Our protocol will be conducted according to the Preferred Reporting Items for Systematic Review and Meta-Analysis Protocols(PRISMA-P) statement.  We have registered the protocol on PROSPERO (registration number: CRD42020206853).
Types of studies
This study will only include randomised controlled trials (RCTs), and the study design will strictly follow population, intervention, control, outcome and study design principles. We will not use non-RCTs, animal experiments, human cell or tissue experiments, literature reviews, retrospective studies, case reports, studies with unavailable data and repeatedly published studies.
Types of participants
We will include patients >18 years of age with neck pain according to the International Classification of Diseases (ICD-11, ME84.0) diagnostic guidelines. Inclusion criteria for patients are as follows: neck pain as the main symptom for at least 3 months, pain lasting >20 min per time and at least once/month and a neck pain score of >3 on the scale of 0−10, which will be estimated using the visual analogue scale (VAS). Patients who have a specific cause for neck pain (e.g. traumatic injury and whiplash-associated disorder), serious cardiovascular disease or a previous history of neck surgery will be excluded.
Studies using tuina therapy, cupping therapy and acupuncture therapy (e.g. EA, MA and FA) alone or in combination as an intervention will be included.
Any comparison is allowed, including comparisons with placebo intervention, no treatment, false stimulation under same conditions or usual care.
Types of outcome measures
Studies reporting one or more of the following outcomes will be included.
Primary outcomes will include the VAS and Hamilton Anxiety Scale scores.
Secondary outcomes will include the Oswestry disability index, short form health survey, Pittsburgh sleep quality index and hospital anxiety and depression scale scores.
We will develop a comprehensive search strategy and search electronic databases established before December 2020, including five English databases (Web of Science, MEDLINE, EMBASE, OVID and Cochrane Library) and four Chinese databases (CNKI, SinoMed, Wanfang Database and VIP) without language restrictions. In addition, we will include gray references (e.g. conference papers and bibliographies) and academic dissertations to reduce publication bias in our data. (See Table 1)
We will select studies and exclude any duplications with NoteExpress V.3.0 (Beijing Aegean Software Co., Ltd, China.). Two reviewers (S-XQ and AL) will read the title and abstract of each study independently to exclude studies that clearly do not meet the inclusion criteria. Any discrepancy will be discussed with a third reviewer (YZ). Studies that are deemed relevant by at least one reviewer will undergo a full-text review. Then, two reviewers will read the full text to determine the final included reviews. After completing the screening phase, the two reviewers will use Microsoft Excel 2019 to extract information from the included reviews, including general information (author and publication year), review characteristics (study setting and design, inclusion criteria and intervention type) and AEs.  (See Figure 1)
Assessment of RoB
RoB in the methodological quality of the RCTs will be assessed by two reviewers(S-XQ and AL) independently using the revised Cochrane RoB tool (RoB 2.0)  RoB will be classified as high risk, low risk and unclear. The two reviewers will discuss all differences with each other and with a third reviewer (YZ), if necessary.
Assessment of certainty of evidence
We will use the GRADE approach for NMA to assess the certainty of evidence.  The evidence quality evaluation of primary outcomes can be regarded as four levels: high quality, moderate quality, low quality and very low quality.  Evidence quality is generally judged on the basis of RoB, inconsistency, indirectness, inaccuracy and dissemination bias.
Measures of treatment effect
Effect sizes for continuous outcomes will be expressed as mean difference or standardised mean difference with a 95% confidence interval (CI). For dichotomous outcomes, effect sizes will be expressed as risk ratios with a 95% CI. We will handle any data issues that arise according to the method described in the Cochrane handbook.  In exceptional cases (i.e. if a less number of RCTs are expressed as negative health outcomes continuously, while most of the RCTs used dichotomous outcomes), outcomes reported as continuous or categorical will be dichotomised. 
Firstly, we will use RevMan 5.3 software to carry out statistical analysis of RoB. We will also conduct a quantitative analysis of the heterogeneity of results through I2. When P > 0.1, we will neglect the heterogeneity, but if P < 0.1 and I2 > 50%, we will determine the possible reasons from both clinical and methodological perspectives as well as provide an explanation or conduct subgroup analysis.
Second, we will use the risk ratio to analyse dichotomous data. Furthermore, we will calculate the corresponding 95% CI for each parameter between the treatment and control groups.
Third, we will use GeMTC V.0.8.1 package of R-3.3.2 software for NMA based on a Bayesian framework to show meta-analysis using the Markov chain Monte Carlo method.  We will use the Brooks–Gelman–Rubin method to assess convergence. We will perform estimation and inference when the convergence state is stable. We will set the GeMTC package’s initial parameters as follows: number of chains, 4; initial value, 2.5; step size, 5; number of simulation iterations, 100,000 and number of adjustment iterations, 20,000. Parameters can be adjusted appropriately according to specific conditions. With the potential scale reduced factor close to 1, we will determine whether the consistency of the model is reliable. Moreover, a ranking figure of all treatment sessions will be generated if multiple intervention times exist.
Sensitivity and subgroup analyses
If possible, we will conduct sensitivity analysis by only including studies rated as low RoB. We will perform sensitivity analysis to monitor the robustness of the primary decision made in the review process. When considerable heterogeneity is detected in an initial analysis, we will perform a subgroup analysis, if necessary. 
Ethics and dissemination
For NMA, there is no direct data collection from human participants; therefore, no ethical approval is necessary. 
Safety and AEs
We will record the number and type of AEs in the included studies.