The protocol for this trial and supporting CONSORT checklist are available as supporting information; see Checklist S1
and Protocol S1
This study was an investigator inititiated study sponsored by Lundbeck. Lundbeck was the funder and supplied the study drug and placebo, but had no role in the study design, data collection, data analysis, study termination, data interpretation, writing or the report, or the decision to submit for publication.
Participants were recruited from nursing or residential care homes and acute psychiatric wards in the United Kingdom, from September 2007 to January 2010. The last follow up was in July 2010. Homes and wards were those who agreed to the study across the sites from which we had ethical and local permission. The study population were people with AD and agitation judged by their clinical team to require intervention and referred to the trial.
Inclusion criteria were: a diagnosis of probable AD 
, with a SMMSE score of ≤19, Hachinski Score ≤4, 
being aged ≥45, and a history ≥two weeks of clinically significant agitation (requiring treatment) with a CMAI score of ≥45. Exclusion criteria were: memantine usage in the four weeks before study commencement; use of a cholinesterase inhibitor for <3 months; dose alteration in the two weeks pre-study of any anti-psychotic, antidepressant, benzodiazepine, hypnotic or lithium; use of antiparkinsonian medication; hypersensitivity to memantine; severe renal impairment; epilepsy, history of convulsions or seizure, or receiving any anti-epileptic treatment; concomitant usage of NMDA antagonists such as amantadine, ketamine or dextromethorphan; recent myocardial infarction, uncompensated congestive heart failure, uncontrolled hypertension; severe, unstable or poorly controlled medical illness; and any disability interfering with the participant completing the study as judged by the recruiting physician.
Care home staff supervised patients taking medication. Medication compliance was monitored by discussing with care staff whether participants were willing to take the medication and pill counts for individuals by the research staff and compared to administration records. Adverse events, vital signs, concomitant medication and compliance were assessed at each visit.
If participants refused medication for >3 consecutive days, medication was stopped. The full co-operation of the participant with testing and physical examination was required for the baseline assessment. If, during the trial, a participant was uncooperative with procedures, another attempt was made during the following seven days. If co-operation was not obtained then information was collected from staff only.
Randomisation and Masking
Participants were randomly assigned with equal probability to twice daily memantine 10 mg (titrated in 5 mg increments over four weeks) or placebo. Randomisation used a secure internet based randomisation service independent of the study team. Minimisation was adopted to maintain balance on key confounding variables; centre; age group; sex; dementia (moderate, moderately severe, severe and very severe); and agitation severity (CMAI score <50, 51–55, 56–60, 61–65, 66–70, 71–75 and >75). Since participants, study personnel, clinicians and carers were blind to allocation, no probabilistic element was introduced into the minimisation procedure. Blinding was achieved by using placebo and active drug identical in appearance and taste. During the study, the randomisation code was broken only after withdrawal or completed follow-up. There were eight code breaks: four at the treating clinician’s request after study participation (two active and two placebo); the other four because ofserious adverse events (three memantine and one placebo).
The trial was conducted in accordance with Good Clinical Practice guidelines, the Declaration of Helsinki, the Clinical Trials Regulations and local laws and regulations. We obtained written ethics approval for the study from South East Multi-Centre Research Ethics Committee-REC reference 06MRE01/82 for the trial including our procedures to assess capacity to consent and written and verbal documentation of assent and written documentation of consent.
Referring clinicians initially approached participants and their legal representatives with study information, obtaining verbal agreement to share information with the researchers who then contacted the care home and the patient’s legal representative asking for consent to assess for trial eligibility.
We assessed capacity to give informed consent to participate. Capacity to give consent according to the Mental Capacity Act 2005 was established and documented by a psychiatrist with a professional qualification in psychiatry. When present, written consent was obtained from the participant and legal representatives or next-of-kin were consulted. When capacity was lacking, verbal or written assent was obtained from the participant and documented. Written agreement was obtained from an appointed guardian if they existed, if not from next–of-kin and if there was no next of kin from carers. Participants and their next-of-kin/carers could withdraw participation at any time.
Physicians assessed trial eligibility, reviewed medical histories, recent blood tests, and undertook physical examinations at baseline to exclude agitation caused by co-morbid physical illness. Full Blood Counts, Liver Function Tests, Urea & Electrolytes, Thyroid Function Tests, B12, folate, glucose and cholesterol were measured unless recent (<3 months) blood tests were available and the clinical situation was unchanged. If new or significant abnormalities were discovered the clinician liaised with primary care and requested further investigations.
The CMAI was assessed at baseline and at weeks 2, 4, 6 and 12. It contains 29 items each scored from 1 –7 with one meaning “never” and seven “several times per hour” and is validated to measure agitation.
In addition we tested a number of secondary outcomes: the effect of memantine versus placebo on the CMAI at 12 weeks; at 6 and 12 weeks on Neuropsychiatric Inventory (NPI; 
), Clinical Global Impression Change (CGI-C; 
), Standardised Mini-Mental State Examination (SMMSE; 
) and Severe Impairment Battery (SIB; 
). In addition, we explored a CMAI-based response as a 50% reduction in score between baseline and 6 weeks. We also compared the number of occasions rescue treatment was utilised, and adverse effects.
The NPI is a semi-structured instrument carer rated instrument. All items were rated by the home staff as in the other informant interviews. The 12 domains cover delusions, hallucinations, agitation/aggression, depression/dysphoria, anxiety, elation, apathy, disinhibition, irritability/lability, aberrant motor behaviour, sleep and appetite disturbance. Each item is rated by frequency (score: 0–4) and severity (score: 0–3) and the product is the overall score.
SMMSE is a brief widely used test of cognitive function.
SIB is a cognitive scale designed for severe dementia.
CGI-C is a clinician-rated global measure of change.
To maximise trial retention we used a ‘rescue’ protocol during the titration period only if it was felt that the safety of the participant or caregiver was compromised. This started with non-medication approaches, such as staff reassurance and advice on managing agitation in dementia, using an information sheet and checklist. Then trazodone was used as rescue medication at a dose of 50–150 mg. The protocol was derived from an existing study 
and best practice defined through consensus discussion within the study team.
The sample size was calculated using NCSS PASS and employed a repeated measures approach because the primary outcome is measured at baseline and then weeks 2, 4, 6 and 12. A first order auto covariance structure for the observations was assumed. We pre-specified a clinically important difference of 6 points on the CMAI 
at 6 weeks between the memantine and placebo groups. We estimated a pooled standard deviation (SD) of 16.84 using data from all medication intervention trials using the CMAI. Test-retest scores for CMAI, indicated a correlation of 0.6 between successive observations for an individual patient during treatment. 
With statistical significance level set at 5%, two-tailed tests and 80% power, we required two study groups each with 68 participants. Assuming 20% attrition, this inflated the sample size to 82 per group. During the study, loss to follow-up at 6 weeks was <20% and the sample size was therefore adjusted to 148 participants.
We generated frequencies and presented mean and SD for continuous and proportions for categorical results. The program package R (2.12.2) was used 
The primary analysis was an intention to treat (ITT) analysis; participants were analysed as part of their allocated group irrespective of medication protocol adherence. Linear mixed effects (lme) modeling was used to handle the repeated measurements data. The dependent variables were score at weeks 2, 4, 6 and 12. Group and week were factors and baseline score was used as adjusting covariate. A first order autoregressive model was used for the correlation structure. Model fit was judged by standard residual analyses. In addition bootstrap analyses were performed that confirmed the parametric results.A sensitivity analysis of observed cases was used for the CMAI. An ‘observed case’ is defined as a person who remained in the trial until the assessment date and took medicine according to protocol i.e. did not miss >3 days. A similar approach was taken in the analysis of secondary outcomes. Descriptive statistics were used for safety and tolerability, tabulating adverse events frequency by treatment group.