Data extraction and management  

The following data will be extracted from the included studies:

1. Methodology (study design, blinding, allocation concealment).

2. Participant characteristics (total number, age, type of specific learning disorder, comorbidity).

3. Intervention:

   1. dose and combination of PUFAs;

   2. duration of intervention.

4. Outcomes:

   1. results of reading, writing, spelling and mathematics test;

   2. reported adverse effects.

Two review authors (ML and KH) will independently enter all data onto a specially designed data collection form.

Assessment of risk of bias in included studies

We will use the Cochrane Collaboration tool for assessing the risk of bias in the included studies (Higgins 2008, chapter 8.5). Two review authors will independently assess each study for risk of bias in sequence generation, allocation concealment, blinding of participants, personnel and outcome assessors, incomplete outcome data, selective outcome reporting and other sources of bias. For each of these risk of bias elements, each review author rated them as low, unclear or high risk of bias.

Sequence generation

We will rate the studies as low risk of bias if the method of sequence generation used a random method of allocating the participants such as random number generator or table, toss of coin, roll of dice or drawing lots.

Allocation concealment

We will look for adequate allocation concealment. We will rate studies as low risk of bias if the allocation to intervention is concealed and unpredicted by participants or investigator such as central allocation, sequentially numbered identical drug container or sequentially numbered, sealed, opaque envelopes.

Blinding

All participants, investigators and assessors can be blinded in potential studies for this review. In addition to the pills or capsules of PUFAs being identical in appearance, the smell and taste should be similar as well as PUFAs derived from fish sources have a distinct taste and smell. We will rate the blinding of participants and outcomes assessors for each main outcome, namely results of the standardised tests and results of parent or teacher‐reported outcomes. For each, we will rate them as as low risk of bias if blinding was described adequately and knowledge of the allocated intervention was prevented throughout the study.

Incomplete outcome data

For each outcome of an included study, we will look for a report of attrition and exclusion, including reasons, for each intervention group. If the study addresses the incomplete outcome data adequately, including all intervention groups having similar reasons for the missing data (and not related to the true outcome) and balanced numbers in all groups, then it will be rated as low risk of bias.

Selective outcome reporting

We will check for selective outcome reporting by determining if all prespecified outcomes from the study's protocol were reported or, if the study protocol was not available, that all expected outcomes from the study were reported. If the protocol is available, the study will be rated as low risk if all pre‐specified outcomes were reported.

Other sources of bias

We will also look for other sources of potential bias in the studies, such as inappropriate design, premature stopping of the study, extreme baseline imbalance or suspicion of fraud.

Measures of treatment effect  

Treatment effect will be measured by using standardised reading, writing, spelling or mathematical tests, which are continuous data. If there are various tests used to measure each learning ability (reading, writing, spelling or mathematics) in the studies, we will calculate the standardised mean difference (SMD) and 95% confidence interval for each.

Where dichotomous data are presented, we will use the risk ratio (RR) with 95% confidence interval.

Unit of analysis issues

Cluster‐randomised study

We do not anticipate any cluster‐randomised trials in this review. However, if we do find them, they will be analysed in the unit to which they were randomised and the effective sample size adjusted using an intra‐cluster correlation coefficient (ICC), which we will obtain from the studies themselves, if given, or we will estimate from available published literature.

Cross‐over study

In the analysis of studies with cross‐over design, we will first consider if there will be carry‐over effect, whether only first period data are available, any incorrect analysis and if results are comparable to parallel‐group studies. We will use the methods suggested in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2008 chapter 16.4) when analysing and incorporating the trials for meta‐analysis. If suitable data are available, we will perform a paired analysis and include it in a meta‐analysis using the generic inverse‐variance method in Review Manager 5 (RevMan 2011).

Studies with multiple intervention groups

For studies with more than two intervention groups, we will combine the interventions if they are similar (for example, difference dose of PUFAs or different combination of PUFAs) and control groups to create a single pair‐wise comparison. If any of the intervention groups are completely different (for example visual exercises, music therapy) , then we will exclude the particular intervention group from analysis

Dealing with missing data

If we encounter data that appear incomplete, we will contact the original authors of the particular study to provide missing data such as unreported summary data and details of dropouts. We will assess studies selected for intention‐to‐treat (ITT) analysis and when necessary, make all efforts to obtain additional information that would allow analyses according to ITT principles. We will describe details of participant dropout for each of the selected studies in the 'Risk of bias' table. If we are unable to retrieve the missing data from a study and the missing data will have a significant impact on the final results, we will exclude that study from analysis.

Assessment of heterogeneity  

Variations across the selected studies may be a result of participant factors (for example, different specific learning disorders) or study factors (for example, study design, dropout rates, types and combination of PUFAs used). We will assess heterogeneity across the studies by visual inspection of the forest plots for overlapping confidence intervals. We will also assess for statistical heterogeneity by using Chi² test or I². The I²describes the percentage of the variability is due to heterogeneity rather than chance. A rough estimate to interpretation of I² is as follows (Higgins 2008, chapter 9.5.2):

I² = 0% to 40% might not be important;
I ² = 30% to 60% may represent moderate heterogeneity;

I ² = 50% to 90% may represent substantial heterogeneity;
I² = 75% to 100% represents considerable heterogeneity.

The threshold for interpreting the I² value can be misleading, therefore we will also determine the importance of the observed I² by looking at the magnitude and direction of the effect as well as the strength of evidence for clinical heterogeneity.

Assessment of reporting biases

If sufficient studies are available, we will use funnel plots to assess the relationship of the treatment effect size and standard error. Asymmetry in the funnel plot may be a result of reporting bias, but could also be a true reflection of heterogeneity in the intervention effects. We will examine the clinical heterogeneity of the studies further as a possible explanation. We will also compare results obtained from published studies and results from other sources (for example, correspondence) in order to detect publication bias.

Data synthesis

If two or more studies that are homogenous are found, we will perform a meta‐analysis using Review Manager 5.1 (RevMan 2011). Both fixed‐effect and random‐effects models may be used in the analysis to explore the impact of statistical heterogeneity and, if significant asymmetry is found in the funnel plot, we will present the results of both analyses. For studies that are clinically distinct such as different treatment combinations or different measurements of learning abilities, we will not combine the studies for meta‐analysis and instead will present a narrative description of the study results. The narrative description will include the general direction, size, consistency and strength of evidence of effect of each individual study but we will not attempt to compare the effects of each study or draw an overall conclusion.

Subgroup analysis and investigation of heterogeneity

We intend to perform subgroup analysis only for the primary outcome. The subgroups that we will investigate are as follows.

1. Different combinations of PUFAs, for example, using only DHA or DHA combined with EPA or ALA.

2. Type of specific learning disorder (for example, specific reading disorder compared to specific mathematical disorder).

3. Presence of comorbidity of ADHD or autism.

4. Age when supplementation is given (for example, preschool or early primary school compared to late primary school or secondary school).

5. Duration of intervention (< 3 months, 3‐6 months, > 6 months)

Sensitivity analysis

We will perform sensitivity analysis to assess how much the quality of the studies affects the meta‐analysis results. We will conduct sensitivity analyses to find out if excluding studies with high risk of bias (as assessed using the Cochrane Collaboration tool for assessing the risk of bias) has an impact on the final results. These will include studies with cross‐over design, inadequate sequence allocation and concealment, missing data or use of any imputed values.