Different training durations and styles of tai chi for glucose control in patients with type 2 diabetes: a systematic review and meta-analysis of controlled trials

We performed a search for Chinese and English studies in the following databases: PubMed, Cochrane Central Register of Controlled Trials (CENTRAL), Web of Science, Ovid LWW, Chinese Biomedical Literature Database (CBM), China Knowledge Resource Integrated Database (CNKI), Wanfang Data, and China Science and Technology Journal Database (VIP). The search strategy included terms relating to or describing Tai Chi and type 2 diabetes. Studies published between when the database establishment and April 2018 were retrieved.

Bibliographies of the related published systematic reviews were also reviewed. An illustrative PubMed search strategy is as follows: “Type 2 Diabetes Mellitus, Noninsulin-Dependent Diabetes Mellitus, Ketosis-Resistant Diabetes Mellitus, Non Insulin Dependent Diabetes Mellitus, Non-Insulin-Dependent Diabetes Mellitus, Stable Diabetes Mellitus, Type II Diabetes Mellitus, NIDDM Diabetes Mellitus, Noninsulin Dependent Diabetes Mellitus, Maturity-Onset Diabetes Mellitus, Maturity Onset Diabetes Mellitus, MODY Diabetes Mellitus, Slow-Onset Diabetes Mellitus, Maturity-Onset Diabetes, Maturity Onset Diabetes, Type 2 Diabetes, Adult-Onset Diabetes Mellitus” and “Tai Ji, Tai-ji, Tai Chi, Chi, Tai, Tai Ji Quan, Ji Quan, Tai, Quan, Tai Ji, Taiji, Taijiquan, T’ai Chi, Tai Chi Chuan”.

Studies should meet the following inclusion criteria (PICO format): (1) Participants: no restrictions on patients’ age, gender, disease duration, case source, nationality, or race; with a clear diagnosis of type 2 diabetes; without any serious complications. (2) Intervention: participation in Tai Chi as the major intervention. (3) Control: any type of control group, including usual care or standard treatment, and any kind of exercise was acceptable. (4) Outcomes: primary outcomes were HbA1c and fasting blood glucose. Secondary outcomes were total cholesterol (TC), triglycerides (TG), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), and body mass index (BMI). (5) Study type: RCTs. Nonrandomized controlled trials (N-RCTs), as well as cohort studies and case-control studies, were excluded.

First, two independent investigators reviewed the titles and abstracts. Abstracts that did not meet the eligibility criteria were excluded, and those that did not provide sufficient information about the inclusion criteria were further reviewed. Next, the same investigators analysed the full texts, blinded to each other’s review. Differences between the reviewers were resolved by consensus.

Two investigators independently performed the data extraction using pre-piloted, standardized forms. The collected data included: basic information; methodological characteristics; risk of bias (ROB); participants’ demographic details; interventions; outcomes, fall outs; results of outcomes; and other. All differences were resolved by consensus.

Two investigators independently assessed the methodological quality of the included studies using RevMan 5.3.0, according to the Cochrane Handbook criteria for judging the ROB with the “Risk of bias” assessment tool [10].

From the electronic bibliographic databases, we retrieved 356 original papers published between 1983 and 2018. After 176 duplicates were excluded, we screened the titles and abstracts of 180 publications; another 119 papers did not meet the inclusion criteria and were excluded. We downloaded the full text of the remaining 71 publications for further screening. We subsequently evaluated these retrieved articles by deep reading. Due to an ineligible study design, ineligible participants, inappropriate interventions, lack of controls, or unrelated outcome of primary interest, 54 studies were excluded. Finally, seventeen articles met the inclusion criteria completely and were analysed (Fig. 1).

All seventeen trials were conducted in single centres (Table 1). The settings of the included studies were different, and included China [11‐22], Taiwan (China) [23], Australia [24‐26] and Thailand [27]. The simplified Tai Chi style was used in five RCTs [13, 14, 19, 20], Yang-style Tai Chi was used in two trials [15, 26], Chen-style was used in three trials [21, 22], Sun-style and Yang-style were used in two trials [24, 25], Lin-style was used in one trial [27], Da Yuan Jiang Tang-style was used in one study [16], Tai Chi Ball was used in two studies [12, 18], and two trials did not report the Tai Chi style [11, 17]. The interventions in the control group, which included standard diabetes care (diet control, injection of insulin or anti-diabetic medication, and usual care) [14, 16‐19, 22, 27], active control (jogging, brisk walking, Yangko or social dancing for exercise, and other aerobic exercises) [13, 15, 21, 23, 25, 26], and no treatment (free activity programme, wait list) [11, 12, 24], could be regarded as conventional treatment or relaxation exercises. The sample sizes ranged from 16 to 100 participants. The number of Tai Chi sessions ranged from approximately 36–336. The number of supervised interventions ranged from two to seven sessions weekly, with 30–60 min per session.

The methodological quality of the included studies was generally low (Fig. 2 and Fig. 3). Because participants cannot be blinded to the Tai Chi intervention, performance bias could not be ruled out. Four studies [23‐26] reported the random method of using a computer-generated random sequence, while one study [20] used only a random number table to divide participants into experimental and control groups. Twelve studies lacked descriptions of the method of random sequence generation. Fourteen studies did not describe how the allocation concealment was conducted, while three studies [25‐27] used sealed opaque envelopes to perform the allocation concealment.

All studies lacked details on whether participants and administrators were blinded; however, it was clear that the blinding had been broken due to obvious differences between the intervention and control groups. Four trials used assessor blinding [24‐27]. Fourteen studies reported all patient outcomes, while three studies [14, 25, 26] had a high drop-out rate; however, the drop-outs occurred at random. Additionally, intention-to-treat analyses were performed, one [14] of which provided detailed explanations. Two articles [25, 26] were from the same study and had different outcome indicators, but they had high selectivity bias and incomplete data. Most studies reported all outcomes listed in their methods section. The information necessary for judging the risk of other bias of all studies was insufficient.

Nine studies [12, 15, 16, 18, 20, 21, 23, 24, 27] were included with a total of 517 patients, of whom 266 went through Tai Chi intervention. There was substantial heterogeneity among the included studies (I² = 85%, P < 0.00001), so we first performed a quantitative synthesis using a random-effects model, which showed a significant difference between Tai Chi and the control group in reducing HbA1c (SMD = − 0.68, 95% CI (− 1.17, − 0.19), P = 0.006) (Fig. 7).

Eight trials [11, 13‐15, 18, 21, 22, 24] reported data on the change in TC following Tai Chi practice with 186 participants. The meta-analysis showed that there was significant heterogeneity among the studies (I² = 94%, P < 0.00001). The analysis found that a small sample size was included in one study [14], and there was a large difference in subjects’ baseline values. A sensitivity analysis was used. Quantitative consolidation was performed after the study was removed. The heterogeneity of the seven studies was relatively low (I² = 47%, P = 0.08). Pooled blood TC levels decreased in the Tai Chi group compared with the control group, and the difference was significant (SMD = − 0.35, 95% CI (− 0.54, − 0.16), P = 0.0003) (Fig. 10).

Eight studies [11, 13‐15, 18, 21, 22, 24] reported TG levels in 359 patients, of whom 186 went through a Tai Chi intervention. The heterogeneity of the eight studies was relatively low (I² = 43%, P = 0.09). The TG levels decreased in the Tai Chi group compared with the control group, and the difference was significant (SMD = − 0.19, 95% CI (− 0.31, − 0.07), P = 0.002) (Fig. 11).

Six RCT trials [11, 13, 15, 18, 21, 22] with 290 participants provided data on blood HDL-C levels. The heterogeneity of the 6 studies was low (I² = 0%, P = 0.54). The HDL-C levels improved in the Tai Chi group compared with the control group, but the difference was not significant (SMD = 0.04, 95% CI (− 0.01, 0.09), P = 0.14) (Fig. 12).

LDL-C levels were reported in 6 trials [11, 13, 15, 18, 21, 22] with a total of 290 participants. The heterogeneity of the 6 studies was high (I² = 80%, P = 0.0001). The LDL-C levels decreased in the Tai Chi group compared with the control group, but the difference was not significant (SMD = − 0.49, 95% CI (− 1.06, 0.08), P = 0.09) (Fig. 13).

Six trials [13, 14, 18, 21, 22, 27] reported data on the change in BMI following a Tai Chi practice in 296 participants. The heterogeneity of the 6 studies was relatively low (I² = 38%, P = 0.15). The BMI decreased in the Tai Chi group compared with the control group, and the difference was significant (SMD = − 0.61, 95% CI (− 0.85, − 0.38), P < 0.00001) (Fig. 14).

Safety monitoring: Twelve studies did not report adverse events. Five studies [11, 23, 25‐27] reported that there were no adverse events.

Publication bias: The FBG included in the study was selected as the indicator. It can be seen that the graph is obviously asymmetrical (Fig. 15). There might have been publication bias in the comparison of Tai Chi and the control group, as the results showed that the distribution of the included studies on both sides of the funnel plot was asymmetric.

Overall, the included studies proved that Tai Chi might be beneficial for treating T2DM. This systematic review and meta-analysis showed that blood glucose control was greater among Tai Chi intervention groups than in control groups. Significant differences were found in reducing FBG and HbA1c, although the heterogeneity was high.

Considering the subgroup analysis for FBG, the pooled results showed that 24 movements or Yang-style Tai Chi did not significantly reduce FBG, either after a duration ≤3 months or a duration > 3 months [13‐15, 19, 21]. However, the pooled results showed that other styles of Tai Chi did not significantly reduce FBG after a duration ≤3 months with high heterogeneity, while a significant difference was found after a duration > 3 months with low heterogeneity [16, 18, 21‐23, 27]. There was no significant difference in reducing HbA1c between 24 movements or Yang-style and the control group, and the duration in all studies was > 3 months. However, other styles of Tai Chi significantly reduced HbA1c after a duration > 3 months with low heterogeneity, while a significant difference was not found after a duration ≤3 months with high heterogeneity [12, 16, 18, 21, 23, 27].

The exercise intensity of Tai Chi mainly depends on the training style, posture, frequency and duration. Furthermore, it is a complex, multi-component intervention that integrates physical, psycho-social, emotional, spiritual, and behavioural elements, and not just the utilization of oxygen [34, 35]. Our review first showed that different training durations and styles could result in variable effectiveness. The subgroup analysis was conducted according to the training durations and styles of Tai Chi. The results showed that other styles of Tai Chi significantly reduced FBG and HbA1c after a duration > 3 months. However, 24 movements or Yang-style Tai Chi had no effectiveness in reducing FBG or HbA1c, either after a duration ≤3 months or > 3 months. Our study also focused on TC, TG, HDL-C, and LDL-C levels, and we found that Tai Chi has a beneficial effect on controlling blood lipid levels.

In recent years, more and more research has focused on the Tai Chi training style, posture, frequency and duration. Studies have shown that Tai Chi might affect diabetic patients by improving the motor nerve conduction velocity of the bilateral central and sacral nerves, improving the distal sensory latency of the bilateral ulnar nerves, and increasing vascular function, especially increasing the vascular reactivity index [36, 37]. The results of one study showed that the longer the Tai Chi practice, the stronger the tactile space acumen [38]. Some studies have shown that Tai Chi can enhance the use of the active neural networks through an all-inclusive mind-body approach [39, 40]. Not only has it been shown that the effects of Tai Chi on proprioception in the ankle and knee joints were better than in sedentary controls, but also that the effects on ankle kinaesthesis were better than in swimmers and runners [41]. Such training might decrease the potential for developing joint injuries or foot ulcers, which occur in a large proportion of people with PN [41]. It can be seen that Tai Chi can play an important role in such combinatorial therapies.

The quality of the included studies was not high, with twelve studies lacking a description of the randomization method, and fourteen studies not mentioning allocation concealment. In addition, selection bias existed. The outcome assessors were successfully blinded in four studies [25‐27], while no studies successfully blinded the researchers or participants. Regarding the safety of Tai Chi, twelve studies didn’t reported any adverse events due to Tai Chi. Five studies [11, 23, 25‐27] reported that there were no adverse events. So the evidence was insufficient. More research on the safety assessment of Tai Chi exercise for people with type 2 diabetes is needed.

The greatest concern with this study is regarding the distribution of the included studies in the meta-analysis. Most studies were conducted in Asian countries such as China (Taiwan) or Australia; no studies were from the USA or western countries. We only performed a search for Chinese and English studies, and it is possible that articles may have been published in other languages, which caused the publication bias. Moreover, in our review, the styles of Tai Chi intervention and outcome measurement tools varied. We could not make higher quality and more comprehensive subgroups due to the limited number of included studies. This may have influenced the explanatory effect and the soundness of the pooled effects. The methodological quality of the included trials was not promising; in addition to the fact that Tai Chi is hard to be blinded to, the included studies have other flaws, such as poor randomization. Undoubtedly, the credibility of subgroup effects in this study was low, due to the poor quality of the studies included in this meta-analysis. At the same time, according to the criteria to assess the credibility of subgroup effects [42‐44], the credibility of subgroup effects in this study was still low. Although we have determined the characteristics of the study in advance and tried to select a small number of research features, the reliability of the subgroup effect is not guaranteed. This study was the first discussion of different training durations and styles. The results of the subgroup analysis should be treated with caution.

Heterogeneity among the studies was significant, which may be explained by the inconsistent measurement of exercise intensity. We conducted a sensitivity analysis of the included RCTs and found two studies that could have been the main source of heterogeneity. The target heart rate was used as a measurement of exercise intensity in both studies, which had a great impact on the combined results. The intervention in the control group, which included standard diabetes care, active control, and no treatment, was an important factor in generating heterogeneity. Subject inclusion criteria was also a non-negligible factor in generating heterogeneity. Although the inclusion criteria for the study uniformly included without serious complications, there were some differences in the detailed criteria for each study. For example, some studies have defined blood glucose, blood pressure, and blood lipid levels in the criteria. The variation in training styles, durations and outcome measurement tools resulted in substantial differences in exercise intensity.

We summarized the current condition of Tai Chi for type 2 diabetes and provided information to support a future clinical trial. Although this study showed that Tai Chi might be effective for controlling blood glucose and blood lipid levels, the current evidence and potential findings should be interpreted carefully because of the poor methodological quality of these studies, insufficient evidence for safety and clinical heterogeneity. Future studies should pay more attention to the effects of the Tai Chi styles and exercise intensity measurement methods on patients with type 2 diabetes. Overall, Tai Chi exercise may be used to control blood glucose and blood lipid levels.