Types of studies

We included published randomised controlled trials (RCTs) in full text or abstract form. We included abstracts if sufficient information for analyses could be obtained from study authors. We excluded quasi‐RCTs and clinical controlled trials (CCTs) because we found numerous RCTs for this update. This is different from the approach presented in the protocol.

Types of participants

Participants were adults (18 years or older) with the following neck disorders.

• Mechanical neck disorders (MNDs), including whiplash‐associated disorders (WADs) categories 1 and 2 (Spitzer 1987; Spitzer 1995), myofascial pain syndrome (MPS) and degenerative changes (DCs) (Schumacher 1993).

• Neck disorder with headache (Olesen 1988; Olesen 1997; Sjaastad 1990).

• Neck disorders with radicular symptoms (NDRs), including WAD category 3 (Spitzer 1987; Spitzer 1995).

For the purposes of this review, we defined symptom duration as acute (< 30 days), subacute (30 days to 90 days) or chronic (≥ 90 days).

We excluded studies if they investigated neck disorders with:

• definite or possible long tract signs (e.g. myelopathies);

• neck pain caused by other pathological entities (Schumacher 1993);

• neck pain related to neurological disease (e.g. spasmodic torticollis);

• neck pain related to fracture and dislocation;

• headache not of cervical origin;

• co‐existing headache when neck pain was not dominant or when headache was not provoked by neck movement or sustained neck posture; or

• 'mixed' headache.

Types of interventions

Studies must have used acupuncture techniques involving insertion of needles. Stimulation of needles may involve manual, electrical, heat, laser or other forms of stimulation. Control groups were treated with sham acupuncture (some form of mock or pretend), wait‐list control or inactive treatment control (e.g. sham transcutaneous electrical nerve stimulation (TENS)).

Types of outcome measures

Electronic searches

We searched the following databases from their inception to August 2015.

• Cochrane Central Register of Controlled Trials (CENTRAL, which includes the Cochrane Back Review Group (CBRG) Trials Register; Ovid, August 2015).

• MEDLINE (Ovid, 1950 to August 2015 week 4).

• EMBASE (Ovid, 1980 to August 2015).

• Manual, Alternative and Natural Therapy Indexing System (MANTIS; Ovid, 1980 to November 2013 last available access date).

• Cumulative Index to Nursing and Allied Health Literature (CINAHL; EBSCO, 1982 to August 2015).

• Index to Chiropractic Literature (ICL; August 2015).

• ClinicalTrials.gov (August 2015).

• World Health Organization (WHO) International Clinical Trials Registry Platform (ICTRP) (August 2015).

The Chinese Cochrane Centre searched the Traditional Chinese Medical Literature Analysis and Retrieval System (TCMLARS) in China in September 2005. We did not further update the TCMLARS search, as this search yielded no RCTs previously.

See Appendix 1 for the search strategies used for CENTRAL, MEDLINE, EMBASE, MANTIS, CINAHL and ICL. Subject headings (MeSHs) and keywords included anatomical terms, disorder or syndrome terms, treatment terms and methodological terms consistent with those advised by the CBRG (Furlan 2009).

Searching other resources

We also screened references, communicated with the Trials Search Co‐ordinator of the CBRG, contacted identified content experts and reviewed our own personal files.

Selection of studies

At least two review authors independently identified citations, selected studies and abstracted data. We resolved disagreements through consensus.

If additional information was required to assess the appropriateness of a study for selection, we contacted study authors for clarification. If this was not forthcoming, we allowed a consensus process to determine selection status. If the article or citation posting was written in a non‐English language, one investigator and a translator with a health sciences background conducted study selection in an unblinded manner.

Data extraction and management

Two review authors independently extracted raw data for demographics, descriptions of treatment and all outcomes from full manuscripts onto pre‐designed forms.

Using the Chi² test, we calculated agreement between investigators for study identification, selection and validity processes before reaching consensus (Graham 2012). We used the quadratic weighted kappa statistic (K_w) (Cicchetti 1976) to measure agreement.

Assessment of risk of bias in included studies

At least two review authors independently assessed each selected study for methodological quality using pre‐piloted forms based on the CBRG risk of bias (RoB) assessment tool (12 criteria) (Furlan 2009). See Appendix 2 for operationalisation of the RoB tool. We rated each item as having high, low or unclear risk and entered ratings into the RoB table. A consensus team met to reach a final assessment based on an overview of RoB items; we used no pre‐defined cutoff score for the RoB assessment. We deemed an RCT acceptable for inclusion if we found no fatal methodological flaws (e.g. randomisation, allocation concealment, drop‐outs, intention‐to‐treat (ITT) analyses were done appropriately and were well reported).

Measures of treatment effect

We used descriptive statistics to provide a summary description of groups, interventions, outcomes, adverse effects of treatment and costs of care. We reported results on the basis of the sample size analysed.

We calculated standardised mean differences (SMDs) and 95% confidence intervals (95% CIs) for outcomes reported in a continuous data format. We used the SMD because different measures were frequently used to address the same clinical outcome. Effect size (SMD) is a unitless measure reported in standard deviation units. Generally, effect size can be interpreted as small (‐0.20), medium (‐0.50) or large (‐0.80), as defined by Cohen (Cohen 1988).

For dichotomous outcomes, we calculated risk ratios (RRs) and 95% CIs for outcome rates in the treatment versus control group. For undesirable outcomes, RR less than one represents a beneficial treatment.

When neither continuous nor dichotomous data were available, we extracted findings and statistical significance as reported by authors of the original study and noted them in the Characteristics of included studies tables.

Unit of analysis issues

For clinical trials using a simple parallel‐group design, we collected from each participant and analysed a single measurement for each outcome. For clinical trials using a cross‐over design, we collected from each participant and analysed measurements of each outcome for each participant from each of the interventions. We analysed only the first phase of the cross‐over study and included no cluster‐randomised trials in this review.

Dealing with missing data

When data were not extractable, we contacted the primary authors.

When neither continuous nor dichotomous data were available, we extracted findings and presented statistical significance as reported by study authors.

Assessment of heterogeneity

We considered the following possible sources of heterogeneity: symptom duration (acute vs chronic); subtype of neck pain (e.g. WAD); intervention type (traditional acupuncture vs trigger point needling); characteristics of treatment (e.g. dosage, technique); and outcomes (pain relief, measures of function and disability, patient satisfaction, quality of life).

A P value for the Chi² test (test of heterogeneity) less than 0.05 or an I² value greater than 25% would indicate significant statistical heterogeneity. However, because an acupuncture review includes few studies, the test of heterogeneity might not have the power to detect a difference even when one exists; therefore, we determined that subgroups were statistically heterogeneous if I² values were greater than 25% despite failure to reject the null hypothesis with the Chi² test.

Assessment of reporting biases

When a published protocol was available, we compared outcomes reported a priori with outcomes of interest in the RCT.

Data synthesis

Before calculating a pooled effect measure, we assessed the appropriateness of pooling on clinical grounds. We planned to combine outcome measures from individual trials by performing a meta‐analysis when possible (clinical comparability of populations, interventions and outcomes between trials). We planned to use a random‐effects model for these meta‐analyses.

We considered each subgroup (e.g. 1.1) to be clinically homogeneous. We meta‐analysed each subgroup unless we noted statistical heterogeneity. When a meta‐analysis was not possible, we described qualitatively in the text the results of clinically comparable trials.

Regardless of whether available data were sufficient for use in quantitative analyses to summarise the data, we assessed the overall quality of the evidence for each outcome. To accomplish this, we used the GRADE (Grades of Recommendation, Assessment, Development and Evaluation Working Group) approach, as recommended in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011) and adapted in the updated CBRG method guidelines (Furlan 2009). We considered that the following factors may decrease the quality of the evidence, study design and risk of bias, inconsistency of results, indirectness (not generalisable), imprecision (sparse data) and other factors (e.g. reporting bias). We reduced our rating of the quality of the evidence for a specific outcome by one level according to the performance of studies against these five factors.

• High‐quality evidence: Findings are consistent among at least 75% of RCTs with low risk of bias; consistent, direct and precise data; and no known or suspected publication bias. Further research is unlikely to change the estimate or our confidence in the results.

• Moderate‐quality evidence: One of the domains is not met. Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.

• Low‐quality evidence: Two domains are not met. Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.

• Very low‐quality evidence: Three domains are not met. We are very uncertain about the results.

• No evidence: We identified no RCTs that addressed this outcome.

Subgroup analysis and investigation of heterogeneity

We did not plan to perform a subgroup analysis.

Sensitivity analysis

We planned to conduct a sensitivity analysis on risks of bias or meta‐regression by type of neck pain and duration of symptoms. However, we did not find sufficient data for any of these categories.