Randomized controlled trials (RCTs) evaluating the effect of SA on motor function in stroke with the control group receiving modern standard treatment or conventional treatment were included in this study.

Randomized controlled trials (RCTs) that included patients of any age or sex with acute or chronic stroke were eligible. SA was administered to patients with stroke as diagnosed through CT or MRI. The control group in these RCTs received only standard Western treatment for stroke in a neurological ward.

Randomized controlled trials (RCTs) were included regardless of treatment duration and number of sessions, but acupoint selection was limited to the scalp; ears and body acupoints were excluded. The control group patients received Western standard treatment (medications and rehabilitation). Patients in the trial groups received SA therapy (SA of standard international acupuncture nomenclature) in addition to the same standard treatment as that in the control groups. Concerning SA, acupoints are mostly in the motor areas (the anterior oblique line of the vertex-temporal, MS 6) and sensory areas (the posterior oblique line of the vertex-temporal, MS 7) of the contralateral scalp of hemiplegic limbs. The needle retention time was mostly between 0.5 and 2 h, and the depth of needle insertion was 2–4 cm. The frequency was 5–7 days per week and once a day. The treatment courses ranged from 2 weeks to 6 months. Those who performed SA in these studies were experienced doctors.

We investigated three types of outcome: motor function, sequelae of poststroke hemiparesis, and adverse effects. We evaluated motor function in the upper and lower extremities at the end of the first and third months. The assessment was performed using the Fugl–Meyer Assessment (FMA), a stroke-specific, performance-based impairment index (12, 13). The FMA motor score ranges from 0 (hemiplegia) to 100 (normal motor performance) points, which were divided into 66 points for the upper extremity and 34 points for the lower extremity.

We searched databases, namely, PubMed, Embase, the Cochrane library, Airiti Library, and China National Knowledge Infrastructure (until May 2021) by two reviewers (YJH and CSH). The keywords used for the search were “scalp acupuncture” and “stroke.” The type of article searched for was the RCT. We focused on poststroke hemiparesis or hemiplegia and excluded unwanted outcomes (poststroke dysarthria, poststroke dysphagia, poststroke aphasia, central poststroke pain, poststroke depression, poststroke cognitive impairment, etc.). The reference lists of all relevant articles were searched for further studies. There was no language limitation.

Two reviewers (YJH and CSH) independently assessed all studies and independently extracted eligible data from the trial reports by using a data extraction form; data were cross-checked for accuracy before use. Disagreement was resolved through discussion with a third reviewer (SWC) if necessary. The authors of the trials were contacted and requested to provide missing data.

Quality assessment was conducted by revised Cochrane risk of bias tool 2.0 (RoB2.0) to determine whether the trials had the following concerns regarding internal validity: 1) risk of bias arising from the randomization process; 2) risk of bias due to deviation from intended intervention; 3) missing outcome data; 4) risk of bias in outcome measurements; or 5) risk of bias in the selection of the reported results. We conducted the risk of bias of summary and graph by using Review Manager 5.4.

Heterogeneity between trial results was tested using the standard I² test. Results are reported as odds ratios with corresponding 95% CIs for dichotomous data. If continuous data were available, the weighted mean difference or standardized mean difference was calculated. Statistical analysis was performed using the statistical software R version 3.6.0 and the meta package.

Based on the search strategy, we manually retrieved 1,062 papers found in the five databases and one from another source (manual search using Google Scholar). These studies were published from 1975 to 2021, and 253 papers had duplicate titles. Of the remaining 810 papers, after screening the abstracts, 18 animal studies, six expert opinions, seven case studies, 57 cohort studies, nine protocols, six systematic reviews or meta-analyses, 56 narrative reviews, 326 relevant articles with an unwanted outcome (poststroke dysarthria, poststroke dysphagia, poststroke aphasia, central poststroke pain, poststroke depression, poststroke cognitive impairment, etc.), and 17 irrelevant articles were excluded; the remaining papers with full text available were selected. Among the 306 RCTs selected, 25 papers with incomplete or insufficient data, one paper regarding noninternational-standard SA (Yamamoto new SA), 212 papers regarding impure SA with rehabilitation, 21 papers comparing different SA types, and 17 papers not including FMA scores were excluded. A total of 30 RCT papers containing FMA scores were finally selected (14–43). The process used for the literature search for the application in the systematic review is summarized in Figure 1.

All these 30 RCTs were conducted in China, and the corresponding articles were published from 2007 to 2019 (Table 1). Of these RCTs, 15 mentioned 1-month results; five mentioned 6-week results; six mentioned 2-month results; and six mentioned 3-month results. We assessed both upper and lower extremity strengths, so we selected only papers with complete FMA scores (including the upper and lower extremities with a total score of 100). We selected treatment courses of 1 and 3 months and then conducted the meta-analysis. A total of 1,043 patients with acute or chronic stroke from 12 reports were included.

Based on Figure 2, we determined the overall quality of the included studies to be low by RoB2.0. For the first domain (risk of bias arising from the randomization process), the generally unclear risk was indicated. All trials used a randomization method, including a random number generator (calculator) and a sequencing method. One trial (28) described a grade-A level of adequate concealment of randomization in which the patients were allocated according to calculated random numbers sealed in opaque envelopes. For the second domain (risk of bias due to deviation from the intended intervention), the risk was generally high because only one trial (21) included a control group for which sham acupuncture was implemented. The remaining 29 trials did not mention blinding or intention-to-treat analysis. For the third domain (missing outcome data), the risk was generally high because only two trials (21, 28) recorded missing data. For the fourth domain (risk of bias in outcome measurement), the risk was generally unclear because only three trials (14, 28, 37) mentioned that the assessor did not participate in the therapy. For the fifth domain (risk of bias in the selection of the reported results), the risk was generally low. All trials conducted a comparative analysis between the SA group and control group.

None of the trials used death or dependency as a primary outcome measure.

Among the papers on 30 independent trials, 10 reporting the complete FMA score described the 1-month outcome (Figure 3; mean difference, 10.3 [95% CI, 7.43–12.63]; P < 0.01), and four reporting the complete FMA score described the 3-month outcome (Figure 4; mean difference, 15.18 [95% CI, 8.06–22.31]; P < 0.01). Two papers (17, 29) with a 1-month course mentioned that special rehabilitation may diminish the effect of SA; therefore, we removed these papers and reconducted a sensitivity test of the meta-analysis (Figure 5; mean difference, 11.16 [95% CI, 8.09–14.23]; P < 0.01). There were no homogeneity in the consistency of the trial results (1 month: I² = 78% and 3 months: I² = 83%). SA was found to have significantly improved hemiparesis, as represented using the FMA score and when compared with the control group. The outcome after 3 months of SA (mean difference in FMA score: 15.18) seemed to be better than that after 1 month of SA (mean difference in FMA score: 11.16).

Only two studies (15, 27) recorded the sequelae of hemiparesis, such as shoulder–hand syndrome, shoulder pain, and muscle atrophy. Only one study (15) recorded complete muscle spasticity and bedsores. After pooling the results of the two papers, we found a significant difference in the incidence of shoulder–hand syndrome (Figure 6; odds ratio, 0.39 [95% CI, 0.16–0.94]) between the experimental and control groups, but no difference in that of shoulder pain (Figure 7; odds ratio, 0.43 [95% CI, 0.17–1.11]) or muscle atrophy (Figure 8; odds ratio, 0.39 [95% CI, 0.13–1.15]).

Adverse events were mentioned in one trial (38). In the SA groups, the abnormal response rate for dizziness and local skin redness was 6.67% (3/45), whereas it was 8.89% (4/45) in the control group. The difference was not significant.

We used a contour-enhanced funnel plot to differentiate between asymmetry due to publication bias and other reasons. Figure 9 demonstrates that almost all trials obtained a significant result favoring the experimental group, with P < 0.01 (light gray region) in nine trials and 0.01 < P < 0.05 (dark gray region) in one trial. However, publication bias was noted in five trials from the funnel plot.