The Sleep Evidence, Part 4: Does Treating a Sleep Disorder Lower Dementia Risk?

The last post covered three sleep disorders and how each relates to dementia risk. Two of them, apnea and insomnia, can be treated. So the next question is whether treating them protects the brain.

Nobody has run the trial that would answer it. What we have instead is observational data, some mechanism, and a handful of randomized trials that get cited as proof it does not work, when most of them were never built to detect whether it does.

Obstructive sleep apnea

People treated with CPAP are diagnosed with dementia less often. In a study of 53,321 older adults on Medicare, treatment was linked to fewer new Alzheimer’s diagnoses over three years, and the link was stronger in people who used the machine consistently. A ten-year study found faster cognitive decline in untreated patients. A review of eleven studiesfound nine pointing the same protective way.

The randomized trials look worse for treatment, and this is where most accounts get it wrong.

Only APPLES was actually built to measure cognition, and sized properly to do it. It randomized 1,105 patients to real or sham CPAP, found a small difference in executive function at two months, and by six months that difference was gone. That result counts against treatment and should be taken seriously, though the investigators themselves argue it is a limited one.

Their reasoning applies to nearly every trial in this post, so it is the part to understand. Three things have to line up before a trial can detect a benefit like this one. It has to run long enough for the benefit to show up, and dementia builds over decades while these trials ran for months. It has to measure with something sensitive enough to catch early change, and several used brief screening tools like the MoCA, which were designed to flag impairment that is already obvious. And it has to enroll people who have room to decline, because in someone with normal cognition there is no deficit for treatment to fix, so the test hits its ceiling. Miss any one of the three and an empty result tells you about the study, not about the treatment.

The other trials show what happens when you change one of those. A trial across five hospitals in Shanghai found nothing on the MoCA, its main outcome, but the brain imaging it had planned in advance told a different story. Cortical thickness improved, and so did connectivity in the default mode network, the system that stays active when the brain is at rest and that degrades early in Alzheimer’s. Both held at twelve months. Cognition did not improve alongside them, so this shows treatment changing brain structure rather than preserving thinking. A meta-analysis of eleven short trials pooled whatever cognitive tests each study happened to use and found one significant result, on a timed test of mental flexibility.

The observational side has a matching problem, and it deserves the same scrutiny. Researchers call it healthy adherer bias. People who use CPAP every night tend to be healthier in other ways: better educated, more physically active, more engaged with their own care. Each of those lowers dementia risk by itself. In one cohort the treated group was already younger and scoring higher before treatment began. So part of what looks like a treatment effect is really the kind of person who sticks with treatment, and the true benefit is smaller than the raw numbers suggest.

There is also a newer option worth knowing about. Tirzepatide, one of the weight-loss drugs, reduced both the severity of the apnea and the drops in blood oxygen that the last post pointed to as the damaging part. Nobody has measured what it does to cognition yet.

Chronic insomnia

Cognitive behavioral therapy for insomnia

One properly sized trial has tested insomnia treatment against cognition and Alzheimer’s markers. Its headline result is not the interesting part.

SIESTA randomized 200 cognitively normal older adults to cognitive behavioral therapy for insomnia or to an active control, then followed them for a year. Neither cognition nor amyloid burden improved in the treatment arm.

Most accounts stop right there. The investigators also checked whether the therapy had changed anyone’s sleep architecture, and it had not. Slow wave sleep did not go up in either group.

Slow wave sleep, the deep stage, is the whole reason anyone expected a brain benefit here, because it drives the overnight clearance described in the last post. If the therapy never increased it, then the trial never touched the mechanism it was built on. What it tested was one course of therapy that happened not to increase deep sleep, which makes this a result about the program rather than about the idea behind it.

The therapy did do what it is built for. Insomnia severity dropped substantially and people reported sleeping better, which is reason enough to treat insomnia on its own. What nobody has done is test it in people with existing cognitive impairment, where there would be something to preserve.

Hypnotics

CBT-I is the recommended first-line treatment. Medication is the other route, and it is what most people are actually taking.

Benzodiazepines and Z-drugs. For this older class, which includes zolpidem, the observational evidence contradicts itself. One carefully studied cohort found no link to dementia at all. A large national cohort found a harmful link that disappeared once the analysis accounted for how sick people already were. The problem underneath both is what researchers call confounding by indication. People who get prescribed sleeping pills differ from people who do not, both in how they sleep and in their general health, and each of those differences tracks dementia on its own. No randomized trial has sorted it out.

What is settled is that these drugs cause falls and next-day grogginess in older adults, which is reason enough for caution regardless of the dementia question.

Dual orexin receptor antagonists. This newer class works by turning down the brain’s wakefulness signal rather than sedating it. Suvorexant and lemborexant are the two approved in the US.

In a randomized trial of 38 healthy adults, a single dose of suvorexant lowered phosphorylated tau in the spinal fluid by 10 to 15 percent. That is one of the proteins that accumulates in Alzheimer’s. Amyloid fell too, but only briefly, and across the whole night the change was not significant.

A mouse study compared the two classes head to head, in animals bred to develop tau disease. Lemborexant prevented tau from building up and protected the brain against shrinkage. Zolpidem, included specifically as a comparison drug that also promotes sleep, increased deep sleep by about the same amount and delivered none of that protection. Since both drugs delivered the sleep and only one delivered the benefit, whatever was protecting those brains was not the extra sleep itself.

Two limits belong with that finding. The protection appeared in male mice and not female ones, and no equivalent study exists in people. One trial is underway, testing whether suvorexant slows the buildup of Alzheimer’s markers.

REM sleep behavior disorder

There is almost nothing to report here, and the emptiness is itself the point.

Clonazepam and melatonin reduce the dream enactment itself and lower the risk of injury, both on low-certainty evidence. Every one of those studies measured dream enactment and nothing beyond it.

No study of any design has measured whether treatment changes the roughly six percent yearly rate at which these patients go on to develop Parkinson’s disease or Lewy body dementia. Cognition shows up in this literature only as a side effect to watch for, never as a goal, which is striking given that clonazepam is a long-term benzodiazepine prescribed to people already at high risk of dementia.

So the silence is not a treatment that was tried and failed. It is a question nobody has gotten around to asking, and answering it would not be expensive, because the patient groups already exist. One cohort followed 1,280 people for up to nineteen years and never entered treatment into the analysis.

What this adds up to

Nobody has randomized people to treatment and then followed them long enough to see who develops dementia. The observational signal for apnea points toward benefit, though part of it is the kind of person who sticks with treatment. The trials that came back empty mostly measured the wrong thing, and the one that used sensitive imaging favored treatment. The orexin drugs have a plausible mechanism and some human biomarker data, and no outcome data at all.

None of that changes the decision in front of anyone. Treating apnea relieves daytime sleepiness. Treating insomnia restores sleep and daytime function. Both of those are well established, and neither one waits on the dementia question being settled.

References

1. Dunietz GL, et al. Obstructive sleep apnea treatment and dementia risk in older adults. Sleep. 2021;44(9):zsab076. doi:10.1093/sleep/zsab076
2. Kaufmann CN, et al. CPAP treatment and ten-year cognitive trajectories. Alzheimers Dement. 2026;22:e71533. doi:10.1002/alz.71533
3. Shieu MM, et al. Positive airway pressure and cognitive disorders in adults with OSA: a systematic review. Neurology. 2022;99(4):e334-e346. doi:10.1212/WNL.0000000000200383
4. Kushida CA, et al. Effects of CPAP on neurocognitive function (APPLES). Sleep. 2012;35(12):1593-1602. doi:10.5665/sleep.2226
5. Lee MH, et al. Impact of sleep apnea on perivascular diffusivity and cognition. Am J Respir Crit Care Med. 2025;211(12):2382-2392. Link
6. Durtette A, et al. Impact of CPAP on cognitive functions: systematic review and meta-analysis. Sleep Med. 2024. doi:10.1016/j.sleep.2024.08.019
7. Malhotra A, et al. Tirzepatide for the treatment of obstructive sleep apnea and obesity. N Engl J Med. 2024;391(13):1193-1205. doi:10.1056/NEJMoa2404881
8. Siengsukon CF, et al. The impact of CBT for insomnia on cognitive performance and amyloid beta. Alzheimers Dement. 2026;22:e71591. doi:10.1002/alz.71591
9. Lucey BP, et al. Suvorexant acutely decreases tau phosphorylation and Aβ in the human CNS. Ann Neurol. 2023;94(1):27-40. doi:10.1002/ana.26641
10. Parhizkar S, et al. Lemborexant ameliorates tau-mediated sleep loss and neurodegeneration in males. Nat Neurosci. 2025;28(7):1460-1472. doi:10.1038/s41593-025-01966-7
11. Postuma RB, et al. Risk and predictors of dementia and parkinsonism in idiopathic REM sleep behavior disorder. Brain. 2019;142(3):744-759. doi:10.1093/brain/awz030

How I make this. I choose the topic, find the research through Google Scholar, PubMed, and AI tools like OpenEvidence, and decide which findings belong in the post. AI summarizes the full papers and writes the first draft. Those summaries get audited back against the sources. I read the papers to the depth each claim needs, then check the draft against them and direct the revisions. On average five to seven hours of my own time per post. Corrections welcome.

General education, not medical advice. This describes research findings across groups, not what applies to any one person. Do not start, stop, or change a medication based on it, and note that stopping a benzodiazepine abruptly can be dangerous. Talk to a clinician who knows your history.