This research protocol was registered with the International Prospective Register of Systematic Reviews (PROSPERO) on 15 March 2016 (CRD42016036469) [25] . In addition, The Preferred Reporting Items for Systematic Review and Meta-Analysis Protocols (PRISMA-P) 2015 statement was consulted and provided the structure for this review [26] .

Five electronic databases (PubMed, GoogleScholar, Chinese National Knowledge Infrastructure, Chinese Scientific Journal Database, Chinese Biomedical Literature Database [CBM]) were used to conduct literature searches, along with the combination strategy of: Taichi ball, Taichi softball, Taiji ball, or Taichi softball. The studies were included if they met the following criteria: 1) only peer-re- viewed studies in English or Chinese because reviewer authors are proficient in both languages; 2) original randomized controlled studies published from September 2000 to February 2016; 3) Taichi softball as the main intervention in the studies; 4) a minimum of one component relating to handgrip strength, balance, or trunk flexibility as primary outcomes. Except the primary outcomes, other components of the physical health (e.g., systolic and diastolic pressures, resting heart rate, aerobic endurance, shoulder mobility, and bone mineral density and metabolism) were also reported. In addition, because of the limited number of randomized controlled trials carried out, additional study designs investigating the effectiveness of Taichi softball were also displayed to obtain a more comprehensive understanding (Table 1), including non-randomized controlled studies [23] [27] [28] , non-controlled observational studies [29] [30] , and cross-sec- tional studies [31] .

For each included study, two reviewer authors (xx and xx) independently extracted data in a standardized way. The data extraction included the following: the first author, year of publication, sample size, intervention duration and frequency, and details of the control group, sex and age of study participants, study design, and components of the physical health consisting of primary outcome measures (hand-grip strength, balance, or trunk flexibility), secondary outcomes

T = Taichi softball group; C = control group; RCT = Randomized controlled trial; CSS = Cross-sectional trial; NRCT = non-randomized controlled trial; NOT = non-controlled observational trial.

(systolic and diastolic pressures, resting heart rate, aerobic endurance, and mobility), and additional physical health-related components. A third party appeared and had dealt with disagreements between the two reviewers.

With respect to the change scores and standard deviations for the primary and secondary outcomes, if authors did not report the change score data, reviewer (xxx) first tried to contact the authors via email or phone call to obtain the data. In cases where the data was not obtainable, reviewers used one of the following methods: a) if baseline scores between two groups were not significantly different, scores from posttest were used for data analysis; b) if baseline scores were different, reviewers attempted to estimate the change scores and standard deviation through standard formulas provided by Cochrane Handbook for Systematic Reviews of Interventions [32] . If reviewers were unable to find the relevant information for estimating the change scores and standard deviations, the study was excluded. In addition, the study [33] with two interventions and one control groups, review author combined two group (non-Taichi softball intervention and control group) to create a single pair-wise comparison. The method is recommended by Crochrane Handbook for Systematic Reviews of Interventions 16.5.4. (how to include multiple groups from one study).

The methodological quality of each study was independently evaluated according to the Cochrane Collaboration’s for Assessing Risk of Bias from the Cochrane Handbook for Systematic Review Interventions [32] . The type of bias is classified as the following domains: 1) selection bias (sequence generation and allocation concealment); 2) performance bias (blinding of participants and personnel); 3) detection bias (blinding of outcome assessment); 4) attrition bias (incomplete outcome data); 5) reporting bias (selective outcome reporting), and other biases. Judgment of the reviewers on each of seven domains of each included study was based on three categories (low risk of bias, high risk of bias, or unclear risk of bias). The specific requirements of the judgment for each category can be found from Table 8.5. d. of the Cochrane Handbook (criteria for judging risk of bias in the risk of bias assessment tool). A method was used for summary assessment of the risk of bias for each important outcome and across studies on basis of: a) low risk of bias for all key domains (“Low” indicating low risk of bias); b) unclear risk of bias for one or more key domains (“unclear” indicating unclear risk of bias); c) high risk of bias for one or more key domains (high indicating high risk of bias).

Revman 5.3 software within the Cochrane Collaboration for data analysis was used to synthesize a subset of the most commonly reported outcomes from the six randomized controlled trials. The standard mean difference (SMD), along with 95% confidence interval (CI) using more conservative random effects model, was used to calculate heterogeneity for the primary and secondary continuous outcomes [34] . The value of I² was used to test heterogeneity across the included studies. Studies with I² ranging between 25% and 50% for low heterogeneity, I² ranging between 50% and 75% for moderate heterogeneity, and I² > 75% for high heterogeneity were considered respectively [35] . A funnel plot for the primary outcomes (handgrip strength, trunk flexibility, static and dynamic balance) was used to explore publication bias.

As displayed in Figure 1, the flow chart reflects the search strategies (search

process and study selection). A total of 70 titles were identified using the following databases: PubMed (n = 2), Google Scholar (n = 19), Chinese National Knowledge Infrastructure (n =15), Chinese Scientific Journal Database (n = 16), Chinese Biomedical Literature Database (n = 12), and other sources (n = 6). After removing 38 duplicates, 32 articles were screened. 15 article were excluded based on two reasons (no relevance = 10; not a full text article = 5). Full texts of 17 articles were retrieved, and 11 articles were excluded with the following reasons: no components of the physical health reported (n = 4), no Taichi softball intervention (n = 3), and incomplete data extraction (n = 4). The final number of six RCTs were included for meta-analysis, including two studies in English [24] [36] and four studies in Chinese [21] [33] [37] [38] .

The sample size of 411 participants ranged from 32 to 150 in the RCTs, along with a wide age range from 18 to 75. The participants in the studies consisted of healthy college students [33] [38] , patients with Type 2 diabetes [24] [37] , and community dwellers [21] [26] . The control groups either involved no training, [21] [24] [38] jogging and aerobic exercises [26] , regular drug treatment [37] or BaduanJin [33] , and the participants in the experimental group experienced the Taichi softball program [21] [24] [33] [36] [37] [38] ; with an exercise time range from 50-to-120 minute; three-to-six sessions weekly; intervention period ranging from 12 weeks to 12 months. The characteristics of the included articles along with additional relevant non-RCTs [23] [27] [29] [30] [31] [39] were shown in Table 1.

As shown in Figure 2, according to the quality assessment criteria (the Cochrane Collaboration’s for Assessing Risk of Bias from the Cochrane Handbook for Systematic Review Interventions) [32] , the majority of the six RCTs were consi-

dered to be poor methodological quality [21] [24] [33] [36] [37] [38] . Although the randomized allocation of participants in all studies was described, none of the studies included methods of random sequence generation (e.g. referring to a random number table, using a computer random number generator, or coin tossing). Except two items (incomplete outcome data for attrition bias and selective reporting for reporting bias) of all studies with low risk bias [21] [24] [33] [36] [37] [38] , the rest of items (allocation concealment, blinding of participants and personnel, blinding of outcome assessment, and other bias) for all studies remained unclear. In order to obtain the original methods about the five items, one of researchers of the present study (HRW) contacted the authors via email that was found on from the contact information listed on the authors publication, but did not receive any response. Six follow-up phone calls via Skype were made. Only two authors answered but stated that they lost the original data.

Outcomes related to physical health were mainly comprised of three primary components with the six RTCs (hand strength, flexibility using sit-and-reach, and balance), along with secondary components (aerobic endurance, resting heart rate, systolic and diastolic pressures) [21] [24] [33] [36] [37] [38] . In order to obtain a more comprehensive understanding, additional components of physical health from non-RCTS [23] [27] [29] [30] [31] [39] were also depicted in Table 1 (e.g., BMI, leg strength, agility, reaction time, aerobic endurance, shoulder mobility, pulmonary function, and bone mineral density).

Four RCTs (a total of participants = 298) reported handgrip strength with handgrip dynamometer [24] [33] [37] [38] . Meta-analyses were performed to compare the Taichi softball program with non-intervention controls [24] [38] , Baduanjin (one of the Chinese traditional health-promoting exercises) and non-intervention controls [33] , and routine drug treatment [37] . The overall results of these studies suggest that Taichi softball is associated with significantly improved handgrip strength (SMD = −0.6, 95% CI [−0.84, −0.36], P value of 0.55 for heterogeneity with I² = 0%, P value of test for overall effect < 0.00001; Figure 3).

Four RCTs (a total of participants = 298) reported flexibility with sit- and-reach test [24] [33] [37] [38] . Meta-analyses were performed to compare Taichi softball program with non-intervention controls [24] [38] , Baduanjin (one of Chinese traditional health-promoting exercises) and non-intervention controls [33] , and routine drug treatment [37] . The overall results of these studies suggest that Taichi softball is associated with significantly improved trunk flexibility (SMD = −0.4, 95% CI [−0.74, −0.05], P value of 0.13 for heterogeneity with I² = 47%, P value of test for overall effect = 0.03; Figure 3).

Five RCTs (a total of participants = 359) reported static balance [21] [24] [33] [36] [38] . Meta-analyses were performed to compare the Taichi softball program with non-intervention controls [21] [24] [38] , Baduanjin (one of the Chinese traditional health-promoting exercises) and non-intervention controls [33] , and other physical exercise (but the authors did not mention the extract name of the exercise) [36] . The overall results of these studies suggest that Taichi softball is associated with significantly improved balance performance (SMD = −0.73, 95% CI [−0.94, −0.51], P value of 0.79 for heterogeneity with I² = 0%, P value of test for overall effect < 0.00001; Figure 3).

Two RCTs (a total of participants = 104) reported mobility (dynamic balance) with Timed Up and Go test [24] [36] . Meta-analyses were performed to compare the Taichi softball program with non-intervention controls [24] , and exercise (but the authors did not mention the extract name of the exercise) [36] . The overall results of these studies suggest that Taichi softball is associated with significantly improved balance performance (SMD = −0.68, 95% CI [−1.2, −0.17], P value of 0.21 for heterogeneity with I² = 35%, P value of test for overall effect = 0.009).

Two RCTs (a total of participants = 182) reported aerobic endurance [24] [33] . Meta-analyses were performed to compare the Taichi softball program with non-intervention controls [24] , Baduanjin (one of the Chinese traditional health-promoting exercises) and non-intervention controls [33] . The overall results of these studies suggest that Taichi softball is associated with significantly improved aerobic endurance performance (SMD = −0.69, 95% CI [−1.04, −0.34], P value of 0.30 for heterogeneity with I2 = 9%, P value of test for overall effect < 0.0001; Figure 4).

Two RCTs (a total of participants = 202) reported resting heart rate [33] [37] . Meta-analyses were performed to compare the Taichi softball program with non-intervention controls [33] , and routine drug treatment [37] . The overall results of these studies suggest that Taichi softball is associated with significantly improved cardiovascular function (SMD = −1.45, 95% CI [−1.77, −1.12], P value of 1 for heterogeneity with I² = 0%, P value of test for overall effect < 0.00001; Figure 5).

Two RCTs (a total of participants = 202) reported systolic pressure [33] [37] . Meta-analyses were performed to compare the Taichi softball program with non-intervention controls [33] , and routine drug treatment [37] . The overall results of these studies suggest that Taichi softball is associated with significantly decreased systolic pressure (SMD = −0.24, 95% CI [−0.53, −0.05], P value of 0.35 for heterogeneity with I² = 0%, P value of test for overall effect = 0.10; Figure 6). Two RCTs (a total of participants = 202) reported diastolic pressure [33] [37] . Meta-analyses were performed to compare the Taichi softball program with non-intervention controls [31] , and routine drug treatment [37] . The overall results of these studies suggest that Taichi softball is associated with significantly decreased diastolic pressure (SMD = −0.20, 95% CI [−0.49, −0.08], P value of 0.43 for heterogeneity with I² = 0%, P value of test for overall effect = 0.17; Figure 6).

A funnel plot for the primary outcomes was performed, including handgrip strength (n = 4 RCTs) [24] [29] [30] [31] , trunk flexibility (n = 4 RCTs) [24] [33] [37] [38] , static (n = 5) [21] [24] [33] [36] [38] , and dynamic balances (n = 2 RCTs) [24] [36] . The review authors have difficulty to interpret the positive findings because of the limiting number of RCTs in the current study (Figure 7).

None of RTCs mentioned adverse effects (e.g. fall, hospitalization, side effects of medication, or death) resulting from participating in Taichi softball program.