Summary of included studies
On initial online search and relevant references, 248 articles were included. Filter by title, abstract, repetition, and content, most of the unqualified articles have been filtered out. Careful assessment of 21 full-text articles led to the exclusion of 2 studies20, 21 because of lacking a control group, another 3 studies22-24 were excluded due to incomplete data. Therefore, a total of 16 randomized controlled trials (RCTs) with 691 acute or chronic individuals fulfilled the inclusion criteria and were applied to this meta-analysis. The detailed screening process was showed in the form of PRISMA diagram25 in Fig 1.
The summary of the basic characteristics of the included studies is presented in Table 1. The 5 studies13-15, 17, 26 were performed in China, and 2 were conducted in Japan27, 28 and 3 were conducted in America18, 29, 30. The rest of the studies were carried out in Switzerland31, Turkey32, Spain33,Canada34, England35 and Netherlands36, respectively. The shortest duration of intervention among these trials was of 2 weeks. In addition, the types of electro-stimulation used were FES, EMG-triggered NMES and NMES with the forms in which, NMES electrode sheets were stuck to the skin surface or a robot arm was driven by NMES. And the allocation of the electrode sheets were generally attached on the muscles of extensor/ flexor carpi radialis, extensor digitorum communis, anterior/posterior deltoid and triceps muscles.
Data quality:
The assessment regarding the risk of bias among each trial is shown in Supplementary file S1. Amongst all, these selected trials were different in study quality: besides 1217, 18, 26, 28-36 blind trials were at low risk of bias, one study27was assessed as high risk of bias out of not blind. 235, 36trials adopted concealed allocation, which were regarded as low risk of bias. Also, all of the involved trials were at an unclear risk of bias in selective reporting. One study34 was at high risk of bias due to incomplete outcome data. Moreover, 6 studies17, 18, 28, 31, 33, 35 were at low risk of bias due to random sequence generation by computer or pseudorandom number sequence. One study36was at high risk of bias for free of other sources of bias.
Meta-analysis:
Pooled data from 1313-15, 17, 18, 26, 27, 29-32, 34, 36 trials (n = 645 for all experiments and controls) showed obvious efficacy on performing the FMA-ue assessment (SMD=0.44, 95% CI 0.13 to 0.74, P=0.006). Additionally, NMES treatment did not display a positive effect on the wrist MAS (SMD=-0.12, 95%CI -0.43 to 0.19, P=0.434). The meta-analysis of 517, 26, 33, 35, 36 studies implied a distinct benefit in the intervention group (n = 114) in improving ARAT, which had inconspicuous heterogeneity compared with the control group (n = 113) (SMD = 0.31, 95% CI 0.05 to 0.58, P = 0.019, I 2 = 0%). Moreover, NMES therapy was effective when measured by MAL-AOU score (SMD=0.59,95%CI 0.15 to 1.04, P=0.009), which contrasts with BI score (SMD=0.51,95%CI -0.65 to 1.66, P=0.392) that exhibited no benefit in upper limb rehabilitation (Fig 2A-E).
Subgroup analyses:
Subgroups were divided into 2 aspects: (i) duration of intervention and (ii) area. Outcomes of the subgroup analyses are concluded in Supplementary file S2.When grouped by duration, a significant improvement within the duration of ≤4 weeks were observed in the FMA-ue (SMD=0.43,95%CI 0.05 to 0.81,P=0.027), ARAT(SMD=0.42, 95%CI 0.04 to 0.79,P=0.03) and MAL-AOU (SMD=0.55,95%CI 0.07 to 1.02,P=0.025),but not in the duration of longer than 4 weeks. The subgroup analyses conducted to differentiate the area to which the participants belonged to showed a significant improvement in the FMA-ue (SMD=0.57,95%CI 0.14 to 1.01, P=0.009) and ARAT (SMD=0.42, 95%CI 0.04 to 0.79, P=0.03) scores among Asians. And the MAL-AOU score was positive in people from North America (SMD=1.43, 95%CI 0.41 to 2.45,P=0.006).
Publication bias of literature
To evaluate the publication bias, Funnel plots, Egger’s test and Begg’s test were conducted successively, except the condition that selected trials were less than 10 in each meta-analysis. In this meta-analysis, only the FMA-ue scale met the above standard ,which involved total 16 trials. So the Begg ’s and Egger’s approaches can be assessed (P=0.224)( in Fig 3), and was deemed to show no publication bias.