Antiepileptic drug monotherapy for epilepsy: a network meta‐analysis of individual participant data

Sarah J Nevitt; Maria Sudell; Jennifer Weston; Catrin Tudur Smith; Anthony G Marson

doi:10.1002/14651858.CD011412.pub3

. 2017 Dec 15;2017(12):CD011412. doi: 10.1002/14651858.CD011412.pub3

Antiepileptic drug monotherapy for epilepsy: a network meta‐analysis of individual participant data

Sarah J Nevitt ^1,^✉, Maria Sudell ¹, Jennifer Weston ², Catrin Tudur Smith ¹, Anthony G Marson ²

Editor: Cochrane Epilepsy Group

PMCID: PMC6486134 PMID: 29243813

Abstract

Background

Epilepsy is a common neurological condition with a worldwide prevalence of around 1%. Approximately 60% to 70% of people with epilepsy will achieve a longer‐term remission from seizures, and most achieve that remission shortly after starting antiepileptic drug treatment. Most people with epilepsy are treated with a single antiepileptic drug (monotherapy) and current guidelines from the National Institute for Health and Care Excellence (NICE) in the United Kingdom for adults and children recommend carbamazepine or lamotrigine as first‐line treatment for partial onset seizures and sodium valproate for generalised onset seizures; however a range of other antiepileptic drug (AED) treatments are available, and evidence is needed regarding their comparative effectiveness in order to inform treatment choices.

Objectives

To compare the time to withdrawal of allocated treatment, remission and first seizure of 10 AEDs (carbamazepine, phenytoin, sodium valproate, phenobarbitone, oxcarbazepine, lamotrigine, gabapentin, topiramate, levetiracetam, zonisamide) currently used as monotherapy in children and adults with partial onset seizures (simple partial, complex partial or secondary generalised) or generalised tonic‐clonic seizures with or without other generalised seizure types (absence, myoclonus).

Search methods

We searched the following databases: Cochrane Epilepsy's Specialised Register, CENTRAL, MEDLINE and SCOPUS, and two clinical trials registers. We handsearched relevant journals and contacted pharmaceutical companies, original trial investigators, and experts in the field. The date of the most recent search was 27 July 2016.

Selection criteria

We included randomised controlled trials of a monotherapy design in adults or children with partial onset seizures or generalised onset tonic‐clonic seizures (with or without other generalised seizure types).

Data collection and analysis

This was an individual participant data (IPD) review and network meta‐analysis. Our primary outcome was 'time to withdrawal of allocated treatment', and our secondary outcomes were 'time to achieve 12‐month remission', 'time to achieve six‐month remission', 'time to first seizure post‐randomisation', and 'occurrence of adverse events'. We presented all time‐to‐event outcomes as Cox proportional hazard ratios (HRs) with 95% confidence intervals (CIs). We performed pairwise meta‐analysis of head‐to‐head comparisons between drugs within trials to obtain 'direct' treatment effect estimates and we performed frequentist network meta‐analysis to combine direct evidence with indirect evidence across the treatment network of 10 drugs. We investigated inconsistency between direct estimates and network meta‐analysis via node splitting. Due to variability in methods and detail of reporting adverse events, we have not performed an analysis. We have provided a narrative summary of the most commonly reported adverse events.

Main results

IPD was provided for at least one outcome of this review for 12,391 out of a total of 17,961 eligible participants (69% of total data) from 36 out of the 77 eligible trials (47% of total trials). We could not include IPD from the remaining 41 trials in analysis for a variety of reasons, such as being unable to contact an author or sponsor to request data, data being lost or no longer available, cost and resources required to prepare data being prohibitive, or local authority or country‐specific restrictions.

We were able to calculate direct treatment effect estimates for between half and two thirds of comparisons across the outcomes of the review, however for many of the comparisons, data were contributed by only a single trial or by a small number of participants, so confidence intervals of estimates were wide.

Network meta‐analysis showed that for the primary outcome ‘Time to withdrawal of allocated treatment,’ for individuals with partial seizures; levetiracetam performed (statistically) significantly better than current first‐line treatment carbamazepine and other current first‐line treatment lamotrigine performed better than all other treatments (aside from levetiracetam); carbamazepine performed significantly better than gabapentin and phenobarbitone (high‐quality evidence). For individuals with generalised onset seizures, first‐line treatment sodium valproate performed significantly better than carbamazepine, topiramate and phenobarbitone (moderate‐ to high‐quality evidence). Furthermore, for both partial and generalised onset seizures, the earliest licenced treatment, phenobarbitone seems to perform worse than all other treatments (moderate‐ to high‐quality evidence).

Network meta‐analysis also showed that for secondary outcomes ‘Time to 12‐month remission of seizures’ and ‘Time to six‐month remission of seizures,’ few notable differences were shown for either partial or generalised seizure types (moderate‐ to high‐quality evidence). For secondary outcome ‘Time to first seizure,’ for individuals with partial seizures; phenobarbitone performed significantly better than both current first‐line treatments carbamazepine and lamotrigine; carbamazepine performed significantly better than sodium valproate, gabapentin and lamotrigine. Phenytoin also performed significantly better than lamotrigine (high‐quality evidence). In general, the earliest licenced treatments (phenytoin and phenobarbitone) performed better than the other treatments for both seizure types (moderate‐ to high‐quality evidence).

Generally, direct evidence and network meta‐analysis estimates (direct plus indirect evidence) were numerically similar and consistent with confidence intervals of effect sizes overlapping.

The most commonly reported adverse events across all drugs were drowsiness/fatigue, headache or migraine, gastrointestinal disturbances, dizziness/faintness and rash or skin disorders.

Authors' conclusions

Overall, the high‐quality evidence provided by this review supports current guidance (e.g. NICE) that carbamazepine and lamotrigine are suitable first‐line treatments for individuals with partial onset seizures and also demonstrates that levetiracetam may be a suitable alternative. High‐quality evidence from this review also supports the use of sodium valproate as the first‐line treatment for individuals with generalised tonic‐clonic seizures (with or without other generalised seizure types) and also demonstrates that lamotrigine and levetiracetam would be suitable alternatives to either of these first‐line treatments, particularly for those of childbearing potential, for whom sodium valproate may not be an appropriate treatment option due to teratogenicity.

Plain language summary

Antiepileptic drug monotherapy (single drug treatment) for epilepsy

Background

Epilepsy is a common neurological disorder in which abnormal electrical discharges from the brain cause recurrent seizures. We studied two types of epileptic seizures in this review: partial seizures that start in one area of the brain, and generalised onset tonic‐clonic seizures that start in both cerebral hemispheres simultaneously.

For around 70% of people with epilepsy seizures can be controlled, and for the majority, seizures are controlled with a single antiepileptic drug. Currently in the UK, National Institute for Health and Care Excellence (NICE) guidelines for adults and children recommend carbamazepine or lamotrigine as the first treatment options to try for individuals with newly diagnosed partial seizures and sodium valproate for individuals with newly diagnosed generalised tonic‐clonic seizures; however a range of other antiepileptic drug treatments are available.

The choice of the first antiepileptic drug for an individual with newly diagnosed seizures is of great importance and should be made taking into account high‐quality evidence of how effective the drugs are at controlling seizures and whether they are associated with side effects. It is also important that drugs appropriate for different seizure types are compared to each other.

Review methods

The antiepileptic drugs of interest to this review were carbamazepine, phenytoin, sodium valproate, phenobarbitone, oxcarbazepine, lamotrigine, gabapentin, topiramate, levetiracetam, zonisamide. In this review, we evaluated the evidence from 77 randomised controlled clinical trials comparing two or more of the drugs of interest based on how effective the drugs were at controlling seizures (i.e. whether people had recurrence of seizures or had long periods of freedom from seizures (remission)) and how tolerable any related side effects of the drugs were. We were able to combine data for 12,391 people from 36 of the 77 trials; for the remaining 5570 people from 41 trials, data were not available to use in this review.

We performed two types of analysis in this review; firstly we combined data available where pairs of drugs had been compared directly in clinical trials and secondly we performed an analysis to combine all information from the clinical trials across the 'network' of 10 drugs. This analysis allowed us to compare drugs in the network that had not previously been compared to each other in clinical trials.

Key results

Out of the 45 possible pairwise comparisons of the 10 drugs of interest in the review, data from clinical trials were available for just over half of these comparisons but for many only a single trial had made a comparison of the two drugs and the comparison did not include many people.

Our 'network' analysis showed that the oldest drugs in the network (phenobarbitone and phenytoin) were better options in terms of seizure control than the other drugs but that these older drugs were the worst in terms of long‐term retention (withdrawing from the treatment) compared to the newer drugs such as lamotrigine and levetiracetam.

The most commonly reported side effects across all drugs were drowsiness or fatigue, headache or migraine, gastrointestinal disturbances (stomach upsets), dizziness or faintness and rash or skin disorders.

Quality of the evidence

This review provides high‐quality evidence for individuals with partial seizures and moderate‐ to high‐quality evidence for individuals with generalised tonic‐clonic seizures, as less information is available for some of the drugs of interest for people with this seizure type.

Conclusions

The results of this review support the NICE guidelines that carbamazepine and lamotrigine are suitable first treatment options for individuals with partial onset seizures and also show that levetiracetam would also be a suitable treatment. Results of this review also support the use of sodium valproate as the first‐line treatment for individuals with generalised tonic‐clonic seizures and also show that lamotrigine and levetiracetam would be suitable alternatives, particularly for those who are pregnant or considering becoming pregnant, for whom sodium valproate may not be an appropriate treatment option.

How up‐to‐date is this review?

The review authors searched for studies that had been published up to 27 July 2016.

Summary of findings

Background

Description of the condition

Epilepsy is a common neurological condition in which recurrent, unprovoked seizures occur due to abnormal electrical discharges in the brain, with an estimated incidence of 33 to 57 per 100,000 person‐years worldwide (Annegers 1999; Hirtz 2007; MacDonald 2000; Olafsson 2005; Sander 1996), accounting for approximately 1% of the global burden of disease (WHO 1994). The lifetime risk of epilepsy onset is estimated to be 1300 to 4000 per 100,000 person years (Hauser 1993; Juul‐Jenson 1983), and the lifetime prevalence could be as large as 70 million people world‐wide (Ngugi 2010). It is believed that with effective drug treatment, up to 70% of individuals with active epilepsy have the potential to become seizure‐free and go into long‐term remission shortly after starting drug therapy (Cockerell 1995; Hauser 1993; Sander 2004), and that around 70% of individuals can achieve seizure freedom using a single AED (AED) in monotherapy (Cockerell 1995). The remaining 30% of individuals experience refractory or drug‐resistant seizures, which often require treatment with combinations of AEDs or alternative treatments such as epilepsy surgery (Kwan 2000).

Epilepsy is not a single condition, but is in fact a heterogeneous group of conditions ranging from those with a purely genetic cause to those that are symptomatic of a brain injury (e.g. stroke) or other abnormality (e.g. tumour). We also recognise a range of differing seizure types, and epilepsy syndromes that have been classified by the International League Against Epilespy (ILAE), a classification that continues to be revised as our understanding of the genetics and basic biology of epilepsy improves (Berg 2010; Commission 1981; Commission 1989)

The simplest dichotomy in epilepsy is between partial onset (or focal) and generalised onset seizures. Partial onset seizures originate in one part of the brain and include simple partial, complex partial and secondary generalised seizures (Berg 2010). Generalised seizures originate in both cerebral hemispheres simultaneously and include generalised tonic‐clonic seizures, absence seizures and myoclonic seizures. In this review we focus on this dichotomy rather than specific epilepsy syndromes.

Description of the intervention

For the treatment of partial and generalised onset seizures we included in our evidence base the following 10 AEDs, which at the time of publication of the protocol of this review (December 2014) were licensed and used in clinical practice for use as monotherapy in at least one country (eMC 2014; FDA 2014):

carbamazepine;
phenobarbitone;
phenytoin;
sodium valproate;
oxcarbazepine;
lamotrigine;
gabapentin;
topiramate;
levetiracetam;
zonisamide.

Carbamazepine, sodium valproate, phenytoin and phenobarbitone are among the earliest drugs licensed for treating epileptic seizures. Carbamazepine and sodium valproate have been commonly used as monotherapy for partial onset and generalised onset seizures for over 30 years (Shakir 1980), while phenytoin and phenobarbitone have been used in monotherapy for over 50 years (Gruber 1962).

These traditionally used drugs have all been recommended as first‐line treatments due to their effects across a range of seizure types, however they are also associated with a number of adverse effects. Phenytoin and phenobarbitone are no longer considered as first‐line agents in the USA and much of Europe due to worries over adverse events (Wallace 1997; Wilder 1995). Both drugs have been shown to be teratogenic (associated with malformations of an embryo or fetus) and are associated with low folic acid levels and megaloblastic anaemia (a blood disorder marked by the appearance of very large red blood cells (Carl 1992; Gladstone 1992; Meador 2008; Morrow 2006; Nulman 1997)). Phenytoin is particularly associated with fetal hydantoin syndrome, the name given to a group of birth defects associated with exposure to phenytoin (Scheinfeld 2003), and phenobarbitone has been associated with behavioural disturbances, particularly in children (de Silva 1996; Trimble 1988). These agents are however still used as first‐line drugs in low‐ to middle‐income countries (Ogunrin 2005; Pal 1998).

Carbamazepine and sodium valproate are also associated with congenital abnormalities (Canger 1999; Gladstone 1992; Morrow 2006; Nulman 1997; Tomson 2011). Systematic reviews have shown sodium valproate to have the highest incidence of congenital malformations of traditional first‐line AEDs (Meador 2008; Weston 2016), particularly spina bifida, as well as cardiac, craniofacial, skeletal and limb defects known as 'valproate syndrome' (Ornoy 2009). A recent study has shown an increased prevalence of neurodevelopmental disorders following prenatal sodium valproate exposure (Bromley 2013). A recently published Cochrane Review found that levetiracetam and lamotrigine exposure carried the lowest risk of overall congenital malformation, however information regarding specific malformations was lacking (Weston 2016).

In the last 20 years, a second‐generation of AEDs including oxcarbazepine, lamotrigine, gabapentin, topiramate and, most recently, levetiracetam and zonisamide, have been licensed as monotherapy following demonstrations of efficacy, or non‐inferiority within the European Union, compared to the traditional AEDs (for example, Baulac 2012; Bill 1997; Brodie 1995a; Brodie 1999; Brodie 2007; Chadwick 1998; Christe 1997; Dam 1989; Guerreiro 1997; SANAD A 2007, SANAD B 2007; Privitera 2003; Reunanen 1996; Rowan 2005; Steiner 1999; Trinka 2013). Comparative studies have also shown the newer AEDs to be generally well tolerated as monotherapy in both adults and children and related to fewer adverse events, fewer serious adverse events, fewer teratogenic effects and fewer drug interactions with concomitant AEDs and other concomitant medications than the traditional first‐line AEDs (French 2004; French 2007).

Current guidelines from the National Institute for Health and Care Excellence (NICE) for adults and children recommend carbamazepine or lamotrigine as first‐line treatment for partial onset seizures and sodium valproate for generalised onset seizures, on the condition that women and girls of childbearing age are made aware of the potential teratogenic effects of the drug (NICE 2012).

How the intervention might work

AEDs suppress seizures by reducing neuronal excitability, hence reducing the probability that a seizure will occur. Different AEDs have different mechanisms of action; therefore certain AEDs are more effective at treating different seizure types. For example, there are reports of efficacy for sodium valproate in generalised epilepsy syndromes such as juvenile myoclonic epilepsy and absence epilepsy (Bourgeois 1987; Delgado‐Escueta 1984; Grünewald 1993; Jeavons 1977; Penry 1989), while carbamazepine, on the other hand, is reported to exacerbate some generalised seizure types such as myoclonic and absence seizures (Liporace 1994; Shields 1983; Snead 1985).

The majority of traditional AEDs are thought to have multiple mechanisms of action such as blocking ion channels, binding with neurotransmitter receptors or inhibiting the metabolism or reuptake of neurotransmitters. However the precise mechanism of action is not known for all AEDs, particularly sodium valproate. It is thought that one of the mechanisms of action of phenytoin, sodium valproate, carbamazepine, oxcarbazepine and lamotrigine is via blocking of sodium channels (Brodie 1996; Faigle 1990; Granger 1995; Grant 1992; Lees 1993; McLean 1986; Pinder 1977; Ragsdale 1991; Willow 1985), while phenobarbitone binds with gamma‐aminobutyric acid (GABA) A receptors (Rho 1996).

Zonisamide is thought to have multiple mechanisms of action (Endoh 1994; Kawai 1994; Okada 1998; Sackellares 2004; Schauf 1987; Suzuki 1992; Zhu 1999), while the mechanism of actions of gabapentin and topiramate are not fully understood (Brodie 1996; Coulter 1993; Hill 1993; McClean 1995; McLean 1999; White 1997). Levetiracatam has a novel mode of action which is different from that of other AEDs (Cho 2011); it is thought to exhibit its antiepileptic effect by binding to synaptic vesicle protein 2A (encoded within the SV2A gene), influencing excitatory neurotransmitter release (Gillard 2006; Lynch 2004).

Why it is important to do this review

Given that up to 70% of individuals with a new epilepsy diagnosis enter a long‐term remission of seizures shortly after starting drug therapy (Cockerell 1995; Hauser 1993; Sander 2004), the correct choice of first‐line antiepileptic therapy for individuals with newly diagnosed seizures is of great importance. There are currently over 50 AEDs available worldwide for the treatment of all epilepsy syndromes (Epilepsy Foundation of America 2013), and therefore it is important that the choice of first AEDs is based on the highest‐quality evidence regarding potential benefits and harms of various treatments.

We have published a series of Cochrane systematic Reviews investigating pairwise monotherapy comparisons using individual participant data (Marson 2000; Nevitt 2016; Nolan 2013b; Nolan 2013c; Nolan 2015; Nolan 2016a; Nolan 2016b; Nolan 2016d). Each Cochrane Review and meta‐analysis provides high‐quality evidence for each pair of drugs but does not inform a choice among the range of drugs available. Furthermore, direct evidence from randomised controlled trials (RCTs) is not available for some drug comparisons such as between oxcarbazepine and phenobarbitone; therefore it is not possible to make pairwise comparisons of treatment effects between all 10 drugs included in this review. Also, pairwise comparisons between certain drugs are unlikely to be made in the future, such as comparisons with phenobarbitone, which is no longer considered to be a first‐line treatment, so it is unlikely that a RCT will be designed in the future to compare oxcarbazepine with phenobarbitone (Tudur Smith 2007). However, it is possible to estimate an indirect treatment effect size between oxcarbazepine and phenobarbitone using existing evidence comparing oxcarbazepine with phenytoin and phenytoin with phenobarbitone (Nolan 2013b; Nolan 2016d). By similar methodology, an indirect pairwise comparison is possible for all 10 drugs in our treatment network. Indirect comparisons are also valuable in the case that a limited amount of data are available to inform a direct comparison or in the case that evidence informing a direct comparison is of poor methodological quality. The power and precision of a treatment effect estimate can be increased by 'borrowing strength' from the indirect evidence in the network of treatments (Bucher 1997). Eight of the AEDs included in this review have been included in an IPD network meta‐analysis of epilepsy monotherapy drugs (Tudur Smith 2007). We wish to update the information in this network meta‐analysis with new evidence from trials published since 2007 and including evidence for two drugs, which were licensed for use as monotherapy after 2007.

As noted in the series of Cochrane Reviews investigating pairwise monotherapy comparisons, the important efficacy outcomes in epilepsy monotherapy trials often require analysis of time‐to‐event data (for example, time to first seizure after randomisation or time to withdrawal of allocated treatment). Although methods have been developed to synthesise time‐to‐event data using summary information (Parmar 1998; Williamson 2002), the appropriate statistics are not commonly reported in published epilepsy trials (Altman 1995; Nolan 2013a).

Furthermore, although seizure data have been collected in most epilepsy monotherapy trials, we have seen little uniformity in the definition and reporting of outcomes. For example, trials may report time to 12‐month remission but not time to first seizure or vice versa, or some trials may define time to first seizure from the date of randomisation but others use date of achieving maintenance dose. Trial investigators have also adopted differing approaches to the analysis, particularly with respect to the censoring of time‐to‐event data. For these reasons, we performed the pairwise meta‐analyses using IPD, which helps to overcome these problems and is considered to be the 'gold standard' approach to synthesis of censored data (Parmar 1998). We therefore also performed the network meta‐analysis of epilepsy monotherapy drugs as an IPD analysis.

Objectives

Methods

Criteria for considering studies for this review

Types of studies

We included RCTs using either:

an adequate method of allocation concealment (e.g. sealed, opaque envelopes);
a quasi method of randomisation (e.g. allocation by date of birth).

Trials may be double‐blind, single‐blind or unblinded. We included only trials of a monotherapy design; in other words, all participants are randomised to treatment with a single drug. We excluded trials with an add‐on (polytherapy), or withdrawal to monotherapy designs.

We included trials of parallel designs. We excluded trials of a cross‐over design, as this design is not appropriate for assessing treatment decisions at the time of epilepsy diagnosis and the cross‐over design is also inappropriate for measuring our primary time‐to‐event outcome 'time to withdrawal of allocated treatment', as a withdrawal of allocated treatment in the first treatment period would mean than the participant could not cross into the second treatment period, potentially leading to a large amount of incomplete outcome data and therefore a reduction in statistical power. Furthermore, the use of cross‐over designs is no longer recommended in epilepsy due to concerns over trial duration, large proportions of dropouts, unblinding of masked treatments as participants cross into the second period, and potential carryover effects; a particular concern in trials of a monotherapy design that aim to assess the effect of a single treatment (Engel 2008; Wyllie 2006).

Types of participants

Children or adults with partial onset seizures (simple partial, complex partial, or secondarily generalised tonic‐clonic seizures) or generalised onset tonic‐clonic seizures (with or without other generalised seizure types). We did not include participants with other generalised seizure types alone (for example absence seizures alone without generalised tonic‐clonic seizures) as guidelines for the first‐line treatment of other generalised seizure types are different from the guidelines for generalised tonic‐clonic seizures (NICE 2012), and due to documented evidence that certain drugs of interest in our review may exacerbate some generalised seizure types (How the interventions might work). We also considered individuals with a new diagnosis of epilepsy, or who had had a relapse following antiepileptic monotherapy withdrawal.

We excluded trials that considered AEDs as treatment for conditions other than epilepsy.

Types of interventions

We included the 10 AEDs currently licensed and commonly used as monotherapy in our network of treatments: carbamazepine, phenytoin, sodium valproate, phenobarbitone, oxcarbazepine, lamotrigine, gabapentin, topiramate, levetiracetam, zonisamide.

Included trials had to make at least one pairwise comparison between at least two of the 10 AEDs included in our network. For trials with three treatment arms or more, we included treatment arms only of the 10 AEDs included in our network; treatment arms of drugs not included in our network were excluded from analysis. We did not make pairwise comparisons (direct or indirect) between any AEDs not specified above. We made pairwise comparisons (based on direct or indirect evidence, or both) between all 10 drugs (Data synthesis).

We included trials with multiple arms of the same drug as long as at least one arm of another drug from our network was included (e.g. multiple doses of gabapentin compared to carbamazepine in Chadwick 1998). We pooled multiple dose arms of the same drug in our analysis; dose comparisons are outside the scope of this review.

Types of outcome measures

We investigated the following outcomes in this review (Primary outcomes; Secondary outcomes). Reporting of these outcomes in the original trial report was not an eligibility requirement for inclusion in this review.

Primary outcomes

Time to withdrawal of allocated treatment (retention time). This is a combined outcome reflecting both efficacy and tolerability, as treatment may be withdrawn due to continued seizures, adverse effects or a combination of both. This is an outcome to which the participant makes a contribution, and is the primary effectiveness outcome measure recommended by the Commission on Antiepileptic Drugs of the International League Against Epilepsy (Glauser 2006; ILAE 1998).

Secondary outcomes

Time to achieve 12‐month seizure‐free period (remission) after randomisation
Time to achieve six‐month seizure‐free period (remission) after randomisation
Time to first seizure post randomisation
Occurrence of adverse events (to be reported narratively) (Data synthesis)

Search methods for identification of studies

We searched the following databases with no language restrictions:

the Cochrane Epilepsy Specialised Register (26 July 2016) using the search strategy outlined in Appendix 1;
the Cochrane Central Register of Controlled Trials (CENTRAL; 2016, issue 7) via the Cochrane Register of Studies Online (CRSO, 26 July 2016) using the search strategy outlined in Appendix 2;
MEDLINE (Ovid, 1946 to 26 July 2016) using the search strategy outlined in Appendix 3;
SCOPUS (1823 to 09 September 2014) using the search strategy outlined in Appendix 4;
ClinicalTrials.gov searched on 26 July 2016) using the search strategy outlined in Appendix 5;
World Health Organization (WHO) International Clinical Trials Registry Platform (ICTRP) search portal searched on 26 July 2016), using the search strategy outlined in Appendix 6.

We had originally searched SCOPUS as an alternative to Embase, but this is no longer necessary, because randomised and quasi‐randomised controlled trials in Embase are now included in CENTRAL. We have not, therefore, updated the SCOPUS search.

We also reviewed reference lists of retrieved trials to search for additional reports of relevant trials, reviewed relevant conference proceedings and contacted experts in the field for details of any ongoing or unpublished trials.

Data collection and analysis

Selection of studies

One author (SJN) screened all titles and abstracts of all records identified by the electronic searches as described in Search methods for identification of reviews, according to the inclusion criteria specified above (Types of studies; Types of participants; Types of interventions). Subsequently, two authors (SJN and AGM) independently assessed full‐text publications according to the same inclusion criteria specified above. We resolved disagreements by discussion or by consulting a third author (CT) where necessary. We recorded the reasons for exclusion of trials at both stages of screening. We contacted trial authors for clarification if the eligibility of a trial was unclear from the published information.

Data extraction and management

Requesting individual participant data

For all trials meeting our inclusion criteria, two authors (SJN and AGM) sent a data‐request form to the first or corresponding author, or both, of the trial or to the trial sponsor where appropriate (referred to as data providers in this review).

Our data‐request form asked data providers if the following information was available (tick yes or no).

Trial methods:
- method of generation of random list;
- method of concealment of randomisation;
- stratification factors;
- blinding methods.
Participant covariates:
- sex;
- age;
- seizure types;
- epilepsy status (newly diagnosed/relapsed seizures following drug withdrawal);
- time between first seizure and randomisation;
- number of seizures prior to randomisation (with dates);
- presence of neurological signs;
- electroencephalography (EEG) results;
- computed tomography (CT) or magnetic resonance imaging (MRI) results;
- aetiology of seizures (if known).
Follow‐up data:
- treatment allocation;
- date of randomisation;
- dates of follow‐up;
- dates of seizures post randomisation or seizure frequency data between follow‐up visits;
- dates of treatment withdrawal and reason(s) for treatment withdrawal;
- starting dose of treatment;
- dates of dose changes;
- adverse events reported.

We also requested any available, related documents such as case report forms, trial protocols, clinical summaries etc. from data providers.

In the event of no response to our IPD request, we sent a follow‐up email to the original data provider contacted. If we still received no response for a particular trial, we attempted to contact another trial author or sponsor where appropriate. If a data provider was unable to make IPD available to us, we recorded the quoted reason why IPD could not be made available and we requested any aggregate data related to our outcome not reported in the publication.

If data could not be obtained (no response to any requests or IPD was not available), two independent authors (SJN and MS) assessed whether any relevant and appropriate aggregate level data was reported in the trial publication or could be indirectly estimated via the methods described in Parmar 1998 and Williamson 2002. We resolved any disagreements on extracted aggregate data by discussion or by consulting a third author (CT) if necessary.

Management of individual participant data

We stored all obtained data on a secure, dedicated network drive accessible only to the statisticians performing analysis (SJN, MS, CT). We checked all provided data for consistency and prepared them for analysis according to a pre‐specified procedure prepared by one author (SJN) (available on request) and piloted by two authors (SJN and MS). For each trial where IPD were supplied, we reproduced results from trial findings where possible and we performed the following consistency checks:

trial details cross‐checked against any published report of the trial; original trial authors to be contacted if missing data, errors or inconsistencies were found;
review of the chronological randomisation sequence by checking the balance of prognostic factors, taking account of factors stratified for in randomisation procedure.

We discussed any inconsistencies in the provided data with the corresponding data providers. If large or major inconsistencies were present, which could not be resolved by data providers, we did not include the data in any analyses. If minor inconsistencies were present, we analysed the data and conducted sensitivity analyses to test the robustness of results (Sensitivity analysis).

Following consistency checking and data cleaning, we prepared datasets for analysis and calculated outcomes for this review according to the methodology summarised below. We followed a 'standard operating procedure' for the data cleaning and preparation of data for analysis for all datasets to ensure a standardised and consistent approach to analysis throughout this review. Further details of this procedure can be obtained from the corresponding author on request.

Preparation of individual participant data for analysis

For the analysis of time to withdrawal of allocated treatment as a time‐to‐event outcome, we defined an 'event' as either the withdrawal of the allocated treatment due to poor seizure control or adverse events, or both. We also classed non‐compliance with the treatment regimen or the addition of another AED as 'events'. We censored the outcome if treatment was withdrawn because the individual achieved a period of remission, if a participant withdrew from allocated treatment for reasons not related to the treatment (such as loss to follow‐up) or if the individual was still on allocated treatment at the end of follow‐up. Two authors (SJN and AG) independently reviewed reasons for treatment withdrawal for classification as events or censored observations, and we resolved any disagreements by mutual discussion or by involving a third author (CT).

If seizure data were provided or recorded in terms of the number of seizures recorded between clinic visits rather than specific dates of seizures, to enable the calculation of time‐to‐event outcomes, we applied linear interpolation to estimate dates of seizures between follow‐up visits. For example, if the trial recorded four seizures between two visits that occurred on 1 March 2010 and 1 May 2010 (interval of 61 days), then the date of the first seizure would be approximately 13 March 2010. This allowed the computation of an estimate of the time to six‐month remission, 12‐month remission, and first seizure.

We calculated time to six‐month and 12‐month remission from the date of randomisation to the date (or estimated date) the individual had first been free of seizures for six or 12 months respectively. If the person had one or more seizures in the titration period, a six‐month or 12‐month seizure‐free period could also occur between the estimated date of the last seizure in the titration period and the estimated date of the first seizure in the maintenance period

We calculated time to first seizure from the date of randomisation to the date that their first seizure was estimated to have occurred. If seizure data were missing for a particular visit, these outcomes were censored at the previous visit. These outcomes were also censored if the individual died or if follow‐up ceased prior to the occurrence of the event of interest.

Two trials were designed in strata based on whether recommended treatment would be carbamazepine or sodium valproate (Privitera 2003; Trinka 2013). Within the two strata, participants were randomised to topiramate (Privitera 2003) or levetiracetam (Trinka 2013) compared to the recommended treatment of carbamazepine or sodium valproate depending on the strata. To ensure that randomised comparisons were made, we analysed data for these two trials according to the separate strata in this review (i.e. treated as two trials Privitera 2003 carbamazepine branch and Privitera 2003 sodium valproate branch).

Assessment of risk of bias in included studies

Two authors (SJN, JW) independently assessed risk of bias in all included trials using the Cochrane tool for assessing risk of bias (Higgins 2011). The following methodological criteria are assessed according to this tool:

selection bias (sequence generation and allocation concealment);
performance bias (blinding of participants and personnel);
detection bias (blinding of outcome assessment);
attrition bias (incomplete outcome data);
reporting bias (selective outcome reporting).

We resolved any disagreements by discussion. In theory, a review using IPD should overcome issues of reporting biases as unpublished data can be provided and unpublished outcomes calculated. Any selective reporting bias detected could be assessed with the Outcome Reporting Bias in Trials (ORBIT) classification system (Kirkham 2010). As specified in Data extraction and management, we asked the data providers to provide trial methods such as randomisation and blinding methods, and we discussed any missing data and or inconsistencies, or both with them.

Measures of treatment effect

We summarised all time‐to‐event outcomes using the hazard ratio (HR) as the measure of treatment effect. We calculated outcomes from IPD provided where possible or extracted summary statistics from published trials. We did not attempt to analyse or synthesise adverse event data; a large range of different adverse events are thought to be associated with the 10 different drugs and such data were collected and presented in different ways across trials. For these reasons, we believe a synthesis of adverse event data would present only selective, and potentially misleading information, while a narrative description of adverse event data from IPD or extracted from published trials would be the most informative way of presenting these data.

Unit of analysis issues

We did not encounter any unit of analysis issues. For inclusion in the review, the unit of allocation had to be the individual. Trials of a repeated‐measures (longitudinal) nature or of a cross‐over design were not eligible for inclusion.

Dealing with missing data

For all included trials, we conducted an assessment of the proportion of missing outcome, demographic and covariate data and made a judgement regarding the extent and nature of missing data (e.g. missing at random, missing not at random). We attempted to contact all trial authors in order to request relevant data; we included any information regarding missing data in such requests (Data extraction and management). If further information regarding missing data could not be provided and we judged that an important proportion of data (particularly outcome data) were missing, we conducted sensitivity analyses to investigate the potential impact of the missing data (for example, best case scenario or worst case scenario analyses, assuming those with missing outcome data all had a favourable or unfavourable outcome, respectively).

Assessment of heterogeneity

We used a fixed‐effect model for all pairwise and network meta‐analyses in the first instance as we anticipated that our specific inclusion criteria would result in eligible studies of a similar design and populations and our use of IPD to standardise definitions of outcomes. Also, our previous reviews of this topic have not showed any important heterogeneity (Marson 2000; Nevitt 2016; Nolan 2013b; Nolan 2013c; Nolan 2015; Nolan 2016a; Nolan 2016b; Nolan 2016d); see Data synthesis for further details of pairwise and network meta‐analysis.

For each pairwise comparison, we assessed the presence of heterogeneity statistically using the Q test (P value less than 0.10 for significance) and the I² statistic with the following interpretation (Higgins 2003):

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.

We also assessed the presence of heterogeneity by visually inspecting forest plots, particularly in terms of the magnitude and direction of effects. If substantial or considerable heterogeneity (i.e. I² of 50% or over) was found to be present, which we were not able to explain by differences in characteristics of the trials and participants, we planned to perform network meta‐analysis with a random‐effects model.

It was not possible to directly calculate an I² statistic for the network meta‐analysis due to the between‐study covariance structure required for the network meta‐analysis model (see Data synthesis). However, for this model, we were able to estimate an R statistic, which compares the impact of heterogeneity in the fixed‐effect and random‐effects models (Jackson 2012) and it has been previously shown that R can be used to calculate I² as follows: I² = (R² ‐ 1)/R² (Higgins 2002)

Therefore we estimated an I² statistic for the whole treatment network for each analysis and interpreted as above. We also presented an estimate of Tau² (an estimate of the between‐study variance in random‐effects meta‐analysis) for each analysis and we have taken both statistics into account when interpreting the presence of any important heterogeneity in the treatment network.

Assessment of reporting biases

Two authors (SJN and JW) undertook a full 'Risk of bias' assessment for each eligible trial, including risk of reporting biases. In theory, a review using IPD should overcome issues of reporting biases, as unpublished data can be provided and unpublished outcomes calculated. As specified in Data extraction and management, we asked the data providers for trial methods, such as randomisation and blinding methods, and we discussed any missing data and inconsistencies with them.

If we suspected selective reporting bias in the review, we intended to assess the magnitude and impact of this selective reporting bias using the ORBIT classification system (Kirkham 2010), however we did not have any major concerns about selective reporting bias in this review. The approach to this review (re‐analysis of IPD) helps to overcome issues of reporting bias, as unpublished data can be provided and unpublished outcomes calculated.

Data synthesis

Figure 1 and Figure 2 visually present the network of 45 pairwise comparisons from the 10 antiepileptic treatments of interest to this review.

Network plot of pairwise comparisons in all included studies, studies providing individual participant data (IPD) and studies without IPD

Note that the size of the node indicates the number of studies the drug is included in and the thickness of the edges corresponds to the number of participants contributing to the comparison (i.e. larger node = more studies, thicker edge = more participants).

CBZ: carbamazepine; GBP: gabapentin; LEV: levetiracetam; LTG: lamotrigine; OXC: oxcarbazepine; PHB: phenobarbitone; PHT: phenytoin; TPM: topiramate; VPS: sodium valproate; ZNS: zonisamide

To see a magnified version of this figure, please see https://epilepsy.cochrane.org/network‐meta‐analysis‐figures.

Network plot of pairwise comparisons for all included participants (total 17,961 participants), participants with partial seizures and participants with generalised tonic‐clonic seizures with or without other seizure types (shortened to 'generalised seizures' for brevity).

11978 participants were classified as experiencing partial seizures (66.7% of total), 4407 participants were classified as experiencing generalised seizures (24.5% of total) and 1576 had an unclassified or missing seizure type (8.8% of total).

Note that the size of the node indicates the number of studies the drug is included in and the thickness of the edges corresponds to the number of participants contributing to the comparison (i.e. larger node = more studies, thicker edge = more participants).

CBZ: carbamazepine; GBP: gabapentin; LEV: levetiracetam; LTG: lamotrigine; OXC: oxcarbazepine; PHB: phenobarbitone; PHT: phenytoin; TPM: topiramate; VPS: sodium valproate; ZNS: zonisamide

To see a magnified version of this figure, please see https://epilepsy.cochrane.org/network‐meta‐analysis‐figures.

Pairwise and Network meta‐analysis

We used the statistical software package SAS (version 9.3) (SAS 2011) to perform all data cleaning, consistency checking and data preparation (see Data extraction and management) and Stata version 14 (StataCorp 2015) to perform all synthesis of direct and indirect evidence .

We requested data for one trial, Biton 2001, via data sharing portal ClinicalStudyDataRequest.com and the data were provided to us via a remote secure data access system that allowed analysis in SAS‐based statistical software and export of analysis results. We were unable to combine this dataset with the other datasets to perform the analyses described below in Stata version 14, therefore we treated the results exported from the data access system as aggregate data in sensitivity analysis (see Sensitivity analysis).

We took an intention‐to‐treat approach (as far as possible) to analysis; in other words, we analysed participants in the group to which they had been randomised in an individual trial, irrespective of which treatment they had actually received. Therefore, for time‐to‐event outcomes, 'time to six‐month remission', 'time to 12‐month remission' and 'time to first seizure post randomisation', participants were not censored if treatment was withdrawn. For the primary outcome, time to withdrawal of allocated treatment, we considered withdrawals due to lack of efficacy (i.e. recurrent seizures), poor tolerability (i.e. adverse events) or a combination of both poor efficacy and tolerability. Other withdrawals such as losses to follow‐up, non treatment‐related deaths, administrative trial reasons etc. were censored at the time of withdrawal.

For all time‐to‐event outcomes, we investigated the relationship between the time to the event and treatment effect of the AEDs. We fitted a Cox proportional hazards regression model, stratified by trial to preserve the within‐trial randomisation, to the entire individual participant dataset. We fitted this model via the 'mvmeta_make' command in Stata version 14 to produce a dataset in the correct format to perform network meta‐analysis with the 'mvmeta' command (White 2009); in other words, a dataset with trial‐specific estimates of treatment effect (log HR), the associated variance of the treatment effect and covariances where applicable (i.e. correlation between treatment effects for trials with more than two treatment arms).

The Cox proportional hazards model assumes that ratio of hazards (risks) between the two treatment groups is constant over time. To assess the validity of this assumption, we tested the statistical significance of time‐varying covariates for all covariates in the primary model. If we had reason to believe that the proportional hazards assumption had been violated in the primary model, in sensitivity analysis we fitted a parametric, accelerated failure‐time model, stratified by trial, to the entire individual participant dataset via the 'mvmeta_make' command and compared these results to those of the primary analysis (White 2009). An accelerated failure‐time model assumes that treatment effect accelerates or decelerates over time, rather than remains constant as assumed by the Cox proportional hazards model.

We calculated direct pairwise treatment effect estimates (where possible) using the 'metan' command (Palmer 2016) in Stata version 14 to pool trial‐specific log hazard ratios from the Cox proportional hazards model as described above.

We performed network meta‐analysis via the 'mvmeta' command in Stata version 14 assuming equal heterogeneity for all comparisons (i.e. a between‐study covariance structure (variance‐covariance matrix) proportional to unknown parameter Tau²) (White 2009). It was necessary to make an assumption regarding the between‐study covariance structure for a network without pairwise comparisons between all treatments of interest. However, due to this assumption regarding heterogeneity, we could not calculate an I² statistic directly from the model and had to estimated it (see Assessment of heterogeneity). Network meta‐analysis provided treatment effect estimates combining direct and indirect evidence.

We performed pairwise and network meta‐analyses with a treatment by epilepsy type interaction (see Subgroup analysis and investigation of heterogeneity for further details).

For clinical interest and relevance, we have presented HR estimates from the network model (direct and indirect evidence combined) for each AED in the network compared to the current recommended first‐line treatments (carbamazepine or lamotrigine for partial onset seizures and sodium valproate for generalised onset seizures) and for all comparisons by epilepsy type in the main results of this review via forest plots.

Often rankings of treatments (i.e. the probability that each treatment in the network is the best) are presented for network meta‐analysis; however due to the treatment by epilepsy type interaction in this model, we could not calculate rankings by epilepsy type. Instead, we informally 'ranked' treatments by ordering according to their treatment effect sizes compared to the reference treatment (e.g. better or worse than carbamazepine) on the forest plots presented.

Investigation of consistency in network meta‐analysis

A key assumption made in network meta‐analysis is that treatment effect is 'exchangeable' across all included trials; in other words, the indirect comparison made between two treatments is a feasible comparison to make (known as the transitivity assumption) and that the indirect evidence is consistent with the direct evidence where a comparison exists (known as the consistency assumption).

Transitivity requires that all treatments are "jointly randomisable"; in other words, all 10 AEDs could feasibly be randomised in the same trial and those that are not treatment arms in any given trial are "missing at random" (Lu 2006). This assumption cannot be formally tested statistically; transitivity must be judged by careful consideration of trial settings and characteristics, treatment mechanisms and participant demographics to investigate if any differences would be expected to modify relative treatment effects. Given that all of the 10 drugs within this network are licenced as monotherapy treatments for individuals with newly diagnosed partial onset seizures or generalised onset tonic‐clonic seizures (with or without other generalised seizure types) and have all been used within trials of similar designs, we have no concerns over this transitivity assumption in this network.

The consistency assumption can be evaluated statistically comparing the difference between the direct treatment effect estimate and the indirect estimate for each loop of evidence. Given the complexity of the network model fitted (with treatment by epilepsy type interaction) and the number of multi‐arm trials included in analysis, we performed node splitting in Stata version 14 via the command 'network sidesplit' (Dias 2010; White 2015) to formally estimate differences between direct and indirect evidence for each comparison. In order to examine any clinical inconsistency (i.e. important differences in numerical results between direct, indirect and network results), we have presented HR estimates for direct evidence, indirect evidence (from the node splitting model) and direct plus indirect evidence from the network models for each pairwise comparison via forest plots and discuss the potential origins and implications of any apparent inconsistency. Secondly, we fitted a ‘design‐by‐treatment’ inconsistency model in Stata version 14 via mvmeta (White 2009); this method evaluates both loop and design inconsistencies, particularly within multi‐arm trials (Higgins 2012).

Adverse events

Subgroup analysis and investigation of heterogeneity

There are strong clinical beliefs that certain AEDs are more effective in certain seizure types than others, for example carbamazepine is more effective in partial onset seizures and sodium valproate is more effective in generalised onset seizures (Marson 2000), suggesting that there is a treatment‐by‐seizure type (partial or generalised) interaction. Without taking account of this potential interaction in our analysis, we believe that the key assumption of an exchangeable treatment effect across all included trials would be violated.

To account for this, we conducted all analyses separately by epilepsy type (partial onset or generalised onset) according to the classification of main seizure type at baseline and performed all network meta‐analysis with a treatment‐by‐epilepsy‐type interaction. We classified partial seizures (simple or complex) and partial secondarily generalised seizures as partial epilepsy. We classified primarily generalised seizures as generalised epilepsy. We then judged exchangeability of treatment effect separately by analyses of seizure type.

We also performed an analysis adjusted for age at entry into the trial (an interaction between treatment and age (centred) added to initial Cox proportional hazards model described in Data synthesis) and we compared results to primary analysis with adjustment only for seizure type.

We would have liked to explore other participant covariates specified in Data extraction and management as potential modifiers of treatment effect and as potential sources of heterogeneity or inconsistency, or both, such as seizure frequency before randomisation (time since first ever seizure and/or number of seizures before randomisation) and aetiology of seizures (if known according to pre‐treatment investigations such as EEG, CT and/or MRI scan); however, due to large proportions of missing data for most of these covariates and variability in the definitions of data provided to us for these covariates (see Included studies), an additional adjusted analysis was not appropriate. We will consider other options to explore these covariates for an update of this review.

Sensitivity analysis

As described in Data synthesis, we applied a fixed‐effect model principally to pairwise and network meta‐analysis, and fitted a random‐effects model to both pairwise and network meta‐analysis models in sensitivity analysis, and compared the results.

Also as described in Data synthesis, we applied a Cox proportional hazards model principally to pairwise and network meta‐analysis. We fitted an accelerated failure‐time model, which does not make the assumption of constant treatment effect over time, to both pairwise and network meta‐analysis models in sensitivity analysis and compared the results.

As specified in Data extraction and management, we discussed any inconsistencies in the provided data with the corresponding data providers and performed sensitivity analyses to investigate the impact of any missing data (see Dealing with missing data). If large or major inconsistencies were present, which could not be resolved by the data providers, we would not include the data in any analyses. If minor inconsistencies were present, we included the data in analyses and pursued sensitivity analyses to test the robustness of results included in these data. We performed the following sensitivity analyses due to inconsistencies in IPD provided and compared the results of sensitivity analyses to those of the primary analysis:

In Stephen 2007 there were minor inconsistencies between rates of seizure recurrence and reasons for withdrawal between the data provided and the published paper, which the trial authors could not resolve. Therefore we performed sensitivity analysis excluding Stephen 2007 from all analyses.
In Reunanen 1996, participants were considered to have completed the trial and hence treatment was withdrawn if they experienced a seizure after week six. This does not correspond with the treatment withdrawal definition used in this review (see Primary outcomes and Data extraction and management). Therefore, we performed sensitivity analysis excluding Reunanen 1996 for the analysis of 'Time to withdrawal of allocated treatment.'
In Banu 2007, there were minor inconsistencies between rates of seizure recurrence between the data provided and the published paper, which the authors could not resolve. Therefore we performed sensitivity analysis excluding Banu 2007 from analysis of 'Time to first seizure.' (Data for first seizure recurrence only were available, so this trial did not contribute to outcomes of time to six‐month remission and time to 12‐month remission).
Nieto‐Barrera 2001 did not include seizures that occurred during the first four weeks of the trial in efficacy analyses, and dates of seizures before week four were not supplied to us. Therefore, we calculated seizure outcomes as the time to first seizure and time to six‐month remission after week four rather than after randomisation. We performed sensitivity analysis excluding seizure data for Nieto‐Barrera 2001 from analysis of 'Time to first seizure' (this trial was 24 weeks' duration so did not contribute to outcomes of time to six‐month remission and time to 12‐month remission).
In Placencia 1993, there were minor inconsistencies between reasons for withdrawal between the data provided and the published paper. We compared reasons for withdrawal in the data provided with reasons reported in the publication and performed a sensitivity analysis for the analysis of 'Time to withdrawal of allocated treatment', with withdrawals reclassified according to definitions from the published paper (this sensitivity analysis was also performed in a previously published Cochrane Review, see Nolan 2016b for further details).

Given that misclassification of seizure type is a recognised problem in epilepsy (whereby some individuals with generalised seizures have been mistakenly classed as having partial onset seizures and vice versa) and such misclassification did impact upon the results of a review in our series of pairwise reviews for monotherapy in epilepsy comparing phenytoin and sodium valproate in which nearly 50% of participants analysed may have had their seizure type misclassified (Nolan 2016d), we investigated the potential impact of misclassification on results in a sensitivity analysis. Given clinical evidence that individuals with generalised onset seizures are unlikely to have an 'age of onset' greater than 25 to 30 years (Malafosse 1994), we examined the distribution of age at onset for individuals with generalised seizures. We identified 1164 participants classified as experiencing generalised seizures and estimated age of onset as greater than 30 years (age of first seizure provided directly in IPD or estimated to be within one year of age of entry into trial for newly diagnosed participants). We performed two sensitivity analyses to investigate misclassification:

re‐classification of all individuals with generalised seizures and age of onset greater than 30 years as having partial onset seizures. We then repeated network meta‐analysis with the interaction term of treatment by seizure type with the reclassified seizure type.
re‐classification of all individuals with generalised seizure types and age at onset greater than 30 years and those with missing seizure type into an 'unclassified seizure type' group. We then repeated network meta‐analysis with the interaction term of treatment by seizure type, where seizure type is partial epilepsy compared to generalised or unclassified epilepsy.

We were unable to perform network meta‐analysis with a 'three‐way' interaction (i.e. partial epilepsy compared to generalised epilepsy compared to unclassified epilepsy) due to small numbers of participants with unclassified epilepsy on some of the treatments.

Where possible, if IPD were not available for analysis, we attempted to extract aggregate data. Where aggregate hazard ratios and standard errors or confidence intervals could be extracted or estimated from trial publications by seizure type for our outcomes of interest, we incorporated these estimates into network meta‐analysis and compared the results of these sensitivity analyses to those of the primary analysis. As described in Data synthesis, we were provided with IPD for one trial (Biton 2001), in a remote data access system therefore we could not combine this dataset with the other datasets to perform IPD analysis. We also treated our exported results for this trial as aggregate data in sensitivity analysis.

'Summary of findings' table and quality of the evidence

We have presented six 'Summary of findings' tables for our primary outcome and first secondary outcome by epilepsy type and by reference treatment (see Data synthesis for further information);

Time to withdrawal of allocated treatment for individuals with partial seizures (reference treatment carbamazepine) (see Table 1)
Time to withdrawal of allocated treatment for individuals with partial seizures (reference treatment lamotrigine) (see Table 2)
Time to withdrawal of allocated treatment for individuals with generalised seizures (reference treatment sodium valproate) (see Table 3)
Time to 12‐month remission for individuals with partial seizures (reference treatment carbamazepine) (see Table 4)
Time to 12‐month remission for individuals with partial seizures (reference treatment lamotrigine) (see Table 5)
Time to 12‐month remission for individuals with generalised seizures (reference treatment sodium valproate) (see Table 6)

Summary of findings for the main comparison. Summary of findings ‐ Time to withdrawal of allocated treatment for individuals with partial seizures.

Antiepileptic drug monotherapy for epilepsy: time to withdrawal of allocated treatment for individuals with partial seizures
Patient or population: adults and children with partial seizures Settings: outpatients Intervention: phenobarbitone, phenytoin, sodium valproate, lamotrigine, oxcarbazepine, topiramate, gabapentin, levetiracetam and zonisamide Comparison: carbamazepine
Intervention (experimental treatment)^a,b	Comparison (reference treatment)	No of participants  (studies) with direct evidence	Relative effect  HR (95% CI) Direct evidence  (pairwise meta‐analysis)^c Heterogeneity: I²	Relative effect  HR (95% CI) Direct plus indirect evidence  (network meta‐analysis)^c	Proportion of  direct evidence (%)^d	Quality of the evidence  (GRADE)
Phenobarbitone	Carbamazepine	520 (4 studies)	1.57 (1.16 to 2.13) I² = 0%	1.55 (1.18 to 2.04)	52.5%	⊕⊕⊕⊕  high^e,f
Phenytoin	Carbamazepine	428 (3 studies)	1.03 (0.74 to 1.42) I² = 63.6%	1.13 (0.92 to 1.38)	12.8%	⊕⊕⊕⊕  high^e,f,g
Sodium Valproate	Carbamazepine	814 (5 studies)	0.94 (0.73 to 1.19) I² = 0%	1.04 (0.86 to 1.25)	40.1%	⊕⊕⊕⊕  high^e,f
Lamotrigine	Carbamazepine	2268 (9 studies)	0.76 (0.61 to 0.95) I² = 39.3%	0.75 (0.65 to 0.86)	28.9%	⊕⊕⊕⊕  high^e,f
Oxcarbazepine	Carbamazepine	562 (2 studies)	4.62 (0.95 to 22.4) I² = 0%	1.09 (0.84 to 1.42)	5.7%	⊕⊕⊕⊕  high^e,f
Topiramate	Carbamazepine	937 (2 studies)	1.04 (0.52 to 2.07) I² = 0%	1.18 (0.98 to 1.43)	7.4%	⊕⊕⊕⊕  high^e,f
Gabapentin	Carbamazepine	954 (2 studies)	1.14 (0.84 to 1.55) I² = 0%	1.20 (1.00 to 1.43)	87.1%	⊕⊕⊕⊕  high^e,f
Levetiracetam	Carbamazepine	1567 (3 studies)	0.70 (0.52 to 0.94) I² = 0%	0.82 (0.69 to 0.97)	37.9%	⊕⊕⊕⊕  high^e,f
Zonisamide	Carbamazepine	583 (1 study)	1.08 (0.81 to 1.44) I² = NA)	1.08 (0.79 to 1.48)	100%	⊕⊕⊕⊕  high^e,f
Abbreviations: CI: confidence interval; HR: hazard ratio; NA: not applicable
GRADE Working Group grades of evidence  High quality: further research is very unlikely to change our confidence in the estimate of effect.  Moderate quality: further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.  Low quality: further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.  Very low quality: we are very uncertain about the estimate.

Trial\Drug	CBZ	PHB	PHT	VPS	LTG	OXC	LEV	TPM	GBP	ZNS	Total	Total randomised^a
Trials providing individual participant data
Banu 2007	54	54	0	0	0	0	0	0	0	0	108	108
Baulac 2012	301	0	0	0	0	0	0	0	0	282	583	583
Bill 1997	0	0	144	0	0	143	0	0	0	0	287	287
Biton 2001	0	0	0	69	66	0	0	0	0	0	135	136
Brodie 1995a	66	0	0	0	70	0	0	0	0	0	136	136
Brodie 1995b	63	0	0	0	61	0	0	0	0	0	124	124
Brodie 1999	48	0	0	0	102	0	0	0	0	0	150	150
Brodie 2007	291	0	0	0	0	0	288	0	0	0	579	579
Chadwick 1998	74	0	0	0	0	0	0	0	218	0	292	292
Craig 1994	0	0	81	85	0	0	0	0	0	0	166	166
de Silva 1996	54	10	54	49	0	0	0	0	0	0	167	173
Dizdarer 2000	26	0	0	0	0	26	0	0	0	0	52	52
Eun 2012	41	0	0	0	43	0	0	0	0	0	84	84
Guerreiro 1997	0	0	94	0	0	99	0	0	0	0	193	193
Heller 1995	61	58	63	61	0	0	0	0	0	0	243	243
Kwan 2009	0	0	0	44	37	0	0	0	0	0	81	81
Lee 2011	53	0	0	0	57	0	0	0	0	0	110	110
Mattson 1985	155	155	165	0	0	0	0	0	0	0	475	475
Mattson 1992	236	0	0	244	0	0	0	0	0	0	480	480
Nieto‐Barrera 2001	202	0	0	0	420	0	0	0	0	0	622	622
Ogunrin 2005	19	18	18	0	0	0	0	0	0	0	55	55
Pal 1998	0	47	47	0	0	0	0	0	0	0	94	94
Placencia 1993	95	97	0	0	0	0	0	0	0	0	192	192
Privitera 2003 (CBZ branch)^b	129	0	0	0	0	0	0	266	0	0	395	395
Privitera 2003 (VPS branch)^b	0	0	0	78	0	0	0	147	0	0	225	225
Ramsey 1992	0	0	50	86	0	0	0	0	0	0	136	136
Ramsey 2010	0	0	128	0	0	0	0	133	0	0	261	261
Reunanen 1996	121	0	0	0	230	0	0	0	0	0	351	351
Richens 1994	151	0	0	149	0	0	0	0	0	0	300	300
SANAD A 2007	378	0	0	0	378	210	0	378	377	0	1721	1721
SANAD B 2007	0	0	0	238	239	0	0	239	0	0	716	716
Steiner 1999	0	0	95	0	86	0	0	0	0	0	181	181
Stephen 2007	0	0	0	109	117	0	0	0	0	0	226	227
Trinka 2013 (CBZ branch)^b	503	0	0	0	0	0	493	0	0	0	996	999
Trinka 2013 (VPS branch)^b	0	0	0	353	0	0	350	0	0	0	703	703
Turnbull 1985	0	0	70	70	0	0	0	0	0	0	140	140
Verity 1995	130	0	0	130	0	0	0	0	0	0	260	260
Werhahn 2015	121	0	0	0	118	0	122	0	0	0	361	361
Total	3372	439	1009	1765	2024	478	1253	1163	595	282	12,380	12,391
Trials not providing individual participant data
Trial\Drug	CBZ	PHB	PHT	VPS	LTG	OXC	LEV	TPM	GBP	ZNS	Total	Total randomised^a
Aikia 1992	0	0	18	0	0	19	0	0	0	0	37	37
Bidabadi 2009	36	35	0	0	0	0	0	0	0	0	71	71
Brodie 2002	0	0	0	0	151	0	0	0	158	0	309	309
Callaghan 1985	59	0	58	64	0	0	0	0	0	0	181	181
Capone 2008	17	0	0	0	0	0	18	0	0	0	35	35
Castriota 2008	14	0	0	0	0	0	13	0	0	0	27	27
Chen 1996	26	25	0	25	0	0	0	0	0	0	76	76
Cho 2011	15	0	0	0	0	0	16	0	0	0	31	31
Christe 1997	0	0	0	121	0	128	0	0	0	0	249	249
Consoli 2012	66	0	0	0	0	0	62	0	0	0	128	128
Cossu 1984	6	6	0	0	0	0	0	0	0	0	12	12
Czapinski 1997	30	30	30	30	0	0	0	0	0	0	120	120
Dam 1989	100	0	0	0	0	94	0	0	0	0	194	194
Donati 2007	28	0	0	29	0	55	0	0	0	0	112	112
Feksi 1991	152	150	0	0	0	0	0	0	0	0	302	302
Forsythe 1991	23	0	20	21	0	0	0	0	0	0	64	64
Fritz 2006	0	0	0	0	21	27	0	0	0	0	48	48
Gilad 2007	32	0	0	0	32	0	0	0	0	0	64	64
Jung 2015	64	0	0	0	0	0	57	0	0	0	121	121
Kalviainen 2002	70	0	0	0	73	0	0	0	0	0	143	143
Kopp 2007	6	0	0	3	0	0	6	0	0	0	15	15
Korean Lamotrigine Study Group 2008	129	0	0	0	264	0	0	0	0	0	393	393
Lukic 2005	0	0	0	38	35	0	0	0	0	0	73	73
Mitchell 1987	15	18	0	0	0	0	0	0	0	0	33	33
Miura 1990	66	0	51	46	0	0	0	0	0	0	163	163
Motamedi 2013	0	0	0	0	50	0	50	0	0	0	100	100
NCT01498822	0	0	0	0	0	178	175	0	0	0	353	353
NCT01954121	215	0	0	0	0	0	218	0	0	0	433	433
Pulliainen 1994	23	0	20	0	0	0	0	0	0	0	43	43
Ramsey 1983	42	0	45	0	0	0	0	0	0	0	87	87
Ramsey 2007^c	?	0	0	0	0	0	?	0	0	0	37	37
Rastogi 1991	0	0	45	49	0	0	0	0	0	0	94	94
Ravi Sudhir 1995	20	0	20	0	0	0	0	0	0	0	40	40
Resendiz 2004	42	0	0	0	0	0	0	46	0	0	88	88
Rowan 2005	198	0	0	0	200	0	0	0	195	0	593	593
Saetre 2007	92	0	0	0	93	0	0	0	0	0	185	185
Shakir 1981	0	0	15	18	0	0	0	0	0	0	33	33
So 1992	17	0	0	16	0	0	0	0	0	0	33	33
Suresh 2015	30	0	0	0	0	0	30	0	0	0	60	60
Steinhoff 2005	88	0	0	30	121	0	0	0	0	0	239	239
Thilothammal 1996	0	51	52	48	0	0	0	0	0	0	151	151
Total^c	1721	315	374	538	1040	501	645	46	353	0	5570	5570
Grand total^c	5093	754	1383	2303	3064	979	1898	1209	948	282	17,950	17,961

Trial	Gender			Epilepsy type			Epilepsy type reclassified^c
Trial	Male	Female	Missing	Gen^b	Partial	Missing	Gen^b	Partial	Unclassified^d
Banu 2007	61 (56%)	47 (44%)	0 (0%)	49 (45%)	59 (55%)	0 (0%)	49 (45%)	59 (55%)	0 (0%)
Baulac 2012	347 (60%)	236 (40%)	0 (0%)	0 (0%)	583 (100%)	0 (0%)	0 (0%)	583 (100%)	0 (0%)
Bill 1997	174 (61%)	113 (39%)	0 (0%)	105 (37%)	182 (63%)	0 (0%)	75 (26%)	182 (63%)	30 (10%)
Biton 2001	60 (44%)	75 (55%)	1 (1%)	46 (34%)	82 (60%)	8 (6%)	33 (24%)	82 (60%)	21 (15%)
Brodie 1995a	56 (41%)	80 (59%)	0 (0%)	54 (40%)	82 (60%)	0 (0%)	34 (25%)	82 (60%)	20 (15%)
Brodie 1995b	56 (45%)	68 (55%)	0 (0%)	62 (50%)	62 (50%)	0 (0%)	39 (31%)	62 (50%)	23 (19%)
Brodie 1999	83 (55%)	67 (45%)	0 (0%)	45 (30%)	105 (70%)	0 (0%)	0 (0%)	105 (70%)	45 (30%)
Brodie 2007	319 (55%)	260 (45%)	0 (0%)	113 (20%)	466 (80%)	0 (0%)	50 (9%)	466 (80%)	63 (11%)
Chadwick 1998	157 (54%)	135 (46%)	0 (0%)	0 (0%)	292 (100%)	0 (0%)	0 (0%)	292 (100%)	0 (0%)
Craig 1994	71 (43%)	92 (55%)	3 (2%)	86 (52%)	80 (48%)	0 (0%)	2 (1%)	80 (48%)	84 (51%)
de Silva 1996	86 (50%)	81 (47%)	6 (3%)	84 (49%)	89 (51%)	0 (0%)	84 (49%)	89 (51%)	0 (0%)
Dizdarer 2000	21 (40%)	31 (60%)	0 (0%)	0 (0%)	52 (100%)	0 (0%)	0 (0%)	52 (100%)	0 (0%)
Eun 2012	48 (57%)	36 (43%)	0 (0%)	0 (0%)	84 (100%)	0 (0%)	0 (0%)	84 (100%)	0 (0%)
Guerreiro 1997	100 (52%)	93 (48%)	0 (0%)	50 (26%)	143 (74%)	0 (0%)	45 (23%)	143 (74%)	5 (3%)
Heller 1995	117 (48%)	126 (52%)	0 (0%)	141 (58%)	102 (42%)	0 (0%)	82 (34%)	102 (42%)	59 (24%)
Kwan 2009	40 (49%)	41 (51%)	0 (0%)	48 (59%)	29 (36%)	4 (5%)	25 (31%)	29 (36%)	27 (33%)
Lee 2011	57 (52%)	53 (48%)	0 (0%)	15 (14%)	95 (86%)	0 (0%)	6 (5%)	95 (86%)	9 (8%)
Mattson 1985	413 (87%)	58 (12%)	4 (1%)	1 (0%)	474 (100%)	0 (0%)	1 (0%)	474 (100%)	0 (0%)
Mattson 1992	445 (93%)	35 (7%)	0 (0%)	0 (0%)	480 (100%)	0 (0%)	0 (0%)	480 (100%)	0 (0%)
Nieto‐Barrera 2001	329 (53%)	293 (47%)	0 (0%)	3 (1%)	619 (99%)	0 (0%)	1 (0%)	619 (100%)	2 (0%)
Ogunrin 2005	34 (62%)	21 (38%)	0 (0%)	45 (82%)	10 (18%)	0 (0%)	26 (47%)	10 (18%)	19 (35%)
Pal 1998	47 (50%)	45 (48%)	2 (2%)	34 (36%)	60 (64%)	0 (0%)	34 (36%)	60 (64%)	0 (0%)
Placencia 1993	67 (35%)	125 (65%)	0 (0%)	59 (31%)	133 (69%)	0 (0%)	35 (18%)	133 (69%)	24 (13%)
Privitera 2003 (CBZ branch)^a	215 (54%)	180 (46%)	0 (0%)	88 (22%)	285 (72%)	22 (6%)	51 (13%)	285 (72%)	59 (15%)
Privitera 2003 (VPS branch)^a	112 (50%)	113 (50%)	0 (0%)	131 (58%)	78 (35%)	16 (7%)	86 (38%)	78 (35%)	61 (27%)
Ramsey 1992	73 (54%)	63 (46%)	0 (0%)	136 (100%)	0 (0%)	0 (0%)	110 (81%)	0 (0%)	26 (19%)
Ramsey 2010	126 (48%)	135 (52%)	0 (0%)	150 (57%)	53 (20%)	58 (22%)	80 (31%)	53 (20%)	128 (49%)
Reunanen 1996	188 (54%)	163 (46%)	0 (0%)	114 (32%)	237 (68%)	0 (0%)	71 (20%)	237 (68%)	43 (12%)
Richens 1994	153 (51%)	147 (49%)	0 (0%)	154 (51%)	146 (49%)	0 (0%)	87 (29%)	146 (49%)	67 (22%)
SANAD A 2007	922 (54%)	755 (44%)	44 (3%)	25 (1%)	1491 (87%)	205 (12%)	16 (1%)	1491 (87%)	214 (12%)
SANAD B 2007	420 (59%)	282 (39%)	14 (2%)	463 (65%)	52 (7%)	201 (28%)	397 (55%)	52 (7%)	267 (37%)
Steiner 1999	101 (56%)	80 (44%)	0 (0%)	91 (50%)	90 (50%)	0 (0%)	55 (30%)	90 (50%)	36 (20%)
Stephen 2007	114 (50%)	112 (49%)	1 (0%)	32 (14%)	154 (68%)	41 (18%)	29 (13%)	154 (68%)	44 (19%)
Trinka 2013 (CBZ branch)^a	551 (55%)	448 (45%)	0 (0%)	141 (14%)	858 (86%)	0 (0%)	48 (5%)	858 (86%)	93 (9%)
Trinka 2013 (VPS branch)^a	398 (57%)	305 (43%)	0 (0%)	513 (73%)	190 (27%)	0 (0%)	285 (41%)	190 (27%)	228 (32%)
Turnbull 1985	73 (52%)	67 (48%)	0 (0%)	77 (55%)	63 (45%)	0 (0%)	42 (30%)	63 (45%)	35 (25%)
Verity 1995	122 (47%)	138 (53%)	0 (0%)	152 (58%)	108 (42%)	0 (0%)	152 (58%)	108 (42%)	0 (0%)
Werhahn 2015	215 (60%)	146 (40%)	0 (0%)	0 (0%)	361 (100%)	0 (0%)	0 (0%)	361 (100%)	0 (0%)
Total	6971(56%)	5345 (43%)	75 (1%)	3307 (27%)	8529 (69%)	555 (4%)	2130 (17%)	8529 (69%)	1732 (14%)

Trial	Age (years)				Epilepsy duration (years)			Number of seizures in the last 6 months
Trial	Mean	SD	Range	Missing	Median	Range	Missing	Median	Range	Missing
Banu 2007	5.7	3.5	1 to 15	0	1.2	0 to 11.5	0	24	1 to 7200	5
Baulac 2012	36.4	15.9	18 to 75	0	0.2	0 to 17.7	30	2	1 to 30	1
Bill 1997	26.8	10.7	15 to 91	1	0.4	0 to 25	0	3	0 to 252	0
Biton 2001	32	14.5	12 to 76	0	1	0 to 53	27	2	0 to 100	2
Brodie 1995a	34	15.8	14 to 71	0	1	0 to 18	0	4	1 to 960	0
Brodie 1995b	30	14.1	14 to 81	0	0.5	0 to 19.4	0	3	1 to 1020	0
Brodie 1999	76.9	6	65 to 94	0	NA	NA	150	3	0 to 163	0
Brodie 2007	39	16.2	15 to 82	0	NA	NA	579	3	1 to 1410	4
Chadwick 1998	35	16.6	12 to 86	0	0.5	0 to 7.7	5	4	1 to 146	6
Craig 1994	78.2	7.1	61 to 95	3	NA	NA	166	3	0 to 99	3
de Silva 1996	9.9	3.6	3 to 16	6	0.5	0 to 13.7	6	3	1 to 900	6
Dizdarer 2000	10.8	2.3	4 to 15	0	NA	NA	52	3	1 to 60	0
Eun 2012	8.8	2.1	5 to 13	0	0.4	0 to 4.5	0	3	2 to 11	0
Guerreiro 1997	18.6	9.7	5 to 53	1	0.4	0 to 20	0	2	0 to 157	0
Heller 1995	32.3	14.8	13 to 77	3	1	0 to 40	4	2	1 to 579	3
Kwan 2009	33.9	10.9	16 to 56	0	NA	NA	81	1	0 to 540	0
Lee 2011	35.8	12.2	16 to 60	0	NA	NA	110	2	0 to 200	0
Mattson 1985	41	15.5	18 to 82	4	2	0.5 to 59	5	1	1 to 100	7
Mattson 1992	47.1	16.1	18 to 83	0	3	1 to 68	19	12	1 to 2248	38
Nieto‐Barrera 2001	27.2	21.4	2 to 84	1	NA	NA	622	3	1 to 9000	0
Ogunrin 2005	27.5	8.5	14 to 55	0	7	3 to 11.5	18	12	6 to 42	0
Pal 1998	11.4	5	2 to 18	0	2.5	0.5 to 17	2	NA	NA	94
Placencia 1993	29	17.6	2 to 68	0	5	0.5 to 44	0	2	0 to 100	0
Privitera 2003 (CBZ branch)^a	34.4	18.4	6 to 80	0	NA	NA	395	4	0 to 2400	0
Privitera 2003 (VPS branch)^a	32.8	19.4	6 to 84	0	NA	NA	225	4	0 to 20000	0
Ramsey 1992	20.9	14.2	3 to 64	0	0	0 to 3	0	NA	NA	136
Ramsey 2010	34.1	14.8	12 to 78	0	NA	NA	261	4	0 to 570	0
Reunanen 1996	32.1	14.2	12 to 72	2	0.7	0 to 27	3	3	1 to 145	1
Richens 1994	33	14.9	16 to 79	2	NA	NA	300	4	2 to 101	5
SANAD A 2007	38.4	18.3	5 to 86	44	NA	NA	1721	4	0 to 1185	49
SANAD B 2007	22.5	14.1	5 to 77	14	NA	NA	716	3	0 to 2813	17
Steiner 1999	34.1	16.7	13 to 75	1	1.3	0 to 28.5	1	3	1 to 600	0
Stephen 2007	36	16.9	13 to 80	2	NA	NA	227	18	6 to 1080	37
Trinka 2013 (CBZ branch)¹	42.8	17.2	16 to 89	0	NA	NA	999	NA	NA	999
Trinka 2013 (VPS branch)¹	36.5	17.8	16 to 85	1	NA	NA	703	NA	NA	703
Turnbull 1985	35.2	16.1	14 to 70	0	0.75	0.1 to 30	0	2	0 to 60	0
Verity 1995	10.1	2.9	5 to 16	13	0.3	0 to 5.9	32	3	1 to 104	12
Werhahn 2015	71.5	7.2	60 to 95	0	NA	NA	361	2	1 to 96	7
Total (missing)				98			7820			2135

Trial	Electroencephalographic (EEG)			Computerised Tomography (CT) /Magnetic Resonance Imaging (MRI)			Neurological exams
Trial	Normal	Abnormal	Missing	Normal	Abnormal	Missing	Normal	Abnormal	Missing
Banu 2007	49 (45%)	54 (50%)	5 (5%)	21 (19%)	5 (5%)	82 (76%)	0 (0%)	0 (0%)	108 (100%)
Baulac 2012	0 (0%)	0 (0%)	583 (100%)	0 (0%)	0 (0%)	583 (100%)	478 (82%)	103 (18%)	2 (0%)
Bill 1997	126 (44%)	152 (53%)	9 (3%)	173 (60%)	69 (24%)	45 (16%)	0 (0%)	0 (0%)	287 (100%)
Biton 2001	0 (0%)	0 (0%)	136 (100%)	0 (0%)	0 (0%)	136 (100%)	89 (65%)	46 (34%)	1 (1%)
Brodie 1995a	62 (46%)	72 (53%)	2 (1%)	82 (60%)	12 (9%)	42 (31%)	123 (90%)	13 (10%)	0 (0%)
Brodie 1995b	76 (61%)	42 (34%)	6 (5%)	72 (58%)	20 (16%)	32 (26%)	108 (87%)	16 (13%)	0 (0%)
Brodie 1999	0 (0%)	0 (0%)	150 (100%)	62 (41%)	87 (58%)	1 (1%)	90 (60%)	60 (40%)	0 (0%)
Brodie 2007	0 (0%)	0 (0%)	579 (100%)	0 (0%)	0 (0%)	579 (100%)	493 (85%)	86 (15%)	0 (0%)
Chadwick 1998	107 (37%)	179 (61%)	6 (2%)	0 (0%)	0 (0%)	292 (100%)	0 (0%)	0 (0%)	292 (100%)
Craig 1994	28 (17%)	74 (45%)	64 (39%)	0 (0%)	0 (0%)	166 (100%)	0 (0%)	0 (0%)	166 (100%)
de Silva 1996	0 (0%)	0 (0%)	173 (100%)	0 (0%)	0 (0%)	173 (100%)	152 (88%)	15 (9%)	6 (3%)
Dizdarer 2000	18 (35%)	34 (65%)	0 (0%)	50 (96%)	2 (4%)	0 (0%)	0 (0%)	0 (0%)	52 (100%)
Eun 2012	6 (7%)	78 (93%)	0 (0%)	75 (89%)	9 (11%)	0 (0%)	83 (99%)	1 (1%)	0 (0%)
Guerreiro 1997	92 (48%)	99 (51%)	2 (1%)	126 (65%)	12 (6%)	55 (28%)	0 (0%)	0 (0%)	193 (100%)
Heller 1995	0 (0%)	0 (0%)	243 (100%)	0 (0%)	0 (0%)	243 (100%)	222 (91%)	19 (8%)	2 (1%)
Kwan 2009	0 (0%)	0 (0%)	81 (100%)	0 (0%)	0 (0%)	81 (100%)	0 (0%)	0 (0%)	81 (100%)
Lee 2011	58 (53%)	52 (47%)	0 (0%)	74 (67%)	36 (33%)	0 (0%)	110 (100%)	0 (0%)	0 (0%)
Mattson 1985	126 (27%)	343 (72%)	6 (1%)	308 (65%)	119 (25%)	48 (10%)	0 (0%)	0 (0%)	475 (100%)
Mattson 1992	0 (0%)	0 (0%)	480 (100%)	0 (0%)	0 (0%)	480 (100%)	0 (0%)	0 (0%)	480 (100%)
Nieto‐Barrera 2001	0 (0%)	0 (0%)	622 (100%)	0 (0%)	0 (0%)	622 (100%)	0 (0%)	0 (0%)	622 (100%)
Ogunrin 2005	0 (0%)	0 (0%)	55 (100%)	37 (67%)	0 (0%)	18 (33%)	55 (100%)	0 (0%)	0 (0%)
Pal 1998	0 (0%)	0 (0%)	94 (100%)	0 (0%)	0 (0%)	94 (100%)	24 (26%)	70 (74%)	0 (0%)
Placencia 1993	180 (94%)	12 (6%)	0 (0%)	0 (0%)	0 (0%)	192 (100%)	0 (0%)	0 (0%)	192 (100%)
Privitera 2003 (CBZ branch)^a	0 (0%)	0 (0%)	395 (100%)	0 (0%)	0 (0%)	395 (100%)	0 (0%)	0 (0%)	395 (100%)
Privitera 2003 (VPS branch)^a	0 (0%)	0 (0%)	225 (100%)	0 (0%)	0 (0%)	225 (100%)	0 (0%)	0 (0%)	225 (100%)
Ramsey 1992	0 (0%)	0 (0%)	136 (100%)	0 (0%)	0 (0%)	136 (100%)	0 (0%)	0 (0%)	136 (100%)
Ramsey 2010	0 (0%)	0 (0%)	261 (100%)	0 (0%)	0 (0%)	261 (100%)	0 (0%)	0 (0%)	261 (100%)
Reunanen 1996	13 (4%)	13 (4%)	325 (93%)	16 (5%)	5 (1%)	330 (94%)	305 (87%)	46 (13%)	0 (0%)
Richens 1994	0 (0%)	0 (0%)	300 (100%)	0 (0%)	0 (0%)	300 (100%)	0 (0%)	0 (0%)	300 (100%)
SANAD A 2007	0 (0%)	0 (0%)	1721 (100%)	0 (0%)	0 (0%)	1721 (100%)	1267 (74%)	410 (24%)	44 (3%)
SANAD B 2007	0 (0%)	0 (0%)	716 (100%)	0 (0%)	0 (0%)	716 (100%)	595 (83%)	107 (15%)	14 (2%)
Steiner 1999	103 (57%)	71 (39%)	7 (4%)	111 (61%)	33 (18%)	37 (20%)	165 (91%)	16 (9%)	0 (0%)
Stephen 2007	51 (22%)	121 (53%)	55 (24%)	0 (0%)	0 (0%)	227 (100%)	0 (0%)	0 (0%)	227 (100%)
Trinka 2013 (CBZ branch)¹	0 (0%)	0 (0%)	999 (100%)	0 (0%)	0 (0%)	999 (100%)	0 (0%)	0 (0%)	999 (100%)
Trinka 2013 (VPS branch)¹	0 (0%)	0 (0%)	703 (100%)	0 (0%)	0 (0%)	703 (100%)	0 (0%)	0 (0%)	703 (100%)
Turnbull 1985	70 (50%)	70 (50%)	0 (0%)	17 (12%)	10 (7%)	113 (81%)	0 (0%)	0 (0%)	140 (100%)
Verity 1995	0 (0%)	0 (0%)	260 (100%)	0 (0%)	0 (0%)	260 (100%)	0 (0%)	0 (0%)	260 (100%)
Werhahn 2015	117 (32%)	242 (67%)	2 (1%)	78 (22%)	282 (78%)	1 (0%)	0 (0%)	0 (0%)	361 (100%)
Total	1282 (10%)	1708 (14%)	9401 (75%)	1302 (11%)	701 (6%)	10,388 (83%)	4359 (36%)	1008 (8%)	7024 (56%)

Trial	Summary of results^b
Aikia 1992	1. MANOVA revealed no significant interaction effect of group and time 2. MANOVA revealed no significant interaction effect of group and time 3. MANOVA revealed no significant interaction effect of group and time 4. MANOVA revealed no significant interaction effect of group and time
Bidabadi 2009	1. CBZ: 64%, PHB: 63% 2. No statistically significant difference between groups 3. No statistically significant difference between groups 4. Mean seizure frequency: CBZ: 0.66, PHB: 0.8 5. Mean duration (seconds): CBZ: 12.63; PHB: 15
Brodie 2002	1. Median time to exit: GBP: 69 days, LTG: 48 days HR: 1.043 (95% confidence interval 0.602 to 1.809) 2. Proportion of evaluable population completing the study – GBP: 71.6%, LTG: 67.1% No difference between groups for time to withdrawal for any reason 3. No difference between groups for time to first seizure 4. GBP: 76.1%, LTG: 76.8% (ITT population) 5. Withdrawals during titration: GBP: 7, LTG: 10 Withdrawals after titration: GBP: 10, LTG: 13
Callaghan 1985	1a. PHT: 67%; CBZ: 37%; VPS: 53%  1b. PHT: 12%; CBZ: 37%; VPS: 25%  1c. PHT: 21%; CBZ: 25%; VPS: 22% 2. PHT: 10%; CBZ: 8%; VPS:11%
Capone 2008	1. CBZ: 76%, LEV: 76% 2. Proportion with AEs: CBZ: 65%, LEV: 50% 3. CBZ: 2 discontinuations due to failure to control seizures and interactions with other medications LEV: 3 discontinuations – 1 death from stroke and 2 due to AEs
Castriota 2008	1. No significant difference between groups 2. No significant difference between groups
Chen 1996	1. No significant difference between groups  2. No significant difference between groups 3. 2 children from PHB group, 1 child from CBZ group and no children from VPS group withdrew from the study because of allergic reactions 4. No significant difference between groups
Cho 2011	1. Overall effect on sleep parameters was comparable between groups. LEV group PSG significant increase post treatment compared to baseline in sleep efficiency (P = 0.039) and in decrease of wake time after sleep onset (P = 0.047), no significant change in other sleep parameters. CBZ group post treatment compared to baseline significant increases in the percentage of slow wave sleep (P = 0.038), no significant change in other sleep parameters 2. No significant difference between baseline and post‐treatment between the 2 groups
Christe 1997	1. OXC 56.6% ; VPS 53.8% 2. No significant difference between groups 3. OXC 40.6% ; VPS 33.9% 4. Efficacy no significant difference between groups Tolerability no significant difference between groups Therapeutic effect no significant difference between groups 5. Proportion of participants experiencing at least 1 AE regardless of relationship to trial drug OXC 89.8%; VPS 87.6% 6. Seizure frequency per week OXC (n = 106) mean 0.17 median 0, VPS (n = 106) mean 0.40, median 0
Consoli 2012	1. No significant difference between groups 2. Completed study LEV 52/62, CBZ 54/66, withdrawals: 8 poor compliance (LEV 4, CBZ 4); 7 severe adverse effect (LEV 3, CBZ 4); 7 unknown cause (LEV 3, CBZ 4) 3. Attention deficit on digital span end of follow up greater in CBZ group than LEV (P = 0.03) Stroop test worse in CBZ than LEV (P = 0.02) No significant difference between groups for other scales. Impairment of activities of daily living greater CBZ than LEV (P = 0.05) 4. 4 participants (LEV 2, CBZ 2) had abnormal EEG at baseline, normal at end of treatment. Drug dose reduction (LEV 4, CBZ 2). Remaining participants unmodified versus baseline 5. No significant difference between groups
Cossu 1984	1. Significant decrease in visual‐verbal memory for CBZ and acoustic memory for PHB. No significant differences for other tests
Czapinski 1997	1. PHB: 60%, PHT: 59%; CBZ: 62%; VPS: 64%  2. PHB: 33%, PHT: 23%; CBZ: 30%; VPS: 23%
Dam 1989	1. Baseline OXC mean 2.9 (SD 7.0), median 1, range 0‐60 CBZ mean 5.8 (SD 14.7) median 1, range 0‐99 Maintenance phases OXC mean 0.4 (SD 3.0) median 0, range 0‐27 CBZ mean 0.3 (SD 1.4) median 0, range 0‐12 2. Severe side effects CBZ 25, OXC 13, statistically significant difference favouring OXC (P = 0.04) Participants without any side effects CBZ 25, OXC 29 no significant difference between groups (P = 0.22) 3. Global efficacy no significant difference between groups (P = 0.77); global tolerability (P = 0.11) Participants very good/good CBZ 69 (73%), OXC 76 (84%) Participants poor/very poor CBZ 26 (27%), OXC 15 (16%) 4. Nature of side effects same between groups, included tiredness, headache, dizziness, ataxia. Participants withdrawn due to severe side effects CBZ 16, OXC 9 5. Clinically relevant changes observed in 2 participants only, both CBZ group, both stopped treatment
Donati 2007	1. Comparison of cognitive results no significant difference between treatment groups (P = 0.195) No significant difference between treatment groups for secondary variables (psychomotor speed, alertness, memory and learning, attention, intelligence scores) 2. OXC 58%; CBZ 46%; VPS 54% 3. Most common (> 10% reported) side effects OXC fatigue and headache; CBZ fatigue and rash VPS headache, increased appetite, alopecia 4. Good/very good: OXC investigators 84%, participants 82%, parents/carers 86%; Combined CBZ/VPS investigators 77%, participants 73%, parents/carers 80%
Feksi 1991	1. Minor adverse effects reported in PHB: 58 participants (39%) reported 86 AEs, CBZ: 46 participants (30%) reported 68 AEs 2. All withdrawals: PHB: 18%, CBZ: 17% Withdrawals due to side‐effects: PHB: 5%, CBZ: 3% 3. Seizure‐free: PHB: 54%, CBZ: 52% > 50% reduction of seizures: PHB: 23%, CBZ: 29% 50% reduction‐50% increase in seizures: PHB: 15%, CBZ: 13% > 50% increase in seizures: PHB: 8%, CBZ: 6%
Forsythe 1991	1. Significant difference favouring VPS test of speed of information processing No significant differences between treatment groups for any other cognitive tests 2. PHT: 30%; CBZ: 39%; VPS:33%
Fritz 2006	1. Seizure freedom: LTG: 38%, OXC: 44% < 50% seizure reduction: LTG: 48%, OXC: 55% 2. Both groups showed improvement in verbal learning and in 1/4 measures of attention. In addition, participants under OXC improved in word fluency. Improved mood was reported with OXC only.
Gilad 2007	1. Number of participants experiencing early seizures as first event: LTG 2/32, CBZ 3/32 Number of participants remaining seizure‐free in the follow‐up period: LTG 23/32 (72%), CBZ 14/32 (44%) P = 0.05 2. Incidence of side effects: LTG 2/32 (6.25%), CBZ 12/32 (37.5%) P = 0.05 3. Withdrawals from study due to side effects LTG 1/32 (3%), CBZ 10/32 (31%), P = 0.02
Jung 2015	1. No difference between groups in terms of social competence; school competence; internalising behaviour problems; externalising behaviour problems; total behaviour problems and anxiety. Significant decrease in depression in LEV group compared to CBZ group (P = 0.027) 2. LEV 95.7% , CBZ 97.1% , P = 0.686 3. LEV 66.7%, CBZ 57.8% , P = 0.317 4. LEV 33.3%, CBZ 46.9%. Number of AEs not significantly different between groups
Kalviainen 2002	1.CBZ: 53% LTG: 56% 2. No significant difference between groups in overall cognitive score. In terms of individual assessments, only Stroop test B showed a statistically significant advantage for LTG.
Kopp 2007	1. No significant difference between groups 2. No significant difference between groups
Korean Lamotrigine Study Group 2008	1. LTG: 65% CBZ: 70% 2. Total seizure‐free rate LTG: 62% CBZ: 63% Time to first seizure: mean (SD): weeks LTG: 10 (5.09), CBZ: 10.82 (6.44)
Lukic 2005	1. LTG: 54%, VPS: 55 %, no difference by seizure type 2. LTG: 69%, VPS:68 %
Mitchell 1987	1. No significant differences between treatment groups 2. Compliance: trend towards better compliance in CBZ group (not significant) Randomised participants only: trend towards higher rate withdrawal from treatment in PHB group (not significant). More mild systemic side‐effects in CBZ group (significant). 3 children switched from CBZ to PHB and 1 from PHB to CB following adverse reactions 3. 6 months: excellent/good: PHB = 15, CBZ = 13 12 months: excellent/good: PHB = 13, CBZ = 9
Miura 1990	1. Partial seizures ‐ PHT: 32%; CBZ: 40%; VPS : 41% Generalised seizures ‐ PHT :35%; CBZ: 15%; VPS: 7% 1. Partial seizures ‐ PHT: 24%; CBZ: 24%; VPS : 25% Generalised seizures ‐ PHT :13%; CBZ: 0%; VPS: 0%
Motamedi 2013	1. Seizure recurrence at 2 weeks ‐ LTG: 43% LEV: 35%, p=0.42 Seizure recurrence at 4 weeks ‐ LTG: 39% LEV: 33%, p=0.53 Seizure recurrence at 8 weeks ‐ LTG: 35% LEV: 28%, p=0.50 Seizure recurrence at 12 weeks ‐ LTG: 33% LEV: 24%, p=0.35 Seizure recurrence at 20 weeks ‐ LTG: 31% LEV: 13%, p=0.03 2. No significant difference between groups 3. Proportion with AEs ‐ LTG: 53%, LEV: 67%
NCT01498822	1. LEV: 12.7%, OXC: 23.4% 2. Median months: LEV: 7.6, OXC: NA (fewer than 50% of participants in the OXC group had seizure recurrence) 3. LEV: 53.8%, OXC: 58.5% 4. LEV: 34.7%, OXC: 40.9%
NCT01954121	1. LEV: 47.3%, CBZ: 68.4% 2. LEV: 48.4%, CBZ: 70.2% 3. Number of events: LEV: 88, CBZ: 45 4. Number of events: LEV: 87, CBZ: 39 5. Number of events: LEV: 97, CBZ: 57
Pulliainen 1994	1. Compared to CBZ, participants on PHT became slower (motor speed of the hand) and their visual memory decreased. There was an equal decrease in negative mood (helplessness, irritability, depression) on PHT and CBZ 2. 3 participants taking PHT complained of tiredness, and 1 participant taking CBZ complained of facial skin problems, another tiredness and memory problems
Ramsey 1983	1. Incidence of major side effects (proportion of analysed participants): PHT 23%; CBZ: 23% Minor side effects: cognitive impairment and sedation twice as likely on CBZ compared to PHT. Other minor side effects similar between groups. 2. Treatment failures among analysed participants:  PHT 4/35 (11%); CBZ: 5/35 (14%) Seizure control (among analysed participants with no major side effects): PHT: 86%; CBZ: 82% 3. Significantly lower mean LDH level at 24 weeks in CBZ participants than PHT participants. Other laboratory results similar across treatment groups
Ramsey 2007^c	1. 8 discontinuations; due to generalised rash (n = 1), excessive tiredness (n = 1), withdrew consent (n = 2), renal transplant (n = 1), lost to follow‐up (n = 2), died (n = 1) 2. 6 participants reported treatment‐emergent side effects. 3. No participants withdrew due to lack of seizure control
Rastogi 1991	1(a). PHT: 51%, VPS: 49% 1(b). PHT : 24%, VPS: 35% 1(c). PHT: 18%, VPS: 10% 1(d). PHT: 2%, VPS: 6% 2. All reported AEs were minor and similar rates between groups
Ravi Sudhir 1995	1. No significant differences between any tests of cognitive function taken before treatment and after 10‐12 weeks for both treatment groups
Resendiz 2004	1. Six months of seizure freedom: CBZ: 81%, TPM: 91% 50% reduction of seizures: CBZ: 84% TPM: 97% The average number of seizures was significantly less in the TPM group compared to the CBZ group at 6 and 9 months 2. AEs were mild and similar between groups 3. No significant differences between groups
Rowan 2005	1. Significant difference between 3 treatment groups (P = 0.00022) CBZ more early terminators than GBP (P = 0.008) or LTG (P < 0.0001) 2. LTG 51.4%, GBP 47.4%, CBZ 64.3% no significant difference between groups P = 0.09 3. No difference between groups for time to first, second, fifth and tenth seizure (P values = 0.18, 0.13, 0.74, 0.95 respectively) 4. More systemic toxicities on GBP than CBZ or LTG No significant differences in neuro‐toxicities between treatment groups over 12 months 5. Mean serum levels: 6 weeks GBP 8.67 ± 4.83; µg/mL, CBZ 6.79 ± 2.92 µg/mL and LTG 2.87 ± 1.60 µg/mL 52 weeks GBP 8.54 ± 5.57 µg/mL, CBZ 6.48 ± 3.72 µg/mL and LTG 3.46 ± 1.68 µg/mL Overall medical compliance 89% without significant group differences 6. 3 months LTG 49.7%, GBP 43.3%, CBZ 36.0% significant difference between groups P = 0.02 6 months LTG 37.2%, GBP 33.0%, CBZ 28.9% no significant difference between groups P = 0.22 12 months LTG 28.6%, GBP 23.2%, CBZ 22.8% no significant difference between groups P = 0.33
Saetre 2007	1. LTG 68 (73%), CBZ 61 (67%), no significant difference between groups 2. LTG 59 (63%), CBZ 69 (76%), not significant difference P = 0.068 ITT analysis 3. LTG 71 (76%), CBZ 81 (89%), significant difference, P = 0.0234 ITT analysis 4.Hazard ratio (lamotrigine/carbamazepine) 1.50, 95% CI 0.94–2.40, p value 0.092 5. During treatment period LTG 82 (88%) reported 378 AEs, CBZ 79 (86%) reported 310 AEs. No significant differences between groups for any AEs except for immune system Withdrew due to AE LTG 13 (14%), CBZ 23 (25%), P = 0.078 6. No difference between groups even when changes over time corrected for age, gender and baseline score
Shakir 1981	1. PHT: 33%; VPS: 39% 2. All reported AEs were minor and similar rates between groups
So 1992	1. VPS 7/11 (64%), CBZ 9/14 (64%) 2. At least one AE reported VPS 15/16 (94%), CBZ 16/17 (94%)
Steinhoff 2005	1. FE CBZ group 83/88 (94.3%), LTG group 78/88 (88.6%) no significant difference between groups GE VPS group 25/30 (83.3%) LTG group 20/33 (60.6%) no significant difference between groups 2. FE CBZ group 81%, LTG group 91%, not a significant difference between groups GE VPS group 97%, LTG group 88%, not stated as significant or non‐significant difference 3. At least 1 AE FE CBZ 81 participants (91%), LTG 68 participants (77.3%) GE VPS 25 participants (83.3%), LTG 24 participants (72.7%) Serious AEs FE CBZ 8 participants (9%), LTG 6 participants (7%) GE VPS 1 participant (3%), LTG 5 participants (15%) AEs considered related to study drug FE CBZ 65 participants (74%), LTG 38 participants (43%) GE VPS 16 participants (53%), LTG 15 participants (45.5%)
Suresh 2015	1. Mean quality‐of‐life score at baseline CBZ group 31.14 ± 1.83, LEV group 29.76 ± 1.71 (P value = 0.5861) Mean quality of life score after 26 weeks of treatment CBZ group 58.41 ± 1.89, LEV 64.58 ± 2.02 (P value = 0.0302) 2.28 participants in CBZ group, 28 in LEV group Seizure freedom 4 weeks CBZ group 85.72%, LEV group 85.72% (P value = 1); 12 weeks CBZ group 89.29%, LEV group 93.34% (P value = 0.4595); 26 weeks CBZ group 96.43%, LEV group 100% (P value = 0.1212); 6 months CBZ group 71.42% (20 participants), LEV group 78.57% (22 participants) (P value = 0.2529) 3. Participants experiencing at least 1 AE, CBZ group 36.66%, LEV group 40% (P value = 0.77)
Thilothammal 1996	1. PHB: 31%, PHT: 27%, VPS: 21% 2. PHB: 33%, PHT: 63%, VPS: 31%

Trial	Time to withdrawal from allocated treatment^c				Time to first seizure				Time to 12‐month remission^d				Time to six‐month remission^d
Trial	Cens	Event	Total	Missing	Cens	Event	Total	Missing	Cens	Event	Total	Missing	Cens	Event	Total	Missing
Banu 2007^a	0	0	0	108	39	69	108	0	0	0	0	108	0	0	0	108
Baulac 2012	392	191	583	0	388	186	574	9	251	323	574	9	194	380	574	9
Bill 1997	232	55	287	0	137	145	282	5	190	92	282	5	136	146	282	5
Biton 2001	99	36	135	1	64	71	135	1	0	0	0	136	90	45	135	1
Brodie 1995a	78	58	136	0	69	67	136	0	0	0	0	136	122	14	136	0
Brodie 1995b	79	45	124	0	52	72	124	0	0	0	0	124	96	28	124	0
Brodie 1999	95	55	150	0	70	80	150	0	0	0	0	150	106	44	150	0
Brodie 2007	323	256	579	0	350	229	579	0	260	319	579	0	177	402	579	0
Chadwick 1998	69	223	292	0	102	190	292	0	0	0	0	292	193	99	292	0
Craig 1994	0	0	0	166	68	81	149	17	117	30	147	19	58	89	147	19
de Silva 1996	100	67	167	6	18	149	167	6	22	145	167	6	19	148	167	6
Dizdarer 2000	44	8	52	0	40	12	52	0	11	41	52	0	8	44	52	0
Eun 2012	75	9	84	0	52	32	84	0	0	0	0	84	35	49	84	0
Guerreiro 1997	151	42	193	0	106	84	190	3	112	78	190	3	84	106	190	3
Heller 1995	166	77	243	0	66	177	243	0	78	165	243	0	49	194	243	0
Kwan 2009	60	21	81	0	38	29	67	14	68	13	81	0	30	50	80	1
Lee 2011	73	37	110	0	79	31	110	0	0	0	0	110	39	71	110	0
Mattson 1985	267	208	475	0	226	238	464	11	325	149	474	1	281	193	474	1
Mattson 1992	308	172	480	0	165	303	468	12	334	133	467	13	242	225	467	13
Nieto‐Barrera 2001	511	111	622	0	310	312	622	0	0	0	0	622	431	191	622	0
Ogunrin 2005^a	0	0	0	55	29	26	55	0	0	0	0	55	0	0	0	55
Pal 1998	0	0	0	94	41	49	90	4	82	8	90	4	63	27	90	4
Placencia 1993	158	32	190	2	121	71	192	0	132	60	192	0	69	123	192	0
Privitera 2003 (CBZ branch)^b	221	174	395	0	208	187	395	0	316	79	395	0	194	201	395	0
Privitera 2003 (VPS branch)^b	111	114	225	0	119	106	225	0	180	45	225	0	106	119	225	0
Ramsey 1992	113	23	136	0	81	44	125	11	0	0	0	136	78	47	125	11
Ramsey 2010	192	69	261	0	197	64	261	0	0	0	0	261	0	0	0	261
Reunanen 1996	288	63	351	0	216	135	351	0	0	0	0	351	328	23	351	0
Richens 1994	210	76	286	14	91	199	290	10	92	198	290	10	77	213	290	10
SANAD A 2007	857	815	1672	49	383	1261	1644	77	577	1067	1644	77	355	1326	1681	40
SANAD B 2007	400	299	699	17	182	511	693	23	167	526	693	23	96	610	706	10
Steiner 1999	108	73	181	0	100	81	181	0	0	0	0	181	157	24	181	0
Stephen 2007	160	67	227	0	81	140	221	6	172	55	227	0	137	90	227	0
Trinka 2013 (CBZ branch)^b	760	239	999	0	572	427	999	0	780	219	999	0	336	663	999	0
Trinka 2013 (VPS branch)^b	583	120	703	0	456	247	703	0	484	219	703	0	191	512	703	0
Turnbull 1985	91	49	140	0	75	65	140	0	47	93	140	0	36	104	140	0
Verity 1995	187	59	246	14	59	187	246	14	84	162	246	14	19	227	246	14
Werhahn 2015	195	166	361	0	249	96	345	16	211	150	361	0	178	183	361	0
Total	7756	4109	11,865	526	5699	6453	12,152	239	5092	4369	9461	2930	4810	7010	11,820	571

Reason for withdrawal	Classification for analysis	Randomised drug^4b
Reason for withdrawal	Classification for analysis	CBZ	PHB	PHT	VPS	LTG	OXC	TPM	GBP	LEV	ZNS	Total
Adverse events	Event	505 (45%)	24 (20%)	93 (35%)	132 (28%)	235 (41%)	56 (41%)	259 (48%)	73 (20%)	134 (39%)	31 (32%)	1542 (38%)
Inadequate response	Event	232 (20%)	20 (16%)	46 (17%)	140 (29%)	144 (26%)	36 (26%)	148 (27%)	223 (62%)	89 (26%)	23 (24%)	1101 (27%)
Both adverse events and inadequate response	Event	148 (13%)	51 (41%)	54 (20%)	107 (22%)	32 (6%)	11 (8%)	46 (8%)	32 (9%)	0 (0%)	0 (0%)	481 (12%)
Protocol violation/non compliance	Event	102 (9%)	15 (12%)	41 (15%)	11 (2%)	68 (12%)	27 (20%)	0 (0%)	21 (6%)	21 (6%)	3 (3%)	309 (8%)
Withdrew consent	Event	121 (11%)	13 (11%)	25 (9%)	64 (13%)	65 (11%)	2 (1%)	55 (10%)	4 (1%)	68 (20%)	35 (36%)	452 (11%)
Other^a	Event	29 (3%)	0 (0%)	7 (3%)	24 (5%)	26 (5%)	5 (4%)	37 (7%)	9 (2%)	32 (9%)	4 (4%)	173 (4%)
Total events^b		1137 (35%)	123 (38%)	266 (31%)	478 (28%)	570 (29%)	137 (29%)	545 (47%)	362 (61%)	344 (27%)	96 (34%)	4058 (34%)

Illness or death	Censored	34 (2%)	10 (5%)	17 (3%)	7 (1%)	20 (1%)	1 (0%)	10 (2%)	9 (4%)	0 (0%)	0 (0%)	108 (1%)
Remission of seizures	Censored	49 (2%)	4 (2%)	38 (6%)	75 (6%)	40 (3%)	12 (4%)	44 (7%)	21 (9%)	0 (0%)	0 (0%)	283 (4%)
Lost to follow‐up	Censored	81 (4%)	31 (16%)	51 (9%)	63 (5%)	33 (3%)	24 (7%)	18 (3%)	0 (0%)	41 (5%)	0 (0%)	342 (4%)
Other^c	Censored	104 (5%)	6 (3%)	22 (4%)	82 (7%)	31 (2%)	5 (2%)	26 (4%)	26 (12%)	0 (0%)	25 (13%)	327 (4%)
Completed study	Censored	1829 (87%)	139 (73%)	468 (79%)	949 (81%)	1272 (91%)	291 (87%)	501 (84%)	166 (75%)	868 (95%)	161 (87%)	6644 (86%)
Total censored^b		2097 (65%)	190 (62%)	596 (69%)	1176 (72%)	1396 (71%)	333 (71%)	599 (53%)	222 (39%)	909 (73%)	186 (66%)	7704 (66%)

Missing^d		24	7	1	26	12	8	14	11	0	0	103
Total^e		3258	320	863	1680	1978	478	1158	595	1253	282	11,865

Comparision^a	Direct evidence (pairwise meta‐analysis)				Direct plus indirect evidence  (network meta‐analysis)
Comparision^a	Number of  studies	Number of  participants	HR (95% CI)^b,c	I² statistic^d	Direct evidence (%)^e	HR (95% CI)^b,c
CBZ vs PHB	4	520	1.57 (1.16 to 2.13)	0%	52.5%	1.55 (1.18 to 2.04)
CBZ vs PHT	3	428	1.03 (0.74 to 1.42)	63.6%	12.8%	1.13 (0.92 to 1.38)
CBZ vs VPS	5	814	0.94 (0.73 to 1.19)	0%	40.1%	1.04 (0.86 to 1.25)
CBZ vs LTG	9	2268	0.76 (0.61 to 0.95)	39.3%	28.9%	0.75 (0.65 to 0.86)
CBZ vs OXC	2	562	4.62 (0.95 to 22.4)	0%	5.7%	1.09 (0.84 to 1.42)
CBZ vs TPM	2	937	1.04 (0.52 to 2.07)	0%	7.4%	1.18 (0.98 to 1.43)
CBZ vs GBP	2	954	1.14 (0.84 to 1.55)	0%	87.1%	1.20 (1.00 to 1.43)
CBZ vs LEV	3	1567	0.70 (0.52 to 0.94)	0%	37.9%	0.82 (0.69 to 0.97)
CBZ vs ZNS	1	583	1.08 (0.81 to 1.44)	NA	100%	1.08 (0.79 to 1.48)
PHB vs PHT	3	404	0.67 (0.50 to 0.91)	65%	15.2%	0.73 (0.55 to 0.96)
PHB vs VPS	2	75	0.68 (0.34 to 1.36)	23%	8.8%	0.67 (0.48 to 0.92)
PHB vs LTG	No direct evidence				0%	0.48 (0.35 to 0.66)
PHB vs OXC	No direct evidence				0%	0.70 (0.48 to 1.03)
PHB vs TPM	No direct evidence				0%	0.76 (0.55 to 1.06)
PHB vs GBP	No direct evidence				0%	0.77 (0.55 to 1.07)
PHB vs LEV	No direct evidence				0%	0.53 (0.38 to 0.73)
PHB vs ZNS	No direct evidence				0%	0.70 (0.46 to 1.06)
PHT vs VPS	4	168	1.00 (0.60 to 1.64)	58.5%	9%	0.92 (0.70 to 1.21)
PHT vs LTG	1	90	1.10 (0.57 to 2.14)	NA	11.6%	0.66 (0.52 to 0.85)
PHT vs OXC	2	325	0.65 (0.32 to 1.32)	0%	40.4%	0.97 (0.69 to 1.35)
PHT vs TPM	1	53	0.77 (0.38 to 1.57)	NA	10.9%	1.05 (0.80 to 1.39)
PHT vs GBP	No direct evidence				0%	1.06 (0.81 to 1.40)
PHT vs LEV	No direct evidence				0%	0.73 (0.56 to 0.95)
PHT vs ZNS	No direct evidence				0%	0.96 (0.66 to 1.39)
VPS vs LTG*	3	221	1.40 (1.00 to 1.96)	45.1%	5.1%	0.72 (0.58 to 0.90)
VPS vs OXC	No direct evidence				0%	1.05 (0.76 to 1.44)
VPS vs TPM	2	111	1.66 (1.24 to 2.23)	48.1%	33.7%	1.14 (0.88 to 1.48)
VPS vs GBP	No direct evidence				0%	1.15 (0.89 to 1.49)
VPS vs LEV	1	190	1.14 (0.73 to 1.75)	NA	17.2%	0.79 (0.61 to 1.03)
VPS vs ZNS	No direct evidence				0%	1.04 (0.73 to 1.50)
LTG vs OXC	1	506	0.69 (0.12 to 4.14)	NA	4.4%	1.46 (1.11 to 1.92)
LTG vs TPM	1	648	1.18 (0.86 to 1.62)	NA	20.9%	1.59 (1.29 to 1.95)
LTG vs GBP	1	659	0.62 (0.06 to 6.01)	NA	1%	1.60 (1.31 to 1.96)
LTG vs LEV	1	240	0.86 (0.58 to 1.28)	NA	23.7%	1.10 (0.89 to 1.35)
LTG vs ZNS	No direct evidence				0%	1.45 (1.03 to 2.04)
OXC vs TPM	1	496	0.87 (0.16 to 4.73)	NA	4.9%	1.09 (0.82 to 1.44)
OXC vs GBP	1	507	0.90 (0.08 to 9.96)	NA	2.3%	1.10 (0.83 to 1.45)
OXC vs LEV	No direct evidence				0%	0.75 (0.55 to 1.03)
OXC vs ZNS	No direct evidence				0%	0.99 (0.66 to 1.49)
TPM vs GBP	1	649	1.04 (0.12 to 9.33)	NA	1.1%	1.01 (0.82 to 1.25)
TPM vs LEV	No direct evidence				0%	0.69 (0.54 to 0.89)
TPM vs ZNS	No direct evidence				0%	0.91 (0.64 to 1.31)
GBP vs LEV	No direct evidence				0%	0.69 (0.54 to 0.88)
GBP vs ZNS	No direct evidence				0%	0.90 (0.63 to 1.30)
LEV vs ZNS	No direct evidence				0%	1.32 (0.93 to 1.88)

Comparisiona	Direct evidence (pairwise meta‐analysis)				Direct plus indirect evidence  (network meta‐analysis)
Comparisiona	Number  of studies	Number  of participants	HR (95% CI)^b,c	I² statistic^d	Direct  evidence (%)^e	HR (95% CI)^b,c
CBZ vs PHB	3	158	0.56 (0.33 to 0.96)	13.2%	28.2%	1.28 (0.92 to 1.79)
CBZ vs PHT	2	121	1.44 (0.82 to 2.55)	31.4%	13%	0.87 (0.68 to 1.10)
CBZ vs VPS	4	412	1.11 (0.81 to 1.53)	29.9%	30.7%	0.95 (0.84 to 1.09)
CBZ vs LTG	5	254	0.58 (0.25 to 1.32)	0%	0.1%	1.20 (0.99 to 1.49)
CBZ vs OXC	1	9	0.79 (0.17 to 3.56)	NA	4.6%	1.30 (0.42 to 4.00)
CBZ vs TPM	2	101	1.00 (0.55 to 1.79)	0%	32.8%	1.11 (0.78 to 1.59)
CBZ vs GBP	1	6	0.71 (0.07 to 6.90)	NA	10%	1.75 (0.23 to 12.5)
CBZ vs LEV	2	251	1.00 (0.72 to 1.37)	57.9%	26.7%	1.14 (0.85 to 1.52)
PHB vs PHT	3	130	1.31 (0.67 to 2.53)	0%	22.7%	0.68 (0.47 to 0.98)
PHB vs VPS	2	98	1.50 (0.72 to 3.11)	7.5%	15.3%	0.75 (0.53 to 1.05)
PHB vs LTG	No direct evidence				0%	0.94 (0.64 to 1.39)
PHB vs OXC	No direct evidence				0%	1.01 (0.31 to 3.23)
PHB vs TPM	No direct evidence				0%	0.87 (0.53 to 1.41)
PHB vs GBP	No direct evidence				0%	1.37 (0.17 to 11.1)
PHB vs LEV	No direct evidence				0%	0.88 (0.57 to 1.37)
PHT vs VPS	4	394	1.03 (0.68 to 1.54)	0%	36.8%	1.10 (0.85 to 1.43)
PHT vs LTG	1	91	1.96 (0.37 to 10.2)	NA	4.4%	1.39 (1.03 to 1.89)
PHT vs OXC	2	154	0.71 (0.42 to 1.21)	0%	45.1%	1.49 (0.48 to 4.76)
PHT vs TPM	No direct evidence				0%	1.28 (0.84 to 1.96)
PHT vs GBP	No direct evidence				0%	2.00 (0.26 to 16.7)
PHT vs LEV	No direct evidence				0%	1.32 (0.89 to 1.92)
VPS vs LTG	3	387	0.84 (0.48 to 1.47)	0%	43.5%	1.27 (1.03 to 1.56)
VPS vs OXC	No direct evidence				0%	1.35 (0.44 to 4.17)
VPS vs TPM	2	441	0.67 (0.38 to 1.19)	58.7%	12.9%	1.16 (0.81 to 1.67)
VPS vs GBP	No direct evidence				0%	1.82 (0.24 to 14.3)
VPS vs LEV	1	512	0.88 (0.60 to 1.30)	NA	48.6%	1.19 (0.86 to 1.64)
LTG vs OXC	1	10	0.80 (0.20 to 3.26)	NA	7.6%	1.08 (0.35 to 3.33)
LTG vs TPM	1	14	0.59 (0.30 to 1.16)	NA	10%	0.92 (0.62 to 1.37)
LTG vs GBP	1	7	0.73 (0.08 to 6.57)	NA	11%	1.45 (0.19 to 11.1)
LTG vs LEV	No direct evidence				0%	0.93 (0.65 to 1.33)
OXC vs TPM	1	14	1.34 (0.40 to 4.54)	NA	9.4%	0.86 (0.28 to 2.63)
OXC vs GBP	1	7	0.91 (0.10 to 8.20)	NA	10.7%	1.35 (0.15 to 12.5)
OXC vs LEV	No direct evidence				0%	0.88 (0.27 to 2.78)
TPM vs GBP	1	11	0.68 (0.08 to 5.45)	NA	13.9%	1.56 (0.21 to 12.5)
TPM vs LEV	No direct evidence				0%	1.02 (0.65 to 1.61)
GBP vs LEV	No direct evidence				0%	0.65 (0.08 to 5.00)

Drug	Number of participants  randomised	Number of participants  reporting adverse events^a,b	Number of events  reported^a,b
CBZ	5134	3023	9769
PHB	754	271	181
PHT	1384	614	1513
VPS	2303	1294	3599
LTG	3107	1608	6296
OXC	978	623	1000
TPM	1898	920	6316
GBP	1209	506	2580
LEV	948	1441	4258
ZNS	282	182	606
Total	18,045	10,482	36,118

Event (general description)^a,b,c	CBZ	PHB	PHT	VPS	LTG	OXC	TPM	GBP	LEV	ZNS	Total
Accidental injury	100	0	100	28	110	5	95	36	58	8	540
Anorexia or weight loss	126	0	126	24	116	6	394	58	62	25	937
Anxiety/depression	203	0	203	59	171	32	309	82	163	16	1238
Aphasia	59	7	66	11	26	4	106	22	11	2	314
Asthenia	59	1	60	26	41	1	31	33	37	10	299
Ataxia	172	37	209	32	55	17	61	40	32	8	663
Cognitive (memory, concentration, confusion etc.)	321	41	362	100	204	44	439	127	73	19	1730
Dental problems	93	0	93	28	62	5	61	24	70	7	443
Dizziness/faintness	617	0	617	171	348	140	269	160	394	23	2739
Drowsiness/fatigue	1270	1	1271	422	539	233	628	326	477	33	5200
Fever or viral infection	379	0	379	68	172	24	84	58	338	37	1539
Gastrointestinal disturbances	683	20	703	246	394	33	236	142	284	42	2783
Hair loss	47	0	47	130	22	15	39	8	16	3	327
Headache or migraine	843	0	843	264	556	137	315	171	596	47	3772
Impotence	90	24	114	13	17	0	27	32	11	3	331
Increased/worsened seizures	151	0	151	31	164	6	58	48	140	6	755
Infection	121	0	121	19	90	4	56	27	63	5	506
Laboratory results abnormal	367	0	367	103	117	8	47	19	90	32	1150
Menstrual problems	110	0	110	28	31	1	22	18	39	4	363
Mood or behavioural change	279	41	320	128	163	25	415	121	121	15	1628
Nausea/vomiting	413	1	414	167	233	53	132	92	142	20	1667
Pain	345	1	346	65	250	6	154	48	251	25	1491
Paraesthesia or tingling	56	0	56	22	33	2	708	34	28	7	946
Problems sleeping/nightmares	108	1	109	46	197	16	147	31	101	14	770
Rash or skin disorder	701	17	718	46	420	73	163	113	125	31	2407
Renal/urinary disorder	152	0	152	27	78	2	92	57	93	21	674
Respiratory disorder	233	0	233	53	124	4	190	23	131	17	1008
Tremor or twitch	171	1	172	258	219	19	56	23	51	2	972
Visual disturbance/nystagmus	199	0	199	53	96	33	86	59	33	8	766
Weight gain	259	0	259	347	167	22	71	258	70	1	1454

Date	Event	Description
14 December 2017	New citation required but conclusions have not changed	Conclusions remain the same
14 December 2017	Amended	Abstract revised

Methods	Randomised, double‐blinded, parallel‐group trial conducted in Finland 2 treatment arms: OXC and PHT
Participants	Adult participants with newly diagnosed epilepsy and "normal intellectual capacity" with a minimum of 2 seizures in the last 2 years or 1 seizure and an epileptiform EEG Number randomised: OXC = 19, PHT = 18 Number completed and included in analysis: OXC = 14, PHT = 15 (see Notes) 11 male participants (38%) out of 29 included participants 21 participants with partial epilepsy (72%) out of 29 included participants Mean age of included participants (SD): OXC = 33.6 (14) years, PHT = 32.7 (12.5) years
Interventions	Monotherapy with OXC or PHT 4‐ to 8‐week titration period followed by a maintenance phase of 12 months. Doses achieved not stated Range of follow‐up not stated
Outcomes	Neuropsychological assessment and cognitive functioning in 3 major areas at baseline, 6 months' and 12 months' follow‐up: Verbal learning and memory Sustained attention Simple psychomotor speed
Notes	Participants experiencing inadequate seizure control, adverse events or those who were non‐compliant were withdrawn from the trial and excluded from analysis (5 from OXC group and 3 from PHT group). Results presented only for 29 participants (OXC = 14 and PHT = 15) completing the trial Outcomes chosen for this review were not reported; contact could not be made with trial author to provide IPD
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Participants were "randomly assigned" to treatment; no further information provided
Allocation concealment (selection bias)	Unclear risk	No information provided
Blinding of participants and personnel (performance bias)   All outcomes	Low risk	"The study followed a double blind design"
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	"The study followed a double blind design"; no further information provided about whether outcome assessor was blinded
Incomplete outcome data (attrition bias)   All outcomes	High risk	ITT approach not taken: results reported only for 29 participants (OXC = 14 and PHT = 15) who completed 12‐month follow‐up. 8 participants experiencing inadequate seizure control, adverse events or those who were non‐compliant (OXC = 5 and PHT = 3) were excluded from analysis and results
Selective reporting (reporting bias)	Low risk	No protocol available and outcomes chosen for this review not reported. Neuropsychological and cognitive outcomes well reported and treatment withdrawal rates reported
Other bias	Low risk	None identified

Methods	Single‐centre, double‐blind RCT of participants recruited from clinical referral to a multidisciplinary child development centre at a children's hospital in Dhaka, Bangladesh 2 treatment arms: CBZ and PHB
Participants	108 children aged 2‐15 years with 2 or more generalised tonic‐clonic, partial, or secondarily generalised seizures in the previous year Number randomised: CBZ = 54, PHB = 54 61 boys (56%) 59 participants with partial epilepsy (55%) 26 participants had previous AED treatment (24%) Mean age (range): 6 years (1‐15 years)
Interventions	Monotherapy with CBZ (immediate release) or PHB Starting daily dose: CBZ = 1.5 mg/kg/d, PHB = 5 mg/kg/d, maximum daily dose: CBZ = 4 mg/kg/d, PHB = 16 mg/kg/d Trial duration: 12 months, range of follow‐up: 0‐22 months
Outcomes	Seizure control: seizure freedom during the last quarter of the 12‐month follow‐up Time to first seizure after randomisation Time to treatment withdrawal due to adverse events Change in behaviour from baseline according to age‐appropriate questionnaire Incidence of behavioural side‐effects
Notes	We received IPD for all randomised participants from the trial author. We received reasons for withdrawal of allocated treatment as well as the date of the last follow‐up visit, but withdrawal of allocated treatment did not always coincide with the date of the last follow‐up visit (i.e. several participants had the allocated treatment substituted for the other trial drug and continued to be followed up). Dates of withdrawal of allocated treatment could not be provided; therefore, we could not calculate 'Time to withdrawal of allocated treatment'. We received the date of first seizure after randomisation, but dates of other seizures in the follow‐up time could not be provided; therefore, we calculated 'Time to first seizure' for all participants, but we could not calculate the time to 6‐ and 12‐month remission
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Participants were "randomly assigned to treatment"; the method of randomisation was not stated and not provided by the trial authors
Allocation concealment (selection bias)	Low risk	Allocation was concealed by sealed envelopes prepared on a different site to the site of recruitment of participants
Blinding of participants and personnel (performance bias)   All outcomes	Low risk	Participants, a psychologist, and a therapist were blinded throughout the trial. The treating physician was unblinded for practical and ethical reasons
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	A researcher performing outcome assessment was blinded throughout the trial but unblinded for analysis. It was unclear if this could have influenced the results
Incomplete outcome data (attrition bias)   All outcomes	Low risk	Attrition rates were reported. We analysed all randomised participants from the IPD provided (see footnote 2)
Selective reporting (reporting bias)	Low risk	We calculated 1 outcome for this review from the IPD provided (see footnote 2). We could not calculate other outcomes for this review as the appropriate data were not recorded/not available. All cognitive outcomes from the trial were well reported
Other bias	High risk	There were inconsistencies between rates of seizure recurrence between the data provided and the published paper, which the trial authors could not resolve

Methods	Randomised, double‐blind, parallel‐group, non‐inferiority trial, conducted in 120 centres in Asia, Australia, and Europe 2 treatment arms: CBZ and ZNS
Participants	Participants aged 18‐75 years with newly diagnosed epilepsy, at least 2 partial seizures (with or without secondary generalisation) or generalised tonic‐clonic seizures without clear focal origin in the previous 12 months and at least 1 seizure in the previous 3 months, and had not previously received AEDs or had been treated with 1 AED for no more than 2 weeks. Number randomised: CBZ = 301, ZNS = 282 347 (60%) male participants 100% partial epilepsy Mean age (range): 36 (18‐75 years)
Interventions	Monotherapy with CBZ or ZNS Titration over 4 weeks to a target dose of CBZ = 600 mg/d and ZNS = 300 mg/d Range of follow‐up: 0‐29 months
Outcomes	Proportion of participants who achieved seizure freedom for 26 weeks or more (maintenance period) in the per‐protocol population Incidence of treatment‐emergent results Time to 26‐week (6‐month) remission Time to 52‐week (12‐month) remission Proportion of participants with no seizures for at least 52 weeks Time to withdrawal because of absence of efficacy or adverse events
Notes	IPD provided for all outcomes of this review by trial sponsor Eisai
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	The randomisation scheme was generated centrally by computer programme, which produced a randomisation list with a pseudo‐random number generator
Allocation concealment (selection bias)	Low risk	Allocation concealment was achieved by the use of a telephone interactive voice‐response system to dispense the allocated treatment
Blinding of participants and personnel (performance bias)   All outcomes	Low risk	Participants, investigators, and sponsor personnel administering medication, assessing outcomes, and analysing data were masked to the allocation. Masking was maintained by use of matching placebo tablets.
Blinding of outcome assessment (detection bias)   All outcomes	Low risk	Participants, investigators, and sponsor personnel administering medication, assessing outcomes, and analysing data were masked to the allocation. Masking was maintained by use of matching placebo tablets.
Incomplete outcome data (attrition bias)   All outcomes	Low risk	Attrition rates reported, ITT approach, all randomised participants analysed from IPD provided (see footnote 2)
Selective reporting (reporting bias)	Low risk	Protocol provided. All outcomes reported or calculated with IPD provided (see footnote 2)
Other bias	Low risk	None identified

Methods	Six‐month, systematic, simple randomised trial of children referred to a child neurology clinic (the author was from Guilan University of Medical Sciences, Iran, so it was likely that the trial was also conducted there) 2 treatment arms: CBZ and PHB
Participants	Children aged 2‐12 years with partial seizures with secondary generalisation Number randomised: CBZ = 36, PHB = 35 36 boys (53%) 100% of participants with partial epilepsy Percentage newly diagnosed was not stated Age range: 2‐12 years
Interventions	Monotherapy with CBZ or PHB Doses started or achieved not stated Trial duration: 6 months, range of follow‐up not stated
Outcomes	Proportion seizure‐free Response rate and rate of side‐effects Seizure frequency and seizure duration
Notes	The trial was reported in abstract form only with very limited information. Outcomes chosen for this review were not reported; IPD were not available, trial author could not be contacted
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	The trial was described as a 'systematic simple randomised study'; no further information was provided
Allocation concealment (selection bias)	Unclear risk	No information provided
Blinding of participants and personnel (performance bias)   All outcomes	Unclear risk	No information provided
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	No information provided
Incomplete outcome data (attrition bias)   All outcomes	Unclear risk	No attrition rates were reported; it was unclear if all participants were analysed
Selective reporting (reporting bias)	Unclear risk	There was no protocol available; the trial was available in abstract format only. Outcomes for this review were not available
Other bias	Low risk	None identified

Methods	Multicentre, double‐blind, parallel‐group trial conducted in centres in Argentina, Brazil, Mexico, South Africa 2 treatment arms: OXC and PHT
Participants	Participants aged 16‐65 years with newly diagnosed epilepsy with partial or generalised tonic clonic seizures A minimum of 2 seizures, separated by at least 48 h, within 6 months preceding trial entry No previous AED, except emergency treatment of seizures for a maximum of 3 weeks prior to trial entry Number randomised: total = 287, OXC = 143, PHT = 144 174 male participants (61%); 182 participants with partial epilepsy (63%) Mean age (range) = 26 (15‐91) years
Interventions	Monotherapy with OXC or PHT 8‐week titration period started with 300 mg OXC or 100 mg PHT, increased bi‐weekly, based on clinical response After 8 weeks participants were to be on a three‐times‐a‐day regimen with daily doses of 450 mg‐2400 mg OXC or 150 mg‐800 mg PHT Continued during 48‐week maintenance with adjustment according to clinical response A third long‐term, open‐label extension phase followed the maintenance period. Double‐blind results only were reported Range of follow‐up = 0‐19 months
Outcomes	The proportion of seizure‐free participants who had at least one seizure during the maintenance period Time to premature discontinuation due to adverse experiences Rate of premature discontinuations for any reason Overall assessments of efficacy and tolerability and therapeutic effect Individual adverse experiences Laboratory values Seizure frequency during maintenance
Notes	IPD provided for all outcomes of this review from trial sponsor Novartis
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Treatment groups randomised in 1:1 ratio across centres via computer‐generated randomisation numbers over balanced blocks of size 6
Allocation concealment (selection bias)	Low risk	Allocation concealment was achieved with sequentially‐numbered packages that were identical and contained identical tablets (information provided by trial statistician)
Blinding of participants and personnel (performance bias)   All outcomes	Low risk	Trial conducted in 2 phases: 56‐week, double‐blind phase followed by long‐term, open‐label extension. Double‐blind phase results reported only. Blind achieved with divisible OXC and PHT tablets identical in appearance
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	No information provided
Incomplete outcome data (attrition bias)   All outcomes	Low risk	Attrition rates reported in both treatment phases, participants withdrawing from treatment were no longer followed up so seizure outcomes had to be censored at time of withdrawal and therefore analyses for remission and seizure outcomes could not adopt an ITT approach
Selective reporting (reporting bias)	Low risk	All outcomes reported or calculated with IPD provided (see footnote 2)
Other bias	Low risk	None identified

Methods	Randomised, double‐blind, parallel group, multicentre trial conducted in the USA. 2 treatment arms: LTG and VPS
Participants	Participants > 12 years with newly diagnosed or previously diagnosed epilepsy of any seizure type, not currently using an AED Number randomised: LTG = 66, VPS = 69, ITT population: LTG = 65, VPS = 68 (2 participants withdrew before drug escalation phase) 60 male participants (44%) 82 participants with partial epilepsy (60%) Proportion newly diagnosed not stated Mean age (range): 32 (12‐76) years
Interventions	Monotherapy with LTG or VPS Dose‐escalation phase of 8 weeks to target doses of LTG = 200 mg/d and VPS = 20 mg/kg/d Trial duration: 32 weeks
Outcomes	Weight change The proportion of participants seizure‐free during the entire trial Incidence of the most common drug‐related adverse events Time to withdrawal from the trial
Notes	IPD provided for remote analysis by trial sponsor Glaxo Smith Kline for time to treatment withdrawal, time to first seizure and time to six‐month remission. IPD had to be treated as aggregate data in network meta‐analysis due to remote access to data.
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Computer‐generated randomisation scheme was used
Allocation concealment (selection bias)	Unclear risk	No information provided
Blinding of participants and personnel (performance bias)   All outcomes	Low risk	Participants and personnel double‐blinded
Blinding of outcome assessment (detection bias)   All outcomes	Low risk	Results presented to investigator in a "blinded" fashion
Incomplete outcome data (attrition bias)   All outcomes	Low risk	Attrition rates reported, ITT approach, all randomised participants analysed from IPD provided (see footnote 2)
Selective reporting (reporting bias)	Low risk	Protocol provided. All outcomes reported or calculated with IPD provided (see footnote 2)
Other bias	Low risk	None identified

Methods	Randomised, multicentre, double‐blind, parallel‐group trial conducted in the UK 2 treatment arms: LTG and CBZ randomised in a 2:1 ratio
Participants	Adults > 65 years with newly diagnosed epilepsy with ≥ 2 seizures in the previous year with at least 1 seizure in the last 6 months. None had received previous AED treatment. Number randomised: LTG = 102, CBZ = 48 83 male participants (55%) 105 participants with partial epilepsy (70%) Mean age (range): 77 (65‐94) years
Interventions	Monotherapy with LTG or CBZ 4‐week escalation phase leading to LTG = 100 mg/d, CBZ = 400 mg/d Range of follow‐up = 0‐13.5 months
Outcomes	Time to first seizure after 6 weeks of treatment Time to withdrawal Percentage of participants reporting an adverse event Proportion of participants who were both seizure‐free in the last 16 weeks of the trial and did not discontinue treatment
Notes	IPD provided by trial sponsor Glaxo Smith Kline for time to treatment withdrawal and time to first seizure (plus seizure‐freedom rates at 24 weeks)
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Computer‐generated random sequence (information provided by drug manufacturer). Participants randomised in a 2:1 ratio (LTG:CBZ)
Allocation concealment (selection bias)	Low risk	Allocation concealed by individual sealed, opaque envelopes (information provided by drug manufacturer)
Blinding of participants and personnel (performance bias)   All outcomes	Low risk	Double‐blind achieved using LTG tablets formulated to be identical in appearance to CBZ tablets
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	Trial investigator blinded, not stated if other outcome assessors were blinded
Incomplete outcome data (attrition bias)   All outcomes	Low risk	Attrition rates reported, all randomised participants analysed from IPD provided (see footnote 2)
Selective reporting (reporting bias)	Low risk	Protocol provided. All outcomes reported or calculated with IPD provided (see footnote 2)
Other bias	Low risk	None identified

Methods	Randomised, multicentre, double‐blind trial conducted in 41 centres in Europe and Australia 2 treatment arms: GBP and LTG
Participants	Participants > 16 years with at least 2 partial seizures with or without secondary generalisation or primary generalised tonic clonic seizures in the last 12 months. All participants were untreated in the previous 6 months or AED naive Number randomised: GBP = 158, LTG = 151. Evaluable population (exclusions due to protocol violations): GBP = 148, LTG = 143 152 male participants (52%) out of evaluable population 233 participants with partial epilepsy (80%) out of evaluable population Mean age of evaluable population (SD, range): GBP: 35.8 years (16.4, 13‐78), LTG: 37.9 (16.7, 16‐78)
Interventions	Monotherapy with GBP or LTG Titration of 2 weeks for GBP to a dose range of 1200 mg/d‐3600 mg/d and titration of 6 weeks for LTG to a dose range of 100 mg/d‐300 mg/d Titration period followed by 24‐week maintenance period. Range of follow‐up not stated
Outcomes	Time to exit Percentage of completers/time to withdrawal for any reason Time to first seizure Percentage who remained seizure‐free during the final 12 weeks of the 30‐week evaluation period Withdrawal rate due to adverse events
Notes	IPD requested from trial sponsor Pfizer but data could not be provided due to time elapsed since the trial was completed. Additional information provided in a clinical study report. Aggregate data extracted for time to exit from the trial and time to first seizure extracted from the publication
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Randomisation was performed with permuted blocks, stratified within each centre by seizure type.
Allocation concealment (selection bias)	Unclear risk	No information provided
Blinding of participants and personnel (performance bias)   All outcomes	Low risk	Masking was achieved by double‐dummy dosing. A dose range was permitted within the trial to maintain the blind of two drugs with different titration rates (2 weeks for GBP and 6 weeks for LTG)
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	No information provided
Incomplete outcome data (attrition bias)   All outcomes	Low risk	Attrition rates reported, all randomised participants included in an ITT analysis (even though demographics presented for 'evaluable population' only)
Selective reporting (reporting bias)	Low risk	All efficacy and tolerability outcomes specified in the methods sections were reported well in the results section. No protocol was available.
Other bias	Low risk	None identified

Methods	Single‐centre, randomised, parallel‐group trial of participants referred for assessment at Cork Regional Hospital, Ireland 3 treatment arms: CBZ, PHT, VPS
Participants	Adults and children with a minimum of 2 untreated generalised or partial seizures in the 6 months preceding the trial Number randomised: PHT = 58, CBZ = 59, VPS = 64 95 male participants (52%) 79 participants (44%) with partial epilepsy Age range: 4‐75 years
Interventions	Monotherapy with CBZ, PHT or VPS Mean daily dose achieved: PHT = 5.4 mg/kg, CBZ = 10.9 mg/kg, VPS = 15.6 mg/kg Duration of treatment (range in months): 14‐24 months
Outcomes	Seizure control: excellent (complete freedom of seizures) good (> 50% reduction in seizure frequency) poor (< 50% reduction in seizure frequency or no response) Side effects
Notes	Outcomes chosen for this review were not reported. IPD not available
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Randomisation based on two Latin squares without stratification. The first, second and third preference of drug for the participant appears to have been taken into account in the process. Unclear if assignment was completely random
Allocation concealment (selection bias)	High risk	An independent person (department secretary) selected the “drug of first preference” from randomisation list on a sequential basis. Allocation not adequately concealed
Blinding of participants and personnel (performance bias)   All outcomes	Unclear risk	No information provided
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	No information provided
Incomplete outcome data (attrition bias)   All outcomes	Low risk	Attirition rates reported. ITT approach taken, all randomised participants analysed
Selective reporting (reporting bias)	Low risk	Primary outcomes (seizure control) and secondary outcomes (side effects) reported sufficiently
Other bias	Low risk	None identified

PERMALINK

Antiepileptic drug monotherapy for epilepsy: a network meta‐analysis of individual participant data

Sarah J Nevitt

Maria Sudell

Jennifer Weston

Catrin Tudur Smith

Anthony G Marson

Abstract

Background

Objectives

Search methods

Selection criteria

Data collection and analysis

Main results

Authors' conclusions

Plain language summary

Summary of findings

Background

Description of the condition

Description of the intervention

How the intervention might work

Why it is important to do this review

Objectives

Methods

Criteria for considering studies for this review

Types of studies

Types of participants

Types of interventions

Types of outcome measures

Primary outcomes

Secondary outcomes

Search methods for identification of studies

Data collection and analysis

Selection of studies

Data extraction and management

Requesting individual participant data

Management of individual participant data

Preparation of individual participant data for analysis

Assessment of risk of bias in included studies

Measures of treatment effect

Unit of analysis issues

Dealing with missing data

Assessment of heterogeneity

Assessment of reporting biases

Data synthesis

1.

2.

Pairwise and Network meta‐analysis

Investigation of consistency in network meta‐analysis

Adverse events

Subgroup analysis and investigation of heterogeneity

Sensitivity analysis

'Summary of findings' table and quality of the evidence

Summary of findings for the main comparison. Summary of findings ‐ Time to withdrawal of allocated treatment for individuals with partial seizures.

Summary of findings 2. Summary of findings ‐ Time to withdrawal of allocated treatment for individuals with partial seizures.

Summary of findings 3. Summary of findings ‐ Time to withdrawal of allocated treatment for individuals with generalised seizures.

Summary of findings 4. Summary of findings ‐ Time to 12‐month remission for individuals with partial seizures.

Summary of findings 5. Summary of findings ‐ Time to 12‐month remission for individuals with partial seizures.

Summary of findings 6. Summary of findings ‐ Time to 12‐month remission for individuals with generalised seizures.

Results

Description of studies

Results of the search

3.

Included studies

Characteristics of included trials

1. Number of participants randomised to each drug.

Trials with individual participant data

2. Characteristics of participants providing individual participant data (categorical variables).

3. Characteristics of participants providing individual participant data (continuous variables).

4. Characteristics of participants providing individual participant data (baseline investigations).

Trials without individual participant data

5. Summary of results of trials without individual participant data.

Excluded studies

Risk of bias in included studies

4.

Allocation

Trials for which we received IPD (information reported in published papers or provided with IPD)

Trials for which no IPD were available (information reported in published papers only)

Blinding

Trials for which we received IPD (information reported in published papers or provided with IPD)

Methods	Randomised trial of participants with epileptic seizures following stroke conducted in Italy 2 treatment arms: CBZ (controlled release) and LEV
Participants	Participants with "vascular epilepsy", newly onset following stroke. Not stated if participants had been previously treated with AEDs Number randomised: CBZ = 17, LEV = 18 17 male participants (49%) Proportion of participants with partial epilepsy not stated Mean age: 70 (43‐90) years
Interventions	Monotherapy with CBZ or LEV Dose achieved: CBZ: 400 mg/d‐1200 mg/d, LEV = 1000 mg/d‐3000 mg/d Trial duration and range of follow‐up not stated
Outcomes	Seizure freedom Adverse events during the trial Discontinuations of the trial drug
Notes	The trial was published in Italian; the characteristics and outcomes were translated. Outcomes chosen for this review were not reported; IPD were not available (author confirmed that the data had been lost)
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	The trial was described as randomised ('randomizzazione' in Italian); no further information was available
Allocation concealment (selection bias)	Unclear risk	No information provided
Blinding of participants and personnel (performance bias)   All outcomes	Unclear risk	No information provided
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	No information provided
Incomplete outcome data (attrition bias)   All outcomes	Low risk	Attrition rates reported, no formal statistical analysis performed so withdrawals did not influence results
Selective reporting (reporting bias)	Unclear risk	Methods brief, efficacy and tolerability reported in the results. Outcomes chosen for this review not reported. No protocol available so unclear which outcomes were planned a priori
Other bias	Low risk	None identified

Methods	Randomised, open‐label trial to evaluate event‐related potentials on the effect of CBZ and LEV cognitive functions, conducted in Italy 2 treatment arms, CBZ (controlled release) and LEV
Participants	Participants with newly diagnosed partial epilepsy Number randomised: CBZ = 14, LEV = 13 14 male participants (52%) 100% of participants had partial epilepsy Mean age (years): CBZ = 38, LEV = 42, range not stated
Interventions	Monotherapy with CBZ or LEV Fifteen‐day titration to CBZ = 800 mg/d and LEV = 100 mg/d Trial duration: 24 weeks (assessments at baseline and 12 weeks), range of follow‐up not stated
Outcomes	Event‐related potential recordings Neuropsychological assessments
Notes	The trial was published in Italian; the characteristics and outcomes were translated. Outcomes chosen for this review were not reported; IPD were not available
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	The trial was described as randomised ('randomizzazione' in Italian); no further information was available
Allocation concealment (selection bias)	Unclear risk	No information provided
Blinding of participants and personnel (performance bias)   All outcomes	High risk	Open‐label trial
Blinding of outcome assessment (detection bias)   All outcomes	High risk	Open‐label trial
Incomplete outcome data (attrition bias)   All outcomes	Unclear risk	Attrition rates reported (3 dropouts from the CBZ group, 11% of total participants). These participants are excluded from analysis; this is not an ITT approach
Selective reporting (reporting bias)	Unclear risk	Cognitive outcomes described in methods section well reported in results section. No seizure outcomes or adverse events reported and outcomes chosen for this review not reported. No protocol available so unclear if seizure outcomes were planned a priori
Other bias	Low risk	None identified

Methods	Randomised (partially), double‐blind, multicenter trial conducted at 25 sites in Europe, Australia, South Africa and Canada. 4 treatment arms: GBP (3 arms, 300 mg/d, 900 mg/d and 1800 mg/d) and CBZ. Dose of GBP was masked within the treatment arm but CBZ was given open‐label due to difficulties of blinding tablets and capsules and differing titration periods for the two drugs.
Participants	Participants with newly diagnosed partial epilepsy, with at least 2 unprovoked partial or generalised tonic clonic seizures in the 6 months prior to trial entry, who were AED naive or had received fewer than 2 weeks of AED therapy, which had to be discontinued before trial entry. Participants with a seizure recurrence after at least 2 years of remission were also eligible. Number randomised: CBZ = 74, GBP = 218 157 male participants (54%) 100% participants with partial epilepsy Mean age (range): 35 (12‐86 years)
Interventions	Monotherapy with GBP or CBZ Titration period of 7 d for GBP to target doses 300 mg/d, 900 mg/d or 1800 mg/d. Titration period of 21 d for CBZ to target dose 600 mg/d. Titration period followed by an evaluation period of 24 weeks and an optional open‐label period Range of follow‐up: 0‐77 months
Outcomes	Time to exit Time to exit event plus withdrawals because of adverse events Completion rate (percentage of participants attending end‐of‐phase visit) Exit event rate (percentage of participants who experienced an exit event during the evaluation phase) Adverse event withdrawal rate (percentage of participants who withdrew because of adverse events during either titration or evaluation phases) Exit plus adverse event withdrawal rate (the sum of the exit rate plus the adverse event withdrawal rate) Incidence of adverse events
Notes	IPD provided for all outcomes of this review by trial sponsor Pfizer. In primary analysis, three arms of GBP are pooled and compared to CBZ (see Data extraction and management)
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	A randomisation schedule was prepared separately for each trial centre in blocks of four and eight
Allocation concealment (selection bias)	Low risk	Trial medication was distributed centrally via a pharmacy
Blinding of participants and personnel (performance bias)   All outcomes	High risk	The trial was partially double‐blinded (the dose of GBP was blinded but GBP was not blinded compared to CBZ). Given that the main comparison made in this review is GBP compared to CBZ rather than comparisons between the doses of GBP, this trial is be treated as an open‐label trial
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	Not specifically stated
Incomplete outcome data (attrition bias)   All outcomes	Low risk	Attrition rates reported, ITT approach, all randomised participants analysed from IPD provided (see footnote 2)
Selective reporting (reporting bias)	Low risk	All outcomes reported or calculated with IPD provided (see footnote 2)
Other bias	Low risk	None identified

Methods	Randomised, parallel‐group trial conducted in Taiwan 3 treatment arms: CBZ, PHB, VPS
Participants	Children with 2 or more previously untreated unprovoked epileptic seizures Number randomised: CBZ = 26, PHB = 25, VPS = 25; number analysed: CBZ = 25, PHB = 23, VPS = 25 (see notes) 38 boys (52%) 38 participants with partial epilepsy (52%) Mean age (range) for participants analysed: CBZ = 10.8 (7‐15 years), PHB = 9.9 (7‐15 years), VPS = 9.9 (7‐15 years)
Interventions	Monotherapy with CBZ, PHB or VPS Dose started or achieved not stated Trial duration: 12 months, range of follow‐up: not stated
Outcomes	Cognitive/psychometric outcomes: IQ (WISC‐R scale) and developmental delay (Bender‐Gestalt test) Auditory event‐related potentials (neurophysiological outcome) Incidence of allergic reactions Seizure control
Notes	2 children from the PHB group and 1 child from the CBZ group withdrew from the trial because of allergic reactions. Published results were presented for children who completed the trial only. Outcomes chosen for this review were not reported; IPD were not available
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Participants were allocated with "simple randomisation of block size 3"
Allocation concealment (selection bias)	Unclear risk	No information provided
Blinding of participants and personnel (performance bias)   All outcomes	Unclear risk	The cognitive assessor was "single blinded", implying that participants and personnel were unblinded, but no further information was provided
Blinding of outcome assessment (detection bias)   All outcomes	Low risk	The cognitive assessor was "single blinded"
Incomplete outcome data (attrition bias)   All outcomes	Unclear risk	Withdrawal rates were reported; results were presented only for those who completed the trial (3/73 (4%) excluded from analysis). An ITT approach was not taken but unclear whether the exclusion of this small proportion of participants would influence results
Selective reporting (reporting bias)	Low risk	All cognitive, efficacy, and tolerability outcomes specified in the methods sections were reported well in the results section. No protocol was available. Outcomes chosen for this review were not reported
Other bias	Low risk	None identified

Methods	Randomised trial conducted in Republic of Korea 2 treatment arms: CBZ (controlled release) and LEV
Participants	Participants with newly diagnosed partial epilepsy who had their first seizure between 6 and 1 month prior to entry into the trial and had not taken any AEDs previously. Number completing the trial: CBZ = 15, LEV = 16 (number randomised not stated) 22 male participants (71%) 100% of participants had partial epilepsy Mean age (SD, range): CBZ = 29.8 (9.31, 15‐49), LEV = 31.4 (15.3, 15‐66) years
Interventions	Monotherapy with CBZ or LEV Treatment regimens were CBZ = 400 mg/d and LEV = 1000 mg/d Trial duration 4‐6 weeks, range of follow‐up not stated
Outcomes	Change in overnight PSG scores (sleep latency, REM sleep latency, total sleep time, sleep efficiency, percentage of each sleep stage, arousal index, and Wake time After Sleep Onset) from baseline after 4‐6 weeks of treatment Change in sleep questionnaires (sleep diaries, the Pittsburg Sleep Quality Index, the Korean version of the Epworth Sleepiness Scale, Beck’s depression inventory‐2 and the Hospital Anxiety Scale) and National Hospital Seizure Severity Scale (NHS3) from baseline after 4‐6 weeks of treatment
Notes	IPD could not be provided for the trial due to concerns over institutional review board approval (information provided by corresponding author). Outcomes chosen for this review were not reported
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Trial described as randomised, no further information provided
Allocation concealment (selection bias)	Unclear risk	No information provided
Blinding of participants and personnel (performance bias)   All outcomes	Unclear risk	No information provided
Blinding of outcome assessment (detection bias)   All outcomes	Low risk	PSG scores were interpreted by a certified physician who was blinded to treatment
Incomplete outcome data (attrition bias)   All outcomes	Unclear risk	Number randomised not stated, results provided only for those who completed the trial
Selective reporting (reporting bias)	Low risk	All sleep, efficacy, and tolerability outcomes specified in the methods sections were reported well in the results section. No protocol was available. Outcomes chosen for this review were not reported
Other bias	Low risk	None identified

Methods	Multicentre, open‐label randomised trial conducted in two centres in Italy 2 treatment arms: CBZ and LEV
Participants	Participants > 18 years with late post‐stroke seizures (2 weeks to 3 years after stroke) seen in the Cerebrovascular Unit between September 2008 and March 2009. No previous AED treatments were allowed except for emergency treatments Number randomised: CBZ = 66, LEV = 62. Number completing the trial: CBZ = 54, LEV = 52 58 male participants (55%) of those completing the trial 74 participants with partial epilepsy (74%) of those completing the trial Mean age of those completing trial (SD): CBZ = 69.7 (13.2), LEV = 74.1 (11.3)
Interventions	Monotherapy with CBZ or LEV 2‐week titration period to CBZ: 600 mg/d or LEV: 1000 mg/d Titration period followed by 52‐week maintenance period. Range of follow‐up not stated
Outcomes	Frequency of seizures during the treatment period Rentention of treatment from the first intake Changes in cognitive measures and quality‐of‐life measures at the end of the treatment period: Mini Mental Scale Examination to evaluate global cognitive functioning Logical Memory from the Wechsler Memory Scale‐Revised Visual Memory was assessed with the Benton visual memory test Digital Span Test for attention and some executive functions Stroop Test to investigate the inhibition process Raven’s Coloured Progressive Matrices Test for nonverbal reasoning Corsi span and supraspan learning test ADL index and the Instrumental‐ ADL (IADL) depression was assessed with the Geriatric Depression Scale Changes in EEG assessments at the end of the treatment period Tolerability of treatment
Notes	Contact made with trial author who provided additional information for one of the trial centres but full IPD dataset unavailable. Aggregate data extracted from graph of time to seizure recurrence in the publication. Trial was terminated early due to financial reasons when 128 out of a target 630 participants had been recruited
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Randomisation numbers were sequentially assigned across centres, and a computer‐generated randomisation scheme was used to provide balanced blocks of participants for each treatment group within each centre
Allocation concealment (selection bias)	Unclear risk	No information provided
Blinding of participants and personnel (performance bias)   All outcomes	High risk	Open label trial
Blinding of outcome assessment (detection bias)   All outcomes	High risk	Open label trial
Incomplete outcome data (attrition bias)   All outcomes	High risk	Attrition rates reported, only those who completed the trial were included in efficacy analyses. This is not an ITT approach
Selective reporting (reporting bias)	Low risk	All efficacy, cognitive and tolerability outcomes specified in the methods sections were reported well in the results section. No protocol was available.
Other bias	High risk	Likely that trial is underpowered from the early termination with 20% of target sample size recruited

Methods	Randomised, double‐blind trial to assess short‐term therapy of CBZ and PHB on cognitive and memory function conducted in Italy Three treatment arms: CBZ, PHB, and placebo
Participants	Participants with newly diagnosed and untreated temporal lobe epilepsy with no seizures in the previous month Number randomised: CBZ = 6, PHB = 6 1 man and 5 women in each group 100% partial (temporal lobe epilepsy), 100% newly diagnosed Mean age (SD): CBZ = 26.33 (9.73) years, PHB = 18.5 (2.56) years. Age range: 15‐45 years
Interventions	Monotherapy with CBZ or PHB Dose started and achieved not stated Trial duration: 3 weeks; all participants completed in 3 weeks
Outcomes	Changes in memory function from baseline after 3 weeks of treatment (verbal, visual, (visual‐verbal and visual‐non‐verbal), acoustic, tactile, and spatial).
Notes	The trial was published in Italian; the characteristics and outcomes were translated. Outcomes chosen for this review were not reported; contact could not be made with trial author to provide IPD
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	The trial was described as randomised ('randomizzazione' in Italian); no further information was available
Allocation concealment (selection bias)	Unclear risk	No information provided
Blinding of participants and personnel (performance bias)   All outcomes	Low risk	Trial was described as double blind ('condizioni di doppia cecità' in Italian), we assume this refers to participants and personnel
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	No information provided
Incomplete outcome data (attrition bias)   All outcomes	Low risk	All participants completed this short trial and contributed to analysis
Selective reporting (reporting bias)	Unclear risk	Cognitive and memory outcomes described in methods section well reported in results section. No seizure outcomes or adverse events reported and outcomes chosen for this review not reported. No protocol available so unclear if seizure outcomes were planned a priori
Other bias	High risk	Very small participant numbers and very short‐term follow‐up. Unclear if this trial was adequately powered and of sufficient duration to detect differences

Methods	Parallel design, RCT conducted in the UK 2 treatment arms: PHT and VPS
Participants	Participants > 60 years with newly onset seizures (1 or more generalised tonic‐clonic seizures or 2 or more partial seizures) Number randomised: PHT = 81, VPS = 85 71 male participants (43%) 80 participants with partial epilepsy (48%) Mean age (range): 78 (61‐95 years)
Interventions	Monotherapy with PHT or VPS Starting doses: PHT: 200 mg/d, VPS: 400 mg/d Median daily dose achieved: PHT 247 mg (range 175‐275); VPS: 688 mg (range 400‐1000) Range of follow‐up: 0‐22 months
Outcomes	Psychological tests (cognitive function, anxiety and depression) Adverse event frequency Seizure control
Notes	Trial paper reports on a subset of 38 participants. Full individual participant dataset provided by trial authors and used for this review includes all 166 participants randomised in the trial. IPD provided for 3/4 outcomes of this review ('withdrawal from allocated treatment' not available).
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Computerised stratified minimisation programme, stratified for age group, gender and seizure type
Allocation concealment (selection bias)	Low risk	Pharmacy‐controlled allocation, prescription disclosed to general practitioner and consultant
Blinding of participants and personnel (performance bias)   All outcomes	High risk	Participants and personnel unblinded
Blinding of outcome assessment (detection bias)   All outcomes	Low risk	The main investigator performing cognitive testing was blinded to allocation
Incomplete outcome data (attrition bias)   All outcomes	Low risk	Attrition rates reported. ITT analysis undertaken with all randomised participants from IPD (see footnote 2)
Selective reporting (reporting bias)	Low risk	All outcome measures reported in published report or provided in IPD (see footnote 2)
Other bias	Low risk	None identified

Methods	36‐month randomised, comparative trial conducted in Poland 4 treatment arms: CBZ, PHB, PHT, VPS
Participants	Adults with newly diagnosed epilepsy Number randomised: CBZ = 30, PHT = 30, PHB = 30, VPS = 30 100% of participants had partial epilepsy Age range: 18‐40 years Percentage male and range of follow‐up not mentioned (outcome recorded at 3 years)
Interventions	Monotherapy with CBZ, PHT, PHB or VPS Starting doses CBZ = 400 mg/d, PHT = 200 mg/d, PHB = 100 mg/d, VPS: 600 mg/d Dose achieved not stated
Outcomes	Proportion achieving 24‐month remission at 3 years and exclusions after randomisation due to adverse effects or no efficacy
Notes	Abstract only. Outcomes chosen for this review were not reported, contact made with trial authors but IPD not available
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Trial randomised but no further information provided
Allocation concealment (selection bias)	Unclear risk	No information provided
Blinding of participants and personnel (performance bias)   All outcomes	Unclear risk	No information provided
Blinding of outcome assessment (detection bias)   All outcomes	Unclear risk	No information provided
Incomplete outcome data (attrition bias)   All outcomes	Unclear risk	"Exclusion rates" reported for all treatment groups, no further information provided
Selective reporting (reporting bias)	Unclear risk	No protocol available, trial available in abstract format only. Outcomes for this review not available
Other bias	Low risk	None identified

Methods	Randomised, parallel‐group, open‐label paediatric trial conducted in 2 centres in the UK 4 treatment arms: CBZ, PHB, PHT, VPS
Participants	Children with newly diagnosed epilepsy (2 or more untreated partial or generalised tonic‐clonic seizures in the 12 months preceding the trial) Number randomised: CBZ = 54, PHB = 10, PHT = 54, VPS = 49 86 boys (50%) 89 children with partial epilepsy (51%) Mean age (range): 10 (3‐16) years
Interventions	Monotherapy with CBZ, PHT, PHB or VPS Median daily dose achieved: CBZ = 400 mg/d, PHT = 175 mg/d, PHB = not stated (see notes), VPS= 600 mg/d Range of follow‐up (months): 10‐164
Outcomes	Time to first seizure recurrence after start of therapy Time to 12‐month remission from all seizures Adverse effects and withdrawals due to adverse events
Notes	IPD provided for all randomised participants. All outcomes in this review calculated from IPD. 6 of the first 10 children assigned to PHB had unacceptable adverse effects, so no further children were assigned to PHB. The 10 children randomised to PHB were retained in analysis. IPD provided for all outcomes of this review by the Medical Research Council
*Risk of bias*
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Randomisation list generated using permuted blocks of size 8 or 16 with stratification for centre, seizure type and presence of neurological signs
Allocation concealment (selection bias)	Low risk	Allocation concealed via 4 batches of concealed, opaque envelopes
Blinding of participants and personnel (performance bias)   All outcomes	High risk	Unblinded, authors state masking of treatment would not be “practicable or ethical” and would “undermine compliance.” Lack of masking could have led to early withdrawal of the PHB arm from the trial, which was likely to have influenced the overall results
Blinding of outcome assessment (detection bias)   All outcomes	High risk	Unblinded, authors state masking of treatment would not be “practicable or ethical” and would “undermine compliance.” Lack of masking could have led to early withdrawal of the PHB arm from the trial, which was likely to have influenced the overall results
Incomplete outcome data (attrition bias)   All outcomes	Low risk	Attrition rates reported, all randomised participants analysed from IPD provided (see footnote 2)
Selective reporting (reporting bias)	Low risk	Protocol provided. All outcomes reported or calculated with IPD provided (see footnote 2)
Other bias	Low risk	None identified