With the extension of life expectancy, chronic health problems related to aging have become a priority for the cause of healthy aging. Mild cognitive impairment (MCI), a heterogenous clinical syndrome, is common among the elderly over 60 years old [1], with an estimated prevalence of 16–20% [2]. It is characterized by subjective cognitive dysfunction and no significant impairment of instrumental activities in daily life (ADL) [3]. MCI is considered the intermediate stage from normal aging to dementia, and patients with MCI are at high risk of dementia [4]. It is reported that about 5–17% of patients with mild cognitive impairment are diagnosed with clinical dementia yearly [5]. Dementia is a serious social issue. The extended care of dementia patients will cause a significant burden on the family, the community and the national healthcare system [6]. Consequently, developing approaches to improve cognitive function in older adults is the focus of work in research.

As no strong evidence exists that medicine is effective for dementia-related cognitive impairment [7], the management of MCI is emerging as a global public health challenge. Nonpharmacological interventions are emerging as a research hotspot, which focus on cognitive compensation and lifestyle changes. Early nonpharmacological interventions for MCI patients can improve cognitive function, improve quality of life, and delay their progression to dementia [8]. The effectiveness of multiple nonpharmacological interventions on cognitive function in patients with MCI has been proved in previous randomized controlled trials, such as physical exercise, cognitive training and Taiji [9, 10]. These interventions have been free of adverse effects compared to pharmacological therapies and are widely used in the adjunctive treatment of cognitive function in patients with MCI. However, although previous conventional pairwise meta-analyses have investigated the effectiveness of nonpharmacological interventions for cognitive function in patients with MCI [11, 9], it does not allow for indirect comparisons of data. The optimal option for nonpharmacological interventions for MCI is still unclear, making decision-making difficult for physicians.

Network meta‐analysis (NMA) is a statistical method that compares one intervention with another based on head to head data from randomized controlled studies and ranks them. The advantages of network models are that they integrate direct and indirect evidence and analyze the efficacy of these interventions rather than simply investigating whether each intervention is more effective than placebo or usual care. Hence, this systematic review and network meta-analysis aimed to evaluate the comparative effects of nonpharmacological interventions for global cognitive function and to rank the interventions for adults aged 60 years and over with MCI.


The Preferred Reporting Items for Systematic Reviews and Network Meta-Analyses (PRISMA-NMA) was used to guide this systematic review and network meta-analysis. A PRISMA checklist was provided in the supplementary file. This review was registered with PROSPERO (CDR 42022369607).

The eligible criteria in this review was described via the PICOS framework

P: patients with MCI (amnestic MCI and non-amnestic MCI) and 60 years old and above. Patients with psychiatric, traumatic brain injury, neurological diseases, and alcohol abuse were excluded. I: nonpharmacological interventions, such as exercise, cognitive training, and Taiji. C: the control group can be inactive or active. O: studies are required to report nonpharmacological interventions’ effects on global cognitive function. S: this review only included randomized controlled studies.

Studies were excluded for four reasons: (1) full text was inaccessible; (2) necessary data were not acquired; (3) repeated published literature; (4) published in non-English languages.

Literature search

We systematically searched six electronic databases, including PubMed, the Cochrane Library, Embase, Web of Science, PsycINFO and CINAHL, from inception until September 2022. The retrieval strategy adopted the combination of subject terms and free words to acquire better retrieval results. The specific retrieval strategy was shown in the supplemental file. In addition, we retrieved reference sections of all studies identified to access all relevant literature. Based on the available studies, we classified the interventions into 18 categories: (1) health education (HE), (2) usual care (care with nursing staff, UC), (3) as usual, (4) stretching, (5) social activity, (6) physical exercise (single-format exercise, PE), (7) multicomponent intervention (MI), (8) risk factor modification (RFM), (9) cognitive training (CT), (10) mind–body exercise (MBE), (11) dual-task exercise (DTE), (12) multicomponent exercise (ME), (13) multicomponent exercise combined cognitive training (ME + CT), (14) acupressure, (15) acupressure combined cognitive training (acupressure + CT), (16) visual art therapy, (17) mindful awareness (MA), and (18) music.

Study selection

This review followed the literature screening process. After removing duplicate literature, researchers (XY L, GP W) independently read the title and abstract according to the inclusion and exclusion criteria and then reviewed the remaining literature. Finally, qualified literature was screened out. The original author would be contacted if essential data were missing or the full text was unavailable. Disputes during the screening process were discussed with a third researcher (YJ C).

Quality assessment

Two review authors (XY L, GP W) independently assessed the risk of bias in included studies using Cochrane’s ‘Risk of bias’ tool, which included five domains: selection bias (random sequence generation and allocation sequence concealment), performance bias (blinding of participants and personnel), detection bias (blinding of outcome assessment), attrition bias (incomplete outcome data), reporting bias (selective reporting) and other bias. Each domain can be rated as “low risk of bias,” “high risk of bias” or “unclear risk of bias”. Differences of opinion were resolved by consulting a third review author (YJ C). In addition, we used the Grading of Recommendations Assessment, Development and Evaluation (GRADE) methodology to assess the quality of the evidence, which can be of high, moderate, low or very low.

Data extraction

Data extraction was performed by two review authors (XY L and GP W). A unified Excel form was used to extract information, including the first author, year, country, age, gender (female), study site, treatment strategy, number of tread patients and study duration.

Statistical analysis

A pairwise meta-analysis of comparisons reported in two or more was done by RevMan 5.4 software. The mean change and standard deviation were used to analyze cognitive function changes. If these were not reported, we would calculate them based on means and standard deviation for baseline and post-intervention. The Cochrane Handbook for Systematic Reviews of Interventions described the calculation methods. The random-effects model was used in pairwise meta-analysis. Due to different assessment tools, the pooled effect size was reported as standardized mean difference (SMD) and a corresponding 95% confidence interval.

We performed a Bayesian random-effects network meta-analysis to combine direct and indirect evidence using Stata 15.1 software. Comparative relationships between interventions were analyzed by drawing a network geometry plot. The consistency hypothesis test was analyzed across global consistency and local consistency. If P > 0.05, the consistency model is used. SMD and corresponding 95% confidence interval were reported to assess the comparative effectiveness of interventions. τ2 assessed heterogeneity among all comparisons. An τ2 of 0.04 was considered a low heterogeneity; 0.16, a moderate heterogeneity; and 0.36, a large heterogeneity [12]. The surface under the cumulative ranking curve analysis (SUCRA) was used to estimate the ranked probability of comparing efficacy between different interventions. SUCRA values range from 0% to 100%, and higher SUCRA values indicate better intervention performance [13]. A comparison-adjusted funnel plot was also produced to explore small-study effects and publication bias. Data points were symmetrically distributed around the centerline, and the auxiliary correction line was perpendicular to the center line, indicating no published deviation in this review.


Study selection

In this review, a total of 3636 studies were initially retrieved (Fig. 1). After the removal of duplicates, 2012 studies were left for examination. One hundred and one studies remained after reading titles and abstracts. Finally, 28 studies were enrolled in this review based on the inclusion and exclusion criteria.

Fig. 1
figure 1

PRISMA flow diagram

Study characteristics

The characteristics of the studies are summarized in Table 1. A total of 28 randomized controlled studies (21 two-arm trials, 5 three-arm trials, and 2 four-arm trials) were included in this review, which was from ten countries, with 13 from China [14, 29,30,31,16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26], 4 from Japan [27,28,29,30], 3 from Korea [31,32,33], 2 from Spain [34, 35], 2 from Brazil [36, 37], and 1 each from America [38], Tanzania [39], Singapore [40], and Turkey [41]. Of these, the earliest study was published in 2010. Fifteen studies investigated the effects of HE, 4 studies investigated the impact of UC, 4 studies investigated the effects of as usual, 3 studies investigated the effects of stretching, 6 studies investigated the impact of PE, 5 studies investigated the effects of MI, 5 studies investigated the effects of CT, 6 studies investigated the impact of MBE, 3 studies investigated the effects of DTE, 6 studies investigated the effects of ME, and 1 study investigated the impact of social activity, acupressure, RFM, acupressure + CT, ME + CT, visual art therapy, MA and music, respectively. Inactive control groups included health education, as usual, and usual care. Health education was the most frequently investigated intervention and was included in 15 studies. Duration time of the studies ranged from 6 weeks to 24 months, and 12 weeks in six studies, 6 months in six studies, 3 months in four studies, 1 year in three studies, 6 weeks in two studies, 40 weeks in two studies, 16 weeks in one study, 8 weeks in one study, 10 weeks in one study, 20 weeks in one study and 24 months in one study. Of a total of 3319 patients over 60 years, there were 1924 females. They were drawn from the community in 16 studies, hospitals in 3, and aged service institutions in 9 studies.

Table 1 Characteristics of selected randomized control trials

Risk of bias

The risk of bias in these enrolled studies is shown in Fig. 2. Regarding these studies, 82% (23/28) studies did not report the method of random sequence generation. 82% (23/28) of studies did not describe allocation concealment methods. All participants and those delivering the intervention were aware of the group allocation, and it was difficult for them to be blinded. Only 21% (6/28) of the studies had a low risk of performance and detection bias. Nearly half of the proportion (32%, 9/28) in these studies was at a high risk of attrition bias, while 61% (11/28) were at low risk of reporting bias. We cannot judge if other biases exist due to insufficient information.

Fig. 2
figure 2

Risk of bias graph: review authors' judgments about each risk of bias item presented as percentages across all included studies

Primary outcomes

In the pairwise meta-analysis, we used a random-effects model to assess the direct comparative effects of nonpharmaceutical interventions on cognitive function. The results were presented in a forest plot (Fig. 3). Only interventions on MI (5 studies), CT (5 studies), ME (6 studies), MBE (6 studies), DTE (3 studies), and PE (6 studies) were reported in two or more studies. Pooled results showed that the improvement of MBE (SMD: 0.24, 95% CI: 0.08–0.41, P = 0.004), DTE (SMD: 0.61, 95% CI: 0.09–1.13, P = 0.02), PE (SMD: 0.58, 95% CI: 0.04–1.12, P = 0.03) was better in intervention group than that of control group. We further analyzed the certainty of evidence (Supplementary File) for all outcomes. Evidence certainty was considered low regarding CT vs HE, acupressure + CT vs HE, visual art therapy vs HE, MA vs HE, music vs HE, RFM vs HE, acupressure vs acupressure + CT, music vs MBE, as usual vs MBE, PE vs MBE, DTE vs stretching, ME + CT vs ME, ME vs as usual, social activity vs CT and RFM vs MI, which was closely related to the quality of the original study. While the certainty of evidence for PE vs HE, acupressure vs HE, acupressure + CT vs CT, ME vs CT, ME + CT vs CT, CT vs UC, acupressure vs CT, DTE vs as usual, DTE vs PE and ME vs UC was considered moderate, the certainty of evidence for other outcomes was considered very low.

Fig. 3
figure 3

Forest plot of estimated treatment effects of nonpharmacological interventions on global cognitive function in patients with MCI. MI multicomponent intervention; CT cognitive training; ME multicomponent exercise; MBE mind–body exercise; DTE dual-task exercise; PE physical exercise

In the network meta-analysis, the network geometry plot is displayed in Fig. 4, representing the interactions among the studies included in this network meta-analysis. The nodes in the network geometry plot represented different interventions, and the nodes and connections were weighted according to the number of studies containing direct comparative interventions. The larger the node, the more times it occurs in the corresponding direct comparison. The thicker the connection, the greater is the number of direct pairwise comparisons. In terms of consistency, we simultaneously applied global and local consistency analyses to evaluate the consistency between the direct and indirect evidence correctly. No statistical differences were found (global consistency test: Chi-square = 1.84, P = 0.9974; local consistency test: all P > 0.05). The network forest plot (Supplementary file) displayed the effect sizes of all interventions. The pooled effect of each study in the comparison set was expressed in blue, which displayed the result of the inconsistent model. In contrast, the pooled total effects of interventions in the comparison set were expressed in red, which displayed the result of a consistent model. The P value displayed in the lower-left corner of the figure was 0.999, >0.5. This also meant that the consistency model was supported. Table 2 summarizes this network meta-analysis's results in improving global cognitive function in older patients with MCI. These results indicated that cognitive function in elderly MCI participants was significantly improved in acupressure + CT compared with that in CT (SMD: 1.66, 95%CI: 0.24–3.08), ME + CT (SMD: 1.89, 95%CI: 0.02–3.77), social activity (SMD: 1.93, 95%CI: 0.06–3.81), UC (SMD: 2.07, 95%CI: 0.49–3.65), ME (SMD: 2.39, 95%CI: 0.89–3.90), MI (SMD: 2.53, 95%CI: 0.93–4.14), DTE (SMD: 2.80, 95%CI: 1.01–4.58), PE (SMD: 2.80, 95%CI: 1.19–4.42), music (SMD: 3.01, 95%CI: 1.04–4.99), RFM (SMD: 3.16, 95%CI: 0.96–5.36), MBE (SMD: 3.07, 95%CI: 1.40–4.74), MA (SMD: 1.66, 95%CI: 0.24–3.08), HE (SMD: 3.33, 95%CI: 1.48–5.19), visual art therapy (SMD: 3.38, 95%CI: 1.93–4.82) and as usual (SMD: 3.69, 95%CI: 1.59–5.80). Compared to MA, there were significant improvements in acupressure + CT (SMD: 1.66, 95%CI: 0.24–3.08), acupressure (SMD: 1.90, 95%CI: 0.49–3.31), CT (SMD: 1.71, 95%CI: 0.69–2.74), ME + CT (SMD: 1.48, 95%CI: 0.01–2.96), UC (SMD: 1.31, 95%CI: 0.14–2.47), ME (SMD: 0.98, 95%CI: 0.24–1.72), MI (SMD: 0.84, 95%CI: 0.14–1.55). Compared to HE, there were significant improvements in acupressure + CT (SMD: 3.33, 95%CI: 1.48–5.19), acupressure (SMD: 1.86, 95%CI: 0.03–3.69), CT (SMD: 1.67, 95%CI: 0.13–3.21). Compared to MBE, there were significant improvements in acupressure + CT (SMD: 3.07, 95%CI: 1.40–4.74), CT (SMD: 1.41, 95%CI: 0.09–2.72). We ranked the superiority of all interventions by SCURA. As shown in Fig. 5, the probability of acupressure + CT being the best intervention mode was 93.6%. The surface area of SCURA for the intervention mode acupressure + CT reached nearly 100%, which also supported that this intervention mode was optimal. Acupressure + CT was followed by acupressure, CT, ME + CT, social activity, UC, ME, MI, DTE, PE, music, RFM, MBE, visual art therapy, stretching, HE, MA, and as usual. Publication bias for included studies was detected by comparison-adjusted funnel plot (Fig. 6), which was basically symmetrical. The adjusted auxiliary line was almost perpendicular to the centerline, indicating a limited small sample effect and publication bias in included studies.

Fig. 4
figure 4

Network geometry plot. CT cognitive training; PE physical exercise; DTE dual-task exercise; MBE mind–body exercise; ME multicomponent exercise; RFM risk factor modification; MI multicomponent intervention; MA mindful awareness; UC usual care

Table 2 League table representing summary estimates from network meta-analysis
Fig. 5
figure 5

Results of network rank test. CT cognitive training; PE physical exercise; DTE dual-task exercise; MBE mind–body exercise; ME multicomponent exercise; RFM risk factor modification; MI multicomponent intervention; MA mindful awareness; UC usual care

Fig. 6
figure 6

Comparison-adjusted funnel plot. CT cognitive training; PE physical exercise; DTE dual-task exercise; MBE mind–body exercise; ME multicomponent exercise; RFM risk factor modification; MI multicomponent intervention; MA mindful awareness; UC usual care


This systematic review and network meta-analysis was performed based on 28 randomized controlled studies to evaluate the effectiveness of nonpharmacological interventions on cognitive function in adults aged over 60 years with MCI. The interventions included CT, PE, MI, RFM, MBE, DTE, ME, MA, ME + CT, stretching, social activity, acupressure, acupressure + CT, visual art therapy, and music. To the best of our knowledge, this review was the most comprehensive network meta-analysis of nonpharmacological interventions for older adults with MCI.

Summary of evidence

In the pairwise meta-analysis, the pooled effect size showed that nonpharmacological interventions could significantly improve cognitive function for patients with MCI, which was consistent with previous evidence [42]. Improvements in cognitive performance for the MBE [31, 43,44,45, 16, 18, 22], DTE [45, 16, 18, 33] and PE [41, 31, 32, 33, 21, 22] group were more significant than that of the control groups. However, this result must be viewed with caution because of the limited number of studies and the large heterogeneity. In addition, since some of the intervention types, including stretching, social activity, acupressure, RFM, acupressure + CT, ME + CT, visual art therapy, MA and music, were reported in only one study, we could not judge their pooled direct effects on cognitive function in elderly MCI patients. However, they proved to be significant in the original research. This was a common inherent limitation of pairwise meta-analysis.

With the deepening of medical research, interventions have gradually diversified. In the face of multiple interventions, pairwise meta-analysis cannot provide effective methodological support for selecting optimal interventions due to the limitation of pairwise comparisons. Network meta-analysis combined all direct and indirect evidence, thus improving precision and fully respecting randomization [46]. This meant that multiple intervention approaches for the metric of interest could be compared simultaneously and ranked for superiority, even if not all possible comparisons have been made. Although previous studies synthesized the results of cognitive interventions for patients with MCI, most of them focused on directly comparing the interventions with usual care or as usual [11, 47] and did not take into account the special characteristics of the elderly individuals. Our study used network meta-analysis to evaluate the effectiveness of nonpharmacological interventions, and the intervention type of acupressure + CT was ranked as the most effective intervention in this review.

Acupressure was an important part of traditional Chinese medicine, which was guided by traditional Chinese medical theory, based on the doctrine of meridians and acupoints, with massage as the primary treatment used to prevent and treat diseases. Acupressure can relax muscles, relieve fatigue, regulate body functions, and improve human immunity, unblocking meridians, balance yin and yang, and prolong life [46]. Acupressure acts as a complementary and alternative medical therapy. It was reported that acupressure could be an effective approach to improve sleep quality, psychological distress and physical function and relieve pain in the elderly [48,49,50]. However, there were limited studies using acupressure to improve cognitive function in individuals with cognitive impairment. A previous study noted that acupressure was effective in improving cognitive function in patients with traumatic brain injury compared to a placebo group [51]. The mechanism of action of acupressure to improve cognitive function remained unclear. From a traditional Chinese medicine perspective, Qi deficiency may be the cause of MCI in the elderly [52]. Acupressure can stimulate the relevant meridians to maintain the balance of Qi in the body [53]. From the perspective of modern medicine, this may be related to acupressure stimulating the transmission of signals in nerve pathways. The stress and cognitive functioning model may explain this conjecture. This model stated that stress can affect cognition in a variety of ways and that chronic exposure to stress can lead to neuronal loss, particularly in the hippocampus [54]. Recent studies have also pointed to stress-induced cortisol arousal as a potential marker of cognitive impairment [55]. In other words, acupressure works by relieving individual stress to reduce its damage to brain function. Acupressure is noninvasive and can be performed by laypersons regardless of equipment conditions, which is cost-effective for older adults with MCI and can be taught to caregivers.

CT has attracted considerable attention as a safe, relatively inexpensive, and scalable intervention. CT refers to repeated standardized exercises in specific cognitive areas to establish or restore cognitive reserves, compensate for brain damage and improve cognitive function [56].Nerves are plastic, which means that they can be stimulated by external environment, experience and other factors to have neural structure and function remodeling. Animal and human studies have shown that the sensory system of the cerebral cortex can be improved by learning and practicing tasks, while brain changes in cortical areas mediate cognitive improvement, and task-specific training also increases gray matter volume and delays cognitive decline [57]. There have been numerous studies exploring the effects of CT on cognitive function in older adults, with mixed results [58, 59], but few studies have demonstrated CT to be effective in individuals with low baseline cognitive levels [70]. Our network meta-analysis found that CT's effect on improving cognitive function was significant only when compared to the MBE, HE, MA, and as-usual group. This suggested that the advantages of CT in elderly individuals with MCI were not prominent. A recent study also noted that CT has shown no significant benefit in improving cognitive function in older adults and that other more rational practices should be adopted to enhance cognitive performance [60]. In recent years, there have been various forms of CT. Advances in computer science and information and communication technology have increased the availability and accessibility of computerized CT. For example, virtual reality can provide intervention in a flexible and real world-like environment and promote visual spatial functions through learning and empathy results [61]. Computer-based CT is more engaging and immersive than traditional cognitive training, complete with pictures, sounds, actions and feedback. The different forms of CT may also be responsible for the differences in results, but they were not further analyzed in this study. Previous studies have reported that interventions delivered using technology showed better results in improving cognitive function and mobility compared to traditional cognitive training, possibly related to the real-time feedback and motivational factors received from the training system [62]. Acupressure + CT was one of this study’s most effective interventions to improve cognitive function. To our knowledge, this was the first network meta-analysis showing improved cognitive function after acupressure + CT. Further studies on acupressure + CT as adjunctive therapy for older adults with MCI are warranted because the mechanism of action behind it remains unknown.

The effectiveness of mindfulness-based interventions in improving the mental health of individuals has been widely recognized [19, 20, 63]. Recently, studies have found that mindfulness training can benefit cognitive function [57]. The mechanism behind this may be related to enhanced immune regulation, changes in gene expression and activity, and reduced levels of inflammatory markers [64]. Moreover, recent studies suggest that mindfulness training may play a role in improving cognitive function by modulating the intestinal microbiome [65]. However, due to the limited number and sample of studies and their generally low quality, the results of cognitive improvement were inconclusive. In our network meta-analysis, the effects of MA were not satisfactory among all the interventions in this study, which was consistent with the results of other reviews. It was reported that there was no significant difference in the effect of mindfulness-based interventions on improving cognitive function compared to the active control group [66, 67]. Therefore, high-quality studies comparing mindfulness training with other interventions are still needed to determine the effectiveness and priority of mindfulness training.

Although the sample size for indirect comparisons was limited, consistency analysis showed that the results of direct and indirect comparisons were consistent, implying that the results of this study are plausible.


Despite the strengths of this network meta-analysis, there are also limitations. (1) Due to the limited number of studies, only 6 of the 18 nonpharmacological interventions were analyzed for direct comparison. (2) Only 15% of the studies had more than 100 participants per group, which may introduce bias. (3) The methodological quality of the included studies was overall low and characterized by a high risk of bias. (4) Although the SUCRA allows comparison of the effectiveness between interventions, it also has limitations, and results should be interpreted cautiously. SUCRA did not show whether the difference between interventions was clinically significant. (5) We grouped the interventions into 18 categories based on published studies. However, we did not further analyze the effectiveness of the different interventions in each category. For example, we categorized treatments with three or more simultaneous interventions as MI, but did not specifically analyze the differences and it is undeniable that heterogeneity exists between these interventions.

Implications for clinical practice

With regard to the treatment of mild cognitive impairment, the safety of nonpharmacological interventions contrasts with that of pharmacological treatments. Medications inevitably produce side effects that even counteract the therapeutic benefits they provide. Hence, healthcare workers, especially community caregivers, should be encouraged to apply promising nonpharmacological interventions to patients with MCI during usual care. In addition, family caregivers are expected to be essential intervention implementers because most nonpharmacological interventions are easy to learn and conduct. Future studies with larger sample sizes and multiple arms of nonpharmacological interventions are needed to determine the effectiveness of cognitive function in older adults with MCI.


We conducted a comprehensive analysis of the effectiveness of 18 different nonpharmacological interventions on cognitive function in adults aged over 60 years with MCI. The results of this network meta-analysis suggested that acupressure + CT was the optimal nonpharmacological intervention to improve cognitive function in older adults significantly. Therefore, healthcare workers should select appropriate interventions based on older adults’ preferences and cognitive levels to improve cognitive function and reduce caregivers' burden.