Recently, rehabilitative exercise therapies have been described as an important method of overcoming the limitations of the conventional therapies for Parkinson’s disease. The present study aimed to evaluate efficacy and safety of exercise therapies for Parkinson’s disease. Randomized controlled trials that evaluated exercise therapies in patients with Parkinson’s disease until December 2016 were searched for in five electronic databases: PubMed, CENTRAL, EMBASE, OASIS, and CNKI. Eighteen studies (1,144 patients) were included. The overall methodological quality was not high. Patients who underwent exercise therapies exhibited statistically significant improvements in the total UPDRS, UPDRS II and III, Berg Balance Scale, preferred walking speed, and Timed Up and Go Test compared to patients who underwent nonexercise therapies. In comparison to patients who performed regular activity, patients who underwent exercise therapies exhibited statistically significant improvements in the total UPDRS, UPDRS II, and UPDRS III. Exercise therapies were found to be relatively safe. Exercise therapies might promote improvements in the motor symptoms of Parkinson’s disease. However, due to the small number of randomized controlled trials and methodological limitations, we are unable to draw concrete conclusions. Therefore, further studies with better designs will be needed.

1. Introduction

Parkinson’s disease (PD) is a neurodegenerative neurological disease characterized by a decrease in dopaminergic neurons in the substantia nigra pars compacta (SNpc) and lowered dopamine concentrations in the basal ganglia [1]. Symptoms are divided into motor and nonmotor symptoms. Motor symptoms are characterized by bradykinesia, rigidity, resting tremor, and postural instability. There are also several nonmotor symptoms such as anosmia, sleep disorders, psychiatric symptoms, cognitive impairment, autonomic dysfunction, fatigue, and pain [2].

The motor symptoms of PD begin to appear in the early stage of the disease, leading to a decrease in the quality of life (QOL) [2, 3]. PD patients stay in hospital for about 1.45 times longer than healthy persons for about 2–14 days. Furthermore, they are more likely to be exposed to emergency situations such as falls [4]. The prevalence and incidence of PD has been increased gradually [5, 6]. According to the statistics up to 2016, 6.1 million patients suffer from PD globally [6].

In general, anti-Parkinsonian medications such as levodopa, dopamine agonists, monoamine oxidase type B inhibitors (MAOBIs), amantadine, and anticholinergics are administered as first-choice treatment. However, long-term use of dopaminergic medications could lead to adverse effects such as peak-dose dyskinesia, on-off phenomenon, and wearing off [7]. Surgical treatment such as thalamotomy, chemopallidectomy, and deep brain stimulation has been used to reduce the physiological changes of brain tissue caused by PD [8, 9]. However, it is expensive, it has high risk of side effects [10], and the possibility of reoperation cannot also be ruled out. Therefore, complementary therapies such as rehabilitation exercises could be considered in a long-term perspective. Previous studies suggested that rehabilitation exercise therapies could activate the central and peripheral nervous systems, thereby maximizing body function and slowing the progression of the disease [11].

Recently, there has been a growing interest in constructing rehabilitation strategies for PD patients in a comprehensive and diverse manner, one of which is exercise. According to animal studies, exercise therapies have neuroprotective effects and an inhibitory effect on the progression of PD or the restoration of the disease in animals. The neuroprotective effect of exercise on humans has not yet been clearly reported, but exercise therapy is most likely to be used in clinical practice [12].

The purpose of this systematic review and meta-analysis study was to investigate the effect and safety of exercise therapies on PD. To reflect the differences in exercise interventions used in each of the existing studies, we performed a meta-analysis by grouping them according to the nature of the exercise interventions.

2. Materials and Methods

2.1. Study Design

This study is a systematic review and meta-analysis to examine the effect and safety of exercise therapies on patients with PD.

2.2. Data Sources and Search Strategy

This study was carried out according to the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines [13] and the Cochrane Handbook for Systematic Reviews of Interventions [14]. The systematic literature search was conducted using Pubmed (Medline), Excerpta Medica dataBASE (EMBASE), Cochrane Central Register of Controlled Trials (CENTRAL), the Oriental Medicine Advanced Searching Integrated System (OASIS), and Chinese medical databases (CNKI- Chinese Academic Journal). The articles reported until December 2016 were searched, and there was no language limitation. Various exercise terms and MeSH terms were used for searching. The search strategies used in each database are presented in Table 1.

2.3. Study Selection

The criteria for the selection of the literature were as follows: randomized controlled trials which evaluated the effect of walking training, strength or flexibility training, balancing training, and aerobic training on patients with PD. We excluded nonrandomized or uncontrolled trials, in vivo or in vitro studies, statistical studies, or protocol papers. In the case of duplicate documents, when more than two studies were available, the most recently reported or more complete literature was selected.

2.4. Type of Participants

Studies involving patients with PD were selected. UK Parkinson’s Disease Society Brain Bank clinical diagnostic criteria was used as the PD diagnostic criteria [15]. There were no restrictions on sex, age, race, or disease duration. Patients with other diseases such as dementia, chronic medical illnesses, and atypical or secondary Parkinsonism were excluded.

2.5. Type of Interventions

Studies that used exercise therapy as an intervention for PD were included. We also included studies that used an auxiliary device for exercise such as a treadmill, but excluded studies in which the device was used as a core intervention, such as Nintendo or robot. There were no limitations on program content, such as exercise treatment methods, progress, frequency, duration, and intensity, but we only included studies which had the program of activity developed in detail. Qigong therapy in East Asian traditional medicine such as tai chi was not included neither.

In this study, the studies were classified according to each type of exercise treatment; meanwhile, they were classified as complex exercise when two types of exercises were used. The types of exercise are as follows: walking exercise either on a treadmill or on flat ground; balancing exercise, referring to the movement that shifts from one movement to another, holds a posture, and delays adjustment through physical cooperation; aerobic exercise, referring to the movement that involves stepping with a partner, tapping the ground, crossing the foot, or moving weight from one leg to the other; and dancing were included in this category; strength exercises, referring to training that prevents muscle weakness through the contraction of muscle fibers by external loads [16], and exercises that strengthen the quadriceps, hamstrings, gastrocnemius, and rectus abdominis muscles were included in this category.

2.6. Type of Comparisons

According to the type of the control group, the studies were divided into two categories. The first was for the conventional drug treatment (standard of care) and no exercise treatment. In this case, the experimental group performs the exercise therapy as an adjunctive intervention, and the control group continues the usual medication just as before the trial. Second, the control group performed a regular activity with a regular program. This program included any simple activity, physiotherapy, or cognitive activity without exercise, except for walking, balancing, aerobics, and strength training. The control group proceeded regularly in the same way during the trial process. However, studies with an active control group which performed similar intervention with the experimental group and studies with two different experimental groups (for example, walking vs. strength exercise) were excluded.

2.7. Type of Outcome Measures

The symptoms of PD were evaluated and divided into motor function, balance function, gait, quality of life (QOL), and general symptoms. In general, the Unified Parkinson’s Disease Rating Scale (UPDRS) and the Movement Disorder Society Unified Parkinson’s Disease Rating Scale (MDS-UPRDS [17, 18]) were used to evaluate the symptoms of Parkinson’s disease. To assess motor function, the UPDRS and MDS-UPRDS part III were used, and total UPDRS and UPDRS part I/II were used to evaluate general symptoms in this study. The evaluation of balance function was carried out using the Berg Balance Scale (BBS) [19] and Timed Up and Go Test (TUGT) [20]. Gait function was assessed by the gait velocity and the 6-minute walk test. The gait velocity was evaluated in two ways: the preferred walking speed (m/s) and the fast walking speed (m/s). The preferred walking speed (m/s) was the measurement of the patient’s most comfortable walking speed, while the fast walking speed (m/s) was the patient’s maximum walking speed. Reported adverse effects were also extracted.

2.8. Data Extraction

Data extraction was conducted by two researchers (Hyun-young Choi and Seungwon Kwon), and an arbiter (Ki-Ho Cho) made the final decision if there was a disagreement between the 2 researchers. The first author, characteristics of the study (i.e., year, nation (English/Chinese), setting, and design), characteristics of participants (i.e., sex, sample size, Hoehn and Yahr scale (H&Y scale), disease duration, and medication), intervention details of the experimental and control groups, measured outcome, intergroup differences, and adverse events were extracted. If any of the abovementioned data was unclear, efforts were made to contact the authors of the study.

2.9. Quality Assessment in Individual Studies

Cochrane’s risk of bias tool was used for the quality evaluation [21]. It is a tool for evaluating the bias of research included in the creation of systematic reviews and meta-analyses. It consists of 7 sections, and each was divided into “low risk of bias,” “unclear risk of bias,” and “high risk of bias.” The quality of the literature was assessed based on what is described in the literature. Risk of bias (ROB) assessment was conducted by two independent authors (Hyun-young Choi and Seungwon Kwon). In the event of a disagreement while extracting data or assessing the ROB, the third author (Ki-Ho Cho) resolved the discrepancy.

2.10. Synthesis of Data and Meta-Analysis

Meta-analysis was performed using Cochrane review manager software version 5.3 (RevMan 5.3). Based on the study design, a meta-analysis was conducted on the comparative study of the exercise with conventional drug treatment combination and conventional drug monotherapy groups. Separately, a meta-analysis was conducted on the comparative study of exercise and conventional drug treatment combination and on the regular activity and conventional drug treatment combination groups. The efficacy estimates were obtained from the relative risk (RR) for dichotomous variables and from the mean difference (MD) for continuous variables. A random effect model was used based on clinical heterogeneity between studies. The statistical significance of the effect estimates was verified based on the total effect test, 95% confidence interval (CI), and significance level of 5%. Meta-analysis was conducted by the classification of each outcome.

The Chi-square test and the Higgins I2 statistics were used to assess statistical heterogeneity. In the Chi-square test, it was verified that there was significant heterogeneity when the value was less than 0.05 or the I2 value was greater than 50.

3. Results

3.1. Description of the Included Studies

A total of 4,047 studies were retrieved by electronic search. After eliminating duplicates, the 2,795 studies left were screened by abstract. Among them, 71 studies were selected for eligibility assessment. After reviewing the full texts, 18 studies (1,144 patients) were finally selected for the meta-analysis. Fifty-three studies were excluded due to the following reasons: improper interventions such as robot therapies (n = 9), inappropriate outcome measures (n = 27), ineligible study design (n = 3), and inappropriate control group which contained more than 2 active control groups (n = 14) (Figure 1, Table 2).

Among the 18 final studies, 12 were reported in English and 6 were reported in Chinese (Table 3).

Disease duration and symptom severity (Hoehn and Yahr scale) showed large variations among the included literature. All studies [2340] used anti-Parkinsonian medications as usual therapies regardless of the intervention and performed the outcome measurements in one period.

The intervention period ranged from a minimum of one month [34] to a maximum of 14 months [27]. The frequency of the intervention performed was different for each study. The types of interventions were as follows: walking exercise (n = 3) [2325], strength exercise (n = 4) [2629], balancing exercise (n = 2) [30, 31], aerobic exercise (n = 4) [3235], and complex exercise (n = 5) [3640]. In all the studies [2340], participants maintained their usual daily activities outside of the trial, including anti-Parkinsonian medication. The exercise duration in each session varied from 10 minutes [34] to 3 hours [37]. Most studies were performed 30–60 minutes (n = 13) regardless of the type of exercise [2325, 2833, 35, 36, 38, 39].

The frequency of the exercises also varied. Most literature (n = 4) [24, 25, 30, 31] conducted exercise three times a week. Three studies carried out exercise once a week [26, 33, 38], two studies twice a week [32, 35], other three studies five times a week [28, 37, 39], two studies every day [27, 36], one study four times a week [23], and another study five or six times a week [29]. However, there was one study that did not report exercise frequency [40].

There were two types of comparisons. (1) exercise + conventional medications vs. conventional medications only (nonexercise) (n = 15) [23, 2740] and (2) exercise + conventional medications vs. regular activity + conventional medications (n = 3) [2426]. The types of regular activities varied, such as social interaction, life skill program, and conventional physical therapy. These were performed in the same manner as in the exercise treatment group.

Regarding the evaluation scale, UPDRS part III was the most common evaluation scale in 9 studies [23, 24, 2628, 31, 32, 35, 37]. For balance evaluation, the BBS was used in a total of 6 articles [28, 29, 33, 36, 38, 39], and the TUGT was used in 5 articles [26, 33, 35, 38, 39]. Gait velocity and the 6-minute walk test (6 MWT) were the evaluation scale for walking ability. The gait velocity was evaluated by two methods: the preferred speed (m/s) and the fast speed (m/s). Four articles [2426, 30] used the preferred speed (m/s), and one article [34] used the fast speed (m/s). Two articles [23, 25] used the 6 MWT as the evaluation scale.

3.2. Risk of Bias within Studies

In most studies, the risk of bias was not high. Among the risk of bias domains, blinding of the participants and personnel and selective reporting revealed methodological concerns. Nine articles [2325, 27, 29, 31, 33, 39, 40] were classified as ‘unclear risk of bias’ in the random sequence generation because there was no specific description of the randomization method. Eight studies [24, 2729, 32, 33, 39, 40] were classified as “unclear risk of bias” in the allocation concealment. Another study [35] that did not conceal the assignment order was classified as “high risk of bias.” Most studies were classified as “high risk of bias” in the blinding of participants (performance bias) [2326, 2840]. In the incomplete outcome data (attrition bias), one study [32] was evaluated as “high risk of bias” and all the remaining studies were evaluated as “low risk of bias”. In the selective reporting (reporting bias), one study [26] was rated as “high risk of bias” and the rest of the studies were evaluated as “unclear risk of bias.” A summary of the risk of bias is shown in Figure 2.

3.3. Total Unified Parkinson’s Disease Rating Scale (UPDRS) Scores

Three studies [24, 33, 40] used the total UPDRS scores, five articles [24, 26, 27, 30, 31] used UPDRS II, and 9 studies [23, 24, 2628, 31, 32, 35, 37] used UPDRS III.

Two studies [33, 40] compared the total UPDRS score in the exercise therapy group (ET) to that of the nonexercise group (NE). ET showed a significant effect on the total UPDRS score (MD −16.84, 95% CI (−22.52, −11.16)). In the subgroup analysis based on the type of exercise, there were significant results in the ET (aerobic exercise [33]: MD −14.20, 95% CI (−22.66, −5.74); complex exercise [40]: MD −19.00 and 95% CI (−26.66, −11.44)) (Figure 3(a)).

One study [24] compared the exercise therapy group (ET, walking exercise) with the regular activity group (RA). ET showed a significant effect in the total UPDRS score (MD −2.90, 95% CI (−5.44, −0.36)) [24] (Figure 3(b)).

3.4. UPDRS I Scores

Two studies [24, 31] used UPDRS I. There was a study [31] evaluating the effect of balancing exercise and it compared the UPDRS I subscore in ET and NE. ET showed a positive effect in the total UPDRS I (MD −0.10, 95% CI (−0.26, 0.06)) (Figure 3(c)). There was another study [24] that used walking exercise and compared the UPDRS I subscore between ET and RA. In this study, ET did not show a positive effect in the UPDRS I (MD 0.50, 95% CI (0.36, 0.64)) (Figure 3(d)).

3.5. UPDRS II Scores

Three studies (including 166 patients) [27, 30, 31] evaluated the UPDRS II score in ET and NE. ET showed a significant effect on the UPDRS II score (MD −4.12 and 95% CI (−8.23, −0.02)). In the subgroup analysis based on the type of exercise, there was also a significant effect in strength exercise (MD −8.15, 95% CI (−14.96, −1.34)) [27] (Figure 3(e)).

Additional two studies [24, 26] compared ET with RA. ET showed a significant effect on the UPDRS II score (MD −1.82, 95% CI (−3.56, −0.07)) (Figure 3(f)).

3.6. UPDRS III Scores

Five studies (including 179 patients) [23, 27, 28, 31, 37] compared UPDRS III scores in ET and NE. ET showed a significant effect on the UPDRS III score (MD −6.09 and 95% CI (−7.79, −4.38)). In the subgroup analysis, there were inconsistencies depending on the exercise type. Walking [23] and strength [27, 28] exercise did not show a positive effect, and balancing [31] and complex exercises [37] showed significant effects (Figure 3(g)).

Eighty-five Parkinson’s disease patients in two studies [32, 35] divided into ET and NE were evaluated by MDS-UPDRS III. ET showed a significant effect in the UPDRS III score (MD −11.69 and 95% CI (−16.96, −6.42)) (Figure 3(h)).

On the other hand, two studies (144 patients) [24, 26] compared ET with RA. ET showed a significant effect in the UPDRS III score (MD −2.53, 95% CI (−3.75, −1.31)) (Figure 3(i)).

3.7. Berg Balance Scale (BBS)

Six studies (463 patients) [28, 29, 33, 36, 38, 39] compared ET with NE. ET showed a significant effect on the BBS scores (MD 2.72 and 95% CI (1.63, 3.80)). In the subgroup analysis, strength [28, 29] and aerobic exercise [33]showed significant effects on BBS scores (Figure 4).

3.8. Preferred Walking Speed

There were 4 studies [2426, 30] that used preferred walking speed to evaluate the gait function. Among them, two studies (107 patients) [23, 30] compared ET with NE. There was a significant difference in the preferred walking speed between the two groups (MD 0.11, 95% CI (0.10, 0.12)). In the subgroup analysis, only balancing exercise [30] revealed a significant effect (Figure 5(a)).

Two other studies (141 patients) [25, 26] compared ET with RA. No significant differences in the preferred walking speed were shown between the two groups (MD -0.54, 95% Cl (−2.15, 1.07)) (Figure 5(b)).

3.9. Fast Walking Speed

One study (including 33 patients) [34] compared ET with NE (placebo exercise). No significant differences in fast walking speed were shown between the groups (MD 0.18, 95% CI (−0.09, 0.45)) (Figure 6).

3.10. Timed Up and Go Test (TUGT)

Five studies used the Time Up and Go Test (TUGT) for evaluating gait function. Among them, four studies (235 patients) [33, 35, 38, 39] compared between ET and NE. ET showed a significant difference between groups (MD −1.44, 95% CI (−2.41, −0.47)) (Figure 7(a)).

Another study (including 124 patients) [26] compared ET with RA. No significant difference in the TUGT was shown between the groups (MD −3.50, 95% CI (−8.92, 1.92)) (Figure 7(b)).

3.11. Six-Minute Walk Test (6 MWT)

Two studies used the 6-minute walk test (6 MWT) to evaluate gait function. One study (including 17 patients) [23] compared ET with NE. ET did not show a significant effect in the 6 MWT (Figure 8(a)). Another study (including 17 patients) [25] compared ET with RA. In this analysis, ET did not show a significant effect in the 6 MWT (Figure 8(b)).

3.12. Safety

Among the 18 articles, only seven articles [23, 25, 26, 30, 34, 35, 38] investigated adverse effects due to the interventions. Of these, there were no adverse events reported in four studies [23, 25, 26, 38]. Falls (n = 14) and fatigue (n = 2) were reported as adverse effects in three studies [30, 34, 35], and respiratory infection (n = 1) which was not related to the intervention was reported in one study [35].

4. Discussion

The results of this systematic review and meta-analysis show that ET improved motor and nonmotor symptoms in PD compared with NE or RA. ET showed a significant improvement in UPDRS (total, II, III, and MDS-UPDRS III) scores, BBS, preferred walking speed, and TUGT compared to NE and UPDRS (total, II, and III) compared to RA.

Previously, several meta-analyses had been reported which evaluated exercise interventions in patients with PD [10, 4145]. The differences between the previous studies and the present study are as follows. First, in this study, various interventions and outcomes were investigated. Most of the previous studies were limited in the specific type of exercise therapy [10, 43, 44] or the specific type of outcome measure [45]. Another study only showed the characteristics, intervention delivery, retention rates, adherence, and adverse events of exercise therapies [42]. However, there was no report about improvement of PD symptoms. In this study, we aimed to comprehensively evaluate the effects of various types of exercise therapies on PD and to evaluate the effects of each type of exercise therapy through subgroup analysis. Therefore, we tried to classify exercise interventions into 5 groups according to four types of exercise therapy and to provide a summary effect estimate of the individual exercise types. At the same time, we extracted various outcome measures such as total UPDRS and UPDRS part I, II, III, and IV, BBS, TUGT, and gait velocity (preferred speed, fast speed, 6MWT). Therefore, the relationship between the improvement of motor or nonmotor symptoms of PD and exercise therapies could be evaluated. The degree of improvements of symptoms according to the type of exercise could be known. Based on these results, it is possible to apply it to clinical care. Second, this study analyzed a larger number of literature, being an update of existing studies. There were 495 PD patients included in 14 articles in the previous studies of various exercise interventions as classified in this study [41]. In this study, we included 1,144 PD patients in 18 studies. Therefore, it could be assessed that the reliability of our study was increased. Finally, we tried to reduce the heterogeneity and to obtain accurate results by dividing the results into two groups according to whether regular activity was performed in the control group or not. In addition to the nonexercise group (NE), we evaluated the regular activity (RA) group to confirm that the results are different. We found that even simple activities could also help to improve symptoms of PD.

In this study, exercise therapies have been shown to be effective in improving the overall symptoms of PD, the activities of daily life (ADLs) related to motor function, overall motor symptoms, balance, and gait disturbance. The effects of each exercise type are as follows: walking exercises showed significant effects on ADLs related to motor function and motor symptoms compared with RA; strength and flexibility exercises revealed significant effects on ADLs related to motor function (compared with NE and RA) and balance (compared with NE); balancing exercise has significant effects on motor symptoms and gait disturbance (compared with NE); aerobic exercise showed significant effects on motor symptoms and balance (compared with NE); and complex exercise revealed a significant effect on motor symptoms. As mentioned above, depending on the type of exercise, we could see the difference in the degree to which PD symptoms were improved. There is a high heterogeneity of the resulting values because of differences in the duration and method of exercise therapy for each study included in the meta-analysis. Nonetheless, in clinical applications, clinicians will be able to make appropriate and flexible use of the results of this study, depending on the circumstances and experience (Table 4). Adverse events such as falling and fatigue have been reported. Among them, the most common was falling in two articles [30, 35]. Falling was observed only in balancing and aerobic exercise, 13 out of 14 occurred during balancing exercise. According to a review article [46], postural instability is known to be observed in 16% of PD patients. Postural instability gradually deteriorates as the disease progresses, which is the main cause of falling [47]. If there are patients with severe postural instability, the balancing exercise should be considered carefully. Other types of exercise (besides the balancing exercise) did not show any severe adverse effects other than fatigue. Therefore, they could be applied to PD patients relatively more safely.

The limitations of this study are as follows: first, there is a bias in the literature included in the aspects of qualitative research methodology. Selection bias may exist because random sequence generation or allocation concealment were not specifically addressed [2325, 2729, 3133, 39, 40] or one study [35] was evaluated as “high risk of bias” because of the absence of blinding of participants (performance bias). Only one study [27] conducted blinding of participants. Also, in the blinding of outcome assessment, two studies [35, 38] were evaluated as “high risk of bias.” Therefore, selection bias and detection bias may have influenced the result of this study. Second, heterogeneity is high. This is thought to be because there was a huge difference in the quality of the studies, the patients participating in the study, and the exercise treatment and regular activity in each literature. Third, the sample sizes of the literatures included in this study are still small. This study was divided into two groups according to whether regular activity was performed in the control group and subgrouped by the types of intervention (walking, strength, balancing, aerobic, and complex exercises). Therefore, since a total of 18 articles were divided, the number of articles included in each evaluation index was quite small. This might have led to the lower test effectiveness. Therefore, further follow-up research with additional exercise therapy intervention clinical papers should be conducted.

5. Conclusions

Exercise therapies might promote improvements in the motor symptoms and related ADLs of PD. Overall, exercise therapies showed significant effects on motor function of PD patients compared to NE group or RA group. In contrast, exercises did not show a statistically significant effect on nonmotor symptoms compared to the NE group or RA group. These results suggest that exercise therapy is more effective for motor symptoms of PD patients rather than nonmotor symptoms. However, due to the small number of randomized controlled trials and methodological limitations, we are unable to draw concrete conclusions. Therefore, further studies with better designs will be needed.

Data Availability

Data can be obtained from the corresponding author on request.


This paper is based on Hyun-young Choi’s theses for the Master’s degree.

Conflicts of Interest

The authors declare that there are no conflicts of interest regarding the publication of this paper.

Authors’ Contributions

Hyun-young Choi and Ki-Ho Cho contributed equally to this manuscript as the first authors.


This study was supported by a grant from the Traditional Korean Medicine R&D Program and funded by the Ministry of Health & Welfare through the Korea Health Industry Development Institute (KHIDI, HB16C0051).