Serfaty 2002.

Methods	Randomised, 2‐period cross‐over design
Participants	Number of participants: 44 randomised, 25 completed study Country: UK Setting: 16 participants resident in nursing home, 4 in hospital, 5 at home (25 study completers) Diagnosis: dementia (DSM‐IV criteria). 35/44 (80%) of those randomised and 21/25 (84%) of those completing the study had AD or mixed (AD + vascular) pathology Sleep‐related inclusion criteria: sleep disturbance identified by the main carer, defined as shouting, agitated behaviour, wandering, or both, on at least 2 nights/week Gender: 16 women, 9 men (25 study completers) Age: 84.2 ± 7.6 years (25 study completers) Severity of dementia: MMSE 13.4 ± 8.5 (25 study completers) Other information: 9/25 were regularly taking other sleep medication (stable for at least 4 weeks prior to trial entry)
Interventions	Duration of treatment: 1‐week baseline period, then 2 treatment periods of 2 weeks, each followed by a 1‐week washout period Oral 6 mg melatonin slow‐release, or placebo, at usual bedtime
Outcomes	Outcomes measured for 3 nights at each of: baseline, the end of both treatment periods, and the end of both washout periods. Outcome was the mean of the 3 measurements at each time point 'Main' outcomes (measured with wrist actigraphy): sleep onset time wakening rest period sleep efficiency 'Subsidiary' outcomes: carers' daily diary recordings of bedtime, sleep onset time, estimated total sleep, wake time, and comments on sleep quality Sleep Evaluation Questionnaire (VAS) Likert scales to assess the quality and duration of sleep for participants and carers cognitive function (MMSE)
Notes	Pill count was used to assess compliance: 4.9% doses missed in first treatment period, 10.6% in second; no differences between placebo and melatonin groups. Paper reported "no adverse effects", but this was not a listed outcome and it is not clear how data were gathered. 308/352 patients who met inclusion criteria had to be excluded (207 due to absence of suitable carer, other reasons not given) Results were presented as medians and interquartile ranges Analysis method was not described in any detail Non‐commercial funding
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Treatment order was random and determined individually by a computer‐generated algorithm as soon as consent was obtained."
Allocation concealment (selection bias)	Low risk	As above
Blinding of participants and personnel (performance bias) All outcomes	Low risk	"Both the pharmacist dispensing the trial medication and the researchers were blind to the treatment received. The code for treatment allocation was only broken once the trial was completed." "Melatonin and placebo were identical with respect to preparation and packaging." Blindness of participants, carers, and researchers was rated with a VAS. None of the participants could remember taking medication. Neither carers nor researchers could distinguish between treatments
Blinding of outcome assessment (detection bias) All outcomes	Low risk	As above. Main outcomes were actigraphic (objective)
Incomplete outcome data (attrition bias) All outcomes	Unclear risk	10/44 randomised participants excluded from analysis because no actigraph data. Further 9 participants excluded due to insufficient actigraph data (7), stroke (1), or poor compliance with medication (2). Results reported on only 25/44 randomised. (ITT analysis conducted on 34 participants, but not reported: "as this did not alter the results"). In main analysis, group median baseline scores used in 4 cases, and LOCF in 7 cases.
Selective reporting (reporting bias)	High risk	Sleep results reported (total time asleep, number of awakenings, and sleep efficiency) differed from those described in methods section. Compliance with diary sheets poor and data not analysed. Participants' sleep quality by VAS and cognitive function reported as "unchanged with melatonin", but no data presented. Carers' sleep not reported because only 5 participants lived with informal carer.
Other bias	Unclear risk	Authors reported conducting an analysis that excluded carry‐over effects on sleep outcomes. "In order to detect carry‐over effects the ‘summed’ scores for each participant were compared across the two groups (Mel/Plac and Plac/Mel; Everitt 1994). No significant carry‐over effects were observed."