Systematic review and meta-analysis were performed following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guideline²¹⁾. The protocol of this review was registered in INPLASY (https://inplasy.com/inplasy-2024-7-0076/).

We searched the following electronic databases from their inception to June 4, 2024: three core databases (PubMed, Embase, and Cochrane Library), China National Knowledge Infrastructure (CNKI), and four Korean databases (Science on, Korean Studies Information Service System (KISS), KoreaMed, and Oriental Medicine Advanced Searching Integrated System (OASIS)). We also searched clinical trial registration platforms, including the International Clinical Trials Registry Platform (ICTRP), ClinicalTrials.gov (CTG), Chinese Clinical Trial Register (ChiCTR), and Clinical Research Information Service (CRIS). The reference lists of the selected articles and related SRs were also manually searched. Only articles published in academic journals were included, while theses, dissertations, and conference abstracts were excluded. For search terms, key terms corresponding to stroke and bloodletting in the English, Chinese, and Korean languages were used in combination according to the setting of each database (Supplement 1).

The retrieved publications were imported into the reference manager software (EndNote) to remove duplicate records. Two authors screened the titles and abstracts of the articles to remove studies unrelated to the topic of this SR, after which they obtained and read the full texts of the screened articles to select the studies for analysis. Any disagreements between authors were resolved through discussion.

Data extraction was performed independently by two authors. Data from the selected articles, including the first author, publication year and country, sample size, sex, age, stroke type, time since onset, participant stratification, treatment and control intervention, acupoints and other details of bloodletting, outcome measures, and measurement time points, were extracted and entered into a Microsoft Excel Spreadsheet. When the data provided in the article were insufficient, the authors of the original article were contacted via email.

The risk of bias (RoB) of the included RCTs was assessed using version 2 of the Cochrane RoB tool for randomized trials (RoB 2)²³⁾. The RoB2 consists of five domains: randomization process, deviations from the intended interventions, missing outcome data, measurement of the outcome, and selection of the reported results. Each domain was rated as ‘low RoB,’ ‘high RoB,’ or ‘some concerns.’ The overall RoB of each study was determined based on the assessment results of the five composite domains. The RoB assessment was conducted independently by two authors using Excel macro file for RoB2 implementation downloaded from the websites (https://www.riskofbias.info/welcome/rob-2-0-tool). Disagreements were resolved through discussion.

When two or more RCTs provided similar outcome measures, data were synthesized. Review Manager 5.4 software was used for meta-analysis. For continuous variables, the mean difference (MD) with a 95% confidence interval (CI) was calculated using the inverse-variance method. For dichotomous variables, the risk ratio (RR) with a 95% CI was calculated using the Mantel–Haenszel method.

For meta-analysis, a random-effects model was first considered, but when the number of studies included in the meta-analysis was three or fewer, which was too small to accurately estimate the variance between studies, a fixed-effects model was used²⁴⁾.

Statistical heterogeneity was measured using Cochrane’s I² value²⁵⁾. When significant heterogeneity between studies was detected, a leave-one-out approach was used for sensitivity analysis. If more than ten studies were synthesized, a funnel plot was drawn to explore the possibility of publication bias.

The certainty of the evidence derived from the meta-analysis was assessed based on the GRADE framework²⁵⁾. The GRADE consists of five domains: RoB, inconsistency, indirectness, imprecision, and publication bias; using this scale, the certainty of evidence is rated as high, moderate, low, or very low based on the assessment results of the five domains. The assessment was performed independently by two authors using the GRADEpro website (https://www.gradepro.org/) and any disagreements were resolved through discussion.

The initial search of electronic databases retrieved 459 records (PubMed, 42 Embase, 48; Cochrane Library, 71; CNKI, 263; Science On, 13; KISS, 6; KoreaMed, 2; and OASIS, 14), and clinical trial registry retrieved 18 records (ICTRP, 2; CTG, 14; ChiCTR, 2). Of the 477 retrieved records, 52 duplicates were removed, and the titles and abstracts of the remaining 425 records were screened. After removing 203 records irrelevant to the topic of this SR, the full texts of 222 records were reviewed for eligibility. Finally, three RCTs^6,19,20) were identified (Fig 1).

The characteristics of the included studies are tabulated in Table 1. All three of the included studies^6,19,20) were conducted in China. The sample sizes of the studies ranged from 52 to 360 participants, and the average age of the participants was 60–70 years. All studies included patients with either ischemic or hemorrhagic stroke in the acute phase within 48–72 hours of stroke onset. Two studies^19,20) stratified participants into three subgroups according to brain lesion volume (small, medium, and large) and reported outcomes by subgroup, whereas the other⁶⁾ divided the participants into two subgroups based on GCS scores (moderate and severe). All three studies^{6,19, 20)} compared the outcomes of participants receiving CT with or without bloodletting, pricking 12 Jing-well acupuncture points on both hands, and withdrawing one drop of blood per acupoint in the treatment group. Yu et al.⁶⁾ previously reported that bloodletting was performed immediately after admission and before CT. However, the other two studies^19,20) did not clearly report the timing of bloodletting. Guo et al.¹⁹⁾ reported differences in GCS scores and vital signs before and after treatment in three subgroups. Ding et al.²⁰⁾ reported differences in GCS scores before and after treatment in the entire study population and in subgroups similar to those established by Guo’s trial¹⁹⁾. Yu et al.⁶⁾ graphically presented the GCS score and vital sign values measured after treatment in the entire population and subgroups. All three studies^6,19,20) presented measurement values at 15 and 30 minutes after the intervention. The last measurement time point in each study were 45¹⁹⁾ and 80 minutes^6,20).

No studies used other endpoints than GCS to assess the participant’s level of consciousness. No studies reported adverse events observed during the study.

The results of the RoB assessment of three studies^6,19,20) are presented in Fig 2. For the first domain, the study by Yu⁶⁾ was rated as having a low RoB because it clearly reported the procedures of randomization and allocation concealment, while the other two studies^19,20) were rated as having ‘some concerns’ because of the omitted related description. The RoB of the second and third domains of all studies was evaluated as low. Although none of them performed patient and practitioner blinding owing to the nature of the bloodletting intervention, there were no cases of changes from the initially intended intervention, and the numbers of participants who were randomized and in whom the primary outcome measure was reported were the same. For the fourth domain, the RoB in Yu’s study⁶⁾ was assessed as low because assessor blinding was implemented. However, the remaining two^19,20) omitted reporting on whether assessor blinding was implemented, and because the possibility that this may have affected the outcome measure cannot be ruled out, the RoB of this domain was rated as high. For the last domain related to the selection of the reported results, Yu’s study⁶⁾ as having a low RoB because we confirmed the consistency between the planned prepublished protocol, and the results reported in the article. Conversely, the remaining two studies^19,20) were evaluated as having ‘some concerns’ because there was no evidence to confirm the preplanning, such as protocols or clinical trial registration data. As a result, the overall RoB of Yu’s study⁶⁾ was low, whereas those of the two^19,20) were high.

The GRADE framework was used to assess the certainty of the evidence derived from the two studies^19,20) in the meta-analysis (Table 2). The overall RoB of both studies was high, and the sample size was small; therefore, the relevant domain was downgraded. Consequently, the certainty in the evidence of this study was rated as very low.

Overall, the results of this study indicate that bloodletting is a promising intervention for improving impaired consciousness in patients with acute stroke. Although it was difficult to identify a significant response 15 minutes after the intervention, the GCS scores significantly improved in the small- and medium-volume lesion subgroups at 30 minutes, and the effect was further strengthened at 45–50 minutes. In the large-volume lesion subgroup, a significant effect of bloodletting was initially found later, at 45–50 minutes after the intervention. In summary, bloodletting is effective at improving impaired consciousness in patients with acute stroke, and the effect tends to strengthen as the volume of the lesion decreases and as time passes within the range of 30–50 minutes after intervention.

This finding is consistent with the results of the study by Yu⁶⁾, which were not included in the meta-analysis. Indeed, Yu⁶⁾ observed the GCS of participants for 80 minutes after the intervention and reported that the difference between the treatment and control groups gradually widened over time, and a statistically significant difference between the groups was finally observed only after 80 minutes, which is consistent with the trend of time-dependent effect enhancement found in our meta-analysis. Yu⁶⁾ also reported that the subgroup with a less severe level of impaired consciousness responded better to bloodletting, which is consistent with the findings of our study that smaller lesion volumes are associated with a better response.

The largest effect size identified in this meta-analysis of our study as an increased GCS score after bloodletting was 0.68. Therefore, we need to examine whether such a small value is clinically meaningful. The minimal clinically important difference in GCS has not been officially recognized; a decrease of even one point may clinically be considered as neurological deterioration, and thus may require urgent brain imaging²⁶⁾. As such, the findings of this study can be interpreted as being clinically significant.

A previously published SR entitled ‘Acupuncture for post-stroke coma’ reported the results of a meta-analysis of relevant RCTs¹⁶⁾. The results of this prior SR indicated that acupuncture including bloodletting was effective at improving GCS scores, but the bloodletting subgroup did not show significant improvements, which is inconsistent with our findings. Wu¹⁶⁾ combined data from three RCTs^6,19,20), unlike our study, which excluded Yu’s data⁶⁾ from the meta-analysis because of the significant differences in the way that the data were presented compared with the other two studies^19,20). To obtain as much synthetic data as possible, we emailed the authors of the original RCTs requesting raw data, but did not receive any response. Therefore, unlike our SR, which combined synthetic data from only two RCTs^19,20) and presented multiple forest plots by subgroups according to lesion volume and time points, Wu’s study¹⁶⁾ combined heterogeneous data from all three RCTs^6,19,20), deriving a single forest plot. As a result, Wu¹⁶⁾ failed to show a significant effect on bloodletting, and did not show a trend of changes in the treatment effect according to lesion volume or time.

We also found it difficult to accept the data from Yu’s study⁶⁾ in Wu’s meta-analysis¹⁶⁾. As explained above, Yu⁶⁾ presented the posttreatment outcome as a figure, not as a numerical value. We attempted to estimate the average GCS score based on the figure in a published article, estimating that the mean GCS score of the entire population of the bloodletting group changed from approximately 6.3 at baseline to 7.9 at the end, and in the case of the moderate-severity subgroup with the best response, it seemed to change from 9.9 to 11.1⁶⁾. As such, the estimated change before and after bloodletting was approximately 1.2–1.6 point. However, the GCS value of the bloodletting group of Yu’s study⁶⁾ entered in Wu’s meta-analysis¹⁶⁾ was 5.55 ± 2.23, which does not match any of the post-treatment values nor pre- and post-treatment changes. We emailed the authors of the previous study in attempt to resolve this issue, but did not receive a reply.

In this situation, we concluded that, based on the currently available clinical evidence, it was best to combine data from only the two RCTs^19,20), despite the small number of studies, to correctly answer our review question. As a result, we reached a different conclusion from a previous study¹⁶⁾, which stated bloodletting significantly improved the consciousness of acute stroke patients, and also obtained unknown information that the effect depended on the lesion volume and the time passed after the intervention. Therefore, we decided to conduct our study with readers, even though we included only two small RCTs in our meta-analysis.

The mechanism by which bloodletting improves consciousness impairment in acute stroke has not yet been clearly investigated. Similar interventions, such as venesection, phlebotomy, or hemodilution, are known to improve conditions associated with several diseases by removing a relatively large amount of blood from the body to relieve blood flow occlusion caused by vasospasm, reduce blood viscosity, or change blood volume⁵⁾. However, bloodletting as defined in our study is thought to have a different mechanism of action, as it involves pricking the skin at acupoints with a small acupuncture needle to release only a few drops of blood.

One preclinical study previously showed that bloodletting contributes to prolonged survival, reduces brain water content, and increases brain density in a rat model of middle cerebral artery occlusion²⁷⁾. Potential mechanisms to explain these effects include suppression of local nitric oxide synthesis in the brain lesion area, reduction of blood brain barrier permeability through protection of tight junctions, and alleviation of cerebral edema^28,29). Another study speculated that bloodletting may exert a neuroprotective effect by reducing the flow rate of cerebral interstitial fluid, thereby downregulating the metabolic rate of neurons within the lesion³⁰⁾, while another study suggested that the regulatory effect on coagulopathy resulting from acute brain injury may contribute to the promotion of recovery³¹⁾.

Increased blood pressure also needs to be considered as one of the mechanisms of improvement underlying the consciousness associated bloodletting, considering the finding of Guo¹⁹⁾, which demonstrated a significant increase in SBP and HR. In particular, these changes were more evident in the small lesion subgroup, where the GCS changes were also prominent¹⁹⁾. As such, Guo assumed that the increase in SBP accompanied by an increase in HR would increase the blood supply to the brain lesion and contribute to the improvement of consciousness¹⁹⁾. However, unlike Guo’s study¹⁹⁾, Yu did not observe any significant changes in any items of the vital signs⁶⁾. Another SR evaluating the effect of repeated bloodletting on hypertension revealed that the proportion of patients with a significant reduction in BP was higher in the bloodletting group, suggesting that vasorelaxation due to increased serum nitric oxide concentration could be a contributing mechanism³²⁾. When reviewing the above results, it cannot be ruled out that single session of bloodletting may temporarily increase BP, while repeated session of bloodletting may actually lower BP in the long term, particularly in patients with hypertension. Further research is required to determine the role of BP in explaining the immediate effect of a single session of bloodletting on acute stroke.

Another hypothesis to explain the mechanism underlying the utility of bloodletting is that peripheral somatic nerve stimulation by bloodletting affects the central nervous system through neurotransmitters in the afferent pathway³³⁾. All RCTs^6,19,20) included in our SR involved bloodletting induced by pricking the Jing-well acupoints of hand with a sharp needle tip. The hand Jing-well acupoints are located at the end of the finger, and the fingertip is rich in peripheral nerve distribution. In addition, the projection area of the fingers in the somatosensory area of the cerebral cortex is much larger than that of other parts of the body³⁴⁾. As such, peripheral stimulation of the fingertip may have a more maximized effect on the cerebral cortex than stimulation of other parts, and Jing-well acupoints may have acted as a portal between the peripheral and central nervous systems^6,20). The theory of TM also interprets the acupoint as the place where the energy of the meridian first wells up and connects the energies of the 12 meridians. However, because none of the included RCTs performed bloodletting on acupoints other than the Jing-well points or non-acupoints, it is not certain whether these effects could be derived only from hand Jing-well points. From this perspective, it is not yet possible to conclude whether the effect of bloodletting comes from the release of blood drops, from acupoint simulation, from simple peripheral stimulation, or from a synergistic effect of all these stimulations combined; as such, further research and discussion are warranted.

The following limitations should be acknowledged when interpreting the results of this study. First, only two small RCTs were included in the meta-analysis. Furthermore, because both studies were conducted and published in China, the possibility of selection bias cannot be ruled out. Further, researchers have previously claimed that the methodological quality of TM RCTs published in China is relatively poor³⁵⁾. Indeed, the two RCTs^19,20) included in our meta-analysis were also assessed as poor in their reporting quality, as they omitted important methodological information, including randomization, allocation concealment, and blinding. Further, most recently published, well-designed study⁶⁾ with high reporting quality was not included in the meta-analysis because the data presented were too heterogeneous from the other two^19,20). As a result, the findings from this study are based on low-certainty evidence and are likely to change with future well-designed, high-quality, large-scale studies, which makes the authors reluctant to make strong recommendations for the use of single session bloodletting for acute stroke.

Second, this study could not explain any effects on the outcomes other than the immediate change in the GCS total score. For example, through subdomain assessment of the GCS, Yu’s study⁶⁾ showed that bloodletting improved eye opening and verbal response in patients with impaired consciousness, but did not significantly influence motor responses, indicating that a single session of bloodletting may not be sufficient to improve motor responses. A few RCTs have reported that repeated sessions of bloodletting for weeks improved neurological deficits and functional dependency as well as GCS in patients with acute stroke^36,37), however, more supporting evidence is required.

Third, we could not determine the effects of other important clinical factors, such as stroke type or lesion location, on the immediate effects of bloodletting other than lesion volume and time point after the intervention. Yu’s study⁶⁾ previously suggested the following interesting possibility regarding the time elapsed from stroke onset to intervention: when bloodletting was applied within 6 hours after onset, the response to bloodletting was faster and better than that after 6 hours. Therefore, there may be a time window for optimizing the immediate effect of bloodletting. However, we cannot be sure of this, as there are no other studies supporting this finding. Moreover, we are not sure whether bloodletting must be applied to all 12 acupoints on both hands, whether only some of the points can be selected, or whether the same effect can be expected from the stimulation of other acupoints or even non-acupoints. Further research is needed to determine whether this effect is necessarily achieved by following the complete procedure in traditional medicine or whether sharp pain stimulation is sufficient.

The results of this study should be interpreted with caution because of the quantitative scarcity, geographical bias, and qualitative weaknesses of the included RCTs. Further research on a larger scale and with greater geographical diversity is warranted to address the methodological shortcomings of previous studies. that is larger in scale and more diverse in geographical locations to address the methodological shortcomings of the previous studies. In addition, future studies need to be designed to provide more information to clarify the indications for when bloodletting is most effective, and to identify detailed intervention techniques to optimize the effects of bloodletting.