Types of studies

We will include randomized or quasi‐randomized trial (controlled clinical trial in which assignment to groups uses alternate allocation, case record numbers, dates of birth, patient record or social security number, etc.), in any language. Quasi ‐randomized trials will be included as it is anticipated that a limited number of trials exist for some orthoses.

Types of participants

The participants will be adults, 18 years or older, who suffered from acute (less than 30 days), subacute (30 to 90 days) or chronic (longer than 90 days) neck disorders categorized into three groups:
1) Mechanical neck disorder (MND) that is non‐specific, including whiplash associated disorders (WAD I to II) (Spitzer 1987; Spitzer 1995), myofascial neck pain, and degenerative changes (Schumacher 1993).
2) Neck disorder with headache (NDH) corresponding to diagnostic classification 11.2.1 proposed by the International Headache Society classification and diagnostic criterion for headache disorders (Olesen 1988; Olesen 1997; Sjaastad 1990).
3) Neck disorders with radicular findings (NDR) including WAD III (Spitzer 1987; Spitzer 1995).

We will exclude studies if they investigated neck disorders with

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

• neck pain caused by other pathological entities (e.g. tumours) (Schumacher 1993);

• headache not of cervical origin but associated with the neck (e.g. migraine, tension‐type headache)

• co‐existing headache when either neck pain was not dominant or the headache was not provoked by neck movements or sustained neck postures (e.g. migraine and cervicogenic headache); or

• 'mixed' headache (e.g. a study group with a variety of headaches where the subgroup with cervical headache cannot be separated from the other headache group(s)).

Types of interventions

Eligible studies will have used at least one type of orthosis to treat neck disorders. The most common orthoses will likely include cervical collars and cervical pillows, but there may be other medical devices that have been used to treat neck disorders such that the outcomes of pain, function, and satisfaction are affected. The comparison groups could be either a control treatment or another treatment. The following comparisons will be included:

1) orthosis versus sham or placebo (e.g. collar versus placebo; collar + sham ultrasound vs sham ultrasound),

2) orthosis versus no treatment or wait list control (e.g. collar versus wait list)

3) orthosis plus another intervention (single or multimodal) versus that same intervention (e.g. collar + exercise versus exercise)

4) orthosis versus another intervention (e.g. collar versus exercises)

5) one type of orthosis (i.e. standard pillow) versus another type of similar orthosis (i.e. orthopaedic pillow)

6) dose of orthosis treatment versus another dose of the same orthosis treatment (i.e. three weeks with nine sessions of collar use versus three weeks with three sessions of collar use).

We anticipate that orthoses will be used in conjunction with other therapies to treat neck disorders. However, studies that compare multimodal therapy that includes the use of orthoses and a single treatment modality (e.g. exercise therapy) and studies where the same orthoses are included in both treatment arms will be excluded.

Types of outcome measures

We selected the outcomes of pain, disability or function (including neck specific disability, work disability, or related work productivity measure), patient satisfaction, quality of life, and global perceived effect; these are the outcomes of primary interest to clinicians and third party payers (Turk 2005). As there is no consensus on which measures best reflect these domains for mechanical neck pain, we will not restrict the type of measures used in studies to capture these domains. In addition, information on unintended harms and cost will be retrieved when available.

The duration of the follow‐up period after treatment will be defined as follows:
1) immediately post‐treatment: up to one day
2) short term follow‐up post‐treatment: more than one day, but less than three months
3) intermediate term follow‐up post‐treatment: three months to one year
4) long term follow‐up post‐treatment: one year or longer

Study Selection

The same eligibility criteria for population, study design, and outcomes will be applied to both title and abstract screening and full text review. Two review authors with expertise in medicine, occupational health, physiotherapy, or clinical epidemiology will independently evaluate the titles and abstracts of citations from the computer searches using pre‐piloted forms and label them "relevant," "not relevant," or "unsure." All citations classified as "relevant" or "unsure" will be retrieved for further scrutiny. Focused selection criteria will be used for final selection of studies, based on the full text of the retrieved article. Disagreement will be resolved through consensus in both stages of selection.

If the article or citation posting is in a language other than English, one investigator and a translator with a medical background will conduct the study selection. Letters will be sent by surface or e‐mail, to authors of articles published only in abstract form to request full manuscripts and data sets.

Data Extraction

Two review authors will independently extract data on continuous (means and SD) or dichotomous data (e.g. proportions) using standardized pre‐piloted forms. We will also extract demographic data and details of the reported treatment characteristics or dosages. Finally, we will extract trial sponsorship and the authors' interpretations of the study results. A consensus process will be used to resolve disagreements. If pain, function or disability, global perceived effect, quality of life, or patient satisfaction data are not reported in a form suitable for quantitative analysis, we will request raw data from the author(s) or assistance with raw data transformation. Authors will also be contacted of there are insufficient detail within eligible studies. These contacts will be recorded in the Characteristics of included studies Table.

Methodological quality of included studies

Two review authors with varied professional background will independently assess methodological quality. Methodological quality will be judged using these three sets of criteria:
1) the validated Jadad 1996 criteria (maximum score: five, high score: three or more, See Table 1),
2) the Cochrane Back Review Group criteria (van Tulder 2003) (maximum score: 11 criteria met, high score: six or more criteria met. See Table 1);
3) the Cochrane grading system for quality of allocation concealment (A to D ‐‐ see Characteristics of included studies table)

Because they are validated, we will use the Jadad criteria for the primary classification of methodological quality. It may be difficult to double blind some orthotic devices; for example, a patient who is using a "different" orthopaedic pillow, relative to the standard one they normally use, may be aware that this is one of the "pillow interventions" he or she is evaluating. The quality assessment criteria met for each study will be listed in a separate Additional Table.

Data Analysis

We will calculate agreement between investigators for study selection prior to consensus. The Kappa statistic will be used to measure agreement. Results measuring less than zero reflect poor agreement, 0 to 0.20 slight agreement, 0.21 to 0.40 fair agreement, 0.41 to 0.60 moderate agreement, 0.61 to 0.80 substantial agreement and 0.81 to 1.00 almost complete agreement (Landis 1977). We will use kappa assessment to check and calibrate our investigators and to ensure consistency is maintained. Review authors will try to reach consensus in cases of substantive disagreement. If disagreement persists, a third review author will be consulted.

Prior to calculation of a summary effect measure, the reasonableness of combining across studies will be assessed on clinical and biological grounds; specifically the rationale with respect to clinical homogeneity. Possible sources of clinical heterogeneity can include: 1) duration of symptoms (i.e. acute versus chronic); 2) subtype of mechanical neck disorder (i.e. neck disorder with radicular findings, whiplash injury, etc.); 3) intervention type (specific orthosis); 4) outcomes (i.e. subject reports of pain and pain relief, range of motion, other measures of performance (i.e. activities of daily living, disability, function, or employment status)). We will also evaluate methodological heterogeneity by examining the methodological quality of the studies.

When sufficient data are available, the pooled or common effect measure will be calculated for the main outcomes [pain, function and disability, global perceived effect, quality of life, and patient satisfaction]. Before data are extracted from eligible studies, we will not know for sure which outcome measures (if any) will be combined into a summary estimate.

When neither continuous nor dichotomous data are available, the findings and the statistical significance as reported by the author(s) in the original study will be extracted.

For outcomes reported in a continuous data format, we will use a random effects model and a standardized mean differences with 95% confidence intervals (SMD; 95% CI) will be calculated. SMD is a unit‐less measure reported in standard deviation units with a conversion to a common scale; this is appropriate for effect size estimates where different scales are used to measure the same clinical domain, a situation often seen in the neck pain literature.

Generally, an effect size measured by the SMD, can be interpreted as small (0.20), medium (0.50) and large (0.80), as defined by Cohen 1988. A 10‐mm change in pain on a 100‐point pain scale (10%) is probably the minimum clinically significant difference for pain scores, as suggested by Farrar 2001 in other pain trials and Felson 1995 in rheumatoid arthritis trials. A change of five units on the Neck Pain Disability Index is likely the minimum clinically important difference for neck pain, as suggested by Stratford 1999. For some outcomes, a negative SMD indicates the positive direction of effect (for example: decreased pain level or increased function (smaller neck disability index score)). For other outcomes, a positive SMD depicts a beneficial treatment effect (for example: improved disability, change in headache intensity, and improved patient satisfaction).

For continuous outcomes reported as medians, the effect size will be calculated as follows: The variance of the median equals ((1.2533² times sigma²) divided by n (Kendal 1963)). The standard deviation (SD) equals three quarters of the inter quartile range (IQR). Thus, the median effect size equals the difference between median₁ and median₂, divided by the SD (median₁ and median₂). The IQR will be transformed to SD.

Dichotomous data will be extracted for outcomes when continuous data are not available. The relative risk (RR), the outcome rate in the treated versus control group, will be calculated. For undesirable outcomes (for example, no improvement in pain), a RR measuring less than one represents a beneficial treatment.

The number needed to treat, absolute benefit, and treatment advantage will be calculated for primary findings of studies that show improvement (benefit) for the outcomes of interest. End point RR, SMD, NNT, and treatment advantage will be recorded in the Characteristics of included studies tables, under outcome.

Where data are not available for meta‐analysis, descriptive information will be reported. To check for the presence of adverse events, the relative risk will be calculated where possible.

Using a random effects model, statistical heterogeneity between the SMD and RR for cumulated studies will be formally tested. If the Q test for heterogeneity is not significant at the P = 0.05 level and the I² is less than 40% (i.e. low to moderate), then a common SMD or RR (based on a weighted average of the results of all the available relevant studies) will be considered in the synthesis as a valid summary estimate. The weight given to each study is the inverse of its variance; for example, more precise estimates (from larger studies with more events) are given more weight. Tests of homogeneity consist of formal statistical analyses, used for examining whether the observed variation in study results is compatible with the variation expected by chance alone. The more significant the results of the test (the smaller the p‐value), the more likely it is that the observed differences were not due to chance alone. Review Manager 4.2 will automatically test the homogeneity of the results of the individual studies being combined for each comparison of dichotomous or continuous data.

Sensitivity analysis or meta‐regression will be conducted, if possible, to assess the extent to which the methodological quality of studies, disorder subtype, and duration (acute, subacute, chronic) of the disorder would lead to variation in the primary outcomes. This logistical model will be used to explore the relationship between these key characteristics of included studies and the results for each study. Sensitivity analysis is only feasible if meta‐analyses have sufficient numbers of studies (four or more).

Descriptive methods will be used to check for publication bias and language bias and to provide a summary description (for the groups, interventions, outcomes, and adverse effect of treatments) where possible. All results reported will be based on the sample size analysed; specifically, the sample that completed the study rather than the sample that entered the study. The baseline mean, end of study mean, and absolute benefit will be noted in the Characteristics of included studies table.

We recognize that post‐hoc power analyses may sometimes be problematic, as power is not a property of a study. However, power analyses will be conducted for articles reporting non‐significant findings to determine if lack of power is a factor in these non‐significant findings. We will assume that the desired level of significance was set at alpha equal to 0.05; adequate power was defined as at least 80%. As stated above, for the results to be considered clinically important, a change of at least 10% between means of the two groups must be detected (Dupont 1990).

Regardless of whether there are sufficient data available to use quantitative analyses to summarize the data, we will assess the overall quality of the evidence for each outcome. To accomplish this, we will use an adapted GRADE approach, as recommended by the Back Review Group (Furlan 2008). The quality of the evidence on a specific outcome is based on the study design, methodological quality, consistency of results, directness (generalizability), precision (sufficient data) and reporting of the results across all studies that measure that particular outcome. The quality starts at high when high quality RCTs provide results for the outcome, and reduces by a level for each of the factors not met.

• High quality evidence = there are consistent findings among at least two (high quality) RCTs with low potential for bias that are generalizable to the population in question. There are sufficient data, with narrow confidence intervals. There are no known or suspected reporting biases.

• Moderate quality evidence = one of the factors is not met

• Low quality evidence = two of the factors are not met

• Very low quality evidence = three of the factors are not met

• No evidence = no evidence from RCTs