1. Additional methods for future updates.

Issue	Method
Data extraction and management	It is planned to use paper data collection forms specially devised for use in this review to include information regarding study location, study quality, methods, participant characteristics at baseline (such as age, degree of disability and comorbidities), recruitment period, intervention, length of follow‐up, loss to follow‐up and outcome measures. The form will be piloted using one or two studies. Data extraction will be performed by two independent review authors for each study and will be scrutinised for disparity (Higgins 2008, section 7.2). Any differences will be resolved by discussion, contacting the authors of the study or independent arbitration as appropriate. Data will be entered into Review Manager software by one review author and checked by another.
Assessment of risk of bias	Included studies will be independently evaluated for risk of bias by two review authors, each of whom will independently assess each study at as high, low or unclear risk of bias using the categories and guidance of the Cochrane Handbook for Systematic Reviews (Higgins 2008). Allocation concealment Low risk of bias indicates that the report gives a clear description of the method used for random allocation and i) the method was adequate to prevent both the person assessing eligibility for trial entry and the participant from knowing what the allocation would be (for example, through allocation by a central office unaware of the subject characteristics, or use of sequentially numbered, sealed, opaque envelopes); ii) the method was such that after assignment the allocation could not be altered and the decision about eligibility could not be changed. Unclear risk of bias indicates uncertainty about whether allocation was adequately concealed, for example the description of the method for allocation was not clearly described in the study report. If the concealment method is unclear, the authors of the study will be contacted to obtain precise information where possible. High risk of bias indicates that the description of the method for allocation was clearly described but the method was inadequate to guarantee allocation concealment (for example, open random number lists, or quasi‐randomisation such as alternate days, odd/even days of birth or hospital numbers). Blinding We do not anticipate that blinding to the study intervention for participants (child and family and physician) will ever be possible for this intervention, and blinding is difficult even for the outcome assessor. This is because the intervention is a surgical procedure and the stoma through which the feeds are given and either jejunostomy tube, gastrostomy tube or 'button' (a device at skin level through which feeds are administered) cannot easily be concealed during physical assessment such as weighing. It also involves giving different types of food to the children in the intervention and control groups. The main source of nutrition for the intervention group would usually be commercially prepared 'feeds' whereas the control group would usually be eating ordinary food (even if pureed or commercially prepared baby foods). Thus it would be very difficult for many of the outcomes to guard against performance and detection bias. Use of reliable, valid assessment scales where possible would thus be particularly important (see above under outcome measures). Study quality for this review must therefore be assessed primarily on concealment allocation and analysis using an 'intention to treat' basis. If authors report attempts at assessment blinding this will be discussed. Selective outcome reporting We will locate the protocols of any included studies to assess whether all outcomes measured have been reported on and the plan for analysis has been followed. Incomplete outcome data Studies should be able to account for all participants at follow‐up. If not clearly reported, an attempt will be made to contact the authors for further information. We will establish whether participants were analysed in the groups to which they were randomised, that is, on an intention‐to‐treat basis and on percentage loss to follow‐up.
Measures of treatment effect	Where included studies have measured similar outcomes, we plan to conduct a meta‐analysis. For continuous data, and where studies have used the same measure for the outcome, the mean difference and its 95% confidence interval (CI) will be calculated. Where a similar outcome has been measured using different instruments, the standardised mean difference with its 95% CI will be calculated, provided it is considered that combining these results makes clinical sense. For dichotomous data, the odds ratio and its 95% CI will be calculated.
Dealing with missing data	We will analyse data (where practical) on an intention‐to‐treat basis. Where insufficient data are reported, trialists will be contacted for further information.
Assessment of heterogeneity	Consistency of results will be assessed visually and by examining I² (Higgins 2002), a quantity which describes approximately the proportion of variation in point estimates that is due to heterogeneity rather than sampling error. This will be supplemented with a test of homogeneity, to determine the strength of evidence that the heterogeneity is genuine. If heterogeneity is suggested by a marked difference of effect shown on the plotted results or if there are differences in the method, population, intervention or outcomes chosen that suggest important heterogeneity, a random‐effects model will be used in addition to a fixed‐effect model. The possible reasons for heterogeneity will be explored by scrutinising the studies and, where appropriate, performing subgroup analyses.
Assessment of reporting biases	Should sufficient studies be identified in future, funnel plots will be drawn to investigate any relationship between effect size and study precision (closely related to sample size). Such a relationship could be due to publication bias, but may also be due to poor methodological quality of smaller studies or other systematic differences between small and large studies or may occur by chance (Egger 1997). If a relationship is found, clinical diversity of the studies will be further examined as a possible explanation.
Data synthesis	Assuming two or more studies that are suitable for inclusion are found in future, and assuming also that the study results are similar enough that they can be sensibly grouped together, a meta‐analysis will be performed on the results. Both fixed‐effect and a random‐effects analyses will be performed as part of a sensitivity analysis.
Subgroup analysis and investigation of heterogeneity	Differences that might influence the effectiveness of gastrostomy feeding compared to oral feeding and that we will explore are: *age *presence of symptomatic gastroesophageal reflux or anti‐reflux procedure *level of disability (including learning disability) *environmental factors
Sensitivity analysis	Sensitivity analysis will be conducted to determine the impact of risk of bias on outcome if studies of different quality are identified and included (for example, studies whose allocation concealment remains unclear after trying to contact the authors).