If sufficient studies are available (3 or more studies), we will perform subgroup analyses based on characteristics thought to modify treatment effects or possible sources of heterogeneity between studies such as age, gender, educational level, comorbidity, patient groups, types of intervention, or types of study. Our a priori hypothesis will be that older, lower educational level, and more comorbidity subgroups may show less improvement in the primary and secondary outcomes. We will also stratify our results by additional study method variables as appropriate to include supervised (exercise observed by clinical or exercise professional) versus unsupervised protocols, high versus low risk of bias, measurement through fitness versus heart rate, quality of the cognitive assessment used, and adherence over 80%.

We will use sensitivity analyses to assess the appropriateness of including various modes of exercise, such as aerobic and resistance training intervention protocols. In addition, we will use sensitivity analyses to assess whether or not the inclusion of studies with missing outcome data of varying degrees meaningfully impacts the results of the meta-analysis.

For each comparison with 10 or more studies, we will visually assess publication bias and small-study effects using funnel plots (using study’s effect estimates for the primary outcomes against their standard errors). We will utilize GRADE approach to determine the quality of evidence of included articles, designating all articles to one of four levels of evidence according to their study and other variables which have the potential to impact the quality of evidence (such as indirect evidence, specific types of bias, and effect size) [28].