Criteria for considering studies for this review

Search methods for identification of studies

Relevant trials will be identified searching the following sources:
1. Cochrane Multiple Sclerosis Group specialised register
2. The Cochrane Central Register of Controlled Trials (CENTRAL)
3. MEDLINE (1966 ‐ )
4. EMBASE (1974 ‐ )
5. Reference list of relevant papers on diet and MS
Our search strategy will be drawn from the search strategy developed for the Multiple Sclerosis Group (See: Collaborative Review Group Search Strategy).
Pertinent trials will be identified using the terms: "Eating.mh", "Nutrition.mh", "Diet.mh", "Food and Beverages.mh", "Lipids.mh", "Vitamins.mh", "Minerals.mh".
In addition the following terms will be used:
food and beverage.tw, diet* or nutri* or eating* or lipid* or fat* or oil* or margarin* or butter or cheese* or dairy or egg* or fiber* or fibre* or vegetable* or nut* or seed* or spice* or barley or wheat or rice or corn or bread or cereal* or vitamin* or mineral*.tw
The search strategy will be formulated in PubMed and will be adapted by an experienced medical librarian to make it applicable for the other databases.
There will be no language restriction in the selection of papers.

Data collection and analysis

Selection of studies
Using titles and after reading the abstracts we will select the studies fulfilling our inclusion criteria. Once all potentially appropriate studies will be obtained, each study will be evaluated independently by two reviewers for inclusion. If no consensus is met about the possible inclusion/exclusion of any individual study, a final consensus decision will be made by discussion amongst all the reviewers. Should the reviewers not come to consensus regarding the inclusion/exclusion of the study, the full article will be submitted to the editorial base for arbitration. Reviewers will not be masked to the name(s) of the author(s), institution(s) or publication source at any level of review.

Assessment of methodological quality
The variety in study designs to be included in this systematic review necessitates the use of specific quality assessment tools. The methodological quality of RCTs and CCTs will be rated by a list recommended by van Tulder (van Tulder 1997). The list considers 11 criteria for internal validity, 6 descriptive criteria and 2 statistical criteria (Appendix 1). All criteria are scored as yes, no, or unclear. Equal weight will be applied to all items. Studies are considered to be of 'high quality' if at least 6 criteria for internal validity, 3 descriptive criteria and one statistical criterion are scored positively.
The methodological quality of ODs will be rated using an adapted version of the Van Tulder list made by Steultjens et al. (Steultjens 2003). Some items (concerning randomisation, similarity of patient groups, masking of patient) are considered inapplicable to ODs and are removed from the list. The outcomes assessor could also be masked in ODs, but we will request, at minimum, that the assessor would not be involved in the intervention. The final list of criteria used for ODs consists of 7 criteria for internal validity, 4 descriptive criteria and 2 statistical criteria (Appendix 1). All criteria are scored as yes, no, or unclear. Equal weight will be applied to all items. Studies are considered to be of 'sufficient quality' if at least 4 out of 7 criteria for internal validity, 2 descriptive criteria and one statistical criterion are scored positively. Of course the internal validity of ODs is weaker than the internal validity of RCTs/CCTs.
The methodological quality of the included trials will be independently assessed by two reviewers. Disagreements will be resolved by discussion. If no consensus is met a third reviewer will decide.

Appendix 1: Criteria of methodological quality for RCTs, CCTs, ODs

• Patient selection

a) Were the eligibility criteria specified?
b) Treatment allocation:1) was a method of randomisation performed?*
2) was the treatment allocation concealed?*
c) Were the groups similar at baseline regarding the most important prognostic indicators?*
*not applied to ODs

• Interventions

d) Were the index and control interventions explicitly described?
e) Was the care provider masked for the intervention?*
f) Were co‐interventions avoided or comparable?
g) Was the compliance acceptable in all groups?
h) Was the patient masked to the intervention?*
*not applied to ODs

• Outcome measurement

i) Was the outcome assessor masked to ("not involved in" for ODs) the interventions?
j) Were the outcome measures relevant?
k) Were adverse effects described?
l) Was the withdrawal/drop out rate described and acceptable?
m) Timing follow‐up measurements: 1) was a short‐term follow‐up measurement performed?
2) was a long‐term follow‐up measurement performed?
n) Was the timing of the outcome assessment in both groups ("all patients" for ODs) comparable?

• Statistics

o) Was the sample size for each group ("of the patient group" for ODs) described?
p) Did the analysis include an intention‐to‐treat analysis?*
q) Were point estimates and measures or variability presented for the primary outcome measures?
*not applied to ODs

Internal validity: b, e, f, g, h, i, j, l, n, p; descriptive criteria: a, c, d, k, m; statistical criteria: o, q.

Data extraction
Two reviewers will independently extract the data from each study meeting the inclusion criteria. In the event that insufficient data are available, the authors will be contacted to provide data and clarification. If the data are unavailable or insufficient, the study will be reported but not included in the final analysis. All studies meeting the inclusion criteria will be summarized in the 'Included Studies' table provided in the Review Manager software developed by the Cochrane Collaboration (RevMan 4.1) to include details on design, participants, interventions, and outcomes.

Data analysis
We expect to meet some difficulties in pooling the available data and studies because of the complex nature of dietary interventions. These may involve one nutrient at a time or entire classes of nutrients and study reductions or increases of the nutrients involved.
In our analysis we will pool studies in different dietary categories based on the rationale for the intervention (i.e. allergens hypothesis, low levels of essential fatty acids, antioxidant action etc.), as explained in the background and/or objective section of the full paper. The allocation of studies to the various categories will be validated by a dietician (MF) to ensure its consistence from the nutritional point of view. The allocation of studies that report on a combination of interventions will be decided by consensus among reviewers. If no consensus is reached, doubts about the nutritional or neurological aspects will be submitted to recognized experts in the field to reach a final decision.
Subsequently, the analysis will be performed separately for each intervention category.
Data will be processed according to intention to treat principle, using in the denominator the number of randomised patients. For dichotomous data, results from studies will be summarised as odds ratio (95% confidence intervals) according to the Peto and Mantel‐Haenszel methods. The latter will be used in the case of small and unbalanced trials.
Continuous data will be analysed using either a weighted mean difference where the same outcome measures are reported, or a standardised mean difference where different scales have been employed to measure similar outcomes.
An assessment for study heterogeneity will be made concerning the extent to which between‐study differences are apparent using both Chi squared test of heterogeneity and visual inspection of the data. If some heterogeneity is found but it is not considered to be a serious source of bias then a random‐effects model will be used. If no heterogeneity is found then a fixed‐effects model will be used for the purpose of data‐synthesis.
We might encounter too much diversity among studies with regard to patients (severity of the disease), interventions (duration, frequency and setting) and outcome measures (diversity, presentation of the results) to be able to perform an appropriate quantitative analysis (meta‐analysis). If so, we will perform a best evidence synthesis by rating the effectiveness of dietary intervention according to levels of evidence, taking into account the design of the studies, the methodological quality and the outcomes of the original studies. The best‐evidence synthesis is based upon the one proposed by van Tulder (van Tulder 2003) and adapted by Steultjens (Steultjens 2003).

Levels of evidence:

• Strong evidence

provided by consistent, statistically significant findings in outcome measures in at least two high quality RCTs

• Moderate evidence

provided by consistent, statistically significant findings in outcome measures in one high quality RCT and at least one low quality RCT or high quality CT

• Limited evidence

provided by statistically significant findings in outcome measures in one high quality RCT or:
provided by consistent, statistically significant findings in outcome measures in at least two high quality CTs ( in the absence of high quality RCTs)

• Indicative findings

provided by statistically significant findings in outcome and/or process measures in one high quality CT or low quality RCTs (in the absence of high quality RCTs) or:
provided by consistent, statistically significant findings in outcome and/or process measures in at least two high quality ODs (in the absence of RCTs and CTs)

• No evidence

in case of results of eligible studies do not meet the criteria for one of the above levels of evidence.