The hour is usually somewhere between two and four. A hallway light goes on, a coat is pulled from the hook because it must be morning somewhere, and a daughter who has to be at work in five hours stands in a doorway deciding whether to argue. Nearly every family living with dementia knows this shift. What almost none of them know is how little formal evidence exists behind whatever their clinician suggests next.

A review published in Medicine on August 8, 2025 by Mohammed M. Aldurdunji of Umm Al-Qura University put a number on it. Searching ClinicalTrials.gov on July 21, 2024 for interventional studies that specifically targeted sleep disturbance in people diagnosed with Alzheimer's disease or dementia, that were completed, and that had results posted, he found nine. Together they enrolled 1,139 participants. Three tested nonpharmacological approaches; six tested drugs, among them prolonged-release melatonin (Circadin), zolpidem CR, lemborexant, mibampator, piromelatine and suvorexant. Four were phase 2 studies and one was phase 3. The results, as the review describes them, pull in different directions: some trials improved sleep measures, others did not.

Up close, the ledger looks thinner still. One of the nonpharmacological entries, a nursing home staff huddling programme called LOCK (NCT04533815), was a single-group study of 23 residents run over 15 weeks; the review reports significant gains in total sleep time and sleep efficiency on actigraphy (P <.01), with caregivers describing less severe nighttime disturbance. That is 23 people, with no control arm. Elsewhere, a phase 2 trial of mibampator, an AMPA receptor potentiator, randomised 132 outpatients with Alzheimer's to 3 mg or placebo for 12 weeks; its primary target was agitation and aggression, not sleep, and it showed no significant advantage over placebo. Sleep, in other words, was sometimes a passenger in a trial designed to measure something else.

One registry on one day is not the whole world, and honest reporting has to say so. A systematic review and meta-analysis by Crowley and colleagues at University College Cork, published in the International Journal of Environmental Research and Public Health on June 18, 2025, searched Medline, CINAHL, PsycINFO and Cochrane CENTRAL from inception to October 3, 2023 and found 144 controlled trials involving 13,471 participants, median 50 per trial: 95 nonpharmacological, 46 pharmacological, three both. That literature has its own problem, which is arguably worse than scarcity. The Cork team counted 287 distinct sleep outcome measures, 205 of them used in a single trial only, and no single measure used in even half the studies. Evidence that cannot be pooled is evidence that struggles to change practice.

The gap also swallows good work. DREAMS START, a phase 3, single-masked, multicentre randomised trial reported in The Lancet Healthy Longevity in October 2024, randomised 377 dyads of people with dementia and sleep disturbance and their family carers between February 24, 2021 and March 5, 2023 to a six-session manualised sleep intervention delivered by non-clinically trained graduates, or to usual treatment. At eight months, Sleep Disorders Inventory scores were lower in the intervention arm, an adjusted difference in means of -4.70 (95% CI -7.65 to -1.74; p = 0.002). It was registered with ISRCTN, not ClinicalTrials.gov, and published after the registry search closed, so it sits outside the nine entirely. The counting method, not the science, decides who is in the room.

For a family, the practical shape of this is uncomfortable but useful to know. The 2020 Cochrane review of drug treatments for sleep problems in dementia found nine eligible randomised trials (melatonin in five, n = 222; trazodone in one, n = 30; ramelteon, n = 74; suvorexant and lemborexant in two, n = 323), and noted that only four of them systematically assessed adverse effects. The suvorexant trial that anchors much of the drug evidence (NCT02750306, run at 35 centres in eight countries from May 2016 to September 2018) tested four weeks of double-blind treatment. Four weeks. Dementia lasts years.