Types of studies

We planned to include parallel‐group randomised controlled trials (RCTs).

Types of participants

Participants in the studies eligible for inclusion were children of either sex younger than five years of age with a first episode of complex febrile seizure.

We would include children in the review only:

• if the EEG was performed after the first complex febrile seizure episode; or

• if recruitment of children into the study was delayed and the EEG was performed before the child's second seizure.

We excluded studies in which children had other neurological disorders (e.g. behavioural disorders, cerebral palsy, mental retardation).

Types of interventions

The intervention was EEG investigation. We selected two comparisons.

• Children without administration of an EEG versus children with administration of an EEG (early or late or at any time).

• Children with administration of an early EEG versus children with administration of a late EEG.

Types of outcome measures

Electronic searches

We ran searches for the original review on 18 October 2012, and ran subsequent searches on 17 October 2013, 6 July 2015, and 23 January 2017. For the latest update, we searched the following databases on 12 March 2019 using the search strategies shown in Appendix 1.

• Cochrane Register of Studies (CRS Web), which includes the Cochrane Epilepsy Group Specialized Register and the Cochrane Central Register of Controlled Trials (CENTRAL).

• MEDLINE (Ovid, 1946 to 11 March 2019).

• US National Institutes of Health Ongoing Trials Register ClinicalTrials.gov.

Searching other resources

We searched the references of all studies retrieved in full to identify additional research papers. We contacted authors of relevant articles and experts in the field to ask about additional research papers and unpublished trials. We looked for conference proceedings for trials relevant to our review. We decided not to impose language restrictions.

Selection of studies

Two review authors (PBS (methodology expert), SJ (content expert)) conducted the searches and independently carried out an initial screening of the titles of research papers (Lefebvre 2011). The review authors independently screened the abstracts of papers deemed potentially relevant. We retrieved the full text for all articles whose abstracts were relevant. When full‐text articles were not available, we contacted the study authors and asked them to provide the full text of the article. We decided to use this search methodology during screening of the cross‐references of the full‐text retrieved articles and articles suggested by authors and experts in the field. Two review authors (PBS, SJ) independently reviewed the full‐text articles and classified each into one of the following groups: included studies; excluded studies; or studies pending decision (if required, we contacted study authors to request additional details of the study).

We recorded the information on the eligibility assessment form included in Appendix 2.

We attempted to identify duplicate publications by assessing similar study types, same place, or same authors (maybe with different sequence). We used reference management software to identify and exclude duplicate publications (Mozilla with Zotero). In cases of doubt, we contacted the study authors to avoid inclusion of duplicate publications in this review.

We agreed to resolve any disagreements by discussion and to reach a final decision by consensus or by consultation with the Cochrane Epilepsy Group. We used standard methodological procedures expected by Cochrane.

Data extraction and management

We could not extract the data in the present review, as no study met the inclusion criteria. We planned to conduct the data extraction process as follows.

Two review authors (PBS, SJ) planned to independently extract required data from the full‐text articles of the included studies. (We have included the data extraction form in Appendix 3.) The data extraction form has the following five components.

• Identification of study.

• Characteristics of included studies with a brief description in tabular form of methods, participants, interventions and outcomes, and notes on specific issues (if any).

• 'Risk of bias' table, as per guidelines of the Cochrane Handbook for Systematic Reviews of Interventions (Altman 2011).

• Measurements of treatment effects extracted.

• Information pertaining to any discrepancy noted in records of the clinical trial registry.

We agreed to resolve any disagreements by discussion and to reach a final decision by consensus or by consultation with the Cochrane Epilepsy Group. We agreed to follow the guidelines of the Quality of Reporting of Meta‐Analyses (QUOROM) statement, and PRISMA (formerly QUOROM) (Moher 2009).

We have provided a list of excluded studies with reasons for exclusion in tabular form (see Characteristics of excluded studies table).

We planned to report in tabular form the details of any ongoing studies in terms of methods, interventions and outcomes, or trial registration numbers; however, we found no ongoing studies.

Assessment of risk of bias in included studies

We could not assess the risk of bias in the present review, as no study met the inclusion criteria.

We planned to report risk of bias as recommended in the Cochrane Handbook for Systematic Reviews of Interventions for 'Risk of bias' tables (Appendix 4) (Altman 2011). We planned to fill the description for each domain with a quote from the article or correspondence and then comment with judgement regarding particular bias (yes, no, or unclear). We would undertake a summary assessment of risk of bias for each outcome (across domains) within and across studies and make judgements about the risk of bias as low, unclear, or high. We planned to report a 'Risk of bias' graph and a 'Risk of bias' summary figure. Two review authors (PBS, SJ) intended to independently assess the risk of bias. Any disagreements would have been resolved by discussion and through consensus or by consultation with the Cochrane Epilepsy Group.

Measures of treatment effect

We could not measure the treatment effect in the present review, as no study met the inclusion criteria. We planned to measure treatment effect as follows.

We planned to calculate the proportions of children developing seizures after follow‐up periods of one month, six months, 12 months, and two years. We also planned to calculate risk of recurrence between no EEG and EEG (early EEG, late EEG, or any other time), total number of seizure episodes in each group during the two‐year follow‐up, time to development of seizures as time‐to‐event outcome, and adverse events (if any). We planned to use odds ratio for dichotomous outcomes and mean difference for continuous data. For data provided in other forms, we decided to convert extracted data to the above measures. We planned to extract or calculate 95% confidence intervals for all data. In the case of insufficient information, we would contact study authors to ask for additional details.

Unit of analysis issues

We did not anticipate any unit of analysis issues.

Dealing with missing data

We contacted study authors to ask for missing data such as method of randomisation if not stated in the paper or whether an outcome of interest that was not reported had been analysed. The reasons for missing data would have been useful in the imputation of missing data.

Assessment of heterogeneity

We could not assess the heterogeneity in the present review, as no study met the inclusion criteria. We planned to assess heterogeneity as follows.

We planned to assess clinical heterogeneity by comparing participant factors, interventional factors, outcome factors, and methodological heterogeneity by study methods. We also planned to assess statistical heterogeneity by measuring variability in interventional effects by visually comparing the overlap of confidence intervals on forest plots. If confidence intervals for the results of individual studies have poor overlap, this might indicate the presence of statistical heterogeneity. We also planned to measure statistical heterogeneity using the I² statistic, interpreting it as follows (Deeks 2011).

• 0% to 40%: might not be important.

• 30% to 60%: may represent moderate heterogeneity.

• 50% to 90%: may represent substantial heterogeneity.

• 75% to 100%: considerable heterogeneity.

Assessment of reporting biases

We planned during this stage to determine primary and secondary outcomes of the included studies from clinical trial registries (if possible) to assess any discrepancies in reporting, which we would have reported had we found any. We planned to assess funnel plot asymmetry if we identified more than 10 studies. Reasons for asymmetry include publication bias, outcome reporting bias, and heterogeneity.

Data synthesis

We did not attempt data synthesis in the present review, as no study met the inclusion criteria. We planned to perform data synthesis as follows.

We planned to analyse each trial and record data on the data extraction form. The review authors would independently enter the results of each study using Review Manager 5 (RevMan 2014). We planned to carry out a meta‐analysis according to Cochrane guidelines (Higgins 2011); the most common method available using Review Manager 5 is the Mantel‐Haenszel method. If RCTs were clustered, we would have used the inverse method. In case of heterogeneity, we planned to carry out a thorough assessment (see Assessment of heterogeneity).

Subgroup analysis and investigation of heterogeneity

We did not attempt subgroup analysis and investigation of heterogeneity in the present review, as no study met the inclusion criteria. We agreed upon the following process.

We planned to carry out subgroup analyses based on sex, duration of disease, duration of hospitalisation, and length of follow‐up. In cases of significant heterogeneity, we might have followed the steps outlined below.

• Recheck extraction and recording of data.

• Change from random‐effects to fixed‐effect model.

• Perform sensitivity analysis (see Sensitivity analysis).

• Explore heterogeneity by subgroup analysis.

• Present systematic review without meta‐analysis.

Sensitivity analysis

Sensitivity analysis is a repeat primary analysis in which alternative decisions and ranges of values are substituted for decision making related to assessment of the robustness of conclusions.

We did not attempt sensitivity analysis in the present review, as no study met the inclusion criteria.

We planned to carry out the following steps in a sensitivity analysis.

• Some studies have larger effects than others because random error means that multiple replications of the same study will produce different effect estimates due to sampling variation, even if replications would yield the right answer on average. The results of smaller studies are subject to greater sampling variation and hence are less precise. Imprecision is reflected in the confidence interval around the intervention effect estimate from each study and in the weight given to the results of each study in a meta‐analysis. More precise results are given more weight.

• In cases of missing values, use of the following imputation methods, of which many methods for imputation techniques have been proposed, is suggested. In cases of dichotomous data, best case‐worst case analysis is performed to find out how the risk factor or the result value may vary in different situations. For continuous data, the last value takes the role of the missing value. These imputations are important because RCTs have to be analysed as intention‐to‐treat analyses.

• Use random‐effects rather than fixed‐effect model.

While the above sensitivity analyses were prespecified, it was not possible to specify all sensitivity analyses, as issues may develop only in the course of completing the review. Where sensitivity analysis influenced the robustness of our conclusion, we planned to attempt to resolve the uncertainty by contacting trial authors and consulting the Cochrane Epilepsy Group. As it might not have been possible to report all sensitivity analyses in detail, we planned to provide a summary table. Sensitivity analysis would have helped us to explore the influence of various factors.