Note: ADAS-cog = cognitive subscale of the Alzheimer’s Disease Assessment Scale, CI = confidence interval, MMSE = Mini-Mental State Examination, RT = randomized trial.

↵* Clinical significance is considered to be a change in score of between 1.4 to 3 points on MMSE and 4 points on ADAS-Cog. Negative and positive effects are outcome-measure dependent. For MMSE, increases in score (positive values) indicate an improvement of symptoms; however, for ADAS-cog, decreases in score (negative values) indicate an improvement.

↵† Grading of Recommendations, Assessment, Development and Evaluation (GRADE) rates the continuum of quality of evidence in 4 categories of high, moderate, low or very low. All data can be found in the accompanying evidence review. (10)

↵a Using Cochrane’s risk-of-bias tool, for this outcome 1 study was rated as low and 2 studies were rated as unclear risk. Across the studies, there was a lack of certainty (unclear ratings) regarding sequence generation (33%) and allocation concealment (33%), and a high risk of bias associated with incomplete outcome reporting (33%) and other sources of bias (67%; i.e., industry funding, baseline differences between groups, insufficiently powered and/or sample size < 30 per arm). Given that most of the information is from studies at moderate risk of bias, this body of evidence was downgraded for serious study limitations.

↵b The statistical heterogeneity is minimal (χ² = 0.79, 2 degrees of freedom [df]; p = 0.68; I² = 0%), and the 95% CIs overlap across most studies. This body of evidence was not downgraded for inconsistency.

↵c Three randomized controlled trials (RCTs) provided data for this outcome. All studies included mixed sex samples. The mean age across studies ranged from 69 to 74 years. The intervention arm received donepezil (10 mg/d) in 2 studies and rivastigmine (3–12 mg/d) in 1 study; all control groups received placebo. One study was conducted in the United States and Canada, 1 in the US and 1 study in 14 countries. All studies were published from 2005 to 2009. The length of intervention across studies ranged from 11 to 48 months. There were no serious concerns regarding indirectness for this body of evidence, and it was not downgraded.

↵d The sample size is adequate (i.e., > 300 [1140 intervention arm, 1147 control arm]), but the pooled effect estimate is not precise and the 95% CI includes the null value “0” (mean difference = 0.17 [95% CI −0.13 to 0.47]). This body of evidence was downgraded for serious concerns regarding imprecision.

↵e There were too few studies (< 10) to assess publication bias.

↵F Using Cochrane’s risk-of-bias tool, for this outcome 1 study was rated as low and 3 studies were rated as unclear risk. Across studies, there was a lack of certainty (unclear ratings) regarding sequence generation (50%) and allocation concealment (50%), and a high risk of bias associated with incomplete outcome reporting (25%) and other sources of bias (75%; i.e., industry funding, baseline differences between groups, insufficiently powered and/or sample size < 30 per arm). Given that most of the information is from studies at moderate risk of bias, this body of evidence was downgraded for serious study limitations.

↵g The statistical heterogeneity is minimal (χ² = 4.63, 4 df; p = 0.33; I² = 14%), and the 95% CIs overlap across most studies. This body of evidence was not downgraded for inconsistency.

↵h Four RCTs provided data for this outcome. All studies included mixed sex samples. The mean age across studies ranged from 69 to 74 years. The intervention arm received donepezil (10 mg/d) in 2 studies, rivastigmine (3–12 mg/d) in 1 study and galantamine (16–24 mg/d) in 1 study. The control group across all studies received placebo. Two studies were conducted in the US and Canada, 1 study in the US and 1 study in 14 countries. All studies were published from 2005 to 2009. The length of intervention across studies ranged from 11 to 48 months. There were no serious concerns regarding indirectness for this body of evidence, and it was not downgraded.

↵i The sample size is adequate (i.e., > 300 [2078 intervention arm, 2110 control arm]), but the pooled effect estimate is not precise and the 95% CI includes the null value “0” (mean difference = −0.33 [95% CI −0.73 to 0.06]). This body of evidence was downgraded for serious concerns regarding imprecision.

↵j Using Cochrane’s risk-of-bias tool, for this outcome 1 study was rated as low and 1 as unclear risk. Across studies, there was a lack of certainty (unclear ratings) regarding sequence generation (50%) and allocation concealment (50%), and a high risk associated with other sources of bias (50%; i.e., industry funding, baseline differences between groups, insufficiently powered and/or sample size < 30 per arm). Given that most of the information is from studies at moderate risk of bias, this body of evidence was downgraded for serious study limitations.

↵k The statistical heterogeneity is minimal (χ² = 0.58, 1 df; p = 0.44; I² = 0%), and the 95% CIs overlap across studies. This body of evidence was not downgraded for inconsistency.

↵l Two RCTs provided data for this outcome. Both studies included mixed sex samples. The mean age across studies ranged from 70 to 74 years. The intervention arm received donepezil (10 mg/d) and the control group received placebo. One study was conducted in the US, and 1 in the US and Canada. One study was published in 2005 and 1 in 2009. The length of intervention across studies ranged from 11 to 36 months. There were no serious concerns regarding indirectness for this body of evidence, and it was not downgraded.

↵m The sample size is adequate (i.e., > 300 [632 intervention arm, 637 control arm]), but the pooled effect estimate is not precise and the 95% CI includes the null value “0” (mean difference = 0.24 [95% CI −0.19 to 0.67]). This body of evidence was downgraded for serious concerns regarding imprecision.

↵n The statistical heterogeneity is minimal (χ² = 1.48, 1 df; p = 0.22; I² = 33%), and the 95% CIs overlap across studies. This body of evidence was not downgraded for inconsistency.

↵o The sample size is adequate (i.e., > 300 [632 intervention arm, 637 control arm]), but the pooled effect estimate is not precise and the 95% CI includes the null value “0” (mean difference −0.60 [95% CI −1.35 to 0.15]). This body of evidence was downgraded for serious concerns regarding imprecision.

↵p Using Cochrane’s risk-of-bias tool, for this outcome the included study was rated as unclear risk. There was a high risk of bias associated with incomplete outcome reporting and other sources of bias (i.e., industry funding, baseline differences between groups, insufficiently powered and/or sample size < 30 per arm). Given that the information is from a study with moderate risk of bias, this body of evidence was downgraded for serious study limitations.

↵q The statistical heterogeneity across studies could not be assessed owing to only 1 study providing data for this outcome.

↵r One RCT provided data for this outcome. The study included a mixed sex sample. The mean age was 70.6 years for the intervention group and 70.3 years for the control group. The intervention arm received rivastigmine (3–12 mg/d); the control group received placebo. The study was conducted in 14 countries and published in 2007. The length of intervention was 48 months. There were no serious concerns regarding indirectness for this body of evidence, and it was not downgraded.

↵s The sample size is adequate (i.e., > 300 [508 intervention arm, 510 control arm]), but the pooled effect estimate is not precise and the 95% CI includes the null value “0” (mean difference = 0.10 [95% CI −0.32 to 0.52]). This body of evidence was downgraded for serious concerns regarding imprecision.

↵T The sample size is adequate (i.e., > 300 [508 intervention arm, 510 control arm]), but the pooled effect estimate is not precise and the 95% CI includes the null value “0” (mean difference = 0.0 [95% CI −0.80 to 0.80]). This body of evidence was downgraded for serious concerns regarding imprecision.

↵u Using Cochrane’s risk-of-bias tool, for this outcome the included study was rated as unclear risk. There was a lack of certainty (unclear ratings) regarding sequence generation and allocation concealment, and a high risk of bias associated with incomplete outcome reporting and other sources of bias (i.e., industry funding, baseline differences between groups, insufficiently powered and/or sample size < 30 per arm). Given that most of the information is from studies at moderate risk of bias, this body of evidence was downgraded for serious study limitations.

↵v One RCT provided data for this outcome. The study included a mixed sex sample. The study included results from 2 trials: in the first trial, the mean age was 69.2 years in the intervention group and 70.1 years in the control group; the mean age was 70.6 years and 70.9 years, respectively, in the second trial. The intervention arm received galantamine (16–24 mg/d) in both trials; the control group received placebo. The study was conducted in the US and Canada and published in 2008. The length of intervention was 24 months. There were no serious concerns regarding indirectness for this body of evidence and was not downgraded.

↵w The sample size is adequate (i.e., > 300 [938 intervention arm, 963 control arm]), but the pooled effect estimate is not precise and the 95% CI includes the null value “0” (mean difference = −0.21 [95% CI −0.80 to 0.38]). This body of evidence was downgraded for serious concerns regarding imprecision.

↵x Using Cochrane’s risk-of-bias tool, for this outcome 1 study was rated as low and 3 studies were rated as unclear risk. Across studies, there was a lack of certainty (unclear ratings) regarding sequence generation (50%) and allocation concealment (75%), and a high risk associated with other sources of bias (25%; i.e., industry funding, baseline differences between groups, insufficiently powered and/or sample size < 30 per arm). Given that most of the information is from studies at moderate risk of bias, this body of evidence was downgraded for serious study limitations.

↵y The statistical heterogeneity is minimal (χ² = 1.36, 3 df; p = 0.71; I² = 0%), and the 95% CIs overlap across most studies. This body of evidence was not downgraded for inconsistency.

↵z Four RCTs provided data for this outcome. All studies included mixed sex samples. The mean age across studies ranged from 66 to 77 years. The intervention arm received vitamin E in 1 study, vitamin E and folic acid in 1 study, docosahexaenoic acid (DHA; fish oil) in 1 study, and vitamins E and C in 1 study; all control groups received placebo. One study was conducted in the US and Canada, 1 in the United Kingdom, 1 in Malaysia and 1 in Iran. All studies were published from 2005 to 2014. The length of intervention across studies ranged from 12 to 36 months. There were no serious concerns regarding indirectness for this body of evidence, and it was not downgraded.

↵A The sample size is adequate (i.e., > 300 [511 intervention arm, 519 control arm]), but the pooled effect estimate is not precise and the 95% CIs includes the null value “0” (mean difference = 0.20 [95% CI −0.04 to 0.43]). This body of evidence was downgraded for serious concerns regarding imprecision.

↵B Using Cochrane’s risk-of-bias tool, for this outcome the included study was rated as unclear risk. There was a lack of certainty (unclear ratings) regarding sequence generation and allocation concealment. Given that the information is from a study with moderate risk of bias, this body of evidence was downgraded for serious study limitations.

↵C One RCT provided data for this outcome. The study included a mixed sex sample. The mean age was 72.8 years for the intervention group and 72.9 years for the control group. The intervention arm received donepezil (10 mg/d); the control group received placebo. The study was conducted in the US and Canada and was published in 2005. The length of intervention was 36 months. There were no serious concerns regarding indirectness for this body of evidence, and it was not downgraded.

↵D The sample size is not adequate (i.e., < 300 [257 intervention arm, 259 control arm]), and the pooled effect estimate is not precise with a 95% CI that includes the null value “0” (mean difference = 0.85 [95% CI −0.32 to 2.02]). This body of evidence was downgraded for serious concerns regarding imprecision.

↵E Using Cochrane’s risk-of-bias tool, for this outcome 1 study was rated as low and 4 studies were rated as unclear risk. Across studies, there was a lack of certainty (unclear ratings) regarding sequence generation (80%), allocation concealment (80%), blinding (20%), incomplete outcome reporting (20%) and other sources of bias (20%), and a high risk of bias associated with blinding (20%), incomplete outcome reporting (20%) and other sources of bias (40%; i.e., industry funding, baseline differences between groups, insufficiently powered and/or sample size < 30 per arm). Given that most of the information is from studies at moderate risk of bias, this body of evidence was downgraded for serious study limitations.

↵F The statistical heterogeneity is high (χ² = 16.92, 4 df; p = 0.002; I² = 76%), but the direction of the effect is consistent across studies and the 95% CIs overlap across most studies. The statistical heterogeneity is most likely due to small versus large treatment effects observed across studies. This body of evidence was not downgraded for inconsistency.

↵G Five RCTs provided data for this outcome. All studies included mixed sex samples. The mean age across studies ranged from 65 to 77 years. The intervention arm received multicomponent exercise programs in 3 studies and cognitive training and rehabilitation in 2 studies; the control group across studies received no therapy, wait list, or minimal contact involving education about health promotion. Two studies were conducted in Japan, 1 in China, 1 in Greece and 1 in Argentina. All studies were published from 2009 to 2014. The length of intervention across 4 studies ranged from 6 to 12 months. There were no serious concerns regarding indirectness for this body of evidence, and it was not downgraded.

↵H The sample size is not adequate (i.e., < 300 [221 intervention arm, 187 control arm]), but the pooled effect estimate is precise with a narrow 95% CI (mean difference = 1.01 [95% CI 0.25 to 1.77]). This body of evidence was not downgraded for serious concerns regarding imprecision.

↵I Using Cochrane’s risk-of-bias tool, for this outcome the included study was rated as low risk. There were no serious concerns regarding risk of bias, and this body of evidence was not downgraded for serious study limitations.

↵J One RCT provided data for this outcome. The study included a mixed sex sample. The mean age was 74.8 years for the intervention group and 75.8 years for the control group. The intervention arm received a multicomponent exercise program biweekly; the control group received minimal contact with 2 education classes about health promotion. The study was conducted in Japan and published in 2013. The length of intervention was 6 months. There were no serious concerns regarding indirectness for this body of evidence, and it was not downgraded.

↵K The sample size is not adequate (i.e., < 300 [47 intervention arm, 45 control arm]), and the pooled effect estimate is not precise with a 95% CI that includes the null value “0” (mean difference = −0.60 [95% CI −1.44 to 0.24]). This body of evidence was downgraded for serious concerns regarding imprecision.