Types of studies

We included studies with no language restrictions as follows:

1. RCTs using adequate methods of randomisation;

2. double‐blinded trials in which both participants and treating personnel or outcome assessors were blinded to the treatment;

3. placebo‐ or active‐controlled trials;

4. parallel‐group or cross‐over trials; for cross‐over studies, we planned to use the first treatment period as a parallel trial.

Types of participants

People of any age with drug‐resistant (defined in this review as uncontrolled seizures despite treatment with one or more antiepileptic drugs) focal epilepsy, including: simple focal seizures, complex focal seizures, or secondary generalised seizures.

Types of interventions

1. The treatment groups received carisbamate in addition to one or more existing antiepileptic drugs.

2. The control group received placebo or another antiepileptic agent in addition to one or more existing antiepileptic drugs.

Types of outcome measures

Electronic searches

We searched the following databases on 8 April 2021:

1. Cochrane Register of Studies (CRS Web) using the search strategy shown in Appendix 1;

2. MEDLINE (Ovid 1946 to April 07, 2021) using the search strategy shown in Appendix 2.

CRS Web includes randomised or quasi‐randomised controlled trials from PubMed, Embase, US National Institutes of Health Ongoing Trials Register ClinicalTrials.gov, the World Health Organization International Clinical Trials Registry Platform (WHO ICTRP), the Cochrane Central Register of Controlled Trials (CENTRAL), and the specialised registers of Cochrane review groups including Epilepsy. We did not impose any language restrictions.

Searching other resources

We checked the reference lists of included studies and review articles to identify additional reports of relevant studies. We contacted the manufacturers of carisbamate and the original investigators of relevant studies to obtain additional published or unpublished data. We also searched Chinese Clinical Trial Register (www.chictr.org/cn/proj/search.aspx) for ongoing trials.

Selection of studies

Two review authors (CL and YC until 2019, RB and KMM thereafter) independently screened the titles and abstracts of the articles identified by the search. During the initial title and abstract screening phase, the review authors firstly eliminated obviously ineligible reports. The review authors then retrieved the full texts of any potentially relevant reports and independently evaluated them for inclusion in the review. Any disagreements were resolved by discussion or by consulting a third review author (JZ) if necessary.

Data extraction and management

We did not undertake an individual patient data review. Instead, two review authors (CL and YC) independently extracted the following information and aggregate data from the included trials. Any disagreements were resolved by discussion.

1. Methodological/trial design:

   1. method of randomisation and allocation concealment;

   2. method of double‐blinding;

   3. duration of baseline period;

   4. duration of treatment period;

   5. dose(s) of carisbamate tested;

   6. description of treatment withdrawals.

2. Participant/demographic information:

   1. total number of participants allocated to each treatment group;

   2. age/sex;

   3. number with focal/generalised epilepsy;

   4. ethnicity;

   5. seizure types;

   6. number of background drugs.

3. Interventions:

   1. dosage;

   2. administration method.

4. Outcomes:

   1. number of participants experiencing each outcome per randomised group (see Types of outcome measures).

We predict that an individual patient data analysis approach will not be appropriate for the subsequent review update unless more high‐certainty evidence is available at the time of conduct. The low‐certainty evidence available at present (see summary of findings Table 1) is unable to justify the time‐costs and expenses associated with conducting an individual patient data review (Tudur Smith 2016).

Assessment of risk of bias in included studies

Two review authors (CL and JZ) independently assessed the risk of bias of the included studies using Cochrane's tool for assessing risk of bias as recommended by the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). The tool considers the following seven specific parameters:

1. random sequence generation (selection bias);

2. allocation concealment (selection bias);

3. blinding of participants and personnel (performance bias);

4. blinding of outcome assessment (detection bias);

5. incomplete outcome data (attrition bias);

6. selective reporting (reporting bias);

7. other bias.

Any disagreements were resolved by discussion between the two review authors.

Measures of treatment effect

We managed data according to the intention‐to‐treat principle. For dichotomous data, we presented treatment effect measures as risk ratios (RRs) with 95% confidence intervals (CIs) using the Mantel‐Haenszel statistical method. All the outcomes listed in Types of outcome measures, except for quality of life, are dichotomous data. Had we obtained data for quality of life, a continuous data outcome, we would have calculated mean differences (MDs) with 95% CIs. We considered a P value of ≤ 0.05 to indicate statistical significance.

Unit of analysis issues

Two review authors (CL and JZ) dealt with any unit of analysis issues according to the guidance in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011).

1. For cross‐over trials, we only included data from the first period for meta‐analysis.

2. For multi‐arm trials:

   1. for multiple‐dose‐group trials, we combined groups to create a single pair‐wise comparison;

   2. for multiple‐medication trials, we included one or more correlated comparisons and accounted for the correlation.

Dealing with missing data

There are many potential origins of missing data in a systematic review or meta‐analysis. It is important to investigate the sources of missing data. Data may be missing at random (unrelated to actual values of the missing data) or not at random (related to the actual missing data). We attempted to contact the authors of included studies and manufacturers for any missing data. We planned to make explicit assumptions of any methods used to cope with missing data (e.g. the data were assumed to be missing at random or to have a particular value, such as a poor outcome). When data were assumed to be missing at random, we only analysed the available data. For data judged to not be missing at random, we conducted sensitivity analyses to assess how sensitive results are to reasonable changes in the potential impact of missing data.

Assessment of heterogeneity

We visually assessed clinical heterogeneity by comparing the characteristics of the participants and interventions, and methodological heterogeneity by comparing methodological factors (such as study designs, concealment of allocation, blinding, etc.) between studies that met our inclusion criteria. We planned to assess statistical heterogeneity using the I² statistic. We assessed the percentage ranges of I² statistic as follows (Higgins 2011):

1. 0% to 40%: may not be important;

2. 30% to 60%: represents moderate heterogeneity;

3. 50% to 90%: represents substantial heterogeneity;

4. 75% to 100%: represents considerable heterogeneity.

Visual inspection of the forest plots also helped us to assess whether or not heterogeneity was present.

Assessment of reporting biases

Had we identified more than 10 studies, we would have visually inspected funnel plots for asymmetry. We would have investigated reasons for asymmetry (if any) including publication bias, outcome reporting bias, language bias, citation bias, poor methodological design, and heterogeneity.

Data synthesis

We analysed the data using Cochrane's statistical software, Review Manager 5 (Review Manager 2014). We used a fixed‐effect model where the I² statistic indicated no important or moderate heterogeneity (see Assessment of heterogeneity). Where the I² statistic indicated substantial or considerable heterogeneity, we employed a random‐effects model. We then explored factors that could have produced this heterogeneity and made a determination as to whether to conduct a subgroup analysis.

Subgroup analysis and investigation of heterogeneity

We planned the following subgroup analyses, where possible:

1. different ages of participants: children (less than 16 years old) versus adults (16 years old and over);

2. different doses of carisbamate: high dose (over 800 mg/d) versus intermediate (400 mg/d to 800 mg/d) or low dose (less than 400 mg/d);

3. different interventions in control groups: actively controlled or placebo‐controlled studies;

4. different ethnicity of the participants: white, black, and Asian;

5. different duration of the intervention: short term (less than 12 months) and long term (12 months or longer).

Sensitivity analysis

We planned to carry out a sensitivity analysis to test the robustness of the evidence by repeating meta‐analysis as follows:

1. exclude trials at high risk of bias;

2. exclude studies that were available as abstracts only.

Summary of findings and assessment of the certainty of the evidence

We used the GRADE approach to interpret findings (Schünemann 2011). We used GRADEpro GDT software (GRADEpro GDT) to import data from Review Manager 5 (Review Manager 2014) and to create a 'Summary of findings' table for the main comparison of the review (carisbamate compared to placebo for drug‐resistant focal epilepsy). We GRADE assessed the primary outcome, 50% or greater reduction in seizure frequency, and the secondary outcomes of seizure freedom, treatment withdrawal, and adverse events, to provide an overall certainty of evidence judgement, which was then included in the summary of findings Table 1. This information is of importance for healthcare decision making and considers eight important criteria (risk of bias, inconsistency, indirectness, imprecision, publication bias, effect size, presence of plausible confounding that will change effect, and dose‐response gradient). We used these overall certainty of evidence judgements to guide our conclusions.