News · Sleep
Unsteady breathing in sleep tracked death risk
Across 43,611 overnight sleep studies, researchers writing in the journal SLEEP tested a measure of how steadily people breathe at night and found it tracked death risk in two hospital cohorts, though not in the community one.
Based on a peer-reviewed analysis of 43,611 sleep studies in SLEEP
- 43,611 overnight sleep studies from two hospital cohorts and one community cohort.
- A new stability score was built from chest and abdominal effort, not from counting breathing pauses.
- Each standard deviation of instability carried 1.38 times the death risk in one hospital cohort.
- The mortality link did not appear in the community cohort, which is the finding's weak point.
- Discrimination for disease labels ran 0.509 to 0.642, where 0.5 is a coin toss.
Sleep medicine has spent decades counting. Sleep apnea is a common disorder that causes your breathing to stop or get very shallow, and the number that follows a person through their care is a tally of how often that happened per hour.
A study in the journal SLEEP asks whether the counting misses something. Its authors built a continuous measure of respiratory effort instability out of the same overnight recordings and tested whether steadiness carries information the tally does not.
Reading the rhythm instead of the events
The measure is deliberately simple in concept. It was derived from abdominal and thoracic effort envelopes, which is to say the rise and fall of the chest and belly that a sleep study already records all night.
Breathing pauses can last from a few seconds to minutes. They may occur 30 times or more an hour. Counting those is what a diagnosis rests on. But between the countable events there is a texture, steady in some sleepers and ragged in others, and the conventional number throws it away.
The researchers analyzed 43,611 overnight polysomnography records from two clinical cohorts and one community cohort, which is an unusually large base for this kind of question.
What the stability score added
The first test was whether the new measure is simply the old one wearing different clothes. It largely was not. The index tracked comparator metrics while remaining partly nonredundant, and in mutually adjusted models it remained associated with periodic breathing, apnoea counts, arousals and oxygen dips even with all of those in the model together.
It also behaved the way sleep physiology says it should. Within severity strata, instability was lowest in deeper NREM sleep and higher in lighter sleep and REM sleep, which is the pattern anyone who has watched a sleep trace would predict. A new measure that behaves plausibly is more believable than one that does not.
The mortality signal, and where it vanished
Then the survival analysis, and this is where the article has to slow down.
Instability was associated with all-cause mortality after adjustment for age, sex, and comparator sleep metrics in both hospital cohorts. In one, each standard deviation carried about 1.38 times the risk. In the other, about 1.10.
In the community cohort it was not.
That single failure to replicate does more to define this finding than either positive result. Hospital sleep laboratories see people who were referred because something was already wrong. A measure that sorts risk in that population and then goes quiet in a sample of community-dwelling older men may be detecting illness rather than sleep.
The disease analysis lands in similar territory. The index provided modest discrimination across nine conditions drawn from medical records, with scores running from 0.509 to 0.642. The bottom of that range is a coin toss.
What an abstract cannot tell you
This paper is paywalled, so the authors’ own account of their limitations was not available for this piece and the caveats here are the conservative ones.
The disease comparisons matched a night of sleep against diagnosis labels from the same period, which cannot establish that unstable breathing came first.
The cognitive results were openly inconsistent: cognitive associations were cohort dependent, holding in one cohort and not reliably elsewhere. Reporting that plainly is to the authors’ credit, and it is also a reason to wait.
Why it is still interesting
Doctors diagnose sleep apnea based on medical and family histories, a physical exam, and sleep study results, and the sleep study produces far more data than the single number extracted from it. Most of that recording is currently discarded.
The argument here is not that the tally is wrong. It is that a night of breathing has a shape as well as a count, and the shape may turn out to carry its own information. Whether it survives outside a hospital sleep lab is the question that has to be answered next.
People also ask
What does breathing stability measure that apnoea counts do not?
The standard number is the apnoea-hypopnea index, which counts how many times an hour your breathing stops or goes shallow. It is a tally of events. The Breathing Stability Index instead reads the effort your chest and abdomen make across the whole night and scores how erratic it is. Two people can log the same event count with very different underlying rhythms.
How strong was the mortality link?
Moderate and inconsistent. Each standard deviation of instability carried 1.38 times the risk of death at Massachusetts General Hospital (95% CI, 1.23-1.54) and 1.10 at Beth Israel Deaconess (95% CI, 1.02-1.19), after adjusting for age, sex and the conventional sleep measures. In the MrOS community cohort of older men it was not associated with mortality at all.
Why does the community cohort mattering so much?
Because hospital sleep clinics see a sicker, more selected population than the general public. A measure that separates risk among patients already referred for a sleep study may simply be picking up how unwell they were to begin with. Failing to replicate in a community sample is the result that keeps this a research tool.
What does an AUROC of 0.6 actually mean?
It is a measure of how well a score separates two groups, where 0.5 is a coin toss and 1.0 is perfect. The range reported here, 0.509 to 0.642, spans from essentially no discrimination to modest. That is a signal worth investigating and nowhere near a diagnostic test.
Can I get this measured?
No. The index was derived from research analysis of sleep-study recordings and is not something a clinic reports. Anyone with symptoms of disrupted breathing at night should raise it with a doctor and be assessed by the standard route rather than waiting for this.