Data collection and analysis

The full text articles of the retrieved trials will be reviewed by the two reviewers and the inclusion criteria applied independently. Any differences in opinion about which studies to include in the review will be resolved by discussion.

Data from the studies will be independently extracted by the two authors using standardised forms. Data will be extracted so as to allow an intention to treat analysis. Where data are missing, the reviewers will write to the authors of the study requesting further information.

The quality of all included trials will be assessed independently by the two reviewers, and any differences in opinion will be resolved by discussion. A modification of the method used by Chalmers 1990 will be used. The selected studies will be assessed for the following characteristics:

1. the adequacy of the randomisation process; 2. the potential for selection bias after allocation to study group, i.e. losses to follow up and whether analysis was by intention‐to‐treat; 3. whether there was blinding of outcome assessors to the participants' study group; 4. quality of outcome assessment.

Studies will be graded A, B or C for their overall methodological quality:

A: minimisation of bias in all four categories above, i.e. adequate randomisation; few losses to follow up and intention‐to‐treat analysis; blinding of outcome assessors; high quality outcome assessment; B: each of the criteria in A partially met; C: one or more of the criteria in A not met.

We intend to use study quality for sensitivity analysis.

We plan to analyse data by intention‐to‐treat (i.e. with participants in the groups to which they were originally randomly assigned). Data from adults and children will be analysed separately. If data are comparable and of sufficient quality, they will be combined to give a summary measure of effect, otherwise they will not be combined. However, extreme heterogeneity may be examined by subgroup analysis if deemed appropriate.