This meta-analysis has been registered on PROSPERO, registration number CRD42022348055. PRISMA checklist is shown in Supplementary Materials.

The retrieval strategy adopted a combination of subject words and free words and adjusted the search words according to the search results. The retrieval time was from the establishment of the database to July 2022. The search databases were PubMed, Excellent Medica Database (EMBASE), Cochrane Library, CNKI, Wanfang, and VIP. The search terms included central poststroke pain, CPSP, central neurological pain after stroke, central neurological pain after cerebrovascular disease, central neurological pain after cerebral information, central neurological pain after intracerebral haemorrhage, and thalamic pain. See the appendix for all search formulas.

(1) The study was a randomized controlled trial (RCT). (2) The subjects were identified as stroke patients by brain common therapy (CT), magnetic resonance tmaging (MRI) and other imaging examinations and were diagnosed with CPSP with a pain score of 3 or more. (3) Treatment plan: The control group was treated with routine therapy (including routine rehabilitation training) combined with routine painkillers (gabapentin, pregabalin, and lamotrigine), and the treatment group was treated with noninvasive therapy based on routine rehabilitation training. (4) Outcome indicators: the main indicator was VAS, and the secondary indicators were clinical total effective rate and clinical adverse reaction rate.

(1) Unrelated literature, review, case reports, meta-analyses and other nonrandomized controlled trials; (2) Documents with repeatedly published data, incomplete data or unavailable data; (3) Repeatedly published literature in Chinese and English.

Two researchers independently conducted literature screening and data extraction according to the inclusion and exclusion criteria and cross-checked the results. In case of disagreement, the third researcher participated in the discussion and decision. Endnote (Thomson ResearchSoft, Stanford, CT, USA) and Excel software (Microsoft, Redmond, WA, USA) were used for literature management and data extraction, mainly including the basic characteristics of cases included in the literature, intervention measures, course of treatment and outcome indicators. All continuous data are included in the difference before and after treatment (that is, the difference between indicators after treatment and before treatment). If the original text is not calculated, it will be calculated by itself. The formula is as follows: corr is usually taken as 0.5.

$${\mathrm{SD}}_{\text{E.change}}=\sqrt{{\mathrm{SD}}_{\text{E.baseline}}^2+{\mathrm{SD}}_{\text{E.final}}^2-(2\times \text{Corr}\times {\mathrm{SD}}_{\text{E.baseline}}}$$

Mean${}_{\text{E.change}}$ = Mean${}_{\text{E final}}$ – Mean${}_{\text{E baseline}}$

The quality of the included literature was evaluated according to the quality evaluation scale recommended in the Cochrane Handbook, including 6 parts: reporting bias, detection bias, performance bias, selection bias, loss of visit bias and other bias. Each part of the evaluation content can be judged as high, medium and low risk according to the standard.

This study was based on the framework of frequency. For the continuity index, the mean difference (MD) was used as the effect quantity. If it was a binary variable, the odds ratio (OR) was used as the effect quantity, and the corresponding 95% confidence interval (CI) was calculated. Stata 14.0 software (Stata Corporation, Lakeway, TX, USA) was used to draw a network evidence relationship diagram, forest diagram, grade probability diagram, funnel diagram and corresponding statistics. When testing the global consistency, if the difference was not statistically significant, that is, p $>$ 0.05, no overall inconsistency existed [10]. If the study did not form a closed loop and the consistency test could not be carried out, Review Manager 5.3 software (the Cochrane Collaboration, Nordic Cochrane Center, Copenhagen, Denmark) was used for traditional meta-analysis. The heterogeneity test was mainly judged by I square (I${}^2$). If no heterogeneity existed between the research results (I${}^2$ $\le$50%), the fixed effect model was used for meta-analysis; if heterogeneity existed among the research results (I${}^2$ $>$50%), the source of heterogeneity was further analysed, and the random effect model was used for meta-analysis. In this study, SUCRA was used to calculate the cumulative ranking probability of each treatment scheme. A larger SUCRA value, that is, a larger area under the curve of the cumulative probability ranking diagram, indicated a better effect of this intervention.

The included studies were evaluated using the Cochrane manual 5.1.0 bias risk assessment tool. In terms of the random allocation method, 9 studies were low risk, and the random number table method [11, 13, 14, 15, 16, 17, 20] was used for random allocation. Two studies were high risk [18, 19] and did not mention randomization, and the remaining studies only mentioned randomization without reporting specific random allocation methods. In terms of concealment of the random allocation scheme, one study [22] mentioned that allocation concealment was low risk, and the remaining the studies did not mention it. In terms of the blind method of the random allocation scheme, two studies [11, 22] mentioned the use of the blind method, one of which was single-blinded [11] and the other double-blinded [12]. The measurement of outcome indicators was not mentioned in the literature. In terms of data integrity, the data included in the study were complete. In terms of selective reporting of research results, the included studies were all low risk. Other sources of bias were unclear. The risk assessment of bias in the included studies is shown in Fig. 2.

Fig. 2.

Risk of bias graph. (A) Risk of bias summary. (B) Risk of bias graph.

Among all included studies, five studies [12, 14, 15, 16, 22] reported adverse reactions, which were mainly mild adverse reactions, such as mild headache, fatigue, dizziness, dizziness, drowsiness, dry mouth, temporary scalp numbness or facial muscle twitching. See Table 3 (ref. [12, 14, 15, 16, 22]) for the occurrence of adverse reactions/events.

3.4.1 Evidence Network of the Clinical Adverse Reaction Rate

Five studies reported the clinical adverse reaction rate, including four noninvasive treatment regimens. All four treatments were directly compared, and no closed loop formed. The number of studies comparing rTMS with CT treatment was the largest (3 RCTs). Few studies included this research index, and traditional meta-analysis and funnel analysis were consequently not performed. The evidence network is shown in Fig. 5A.

Fig. 5.

Relevant processing diagram of clinical adverse reaction rate. (A) Network diagram of the clinical adverse reaction rate. (B) Network meta-analysis of clinical adverse reaction rate.

The VAS has obvious advantages in evaluating subjective factors. This method was first used in psychological research [26, 27]. Later, VAS scores ranging from 0 to 10 were used to judge the degree of pain [28]. During the evaluation, patients are asked to select the score according to their own feelings. A higher score indicates a more severe degree of subjective pain, and the score value can be accurate. The clinical manifestations of CPSP are diverse, but the main characteristics are pain and sensory disorders. At present, the diagnosis of the disease lacks specificity, and the diagnosis largely depends on the subjective feelings of patients. Therefore, VAS is widely used in the pain assessment of CPSP. The results of this study show that among all noninvasive interventions, SLI is most effective in improving the VAS indicators of patients. The stellate ganglion is the main pathway of the sympathetic nerve. Blocking the stellate ganglion is generally believed to be able to expand the blood vessels in the area dominated by the stellate ganglion. More studies have shown that [29] stellate ganglion block (SGB) therapy can alleviate many drug-resistant headaches and reduce the original dosage of drugs. Invasive stellate ganglion block has achieved a good curative effect, while noninvasive stellate ganglion irradiation, as an alternative therapy for stellate nerve block, is safe and practical [25], but relevant studies on the clinical effect of treating CPSP are scarce. The VAS results of the main indicators in this study suggest that the pain relief effect of SLI is the best, but only one study was included. Therefore, this result is limited to the data analysis of the included literature and cannot completely explain the advantages and disadvantages of each clinical efficacy. Nevertheless, the results rain significance for clinical guidance. The secondary indicator of this study is the total clinical effective rate. The results suggest that PT has the best curative effect. Many CPSP patients often have a certain degree of anxiety and depression when drug efficacy is poor, and anxiety is often accompanied by prostaglandins, bradykinins and other secretions that exacerbate pain, leading to a decrease in the pain threshold. This effect results in emotional fluctuations and a loss of confidence in treatment and prognosis. Therefore, mindfulness psychotherapy, a type of systematic psychotherapy, can have a positive effect on the treatment of central neuralgia and limb rehabilitation of stroke patients by mobilizing the enthusiasm of patients and relieving tension and negative emotions, but sufficient clinical research to further confirm this hypothesis is lacking [30]. However, clinicians often note unexpected effects on CPSP patients with routine drug treatment and psychological treatment in clinical practice.

Five studies included in this meta-analysis mentioned adverse reactions after the intervention. The intervention measure with the highest adverse reaction rate was rTMS, which manifested as mild headache, facial twitching or numbness. The adverse reaction rate of SLI was the lowest and mainly manifested as lethargy, dry mouth, and dizziness. All adverse reactions reported in RCTs were mild or transient, and no serious adverse reactions were reported. Therefore, the psychological tension of patients is not excluded. rTMS is a noninvasive neuromodulation therapy based on achieving an analgesic effect with the application of a high-intensity pulsed magnetic field to brain tissue. According to current research, rTMS is effective in the short term and has a limited effect on poststroke central neuralgia. Thus, rTMS is generally considered to not be suitable to be used alone, and other treatment measures, such as combined drug treatment, can improve therapeutic effects [31]. However, attention should be given to evaluation before rTMS treatment in clinical practice to identify patients with intracranial metal implants, cochlear implants, and built-in pulse generators (brain pacemakers, cardiac pacemakers). In this case, rTMS easily causes the built-in pulse generator to fail. Patients with serious physical diseases, those who are using drugs that significantly reduce the threshold of seizures or those who abuse alcohol need to consider the risk-benefit ratio after receiving stimulation treatment, and the use of rTMS is recommended with caution.

In this study, the network meta-analysis method was used for the first time, which can not only compare the indirect and mixed effects between different noninvasive therapies but also compare different treatment schemes for CPSP in the absence of direct comparisons between therapies and identify the best scheme with the greatest probability. This study has the following limitations: (1) Most of the included studies were Chinese studies, and the number of studies was small. Because noninvasive therapy intervention CPSP is a relatively new clinical field, most included studies were drug trials or tested invasive therapies. When studies published in English were included, most of the clinical trials involving noninvasive therapy intervention CPSP were non-RCTs. Except for studies of rTMS, studies on noninvasive therapies are scarce, which may be because rTMS, an alternative therapy of neuromodulation, has long been studied. Consequently, it is more prevalent in the literature. Conversely, other noninvasive therapies are relatively new, and they are not well represented in the literature; (2) The overall quality of the studies was not very high. Although most studies mention random methods, others, such as allocation concealment, blinding methods and potential bias, are not mentioned. (3) The number of RCTs included in the study is insufficient, the sample size is small, with some studies reporting sample sizes of less than 10. Thus, negative results may have not been published. (4) Among the included studies, the length of intervention is somewhat inconsistent, which precludes comparisons between studies and increases clinical heterogeneity; (5) Few outcome indicators are included in this paper. Others, such as the numerical rating scale (NRS), Hamilton Depression Scale (HAMD), and Hamilton Anxiety Scale (HAMA), are also common outcome indicators of CPSP, but they were not included mainly because too few other outcome indicators were involved in the RCTs included, and this number was insufficient for meta-analysis; (6) Because few studies were included in this work and their quality was poor, most of the studies did not take the patient’s family genetic history as the basic research index. However, familial genetic factors play a very important role in the onset and prognosis of stroke. I hope that more studies will include the genetic history as a basic index in the future.