Does the combination of exercise and cognitive training improve working memory in older adults? A systematic review and meta-analysis

Search strategy

Electronic searches of online databases including Web of Science, Elsevier Science, PubMed and Google Scholar were conducted to collect relevant literature since the start of each database until April of 2021. The following keywords were used in the searches: exercise intervention terms (“aerobic exercise” OR “multidomain exercise training” OR “physical activity”) AND cognitive intervention terms (“mental training” OR “cognitive training” OR “working memory training”) AND combination of intervention terms (“combined” OR “simultaneous” OR “exergame” OR “dual-task” OR “video dancing” OR “cognitive-motor game”) AND relevant working memory terms (“cognitive function” OR “working memory” OR “executive function”) AND older adults terms (“old” OR “elderly” OR “aging”). Additional articles were obtained through a list of references from published reviews.

Inclusion criteria

Inclusion and exclusion criteria were determined based on PICOS (population, intervention, comparison, outcome, and study design). The following criteria were used for study inclusion: (1) population: participants were elderly (aged 60 and over) and included healthy individuals, along with those with MCI and dementia; (2) intervention: combination of exercise and cognitive trainings; (3) comparison: at least one comparison group (combined intervention group with cognitive intervention alone or physical exercise intervention alone or a control group receiving no intervention); (4) outcome: measurements of working memory included at least one outcome that could be used to calculate an effect size; (5) study design: randomized controlled trial (RCT) design; and (6) other: written in English. Exclusion criteria included: (1) population: participants aged less than 60 years old; (2) intervention: without combination of exercise and cognitive intervention; (3) comparison: studies that did not compare the combined intervention group with other intervention groups alone; (4) outcome: studies without relevant data on working memory pre and post intervention; (5) study design: no RCT design; and (6) others: noninterventional studies, reviews and theoretical articles, case and protocol articles, unpublished studies and papers, studies that were not written in English.

Data extraction and assessment of study quality

In this systematic review, working memory tasks were included based on following criteria: (1) participants are required to store and manipulate information during the task, such as Digit Span Backward test, N-back task, and Corsi block-tapping task, (2) the authors declared the tasks were used to measure working memory (You et al., 2009; Damirchi, Hosseini & Babaei, 2018). Tasks that simply stored information with little manipulation were excluded, such as Short-term Memory task, Immediate Recall, Forward Digit Span, the Word List test (Linde & Alfermann, 2014; Mrakic-Sposta et al., 2018; Romera-Liebana et al., 2018; Jardim et al., 2021).

Data referring to working memory tasks were extracted from combined intervention versus exercise intervention alone, or cognitive intervention alone, and no intervention using a spreadsheet. The mean, standard deviation (SD), and the number of participants (N) in each group at pretest and posttest were extracted. If the means and SDs were not mentioned, the change values of the means and SDs after intervention or mean difference of the 95% confidence interval were extracted.

To assess study quality, two independent researchers evaluated the risk of bias for individual studies using the PEDro scale (Maher et al., 2003). Each criterion met on the scale was scored a point, for a maximum score of 11. Articles scoring 8 or higher were considered high quality while those scoring below 8 which were considered low quality.

The moderator analysis was conducted to assess the effects of combined intervention on working memory in the elderly. The following moderators were analyzed: the mode of combination (separate versus simultaneous), cognitive status (healthy versus MCI versus dementia), intervention length (short versus medium versus long), frequency (<3 sessions per week versus ≥ 3 sessions per week), session length (≤ 60 min versus >60 min), no intervention group (active versus passive), study quality (high quality versus low quality).

Data analysis

The statistical data was quantified using comprehensive meta-analysis (CMA) software and calculated the effect size. The effect of the intervention was measured by the standardized mean difference (SMD) of changes after the combined intervention group versus the no intervention group, single exercise intervention group and single cognitive intervention group (Borenstein et al., 2009). The calculation equation of SMD are as follows: ${\displaystyle SMD=\frac{{\overline{x}}_1-{\overline{x}}_2}{S_{within}}}$ ${\displaystyle S_{within}=\sqrt{\frac{\left(n_1-1\right)S_1^2+\left(n_2-1\right)S_2^2}{n_1+n_2-2}}.}$

If one study had multiple task that measured working memory, outcomes were combined into an average effect size to avoid influencing results. The combined effect size with 95% confidence interval (CI) was calculated to determine the efficacy of CECT on working memory. The I² index was calculated to analyze the heterogeneity of the included studies. 0% indicated that no heterogeneity has been observed and the higher of the I² value reflected on more significant of the heterogeneity. The Egger’s regression intercepts and funnel charts were performed to estimate publication bias. Funnel charts was symmetrical which indicated no risk of bias and studies were outside the funnel sharp reflecting on high risk of bias. If I² <50%, p ≥ 0.05, representing the studies had no statistical heterogeneity, so the fixed effect model was used for analysis and if I² ≥ 50%, p ≥ 0.05, indicating had statistical heterogeneity between the studies, so the random effect model was used for analysis.

Identification of studies

Preliminarily, a total of 1,379 articles were obtained from database searches, of these, 275 articles were removed for duplication and 1,104 were excluded on the basis of titles or abstracts. Of the remaining 55 articles, 38 were excluded based on information found in the full text. Finally, 21 eligible articles were included in this meta-analysis, four of which were included from previous reviews. The specific process of article selection, according to the recommended PRISMA flow diagram is shown in Fig. 1.

Characteristics of included studies

Twenty-one RCT articles were included in the current meta-analysis. The characteristics of included studies are shown in Table 1. Among the included studies, 16 were conducted on healthy older adults (Fabre et al., 2002; You et al., 2009; Legault et al., 2011; Maillot, Perrot & Hartley, 2012; Shatil, 2013; Gschwind et al., 2015; Nishiguchi et al., 2015; Rahe et al., 2015; Eggenberger et al., 2016; Schättin et al., 2016; Ordnung et al., 2017; Kalbe et al., 2018; Norouzi et al., 2019; Adcock et al., 2020; Takeuchi et al., 2020; Dana, 2019), four focused on older adults with MCI (Suzuki et al., 2013; Combourieu Donnezan et al., 2018; Damirchi, Hosseini & Babaei, 2018; Bae et al., 2019), and one focused on older adults with dementia (Karssemeijer et al., 2019). Sample size ranged from 13 to 153. The mean age range for older adults was 65.8 to 79.2. Of all 21 studies, six included four comparison groups, three included three comparison groups, and 12 included two comparison groups. Of the 21 studies, seven studies used separate interventions, while 14 studies adopted a simultaneous combination of exercise and cognitive intervention, (three of the 14 used the dual task and seven of the 14 used the exergame). The length of intervention ranged from four to 24 weeks and the frequency was one to six sessions each week. Five of the studies included active control groups (leisure activity and health education) and the rest included passive control groups (adhered to normal life). Types of exercise interventions included aerobic training, flexibility training, muscle strength training, balance training, endurance training, and resistance training. The types of cognitive interventions included multidomain cognitive training, working memory training, and cognitive games. Only five articles (Maillot, Perrot & Hartley, 2012; Suzuki et al., 2013; Ordnung et al., 2017; Damirchi, Hosseini & Babaei, 2018; Norouzi et al., 2019) explicitly reported the gender distribution of participants, of which there were a total of 134 male and 101 female participants. The other included articles did not clearly report the gender distribution of participants. The quality of studies ranged from six to 11 on the PEDro scale and the specific scoring details are shown in Table S1.

Combined intervention versus no intervention group

Figure S1 presented the funnel plot of included studies and the result showed no significant asymmetry (Egger’s regression intercept = 1.50, p > 0.05).

A total of 15 articles (Adcock et al., 2020; Bae et al., 2019; Combourieu Donnezan et al., 2018; Damirchi, Hosseini & Babaei, 2018; Dana, 2019; Fabre et al., 2002; Gschwind et al., 2015; Karssemeijer et al., 2019; Legault et al., 2011; Maillot, Perrot & Hartley, 2012; Nishiguchi et al., 2015; Norouzi et al., 2019; Ordnung et al., 2017; Shatil, 2013; Suzuki et al., 2013) reported on the effects of CECT compared to no intervention on working memory in older adults. A fixed effect model for meta-analysis was used, as the heterogeneity test showed Q₍₁₄₎ = 17.31, p = 0.24, i² = 19.10. As shown in Fig. 2, the combined intervention group showed significant improvement in working memory compared to the no intervention group (SMD = 0.29, 95% CI [0.14–0.44], p < 0.01).

Combined intervention versus single exercise intervention

The funnel plot revealed that one study had a disproportionately large effect size and Egger’s test was significant (Egger’s regression intercept = 2.71, p < 0.05), so one study was removed from further analysis (Norouzi et al., 2019). The funnel plot after removal of one outlier was presented in Fig. S2 and had no significant asymmetry.

Eleven studies (Combourieu Donnezan et al., 2018; Damirchi, Hosseini & Babaei, 2018; Dana, 2019; Eggenberger et al., 2016; Fabre et al., 2002; Karssemeijer et al., 2019; Legault et al., 2011; Schättin et al., 2016; Shatil, 2013; Takeuchi et al., 2020; You et al., 2009) compared the effect of CECT to single physical exercise. The results of our analysis revealed no significant difference on working memory in the effect of between CECT and single physical exercise (SMD = 0.16, 95% CI [−0.04–0.35], p = 0.12) (Fig. 3). The heterogeneity test was not significant (Q₍₁₀₎ = 8.46, i² = 0, p = 0.58), so a fixed effect model was used.

Combined intervention versus cognitive intervention alone

The funnel plot showed that the result had no significant asymmetry (Fig. S3). Nine articles (Fabre et al., 2002; Legault et al., 2011; Shatil, 2013; Rahe et al., 2015; Combourieu Donnezan et al., 2018; Damirchi, Hosseini & Babaei, 2018; Kalbe et al., 2018; Dana, 2019; Takeuchi et al., 2020) compared the effect of combined intervention to cognitive intervention alone. The results (Fig. 4) of our meta-analysis revealed no significant difference in the effects of combined intervention versus cognitive intervention alone on working memory (SMD = 0.08, 95% CI [−0.13–0.30], p = 0.44). The heterogeneity test was not significant (Q₍₈₎ = 2.13, i² = 0, p = 0.98), so a fixed effect model was used.

Moderator analysis

The results of the moderator analysis are shown in Table 2. A heterogeneity test revealed a significant influence of cognitive status on the effects of CECT on working memory (Q₍₂₎ = 7.67, p = 0.02). The criterion used to categorize participants’ cognitive status was based on explicit declarations made by the authors, identifying individuals as either having mild cognitive impairment (MCI) or dementia. Combined intervention had a significantly influence among healthy participants (SMD = 0.24, 95% CI [0.06–0.42], p = 0.01) and those with MCI (SMD = 0.58, 95% CI [0.29–0.87], p < 0.01) but no significant effect among those with dementia (SMD = −0.17, 95% CI [−0.64–0.31], p = 0.49).

A heterogeneity test revealed a significant influence of intervention frequency on the effects of CECT on working memory (Q₍₁₎ = 5.53, p = 0.02). CECT was significantly effective in low frequency intervention (SMD = 0.52, 95% CI [0.28–0.76], p < 0.01), but no significant effect in high frequency intervention (SMD = 0.15, 95% CI [−0.03–0.34], p = 0.11).

In terms of mode of combination, simultaneous (SMD = 0.31, 95% CI [0.14–0.47], p < 0.01) training were proven effective, but no significance was found for sequential (SMD = 0.23, 95% CI [−0.09–0.55], p = 0.16) training. CECT was significantly effective regardless of intervention time: Short (SMD = 0.48, 95% CI [0.13–0.83], p = 0.01); medium (SMD = 0.27, 95% CI [0.01–0.53], p = 0.05); and long intervention length (SMD = 0.23, 95% CI [0.03–0.44], p = 0.03). Combined intervention was significantly effective in low quality (SMD = 0.44, 95% CI [0.21–0.67], p < 0.01) studies, and had no significance in high quality (SMD = 0.18, 95% CI [−0.01–0.38], p > 0.05) studies. Compared with no intervention, long combined intervention sessions (SMD = 0.31, 95% CI [0.07–0.55], p = 0.01) and short combined sessions (SMD = 0.28, 95% CI [0.09–0.46], p < 0.01) were both significantly beneficial for working memory. In terms of the no intervention group, CECT had a positive effect in passive (SMD = 0.33, 95% CI [0.14–0.51], p < 0.01) but no significant effect in active control group (SMD = 0.22, 95% CI [−0.03–0.47], p = 0.08).