To identify eligible studies, a primary search was conducted using electronic databases (PubMed, Cochrane Library, OvidSP, Web of Science, China Science and Technology Journal Database, Chinese National Knowledge Infrastructure and China Wanfang Standards Database) from the inception dates to October 1st, 2019, using the keywords ‘glutathione’ or ‘GSH’ and ‘Parkinson’ or ‘Parkinson's disease’ or ‘PD’. Specific retrieval strategies were adjusted according to different databases. The procedure was concluded by: i) The perusal of the reference sections of all relevant studies; ii) a manual search for GSH in key journals and abstracts from the major annual meetings in the field of PD; and iii) contact with experts to request unpublished data. The primary search was completed by independent investigators (HLW and JZ) and any discrepancies were resolved by consultation with an investigator (YZC) not involved in the initial procedure.

The inclusion criteria for the present study were as follows: i) Participants were clinically diagnosed with PD; ii) GSH was administered as an intervention treatment; iii) patients treated with GSH were directly compared with a non-GSH or placebo group; iv) outcomes were determined using the Unified Parkinson's Disease Rating Scale (UPDRS) and/or GSH peroxidase (GSH-Px) and/or related adverse events (AEs); and v) the study was a published randomized controlled trial (RCT).

Articles fulfilling the following criteria were excluded from the present study: i) Randomized trials without a placebo or control group; ii) studies lacking original data; and iii) abstracts, conference papers, letters or comments.

The risk of bias in the included studies was assessed by two independent reviewers (WHL and JZ) using the Cochrane Handbook for Systematic Reviews of Interventions (13). Bias was evaluated in the following seven domains: i) Random sequence generation; ii) allocation concealment; iii) blinding of participants and personnel; iv) blinding of outcome assessment; v) incomplete outcome data; vi) selective outcome reporting; and vii) other bias, of which random sequence generation, blinding of participants and personnel, and blinding of outcomes assessment were of most interest. Any disagreements were resolved by discussion among all of the reviewers. The risk of bias in each domain was rated as low, unclear or high, according to methods used to ensure the minimization of each form of bias. Using the following methods, individual studies were categorized as having low, high or unclear risk of bias: i) Low risk of bias (plausible bias unlikely to markedly alter the findings) if the risk of bias was low in all domains; ii) unclear risk of bias (plausible bias that raises certain doubt about the results) if the risk of bias was unclear in one or more domains; or iii) high risk of bias (plausible bias that seriously weakens confidence in the results) if a high risk of bias was present in one or more domains. Any disagreements were resolved through a discussion within the entire review team.

Data were extracted by two independent reviewers (WHL and YPL) using a predefined data extraction method. Disagreements were resolved by discussion or consensus with a third independent author (CYZ). The extracted data included the first author, study characteristics (i.e. year, duration and design), participant characteristics (i.e. age, sample size and systemic therapy) and outcomes (UPDRS/GSH-Px/related AEs). For studies with insufficient information, the reviewers contacted the corresponding authors where possible to acquire the data.

When conditions permitted, the study was divided into three arms based on the administered dose of GSH used to obtain the two-arm data (300 mg/d groups vs. control groups, and 600 mg/d groups vs. control groups). Dichotomous data were analyzed using the risk ratios (RRs) with 95% confidence intervals (CIs). When the result unit, measurement method or measurement time was inconsistent, continuous outcome measurements were analyzed using standard mean differences (SMDs) with 95% CIs; 95% CIs were calculated using the inverse variance (IV) statistical method. I-square (I²) statistics and the Q test were performed to assess the impact of study heterogeneity on the results of the meta-analysis. According to the Cochrane review guidelines (13), if severe heterogeneity was present at P<0.1 or I²>50%, the random-effects model was chosen; otherwise, the fixed-effects model was used. Subgroup analyses were performed according to GSH dose.

According to the aforementioned retrieval strategy, a total of 1,334 related articles were initially retrieved and 918 studies were retained after 416 papers with duplicate data were excluded. Of the identified articles, 901 that did not meet the inclusion criteria were excluded after reading the title and abstract. Of the remaining 17 studies (which were evaluated for applicability by reading the full text), a further 10 were omitted per the exclusion criteria, leaving a total of 7 included studies (10-12,14-17). A flow diagram of the screening process is depicted in Fig. 1. The age distribution of the patients within these studies was 43-84 years and the included studies were published between 2003 and 2019. The studies primarily reported on the outcomes of UPDRS, GSH-Px and related AEs. The specific basic characteristics of the included studies are listed in Table I.

The quality of the included RCTs was assessed according to the Cochrane Handbook (Figs. 2 and 3) (13). In the category random sequence generation, the seven studies had a low risk. There were two articles with sufficient allocation concealment, while the allocation and concealment schemes of the other five articles were not clear. Furthermore, performance bias of three studies were low-risk and four articles were unclear. There were two papers with low detection bias, while another five articles were rated as unclear with regard to this bias. In terms of incomplete data, seven articles were all rated as having low risk, and the risk of selective reporting was low in five articles and was high in two studies. There were seven studies with a low risk of other bias. In conclusion, the overall quality of the seven included studies was moderate.

There were five studies reporting data on UPDRS I, II and III (10-12,14,16). Due to differences in data type (end value and end value minus baseline value), the SMD was applied to determine differences in the UPDRS I, II and III scores between the GSH and control groups. The heterogeneity test did not reveal any differences between studies reporting data regarding the UPDRS I (χ²=3.51, I²=0%, P=0.74); thus, the fixed-effects model was used (Fig. 4A). In addition, heterogeneity between the studies that reported data on UPDRS Ⅱ was low (χ²=7.51, I²=20%, P=0.28) and thus, the fixed-effects model was applied once again (Fig. 4B). However, the heterogeneity test indicated moderate differences between studies reporting data on UPDRS Ⅲ (χ²=14.95, I²=60%, P=0.02), and therefore, the random-effects model was used (Fig. 4C). The pooled SMD was -0.04 (95% CI=-0.25-0.16, P=0.70) for UPDRS I, 0.03 (95% CI=-0.17-0.24, P=0.77) for UPDRS Ⅱ and -0.48 [95% CI=-(0.88-0.08), P=0.02] for UPDRS Ⅲ. These pooled results demonstrated that, compared with the control groups, GSH may slightly improve the motor scores of patients with PD.

In total, four studies (14-17) presented GSH-Px data for the GSH and control groups. The SMD was used to estimate differences in GSH-Px between the two groups. The results of the pooled SMD are presented in Fig. 5. There was significant heterogeneity among the studies (χ²=72.36, I²=96%, P<0.00001) and thus, the random-effects model was used. The pooled SMD was 1.88 (95% CI=0.52-3.24, P=0.007), indicating that compared with the control groups, serum GSH-Px levels were significantly higher in the GSH groups.

There were three studies reporting on the incidence of gastrointestinal reactions (10,12,16); two papers containing data on dizziness or headache (10,16); two articles reporting on involuntary movement (10,16); two papers on labored breathing (10,12); two articles presenting strep throat-associated data (10,12); and two studies reported on insomnia (10,16) in the GSH and control groups (Table II). The heterogeneity test revealed no differences between the studies (separately, I²=0, 0, 0, 41, 41 and 0%, respectively); thus, the fixed-effects model was applied. Separately, the pooled RRs were 0.78 (95% CI=0.28-2.14, P=0.62), 0.99 (95% CI=0.28-3.49, P=0.99), 0.33 (95% CI=0.44-2.99, P=0.32), 1.59 (95% CI=0.29-8.59, P=0.19), 1.59 (95% CI=0.29-8.59, P=0.59) and 1.64 (95% CI=0.23-11.74, P=0.62). These pooled results of AEs demonstrate that the use of GSH appears to be safe.

In total, two studies (11,12) reported data of UPDRS III with the use of GSH (300 mg/d) for PD and two papers (11,12) included data on the use of GSH (600 mg/d) (Fig. 6). The heterogeneity test indicated minimal differences between these studies (individually, χ²=1.06, I² =6%, P=0.30; and χ²=2.87, I²=30%, P=0.24, respectively), and therefore, the fixed-effects model was used. The pooled SMDs were -0.67 [95% CI=-(1.30-0.04), P=0.04] and -0.16 (95% CI=-0.41-0.09, P=0.21), respectively, suggesting that the dose (300 vs. 600 mg) was an influencing factor for UPDRS III. Therefore, it was conservatively hypothesized that in patients with PD, a 300-mg dose of GSH may be more effective than a 600-mg/d dose.