News · Sleep
Nine measures of sleep, none predicted dementia
Researchers writing in Neurology boiled 44 sleep measurements down to nine composites across five US cohorts, then tested them against 1,134 dementia cases. Two tracked with better thinking scores. None tracked with dementia.
Based on a peer-reviewed prospective cohort analysis in Neurology
- Five US cohorts with home sleep studies: 5,958 people for cognition, 5,471 for dementia.
- 1,134 people developed dementia over a median follow-up of 5 to 19 years.
- 44 sleep metrics were reduced to 9 composites, explaining 49% of the original variance.
- Longer, more consolidated sleep and stronger slow oscillation-spindle coupling tracked better cognition.
- None of the nine composites showed a significant association with dementia risk.
The idea that bad sleep helps cause dementia has become close to received wisdom. A large analysis in the journal Neurology set out to find which parts of sleep carry that risk, and came back with an awkward answer: on these measurements, none of them did.
Previously, single sleep measures have been tied to later dementia one variable at a time, which leaves the awkward possibility that the wrong variables were being examined. It remains unclear which aspects are most relevant to cognition and dementia risk, so this team tested the whole of a night at once.
The study did not skimp on the looking. Sleep composites were not significantly associated with dementia risk, and that conclusion rests on nine separate ways of describing a night.
Turning 44 sleep numbers into nine
A modern sleep study produces far more than a duration. The researchers began with 44 sleep metrics drawn from home-based recordings and self-reported habits, then grouped the ones that move together.
That produced nine composites, together explaining 49% of the total variance in the original 44 sleep metrics. They cover the obvious things and the technical ones: sleep quantity and efficiency, sleep fragmentation, N3 predominance, spindle number and duration, REM sleep bouts, respiratory disturbances, and more.
The data came from the Sleep and Dementia Consortium, comprising 5 US population-based cohorts, with everyone free of dementia and stroke at the start.
What did track with thinking
Two composites showed something. Greater sleep quantity and efficiency, meaning longer and more consolidated sleep, went with better global cognition. So did stronger slow oscillation-spindle coupling, a measure of how tightly brief bursts of brain activity align with the slow waves of deep sleep.
Both effects were small. The coefficients were 0.03 and 0.04 per one-unit change on a composite scale, and in each case the lower end of the plausible range sits a hair above zero.
These are the results a headline would normally be built from. They describe test scores measured at a point in time.
What did not track with dementia
Then the part that matters more. The dementia analysis followed 5,471 participants with a median follow-up time of 5-19 years, during which there were 1,134 incident dementia cases.
More than a thousand cases is a substantial number of events. A study with that many outcomes has real power to detect an association if one is there.
No significant associations were identified between the 9 sleep composites and dementia risk.
Not the sleep duration composite. Not fragmentation. Not the breathing-disturbance one. Not the spindle measures that predicted better cognition in the same population.
Why a null result here is worth reading
Null findings get published less often than positive ones, and they get covered even less. This one deserves attention because of what it was equipped to find.
The design is prospective, so sleep was measured before the dementia appeared, which sidesteps the usual objection that early disease disrupts sleep rather than the reverse. The measurements were objective recordings rather than recollection. The follow-up runs to nearly two decades in places.
That combination is roughly what you would want if you set out to prove the sleep-dementia link. It did not produce one.
What the study cannot rule out
A null result is not proof of absence. The composites explain 49% of the variance in the source metrics, which means slightly more than half of the original information was not carried forward, and something could be sitting in the remainder.
Population cohorts also contain relatively few people with severe sleep disorders, so an effect concentrated at the extremes could be diluted into invisibility.
This paper is paywalled, so the authors’ own account of their limitations was not available for this piece, and these caveats are the conservative ones.
Where it leaves things
Aging is the biggest risk factor for dementia, and dementia is not a normal part of aging either. Between those two facts sits a long list of proposed modifiable risks, and sleep has been near the top of it.
The authors do not abandon the question. Their reading is that these composites may be useful in guiding further investigations of sleep-brain health relationships, which is a way of saying the map is now better even though the treasure was not where it was marked.
For anyone reading this over a bad night: sleeping well remains worth doing for the reasons it was always worth doing. This particular promise about it has just become harder to make.
People also ask
Does this mean sleep does not matter for dementia?
No, and that would overread a null result. It means these nine composites, measured this way, in these five cohorts, did not predict who developed dementia. Sleep could still matter through pathways this design cannot see, over timescales longer than the follow-up, or at extremes of disturbance that a population sample contains few of.
What is slow oscillation-spindle coupling?
During deep sleep the brain produces slow waves, and on top of them brief bursts of faster activity called spindles. How precisely those spindles line up with the slow waves is the coupling. It is thought to matter for consolidating memory overnight, and this study found stronger coupling tracked with better cognitive scores, though not with dementia risk.
How were these nine composites chosen?
Not in advance. The researchers fed 44 individual sleep metrics into a clustering method that groups variables travelling together, which returned nine composites explaining 49% of the total variance. That is a data-driven approach rather than a test of pre-specified hypotheses, which makes it good for mapping the territory and weaker as confirmation.
How big were the cognitive effects?
Small. Greater sleep quantity and efficiency carried a pooled coefficient of 0.03 per one-unit change (95% CI, 0.004-0.06) and coupling 0.04 (95% CI, 0.003-0.07). Both lower bounds sit very close to zero. These are population-level slopes on a composite score, not differences anyone would notice.
Should I stop worrying about my sleep then?
That is not the message, and this is general information rather than medical advice. Sleep has well-established links to mood, metabolic health, accident risk and daytime function regardless of what it does to dementia risk. What this study narrows is one specific claim, not the case for sleeping well.