ITEM	DEFINITION
Design‐related characteristics*
Concealment of allocation (avoiding selection bias)	Guidance from the Cochrane Handbook for Systematic Reviews of Interventions will be used to judge bias related to sequence generation and concealment of allocation using the 2 Cochrane 'Risk of bias' items (Higgins 2011). From these, the statistician will derive a single variable to be used in the stratified analysis: allocation concealment will be judged at low risk of bias if the investigators responsible for patient selection were unable to suspect before allocation which treatment was next. Concealment will be downgraded to high risk of bias if there is evidence of inadequate sequence generation (Rutjes 2012)
Blinding of patients and personnel (avoiding performance bias)	Low risk of bias will be judged: ‐ if a credible sham procedure was used; or if a placebo supplement or pill was used that was reported to be identical in appearance to the experimental intervention and the specific outcome or group of outcomes is/are likely to be influenced by lack of blinding ‐ if blinding is absent or suboptimal and the specific outcome, such as mortality, is not likely to be influenced by lack of blinding
Blinding of outcome assessment (avoiding detection bias)	For self‐reported/partner‐reported outcomes: Low risk of bias will be judged if self‐report outcomes were assessed AND blinding of patients was considered adequate AND there was no information to suggest that there was an investigator involved during the process of outcome assessment; OR if blinding of investigators performing the outcome assessment was reported AND an attempt to blind patients was reported For other outcomes: Outcome assessment was considered to be blinded if outcome assessment was reported to be blinded
Statistical analyses (avoiding attrition bias)	For continuous outcomes: Low risk of bias will be judged: ‐ if at least 90% of the patients randomised were analysed AND the difference in percentage of participants not analysed was 5% or lower across trial arms ‐ for trials using imputations to handle missing data: the percentage of participants with missing data did not exceed 20% AND the difference in percentage of participants with imputed data was 5% or lower across trial arms AND applied imputation methods were judged to be appropriate. Multiple imputation techniques will be considered appropriate, simple methods such as 'last observation carried forward' or 'baseline carried forward' will be considered inappropriate For binary outcomes of rare events: Low risk of bias will be judged if the event rate is low (e.g. incidence of dementia) AND at least 95% of the patients randomised were analysed AND there is no evidence of differential reasons for missing data that may alter the estimate AND the rate of missing data does not exceed the expected event rates For binary outcomes of non‐rare events: Low risk of bias will be judged if at least 90% of the patients randomised were analysed AND the difference in percentage of participants not analysed was 5% or lower across trial arms AND there is no evidence of differential reasons for missing data that may alter the estimate AND the rate of missing data does not exceed the expected event rates
Trial size	The cut‐off to distinguish small from larger trials will be determined by a sample size calculation on the primary outcome
Publication status	Full journal article vs other type or unpublished material
Follow‐up duration	For the cognitive outcomes, we will group studies according to these follow‐up cut‐offs to describe immediate results (up to 12 weeks) and short‐term (up to 1 year), medium‐term (1 to 2 years), and longer‐term results (more than 2 years)
Treatment‐related characteristics
Treatment and control Treatment duration	Analyses will be stratified by control intervention (placebo vs no intervention vs usual care, where no intervention refers to RCTs with standardised concurrent treatments in both experimental and control arms training multiple domains (yes/no) mode of delivery training supervision (yes/no) group training (yes/no) Analyses will be stratified into session length > 30 minutes (yes/no), frequency > 3 sessions per week (yes/no), based upon previous findings (Lampit 2014), and total number of sessions. The minimum treatment duration of 3 months is considered short term, 3 to 12 months as medium term, and 12 months as long term. For the outcome all‐cause dementia, only outcome data at 1 year of follow‐up or longer will be considered, and therefore the grouping will include short‐term (up to 1 year), medium‐term (1 to 2 years), and longer‐term results (more than 2 years)
Participant‐related characteristics
Cognition and participant‐related criteria	Gender, level of education (in years), ApoE‐4 (yes/no), baseline age (mid‐life vs late‐life vs other), and time since diagnoses
*The descriptions given in this table are provided in addition to the guidance provided by the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). Stratified analyses are performed only for the primary outcome if about 10 RCTs contributed to the analyses.