control system makes you breathe too hard, which drops carbon dioxide too far, which switches breathing off again. A high loop gain keeps that cycle going, and it helps predict who responds to surgery, mouth-guard devices, oxygen and some medicines. But sleep apnoea also changes from night to night, so one lab night can mislead.

“Loop gain can be estimated without body-worn sensors or scored events and tracked across nights at home.”Phuc Nguyen, Chief Engineer

Teaching a mattress to estimate it

A thin sensor under the mattress picks up the vibrations of every heartbeat and breath. An AI model, pretrained on a large dataset and then fine-tuned on just 13 adults wearing gold-standard airflow and oesophageal-pressure sensors, turns them into estimates of breathing and breathing effort. A standard physiology model of the body’s breathing control then calculates loop gain from those signals.

The AI never guesses loop gain directly, and nobody has to hand-score breathing pauses or awakenings.

How it works

  • One sensor under the mattress feels heartbeat and breathing vibrations
  • AI model pretrained on a large dataset, fine-tuned on 13 adults
  • Physiology model converts breathing and effort into loop gain
  • No wires, no mask, no hand-scoring

One night in the lab

Compared with the standard sleep-study method (called PUP; Terrill et al., 2015; Sands et al., 2018) in 81 independent laboratory studies, the mattress estimates ranked people very similarly (r = 0.86). Sorting everyone into low, moderate or high loop gain, the two methods agreed on 72% of nights and never differed by more than one group.

Overnight loop gain traces comparing PUP and non-contact mattress methods
One night, two estimates. Top: sleep stages and scored events. Middle: loop gain across the night from PUP (grey) and the non-contact mattress method (coral), with low / moderate / high bands. Bottom: a zoomed window of the oscillatory response used to estimate loop gain.
One thing to know: the standard method is itself an estimate rather than a direct measurement, and the mattress values sit on a slightly narrower scale. So this shows agreement between two methods, not proof of accuracy.

Then home, for many nights

Tested at home in 64 people, including children, adults with motor neurone disease, and soft, water and air mattresses, about nine in ten main sleep periods gave a usable result. In adults, readings were consistent from night to night, which makes averaging possible.

Compared with a person’s own multi-night average, a single night wrongly flagged 23% of people as moderate or high loop gain at a fixed 10% missed-case rate. More nights fixed most of it:

False-alarm rate for detecting moderate–high loop gain, at a fixed 10% missed-case rate, versus each person’s own multi-night average (bars scaled to 30%).

How it stacks up

Agreement with the standard method for other ways of estimating loop gain without a full lab study:

ApproachNeedsResult
This studyMattress onlyr = 0.86
Orr 2018Home test sensorsr ≈ 0.47–0.66
Schmickl 2022Clinical datar ≈ 0.48
Dutta 2021PSG variables40% accuracy (chance = 33%)
Different cohorts and measures, so indicative only.

1The cohorts

13 adults to build the model (92% men), 81 independent lab sleep studies to test it, and 64 home participants aged 2 to 80 across seven protocols.

2Open question

The mattress numbers sit on a different scale and agreed on only 2 of 6 high-loop-gain nights, so cut-offs need resetting and direct measurements are needed.

Three things worth remembering

01

Why it matters

Loop gain could be measured at home over many nights, a step toward matching treatment to the cause of each person’s apnoea.

02

Mattress agreement

Across 81 lab nights it ranked people similarly to the standard PUP method (r = 0.86), with no wires on the body.

03

What’s next

Check against direct measurements, set mattress thresholds, and test whether repeated home readings predict treatment response.

Loop gain from under-mattress sensing — research summary
Flinders Health & Medical Research Institute, Sleep Health
Full manuscript: not yet peer reviewed. This page is a plain-language research summary only.

Research-stage summary. Not a medical device.
← Back to Evidence · MedBug home