This systematic review protocol was developed using guidance from the PRISMA Statement . It was peer-reviewed by experts in pharmacology, statistics, and systematic reviews. Our protocol is registered in the PROSPERO database (CRD42012002234) and is closely-related to another systematic review protocol that focuses on cognitive enhancers for AD (CRD42012001948) (Tricco, unpublished paper submitted to BMC SRs).
Studies examining patients diagnosed with MCI using established criteria (for example, Montréal Cognitive Assessment  and DemTect  will be included. Studies must examine cognitive enhancers approved for use for AD in Canada (donepezil, rivastigmine, galantamine, memantine) compared with other cognitive enhancers, memantine or placebo and/or supportive care. These agents are also approved for AD in many other countries, including Australia, United States, and the United Kingdom. Supportive care will include social support, functional assistance, caregiver support, information, education, community service, and other non-pharmacological strategies.
We will include experimental studies (for example, randomized controlled trials [RCTs], quasi-randomized trials, controlled clinical trials), quasi-experimental studies (such as interrupted time series, controlled before and after studies), and observational epidemiology studies (for example, cohort, case–control studies). Inclusion will not be limited by publication status (that is, we will include unpublished material), year, or language of dissemination. Potentially relevant articles not written in English will be translated.
Information sources and literature search
The main literature search will be conducted in the following electronic databases: MEDLINE (OVID interface, 1948 onwards), EMBASE (OVID interface, 1947 onwards), the Cochrane Methodology Register (current issue), Cochrane Central Register of Controlled Trials (CENTRAL, current issue), CINAHL (EBSCO interface, 1980 onwards), and Ageline (EBSCO interface, 1961 onwards). The search strategies will be developed using medical subject headings (MeSH) and text words related to cognitive enhancers for MCI patients.
The electronic database search will be supplemented by searching for difficult to locate or unpublished material (that is, grey literature) using methods (Tricco, unpublished paper submitted to BMC SRs). Briefly, we will search public health and trial registry websites, websites of organizations that produce guidelines, conference proceeding abstracts, Google, key journals, and contact manufacturers to obtain their Scientific Information Packets for the medications. Literature saturation will be ensured by searching the authors’ personal files, scanning the reference lists of included studies and relevant reviews, and contacting researchers and healthcare providers who are active in this field.
The literature searches will be conducted by an experienced librarian (Perrier) and peer-reviewed by another librarian using Peer Review of Electronic Search Strategies (PRESS) . The draft literature search can be found in the Appendix. For simplicity, the literature search was conducted in conjunction with another systematic review on a closely-related topic (Tricco, unpublished paper submitted to BMC SRs). The two reviews will be separated after full-text screening. The results from the literature search will be uploaded to our online SysRev Tool, which will be used for screening the results from the literature search .
Study selection process
We will calibrate the inclusion and exclusion criteria by pilot-testing a random sample of 50 citations resulting from the search. A kappa statistic and the percent agreement will be used to calculate inter-rater agreement for study inclusion between reviewers . If poor to moderate agreement is observed (that is, percent agreement less than 70% or a kappa statistic less than 0.6), the inclusion and exclusion criteria will be revised. All citations and full-text articles will be reviewed by two reviewers independently. Conflicts will be resolved by discussion or the involvement of a third reviewer.
Data items and data collection process
Key stakeholders (for example, patients, healthcare providers, policy-makers) will be engaged to refine the key outcomes. Potential outcomes of interest include:
Cognition: Mini-mental state examination, Goal Attainment Scale, Severe Impairment Battery
Function: Bristol Activities of Daily Living Scale, Caregiver-rated Modified Crichton Scale, Disability Assessment for Dementia, the Interview for Deterioration in Daily living activities in Dementia, Nurses Observation Scale for Geriatric Patients Activities of Daily Living subscale, the Progressive Deterioration Scale
Behavior: Neuropsychiatric Inventory
Global Status: Clinician Interview-Based Impression of Change Incorporating Caregiver Information scale, Clinical Global Impression of Change
Clinical Outcomes: Mortality, Health-Related Quality of Life, Institutionalization, Falls, Balance; Harms (number of adverse events (for example, nausea, vomiting, diarrhea, dizziness, weight loss, hospitalizations, bradycardia), number of withdrawals, number of withdrawals due to adverse events, severity and timing of adverse events); Benefits to caregivers (for example, caregiver stress)
Costs and cost-effectiveness
The data will be abstracted and categorized as study characteristics (for example, study design, year of conduct, setting), patient characteristics (such as type and number of patients, mean age, MCI diagnosis criteria, MCI severity, baseline cognition, co-morbidities), primary outcome results (for example, cognition, function), and secondary outcome results (such as behavior, quality of life, costs, falls, balance, and harms).
We will calibrate the online data extraction form by pilot-testing a random sample of five included studies. Two reviewers will abstract all data independently and conflicts will be resolved by discussion or the involvement of a third reviewer.
We will abstract data based on clinically and methodologically relevant follow-up time periods. These include after 3, 6, 12, and 24 months of follow-up. We will also identify multiple publications reporting data from the same group (that is, companion reports) to ensure that data is not counted twice in meta-analysis. Study authors will be contacted for further information when the data are unclear.
Methodological quality/risk of bias appraisal
We will use the Cochrane Risk of Bias Tool to appraise RCTs  and the Cochrane Effective Practice and Organization of Care Risk of Bias Tool for controlled clinical trials, interrupted time series, and controlled before-after studies . The Newcastle-Ottawa Scale will be used for cohort studies and case–control studies . Summary of findings tables will be compiled to show the relevance and level of evidence across all of the included studies using Grading of Recommendations Assessment Development and Evaluation (GRADE) . The McHarm tool  will be used to specifically examine adverse drug reactions reported in the included studies.
We will first describe our results narratively. We will subsequently conduct meta-analysis using a random-effects model  for each type of study design separately. Clinical, methodological, and statistical heterogeneity will be examined and meta-regression analysis will be conducted if significant statistical heterogeneity is identified (that is, I2 statistic > 60%) . Missing data will be imputed using established methods  and we will examine the effect of the missing data on our results using an established method .
If the data allow, network (such as, indirect comparison) meta-analysis will be conducted to rank treatment efficacy among all the available treatments using the WinBUGS software (MRC Biostatistics Unit, Cambridge, England) . Sensitivity analyses will be conducted to assess the effect of inclusion of trials with high rates of dropouts, missing data imputations, instruments used for the primary outcomes, average adherence between groups, inclusion of observational studies in the network meta-analysis, and use of different priors for variance parameters in the network meta-analysis .