Six sessions, a manual and a psychology graduate: what DREAMS START can and cannot show

Sleep disturbance in dementia at home is where the two errors of this field meet in one room. Either nothing is offered, because broken nights are read as ageing, or a sedative is, and the report cites a review of melatonin and other drugs that found no evidence for any pharmacological treatment, while the harm in older adults is real. The National Institute for Health and Care Research has now published the full project report of the first adequately powered trial of a manualised behavioural alternative. It beat usual care on the outcome it registered. That outcome is the carer's account of the night.

What the six sessions contain, and who runs them

The protocol is written to be reproduced. Six sessions are delivered to the family carer, alone or together with the person with dementia, weekly to fortnightly across roughly three months. The content is information about sleep and dementia; de-arousal in the evening, meaning relaxation, bedroom comfort and no caffeine or alcohol before bed; adaptive stimulus control, in practice keeping a bedtime routine; daytime behavioural activation and light exercise to hold alertness and cut napping; increasing exposure to light; attention to the carer's own sleep; and a tailored action plan the family keeps using after the sessions stop. Each session teaches a different relaxation strategy. Actigraphy recorded at baseline is fed back into the plan to personalise it. Carers keep the manual and the equipment, a light box and a watch, and text-message reminders prompt them to use the strategies and fill in the record forms.

Who delivers it is the point of the design. Not clinicians: supervised graduates without clinical training. They received two days of interactive training on dementia and sleep-wake regulation, empathic listening, facilitating behaviour change, reading actigraphy, using supervision and working with families; each then role-played the whole manual and was signed off as ready by one of three team clinicians before being allocated a family. Clinical supervision was mandatory, fortnightly, in small groups, run by clinical psychologists, with individual supervision available on request. The staffing arithmetic is sobering and the report gives it plainly: 76 people were trained and signed off, only 49 ever delivered anything, and those 49 saw a mean of 3.6 dyads each (standard deviation 3.12, range 1 to 15).

Delivery held up anyway. Because the trial ran through the pandemic, 77 participants (43.8%) were seen in person, 31 (17.6%) by video call, 28 (15.9%) by telephone, 33 (18.8%) by a mix of remote and in-person sessions and 7 (4%) by a mix of video and telephone; the report found no evidence that effects varied by modality. Of surviving intervention participants, 149 of 180 (82.8%) attended at least four of the six sessions and 142 attended all six. One session per participant was recorded where possible, available for 143 of 188 (76.0%), and mean fidelity was 95.4% (standard deviation 0.08), with each of four therapeutic process measures at a median of 5 out of 5 (interquartile range 5 to 5). Eligibility was deliberately wide: any dementia type and any severity, a Sleep Disorders Inventory score of at least 4, and living at home with someone present at night.

Who reported every number, and what nobody measured

This belongs in the main text, not in a footnote, because it governs how far the result can be carried. The primary outcome is a carer-rated instrument. The carer is the person who sat through six sessions, moved the lamp, rebuilt the evening routine and hoped it would help. Masking was applied to the research staff who collected the data and to procedures designed to keep them from learning allocation. It was never applied where it would have mattered most. The report states this without hedging: participants and facilitators could not be masked to group allocation, and unmasked carers reported proxy outcomes, increasing the risk of bias.

That leaves the objective side empty. Seven-day actigraphy with a sleep diary was listed in the registry among the secondary outcomes and was collected, but usable data fell from 267 of 377 dyads (70.8%) at baseline to 132 (35%) at 4 months and 84 (22.3%) at 8 months, so the planned mechanism analysis was abandoned. The report's limitations section describes the absence of actigraphy "as a RCT outcome measure" as a decision taken after the feasibility trial, where actigraphy had proved unfeasible as a primary outcome. The report states the plainer version itself: the team was "unable to collect sufficient data for pre-specified actigraphy data analysis exploring underlying mechanism of action". The wireless EEG substudy is the blunter version of the same problem: of 53 dyads approached, 1 consented, the participant pulled the headband off repeatedly across two nights, and channel data quality ranged from 22% to 74%. There is no direct measure of sleep anywhere in this trial, and the authors say so themselves.

Two things were counted independently of anyone's opinion. Deaths: 17 (9%) in the intervention arm and 17 (9%) under treatment as usual, unrelated to the intervention. Harms and side effects showed no difference between arms at follow-up. On safety, therefore, the evidence is not carer-rated, and it is reassuring.

The numbers, the money, and what held at two years

Among secondary outcomes at 8 months, three reached significance: neuropsychiatric symptoms -4.54 (95% CI -8.71 to -0.37), carer sleep on the Sleep Condition Indicator 1.84 (95% CI 0.32 to 3.35) and carer anxiety -0.86 (95% CI -1.71 to -0.01). That last upper bound is -0.01, significant by the thinnest available margin, on a self-report by the person who received the intervention. No other secondary outcome differed significantly, though the report notes that summary scores consistently favoured the intervention.

A companion paper in Alzheimer's and Dementia (doi 10.1002/alz.71274) reports the 2-year follow-up: 177 of the 377 dyads (46.9%) were followed up, and the adjusted mean difference on the Sleep Disorders Inventory was -5.40 (95% CI -9.14 to -1.67), p = 0.005. Fewer than half the sample remained, and the informant was still the carer.

On cost, the total mean cost of the intervention per dyad, counting sessions, light box, watch, facilitator training and supervision, was £574.38. The incremental difference in health and care costs was £59 less per dyad (95% CI -£5168 to £5050), and £116 less when wider costs were included (95% CI -£5769 to £5536), with a 78% probability of being cost-effective at a £20,000 threshold and no significant difference in quality of life. Read those intervals before repeating the headline: they stretch from roughly £5800 saved to roughly £5500 spent. The report describes the same 78% as "a moderate probability" in one section and "a high probability" in another, and its take-home message states the £116 as a fact without the interval. I followed the interval.

The core did not drift, and it is worth saying because in this field it often does. Registration was prospective: ISRCTN13072268, registered 30 September 2020, enrolment recorded from 1 February 2021, with recruitment from 24 February 2021. The registry names the Sleep Disorders Inventory at 8 months as the primary outcome, which is exactly what was reported. The registered target was 370 dyads and 377 were randomised. One entry did move on paper: the registry corrected the AUDIT-C exclusion threshold from 5 or more to 8 or more, nine months after recruitment closed, and both public documents carrying the higher threshold are dated after recruitment closed, so nothing published places that threshold in force while families were being recruited.

What changes in the room

What you can act on is this. For a family at home with dementia and broken nights, there is now a manualised, six-session alternative to a sedative that beat usual care on its registered primary outcome, held at 4 months and 8 months and, on a halved sample, at two years (reported in the companion paper), cost about £574 per family, produced no harm signal, and could be run by a trained non-clinician over video or telephone. Its ingredients are not exotic. Light, daytime activity, an evening routine, de-arousal before bed, the carer's own sleep, a written plan. NICE guidance had already suggested considering that package. Nobody had shown that it works when it is delivered as a protocol, at scale, by people who are not therapists.

What you should not do with it is tell a family that their relative is objectively sleeping better. The trial does not know that. What it shows is that after six sessions the carer reports fewer and less troubling night-time events for months afterwards, sleeps better herself and is slightly less anxious. That is a legitimate target in its own right, since sleep disturbance is associated with carer depression, burden and care home admission, but it is a different claim, and the difference becomes concrete the moment the family asks whether the sedative can be stopped. There was also no active control arm, so six sessions of contact with a supervised, attentive person remains an untested component of the effect; the authors argue against that explanation on the grounds that the effect was largest on sleep-related measures, which is an argument rather than a finding.

Note finally who was left out. People with dementia living alone were excluded, because someone had to be present at night, and the report names adapting the intervention for them as an open research question. The protocol with evidence behind it is the one for households where the nights are witnessed. For the patients whose nights nobody sees, there is still nothing.