Effects of exercise interventions for specific cognitive domains in old adults with mild cognitive impairment: A meta-analysis and subgroup analysis of randomized controlled trials.

OBJECTIVE: To conduct a meta-analysis and subgroup analysis investigating the effects of exercise on mild cognitive impairment (MCI) patients across specific cognitive domain outcomes. We also analyzed and identified the level of influence of exercise interventions on specific cognitive domains.
METHODS: MEDLINE, EMBASE, Cochrane Library, Web of Science, CNKI, the Wan Fang Database, and CBM were searched from inception to April 2018. Randomized controlled trials of exercise interventions in MCI patients older than 55 years, with an outcome measure of cognitive function were included.
RESULTS: Eleven studies with sufficient data met the inclusion criteria for the meta-analysis. Exercise interventions significantly improved general function (g mini-mental state examination Montreal cognitive assessment = 0.32, 95% the 95% confidence interval (CI) 0.1 to 0.54, P = .005 and g Alzheimer disease assessment scale-cognition = -0.45, 95% CI -0.82 to -0.08, P = .02); executive functions (g digit span forward test, digit span backward test, digit span forward test -B, stroop test-A, stroop test-B = 0.66, 95% CI 0.17 to 1.15, P = .008); memory (g Wechsler memory scale immediate recall and Wechsler memory scale delayed recall = 0.37, 95% CI 0.15 to 0.60, P = .001); language ability (g category verbal fluency test and letter verbal fluency test = 0.55, 95% CI 0.22 to 0.89, P = .001); and visuospatial ability (g block design score = 0.38, 95% CI 0.03 to 0.72, P = .03). However, the improvement exercise conferred on the trail-making test part B-A was not statistically significant (g trail-making test part B-A  = -0.25, 95% CI -0.88 to 0.39, P = .45). The preliminary ranking of the effect on the overall effect was as follows: Z language ability > Z executive functions >   memory > Z visuospatial ability.
CONCLUSION: Exercise improves performance in the 5 cognitive domains. Across cognitive domains, language ability was the domain most affected by exercise. Besides, the kind of ranking (Z value) provides a new perspective for community health care workers to prescribe targeted exercise interventions for MCI patients.PROSPERO registration number: CRD42018093902.

Introduction

The 2018 World Alzheimer Report reported that 50 million people live with dementia worldwide. With the accelerating trend of population ageing, that number is expected to increase to more than 152 million by 2050. Dementia also has a huge economic impact. The total estimated worldwide cost of dementia in 2018 is US$1 trillion, and this figure will rise to US$ 2 trillion by 2030.[ Dementia is a syndrome in which there is deterioration in memory, thinking, behaviour and the ability to perform everyday activities.[ The quality of daily life and activities for those living with dementia are seriously affected.[ However, there is as yet no really satisfactory treatment to delay the progress of dementia.

At present, for the prevention and treatment strategy of dementia, the basic research and clinical research have turned to the preclinical stage—the mild cognitive impairment (MCI) domain. MCI is an intermediate stage between normal brain ageing and dementia characterised by an impairment in memory and/or other cognitive function.[ It has been reported that old people with MCI are at increased risk of conversion to develop dementia, with progression rates of 16% to 20% per year, which is higher than 1% to 2% in cognitively normal people at the same age.[ On average, about 20% to 24%of subjects with MCI will reverse to normal cognition over time.[ Hence, MCI is a progressive, high-risk and uncertain stage. It is particularly important for the prevention of dementia.

At present, treatment of MCI mainly includes pharmacological treatments and nonpharmacological therapies. The American Academy of Neurology plainly recommend against use of cholinesterase inhibitors in MCI.[ Evidence in the guidelines, and in systematic reviews, clearly shows that cholinesterase inhibitors have no reliable benefit on cognition and do not reduce progression to dementia.[ In addition, Patients complained about side effects, including gastrointestinal symptoms and cardiac concern.[ At present, nonpharmacological therapies have been validated and recommended for patients with MCI.[

As 1 of the nonpharmacological therapeutic modalities, exercise intervention plays an increasingly important role in prevention, rehabilitation and health medicine over the past 2 decades. It has been proposed that exercise promotes brain remodelling, including increases brain volume, improves synaptic plasticity,[ triggers several molecular and cellular cascades,[ facilitates growth, development, function, and survival of neurons,[ upregulate of hippocampal brain derived neurotrophic factor levels, reduce accumulation of reactive oxygen species, which promote brain health over time and delay the onset of cognitive decline in aging and dementia.[ Indeed, evidence from neuroimaging works, cross-sectional and epidemiological, which consistently find exercise to be beneficial for cognitive function in the elderly.[ However, the effects of the particular exercise intervention methods, the exercise intensity and the duration of exercise intervention on the functioning in various cognitive domains are different. In terms of exercise intervention method, combining aerobic exercise with resistance training and flexibility training may improve performance on executive tasks, including working memory, processing speed and attention more efficiently than 1 form of exercise by itself.[ In terms of the intensity of the exercise intervention, compared to stretching exercises with a heart rate reserve at or below 50%, exercise with a heart rate reserve of 75% to 85% can significantly improve executive control processes such as cognitive flexibility, multitasking, selective attention, and information processing efficiency.[ In terms of the duration of exercise interventions, a randomized controlled trial shown 6 months of aerobic exercise significantly improved learning performance and verbal memory.[ Lam et al reported that weekly engagement in physical activity for twelve months could improve global cognitive ability, delayed recall and verbal fluency with time (P < .05) in people with MCI.[

From the studies presented above, there were conflicting results on the specific cognitive domains affected, with some studies reporting cognitive gains from exercise interventions in memory function or verbal fluency,[ and other studies reporting effects on executive function.[ The changes in the emphasis on or trends in different cognitive domains in the process of developing exercise interventions needs to be further explored. Up until now, there has been no published literature on examining a subgroup analysis of the effects of exercise interventions on different cognitive domains in the elderly with MCI. Based on the hypothesis that exercise interventions may have some regular, generalized effects on various cognitive domains of the brain, the aim of this study is to explore the influence of exercise interventions on specific cognitive domains in the elderly with MCI, and produce a preliminary ranking of the degree or magnitude of the influence of certain exercise interventions on various cognitive domains. The study contributed to revealing the different effects of exercise interventions on improving various cognitive functions.

Methods and analysis

Methods

The meta-analysis was performed and reported in accordance with the PRISMA guidelines and the checklist was completed.[ PROSPERO registration number: CRD42018093902.

Search strategy

Seven electronic databases were searched (MEDLINE, EMBASE, Cochrane Library, Web of Science (Science and Social Science Citation Index), China National Knowledge Infrastructure (CNKI), the Wan Fang Database, and CBM (China Biology Medicine) from inception to April 2018 were searched. The study given a search strategy example from Cochrane Library in Supplementary Appendix Table 1. The search strategy was modified as necessary for other databases. There was no language restriction.

Relevant randomized controlled trials (RCTs) and reviews for additional studies were hand searched for further eligible studies. Journals containing the highest numbers of included studies were hand searched for recent potentially eligible publications (≤12 months). In addition, the WHO International Clinical Trials Registry Platform (ICTRP) and ClinicalTrials.gov was checked to identify planned, ongoing or unpublished trials.

study inclusion criteria and exclusion criteria

Inclusion criteria were:

Studies were RCTs of exercise intervention (aerobic, resistance, or multicomponent exercise);

Participants were MCI patients based on the adaptations of criteria suggested by Petersen that are commonly used to identify ‘mild cognitive impairment’[ and age 55 years or older;

Outcomes which were general cognition, executive functions, memory, language ability, and visuospatial ability were considered for inclusion;

The control interventions could be sham (such as stretching activities), placebo or no treatment or health education; and

For patients with MCI, a trial lasting for 12 weeks usually shows cognitive enhancement rather than maintenance of cognitive function.[

Therefore, all trials that reported the results after randomization for at least 12 weeks or longer were included in our study. There was no minimum duration of follow up. There were no restrictions on the location of the intervention.

Participants were excluded:

They took medications for cognition, neurological conditions (eg, stroke, multiple sclerosis, Parkinson disease), chronic or acute conditions, or depressive symptoms that would preclude exercise;

They engaged in regular exercise (≥30 min/d, ≥3 d/wk);

They had medical histories that included unstable cardiac disease, significant cerebrovascular disease, musculoskeletal impairment, etc; and

They had other medical conditions with significant psychiatric or metabolic sequelae.

Article screening, data extraction and quality recorded were carried out by 2 authors independently (XLZ and JW). Discrepancies were resolved by discussion or by consulting the senior authors (LNW, LZ and XHS) (see Appendix for detail). Duplicates and manage records were omitted using EndNote software (V.X7.8).[ Articles that were not available were requested from the authors. The detailed information of each included study was extracted using a predetermined data form and includes the following general information, details of study, design; study subjects, intervention characteristics, outcome.

Quality assessment and data analysis

The quality of included studies was evaluated by 2 reviewers(X.L.Z and J.W.) independently using a modified assessment form: Cochrane Handbook for Systematic Reviews of Interventions[ and another 3 items from physiotherapy evidence database (PEDro) scale were added, including “the type of statistical analyses used (true intention-to-treat versus other)”, “eligibility criteria”, and “baseline comparability”.[ The ‘risks of biases’ summary were presented graphically.

All analyses were conducted with Review Manager software (RevMan 5.3.). Using a random effects model and the standardized mean difference (SMD) and the 95% confidence interval (CI) to analyze the outcomes.[ We set P < .05 as statistically significant. For results that were not presented as SMD in the original meta-analysis, RevMan5.3 was used to convert the outcomes to SMD to allow visual comparison of the results in a forest plot. If data were insufficient for conversion, data were extracted from original studies and processed as instructed by the Cochrane Collaboration.[

Heterogeneity was assessed with the I2 statistic and Q statistic for each analysis. Subgroup analysis and investigation of heterogeneity:

A subgroup analysis for specific cognitive domains played an important role in ascertaining the most sensitive cognitive domain that has noticeable improvements by exercise intervention. General cognition and 4 specific cognitive domains were analysed, including executive functions, memory, language ability, and visuospatial ability.

Subgroup analyses was used to explore possible sources of heterogeneity. We tested the heterogeneity by considering the variability in participant factors (eg, age, sex) and trial factors (eg, follow-up period, intervention type, type of control group).

Results

Descriptive results

Qualified studies

The search process is shown in the PRISMA flow diagram in Figure 1. A total of 5676 potentially relevant records were obtained from MEDLINE (1388 records), EMBASE (1360 records), the Cochrane Library (1016 records), Web of Science (1468 records), China National Knowledge Infrastructure (113 records), the Wan Fang Database (220 records), China Biology Medicine (111 records), and 485 records were obtained from other resources. Of these, 311 records were removed as duplicates, and 5809 records were excluded following screening of titles and abstracts. In total, 41 full-text records were assessed. Of these, 9 records did not provide sufficient information regarding the diagnosis of mild cognitive impairment, 2 records did not examine an intervention of interest, 4 records outcomes did not target improving cognitive impairment or maintaining cognitive function, 5 records were non-RCTs, 8 records were protocols/short articles, and 2 records were excluded because of insufficient data. Finally, eleven studies were included.[

Figure 1

Flow diagram for study selection process used in this meta-analysis.

Study characteristics

The characteristics of each eligible study are summarized in Table 1. The characteristics of invention about each eligible study are summarized in Supplementary Appendix Table 2. The studies were from 7 countries. Participates in study meet MCI diagnostic criteria. Intervention methods of all studies is physical exercise. The control groups are health education, no-exercise treatment (recreational activities or memory enhancement training program), or usual (maintained current lifestyle or used a usual stretching and toning exercise group). Primary outcomes are general cognitive performance, executive functions, memory, language ability and visuospatial ability. The intervention durations ranged between 12 weeks and 12 months. There were 357 participants in the experimental group and 319 in the control group. The sample sizes of the eligible studies ranged from 22 to 100.

Table 1

The characteristics of eligible studies.

Methodological quality

Two authors independently assess the risk of bias using the approach recommended by Cochrane Handbook for Systematic Reviews and PEDro. We described what was reported to have happened in each study for each domain of risk of bias and provided the rationale for the judgement of whether this domain was at low, high or unclear risk of bias. All trials were considered to be at low risk of bias regarding the MCI diagnostic criteria and provided detailed reasons justifying use of this term. Ten records[ provided detailed reasons for similar baseline values, but 1 study[ was unclear about this term. We considered 8 trials[ to be at low risk of bias for sequence generation. We judged the remaining 3 trials[ to be at unclear risk of bias for sequence generation. Four trials[ did not describe allocation concealment. The remaining 7 trials[ described relevant methods. In all eleven included trials, blinding of the participants and trainers was not feasible. This was unlikely to introduce bias in trainers, so we considered all eleven trials to be at low risk of bias for blinding trainers. This may have introduced bias in participants, so all eleven trials were judged to be at high risk of blinding of the participants. We judged 2 trials[ to be at high risk of bias for blinding of the assessors for the cognitive outcomes because methods regarding the assessment of cognition were not provided. The other 9 trials[ provided detailed information for this process, so we considered the studies to have a low risk of bias. Four trials[ were judged to be at high risk of bias for addressing incomplete data, based on the dropout being more than 15%, which we considered incomplete data.[ Another 7 trials[ were judged to be at low risk of bias for this item. Three trials[ were judged to be at unclear risk of bias for selective reporting since predetermined outcomes were not reported. All other 8 trials[ were judged to be at low risk of bias in this respect. Seven trials[ reported dropouts per group and analyzed using ITT (intention-to-treat) analysis principles. The other 4 trials[ were judged to be at high risk of bias for this item since they reported dropouts but either lacked information on the group assignment of these dropouts, on the ITT analysis or both. No other apparent bias was identified in each of the studies. According to the criterion in the Modified Assessment Form, 5 studies reported high-quality research, and 6 studies reported mid-quality research (see Table 2).

Table 2

Quality assessment of eligible studies.

According to Cochrane Handbook for Systematic Reviews of Interventions: “the funnel plot asymmetry test can only be used when the meta-analysis is included in at least 10 studies, because if the inclusion of the study is too small, the test performance will be too low, and the opportunity and true asymmetry will not be distinguished. Owing to the limited number (below 10) of studies included in each outcome analysis, the meta-analysis can’t be done.” the funnel plot asymmetry test can only be used when the meta-analysis is included in at least 10 studies, because if the inclusion of the study is too small, the test performance will be too low, and the opportunity and true asymmetry will not be distinguished. Owing to the limited number (below 10) of studies included in each outcome analysis, the meta-analysis can’t be done. so, publication bias was not assessed.

Data Extraction

Posttest effects of exercise interventions on cognitive function

The effects of all subgroup meta-analyses conducted are presented in Figure 2  A∼2G.

Figure 2

A. Forest plot of exercise versus no exercise on general cognition in MCI patients. Outcomes: MMSE, MoCA and ADAS-cog. MMSE, Mini-mental state examination; MoCA, montreal cognitive assessment; ADAS-cog, Alzheimer disease assessment scale-cognition. Figure 2-B. Forest plot of exercise versus no exercise on executive functions in MCI patients. Outcomes: DSF, DSB, DSF-B, ST-A, ST-B and TMT B-A. DSF, digit span forward test; DSB, digit span backward test; DSF-B, digit span forward/backward; ST-A, stroop test A; ST-B, stroop test B; TMT B-A, trail-making test Part B–A. Figure 2-C. Forest plot of exercise versus no exercise on memory in MCI patients. Outcomes: WMS-I and WMS-II. WMS-I, Wechsler memory scale immediate recall; WMS-II, Wechsler memory scale delayed recall. Figure 2-D. Forest plot of exercise versus no exercise on language ability in MCI patients. Outcomes: CVFT and LVFT. CVFT, Category verbal fluency test; LVFT, letter verbal fluency test. Figure 2-E. Forest plot of exercise versus no exercise on visuospatial ability in MCI patients. Outcome: BDS. BDS, block design score.

Mini-mental state examination (MMSE) and Montreal cognitive assessment (MoCA). Using a random effects model, the meta-analysis of 6 studies[ with 355 patients shown that exercise could have improved the MMSE and MoCA scores (g = 0.32, 95% CI 0.1 to 0.54, z = 2.8, P = .005, Fig. 2  -A). There was low heterogeneity among studies (I2 = 0%).

Alzheimer disease assessment scale-cognition (ADAS-cog). Two studies[ were included to estimate the effect of exercise on general cognition assessed with the ADAS-cog. The exercise group outperformed the control group in terms of improving general cognition using a random effects model (g = −0.45, 95% CI -0.82 to -0.08, z = 2.38, P = .02, Fig. 2  -A). Heterogeneity was low between studies (I2 = 20%).

Digit span forward test (DSF), Digit span backward test (DSB), DSF/Backward, Stroop Test A, stroop test B. Five studies[ with 243 patients were included to assess the effects of exercise on executive functions. Exercise could have improved executive functions (g = 0.66, 95% CI 0.17 to 1.15, z = 2.26, P = .008, Fig. 2  -B). Heterogeneity among the 5 exercise versus nonexercised control studies was significant (I2 = 77%).

Trail-Making Test Part B–A (TMT B-A). Three studies[ with 203 patients were included to assess the effect of exercise on executive function. Exercise had a nonsignificant positive effect on executive function assessed using a random effects model (g = −0.25, 95% CI -0.88 to 0.39, z = 0.76, P = .45, Fig. 2  -B). High heterogeneity between studies (I2 = 79%).

Memory

Wechsler memory scale immediate recall, Wechsler memory scale delayed recall. The meta-analysis of 4 studies[ with 295 patients who were suitable for inclusion indicated that exercise had a significant effect on memory (g = 0.37, 95% CI 0.15 to 0.60, z = 3.21, P = .001, Fig. 2  -C). There was low heterogeneity between studies (I2 = 33%). One study[ used the Rey 15-Item memory test as an outcome measure; therefore, no meta-analyses could be conducted.

Language ability

Category Verbal Fluency Test, letter verbal fluency test. Two studies[ with 72 patients were included to assess the effect of exercise on language ability. Exercise could have improved language ability (g = 0.55, 95% CI 0.22 to 0.89, z = 3.23, P = .001, Fig. 2  -D). There was no heterogeneity between studies (I2 = 0%).

Visuospatial ability

Block design score (BDS). The meta-analysis of 2 studies[ with 132 patients who were suitable for inclusion indicated that exercise had a significant effect on the visuospatial ability (g = 0.38, 95%CI 0.03 to 0.72, z = 2.15, P = 0.03, Fig. 2  -E). There was low heterogeneity between studies (I2 = 0%).

Overall effect results

The Z (P value) overall effect results of all meta-analyses conducted are presented in Table 3 and are summarized as follows: general cognitive (MMSE, MoCA, Z = 2.80 [.005]; ADAS-cog, Z = 2.38 [.02]), executive functions (DSF, DSB, DSF/Backward, strop test-A, stroop test B, Z = 2.66 [.008]), memory (Wechsler memory scale immediate recall, Wechsler memory scale delayed recall, Z = 3.21 [.001]), language ability (category verbal fluency test, letter verbal fluency test, Z = 3.23 [.001]), and visuospatial ability (BDS, Z = 2.15 [.03]).

Table 3

The overall effect results of eligible meta-analyses.

Discussion

Exercise intervention is beneficial for cognitive ability in older adults with MCI. Previous studies have examined the results of exercise versus no exercise on cognitive function in MCI patients.[ However, none have focused on conducting a subgroup analysis of the effects of exercise on different cognitive domains in the elderly with MCI. Therefore, this review integrated evidence from a large number of trials that evaluated the effect of exercise on different cognitive domains in the elderly with MCI and produced a preliminary ranking of the degree or magnitude of the influence of certain exercise interventions on various cognitive domains. This meta-analysis included 676 patients who participated in eleven RCTs. Five high- quality and 6 mid-quality research reports are included in this systematic review according to the modified assessment form of Cochrane Collaboration recommendations and PEDro.[ The meta- analysis revealed that exercise had a significant positive effect on general cognition, executive functions, memory, language ability and visuospatial ability in patients with MCI; however, exercise interventions have not shown benefit on executive functions measured by TMT B-A. Only 1 study described processing speed using the digit symbol- coding subset of the Wechsler adult intelligence scale III,[ therefore, no meta-analyses could be conducted.

We used the chi-square test and I2 statistic to assess the heterogeneity among the studies.

The pooled estimate of the effect for the TMT B-A (Fig. 2  -B) was extremely heterogeneous (I-squared 79%). It is probable that there are important clinical and methodological differences among studies. These might include differences in populations since of the 3 articles, two studies[ came from Thailand and one[ came from Canada, or differences in the experimental and control groups such that the experimental and control groups of 2 studies[ were Taiji and health education, and the other study[ had aerobic training and no-treatment groups. At the same time, we consider TMT B-A indicators to be less specific or sensitive to the measurement of executive functions, so it is recommended to use more sensitive indicators to measure the impact of exercise on executive functions. In the future, the effect of exercise interventions on executive functions in MCI patients need to be further explored. The pooled estimate of effect for the DSF and DSB (Fig. 2  -B) shown a high level of heterogeneity (I-squared 77%). Clinical and methodological differences may exist among these studies: three studies[ used health education, 1 study[ used a no-treatment control and one study[ used muscle-stretching exercise. There were also differences in intensity and duration that may have led to different results. Accordingly, we found that different durations of training were used for the included articles, in which 2 studies[ trained for twelve weeks, 1 study[ trained for fifteen weeks and 2 studies[ trained for 6 months.

The preliminary ranking of the effect of exercise interventions across domains of cognitive function based on the overall effect was Zlanguage ability >  memory > Zexecutive functions > Zvisuospatial ability. The effect of exercise interventions on language ability in MCI patients was most significant. Neva and his colleagues (2017) found that a session of exercise decreased intracortical and interhemispheric inhibition and then improved learning skills.[ Furthermore, group activities promoted interpersonal relations. Sports programs in groups have been shown to enrich old people's lives, harmonize feelings and provide more opportunities for communication among the elderly. This study demonstrated that language ability itself as a communication tool in social interactions could be the target, thereby making therapies on language ability more effective. In contrast, the strategy of focusing on the discourse itself seems less effective, at least in the current noncommunicative context of confrontation naming.[ Berthier and his group also found that if verbal utterances are embedded in behaviorally relevant settings (in the context of group exercise) the outcome of speech-language therapy may improve.[

Among the other cognitive functions, visuospatial ability ranked last. In this study, there were only 2 studies,[ which used BDS to measure visuospatial ability. The effect of exercise on visuospatial ability was more prominent at the beginning of the exercise intervention and reached a plateau over time,[ which may be the reason why long-term exercise intervention has no significant effect on visual spatial ability. Therefore, the ability of exercise to improve visuospatial ability may be limited. An alternative explanation may be the limited duration of the exercise and the assessment of visuospatial ability in the maintenance phase.[ A study shown that solving jigsaw puzzles was highly correlated with performance in visuospatial reasoning tasks,[ and engagement in intellectual activities, including jigsaw puzzles, predicted a reduced risk of dementia.[

Limitations

This systematic review has certain limitations. First, the outcome assessors could not be blinded for the comparisons between the exercise group and the no-exercise groups, therefore generating potential response and performance bias. Second, there was no distinction between primary and secondary outcomes related to cognitive functions in this study, so perhaps the results are at higher risk of chance findings due to selective reporting and the multiplicity of testing. Third, different types of exercise may have led to clinical heterogeneity to some extent. Fourth, there were a limited number of the eligible studies with the sample sizes ranged from 22 to 100 which may impact the present findings. Fifth, we did not distinguish assessments during the exercise intervention and during the follow-up time (up to 6 months), and we used the data closest to the sixth month to perform the meta-analysis. The effects of some exercise interventions may have delayed effects, which may cause bias in the comparison of outcome indicators and affect the comparability of outcome indicators.

Implications

First, the study shown that the effect of exercise on the improvement of language ability was most significant, although previous studies seldom used the language ability to evaluate the effectiveness of an exercise intervention. Therefore, future studies of exercise interventions should pay more attention to changes in language ability.

Second, this study shown that there were differences in the overall effect of exercise interventions across various cognitive domains. However, whether the observed differences are significant remains to be verified by a rigorous research design. These differences provide a possibility for medical workers to conduct personalized exercise interventions. For MCI patients with severe memory impairment, the effects of exercise intervention may not be as good as cognitive training; however, for MCI patients with more serious executive function impairment, exercise training will be an effective attempt to delay their cognitive impairment.

Third, this study did not explore the effect of the specific exercise intervention (aerobic exercise, resistance exercise or multicomponent exercise) on the specific cognitive domain. Therefore, to provide more accurate guidance for improving the function of the specific cognitive domains, more high-quality RCT studies are needed to reveal the relationship between the specific exercise methods and effects on specific cognitive domains.

Acknowledgments

This work was supported by the National Natural Science Foundation (NO.71704053);

China Scholarship Council (NO.201908330251); and Zhejiang Province Natural Science

Foundation (NO. LQ17G030002). Thank statistician Xueqiang Wang for providing statistical guidance to this study.

Author contributions

LNW and XHS conceived the study. XLZ and JW independently work on study selection, quality assessment, data extraction, and synthesis. XLZ and LNW wrote the first draft of the manuscript. XHS provided substantial input into all aspects of the manuscript. All authors contributed to and approved the final manuscript.

Conceptualization: Lina Wang, Xin-Hua Shen.

Data curation: Xianglian Zhou, Jie Wang.

Formal analysis: Xianglian Zhou, Lina Wang, Jie Wang.

Funding acquisition: Lina Wang.

Methodology: Xianglian Zhou, Lina Wang, Jie Wang, Xin-Hua Shen.

Project administration: Xin-Hua Shen.

Software: Xianglian Zhou, Jie Wang.

Supervision: Lina Wang, Xin-Hua Shen.

Validation: Lina Wang, Xin-Hua Shen.

Writing – original draft: Xianglian Zhou, Lina Wang.

Writing – review and editing: Xianglian Zhou, Lina Wang.