Results of the search

The original Cochrane Review (2002) identified a single study (Appleton 1995).

The update in 2008 identified three further studies (Ahmad 2006; Lahat 2000; McIntyre 2005).

For this update, we have identified 14 further studies that meet the main inclusion criteria in addition to the single study in the original review and the three studies identified for the 2008 update. (Arya 2011; Ashrafi 2010; Baysun 2005; Chamberlain 1997; Chamberlain 2014; Fişgin 2002; Gathwala 2012; Javadzadeh 2012; Mahmoudian 2004; Momen 2015; Mpimbaza 2008; Shah 2005; Sreenath 2010; Talukdar 2009).

Full details of searches conducted before 2012 are unavailable. Figure 1 shows the study flow diagram for searches completed between 2012 and 2017, in addition to the studies already listed in the 2008 update of the review.

Figure 1

---

---

Study flow diagram.

Searches conducted been 2012 and 2017 identified 140 records, including 41 duplicate records. We screened 99 records (title and abstract) for inclusion in the review and excluded 65 clearly irrelevant records. With the four studies included in previous versions of the review and two studies excluded from previous versions of the review, we assessed 40 full‐text articles or clinical trials registry entries. We excluded 20 studies (see Excluded studies) and included 18 studies (reported in 20 full‐text articles or clinical trials registry entries) in the review.

Included studies

We included 18 trials in this review (Ahmad 2006; Appleton 1995; Arya 2011; Ashrafi 2010; Baysun 2005; Chamberlain 1997; Chamberlain 2014; Fişgin 2002; Gathwala 2012; Javadzadeh 2012; Lahat 2000; Mahmoudian 2004; McIntyre 2005; Momen 2015; Mpimbaza 2008; Shah 2005; Sreenath 2010; Talukdar 2009). All were hospital‐based studies.

This section gives a brief description of the characteristics and participants of each included trial; see Characteristics of included studies for further details.

Ahmad 2006 was a 12‐month, open, randomised study comparing intranasal lorazepam (0.1 mg (100 micrograms)/kg) and intramuscular paraldehyde (0.2 mg (200 micrograms)/kg) as the first‐line treatment of children aged two months to 12 years, presenting to a paediatric emergency centre with a generalised convulsion continuing for at least five minutes. The study was carried out in Malawi, Africa. Intramuscular paraldehyde is commonly used as a first‐line treatment for acute tonic‐clonic seizures in sub‐Saharan Africa but is associated with injury around the injection site, sterile abscesses and is incompatible with plastics. Patient demographics were similar in each group. Because of the geographical location of this study most of the children had acute symptomatic seizures, mainly due to acute brain infection (cerebral malaria or bacterial meningitis in two‐thirds of each of the two study groups). Randomisation was allocated in advance by computer in blocks of 10; after identification and treatment of children with hypoglycaemic seizures, investigators opened an unmarked envelope which contained details of treatment allocation. Primary outcome was the clinical cessation of the seizure within 10 minutes of drug administration. Children with features of hepatic or hypertensive encephalopathy or organophosphate poisoning were excluded, as were children who had received an anticonvulsant agent within one hour of presentation. For children in whom clinical seizure activity continued after 10 minutes, investigators followed a locally‐agreed protocol. The study evaluated 160 children of both sexes.

Appleton 1995 was a one‐year open, quasi‐randomised study, comparing lorazepam and diazepam, with the drugs given either intravenously or rectally depending on ease of venous access. This study evaluated 102 children, aged between one month and 16 years, of both sexes, presenting with an acute tonic‐clonic convulsion including established convulsive status epilepticus to an A&E department of a large children's hospital. The study accepted all causes of the convulsion or status, including symptomatic and idiopathic. No child had evidence of acute head trauma, metabolic encephalopathy, bacterial meningitis or herpes simplex encephalitis as a cause of their presenting convulsion. No children were included with known pseudo‐tonic‐clonic convulsions or pseudo‐convulsive, absence or complex partial status. The demography of the two treatment groups was very similar (age; sex; numbers with pre‐existing epilepsy; numbers with a pre‐existing neurological disorder and duration of the presenting convulsion prior to treatment with the two study drugs). Cessation of the seizure was defined as the seizure or episode of status stopping within seven or eight minutes of administration of the first dose of the study anticonvulsant. If the presenting convulsion had not stopped by eight minutes, then a second dose of either lorazepam or diazepam would be given. If this seizure persisted, then an additional anticonvulsant would be given, based on the hospital's protocol for managing convulsive status epilepticus (Garr 1999).

Arya 2011 was a randomised controlled trial comparing intranasal and intravenous (IV) lorazepam for the treatment of convulsive status epilepticus in children. The trial took place in the emergency room of a hospital in New Delhi, India. Inclusion criteria were children aged six to 14 years who presented convulsing or who developed a seizure during the emergency room attendance. Exclusion criteria were receipt of any anti‐epileptic drug (AED) within one hour of enrolment, the presence of severe cardiovascular compromise, and the presence of cerebrospinal fluid (CSF) rhinorrhoea or upper respiratory infection severe enough to prevent intranasal administration. Fifty‐eight children (41%) had GTC, 77 (54%) had partial seizures and six were described as having "others/unclear". The groups were evenly matched for age, gender, seizure type and prior AED administration. The primary outcome measure was cessation of visible motor activity by 10 minutes. Secondary outcome measures were persistent cessation of seizure activity at one hour, time to IV access, time from drug administration to stopping of the seizure and development of hypotension/respiratory depression. Further seizures were treated with intravenous phenytoin.

Ashrafi 2010 was a randomised controlled trial conducted in two large hospitals in Tehran, Iran, comparing buccal midazolam and rectal diazepam for the control of acute convulsive seizures. Ninety‐eight children aged more than three months with an acute prolonged seizure lasting more than five minutes and those convulsing while attending the emergency rooms were enrolled, irrespective of the cause of the seizure. Patients who already had intravenous access or who were younger than three months were excluded. Most (84 patients or 86%) had GTC, with the remainder being myoclonic, focal clonic and focal tonic seizures. There was no significant difference between the two groups for age, sex or seizure type. Randomisation was by a random‐number table to either buccal midazolam (0.3 to 0.5 mg/kg) or rectal diazepam (0.5 mg/kg). The primary outcome measure was cessation of all motor activity in less than five minutes, without respiratory depression and without another seizure. Further seizures were treated with intravenous diazepam. The outcome measures were further defined as treatment initiation time and drug effect time. The authors also examined the convenience of drug use and parental acceptance of the drug/route of administration for each group.

Baysun 2005 was a prospective randomised study of all children attending the emergency room of a children's hospital in Turkey with a seizure, regardless of type, aetiology and whether the seizure was prolonged (this was assumed). No exclusion criteria were stated. Forty‐three children ranging in age from two months to 12 years were recruited and randomised to buccal midazolam (0.25 mg/kg) on even days of the month and rectal diazepam (0.5 mg/kg for under‐fives and 0.3 mg/kg for those aged six or more) on odd days of the month. The two groups did not differ significantly by sex, age, type of seizures or anti‐epileptic drug used. Ten children in the midazolam group and 10 in the diazepam group had GTC. The remaining participants presented with generalised tonic, simple partial and complex partial seizures. Outcome measures were cessation of convulsive seizure activity within 10 minutes, time to response, and need for a second drug. Those who did not respond within 10 minutes were given the alternative drug, i.e. midazolam given to those who had already received diazepam and vice versa.

Chamberlain 1997 was a prospective, open randomised study of the management of children aged 0 to 18 years presenting to the emergency department of two large hospitals in the USA, with motor seizures of at least 10 minutes' duration (all had tonic‐clonic or clonic seizures ‐ clarified in personal communication). The study compared intramuscular midazolam (0.2 mg/kg) with intravenous diazepam (0.3 mg/kg). Children who had established intravenous access or who had already received treatment for this seizure episode were excluded. Primary outcome measures were seizure cessation within five minutes of administration, seizure cessation between five and 10 minutes after administration (defined as delayed seizure control), and treatment failure (lack of cessation by 10 minutes). Those who had treatment failure were subsequently given intravenous diazepam or phenytoin. Other outcome measures included recurrence of seizures, defined as early recurrence if within 15 minutes, or recurrence if within 60 minutes. Twenty‐eight children were identified for enrolment, but three were excluded as their seizures did not persist beyond 10 minutes. One child who was randomised to diazepam was a protocol violation due to failure to establish intravenous access, necessitating treatment with intramuscular midazolam. Twenty‐three children with 24 seizure episodes were studied (one child had two episodes and appears in the study twice, once in each group). The demographics were similar between the two groups.

Chamberlain 2014 was a large multicentre randomised controlled trial conducted in the emergency departments of 11 North American hospitals, comparing intravenous lorazepam with intravenous diazepam for convulsive status epilepticus in children. Inclusion criteria were children aged three months to 18 years with generalised tonic‐clonic status epilepticus. This was defined as three or more seizures in the previous hour, two or more successive seizures with no recovery of consciousness with an ongoing seizure, or an ongoing seizure lasting at least five minutes. Children who had initial focal seizures rapidly evolving to bilaterally convulsive seizures were included. Patients with the following factors were excluded: known pregnancy, significant cardiac arrhythmia, urgent need for surgical intervention and anaesthesia, known contraindication to benzodiazepines or benzodiazepine use in the previous seven days (including pre‐hospital use by ambulance personnel). "Early terminators" were children removed from the study following administration of the study drug due to discovery of an exclusion factor or a refusal to participate by the family. The study assessed 11,630 patients for eligibility, of whom 11,320 were excluded, mainly because they were not having an acute seizure or did not meet the inclusion criteria. The study randomised 310 children to one of the two study drugs. Twenty‐two and 15 children were early terminators from each treatment arm respectively, and were excluded from the efficacy analysis. Further exclusions largely due to protocol deviations resulted in 102 and 107 children being available for per protocol analysis in each treatment arm. The participants in each treatment arm were well‐matched in demographics and seizure aetiology. Primary efficacy outcome measures were cessation of status epilepticus (defined as cessation of generalised convulsive activity with return of consciousness within the four‐hour observation period) within 10 minutes of the initial dose, and seizure freedom for 30 minutes. Secondary outcome measures included latency of drug response (time to cessation of convulsions), need for a dose of study medication, need for further anticonvulsants and sustained seizure freedom for 60 minutes and four hours. Primary safety outcomes were severe respiratory depression (needing assisted ventilation) within four hours of the study drug administration; secondary safety outcomes were aspiration pneumonia, any degree of respiratory depression, time required to return to baseline mental status, and degree of sedation or agitation as measured by the Riker Sedation‐Agitation scale.

Fişgin 2002 was a prospective, randomised, single‐centre study of children aged between one month and 13 years, presenting with an acute seizure to the emergency room of a children's hospital in Turkey. All children who were seizing on arrival were included, as it was presumed that their seizure had been ongoing for at least five minutes.No exclusion criteria are stated and the aetiology of the seizures is not given. Of 45 enrolled in the study, 28 children (14 per treatment group) had generalised tonic‐clonic seizures, the rest presenting with simple focal (10), secondarily generalised (4), tonic (1) and myoclonic (2) seizures. Both febrile and afebrile seizures were included, although only 10% of the participants had a febrile seizure, and were equally distributed between the two groups. Children were randomised on alternate days to rectal diazepam (0.3 mg/kg) or nasal midazolam (0.2 mg/kg). If the seizure continued beyond 10 minutes, the alternative drug was given, i.e. there was cross‐over between the two groups. Persistent convulsions (not clearly defined) were treated with intravenous midazolam by bolus, then infusion. Outcome measures were stopping of the seizure within 10 minutes, response time, and necessity for a second drug. There was no significant difference between the two groups for age or seizure type.

Gathwala 2012 was a randomised controlled trial undertaken in an Indian teaching hospital, comparing intravenous diazepam, midazolam and lorazepam for the treatment of acute convulsive seizures in children. Children aged six months to 14 years presenting with a convulsion to the emergency department were recruited. Children with liver or renal disease, cardiovascular abnormalities, head injury, diabetes mellitus or hypoglycaemia were excluded, as were those whose seizure had already stopped or where intravenous access could not be established. The study assessed 185 children for inclusion, of whom 65 were excluded; 55 did not meet the inclusion criteria, seven declined, and intravenous access could not be established in three. Participants were randomised into three treatment groups, which were evenly matched for demographics, mean duration of seizure, prolonged seizures, those presenting with first episode, and cause of seizure. The primary outcome measure was time to seizure cessation, defined as cessation of visible epileptic phenomena or return of purposeful response to external stimuli within 15 minutes of drug administration. The secondary outcomes were the effects of the drugs, i.e. vomiting, apnoea, somnolence, respiratory depression and requirement for mechanical ventilation. Other secondary outcomes were the number of participants with seizure recurrence, requirement for a second dose of medication, uncontrolled seizures, and the time to seizure recurrence.

Javadzadeh 2012 was a randomised unblinded study of 60 children aged between two months and 15 years, presenting to the emergency department with an acute seizure. Exclusion criteria were patients with prior IV access, previous anticonvulsant treatment, or concurrent respiratory tract infection. Participants were randomised to intranasal midazolam or intravenous diazepam, although all patients were cannulated on arrival. Outcome measures included time needed to control seizure, oxygen saturations, and heart rate pre‐ and post‐treatment.

Lahat 2000 was a 12‐month single‐centre randomised study comparing intranasal midazolam (0.2 mg/kg) and intravenous diazepam (0.3 mg/kg) in the treatment of prolonged febrile seizures (a seizure of at least 10 minutes duration) in children aged six months to five years. The study was carried out in a paediatric emergency department within a general hospital. Patient demographics were similar in both groups. Treatment was successful if the clinical features of the seizure stopped within five minutes. If the seizure stopped at between five and 10 minutes, this was identified as a delayed but successful treatment. Treatment failures (continued seizure activity after 10 minutes) received intravenous diazepam and then phenobarbital in accordance with local guidelines. Randomisation was allocated in advance by a random‐number table, with investigators receiving an opaque envelope with each allocation at the time of administration. Forty‐four children of both sexes were evaluated, with a total of 52 seizure episodes. Children who had received an anticonvulsant or had an intravenous line sited by paramedics prior to hospital attendance were excluded from the study.

Mahmoudian 2004 is a prospective randomised study of children aged two months to 15 years, presenting with an acute seizure to the paediatric emergency department of a general hospital in Iran over a two‐month period. Seventy children who presented with an acute seizure (length not specified) were randomised by an odd‐ and even‐number table to receive 0.2 mg/kg of intravenous diazepam or 0.2 mg/kg of intranasal midazolam. Fifty children presented with GTC, six with simple partial seizures, twelve with complex partial seizures and five with myoclonic seizures (note that multiple seizure types can occur in a single child). Outcome measures were time from treatment to cessation of seizure, with treatment considered successful if the seizure stopped within 10 minutes. Seizures that did not stop within 10 minutes were defined as treatment failures. Treatment failures in the midazolam group were given intravenous diazepam and those in the diazepam group were given intravenous phenobarbitone. Aetiologies of the seizures were reported, and were not evenly distributed between the groups; 14 of the midazolam group versus one of the diazepam group had febrile convulsions, and 10 of the diazepam group versus four in the midazolam group had central nervous system (CNS) infection.

McIntyre 2005 was a 40‐month, multicentre, randomised, controlled trial comparing buccal midazolam (approximately 0.5 mg/kg) with rectal diazepam (0.5 mg/kg) as the first‐line treatment of children aged six months to 15 years, presenting to a paediatric A&E department with active seizures.The primary outcome measure was clinical cessation of the seizure within 10 minutes of drug administration, without seizure recurrence within one hour and without respiratory depression. Children with partial seizures or non‐convulsive status epilepticus were excluded from the trial. Weekly blocks of treatment of either buccal midazolam or rectal diazepam were randomly selected in each of the four participating centres. Participant demographics were similar between groups. Locally‐agreed guidelines were followed in the event of continued seizure activity after the 10‐minute period. The study evaluated 219 seizure episodes in 177 children of both sexes. Separate results were reported both for total episodes and for first presenting episodes, to minimise potential bias of children with multiple entries. In contrast with the other studies included in the previous review, children were not excluded if they had received anticonvulsant agents prior to their attendance at the A&E department.

Momen 2015 was an unblinded randomised trial of 100 children aged one month to 16 years. Inclusion criteria were children older than one month who were convulsing on arrival. The length of the ongoing seizure was not taken into account, so children with relatively short‐lived seizure may have been included. Exclusion criteria were established IV access, prior administration of rectal or nasal benzodiazepines, lack of consent or serial seizures with no recovery of consciousness. In addition, a history of serious adverse reactions to either of the study medications was an exclusion criterion. Participants were randomised to intramuscular midazolam or rectal diazepam. The main outcome measure was seizure cessation without recurrence within 60 minutes. Respiratory rate and blood pressure were also monitored.

Mpimbaza 2008 was a single‐blinded, placebo‐controlled randomised clinical trial in a paediatric emergency unit in Kampala, Uganda. The inclusion criteria were children aged three months to 12 years who presented while convulsing or who experienced a seizure that lasted more than five minutes while in the unit, and who had no documented evidence of having received intravenous diazepam or phenobarbitone in the 24 hours before presentation. Children aged less than three months or more than 12 years, who had evidence of prior treatment or whose convulsion stopped prior to treatment, were excluded. The study recruited 330 participants (note that multiple seizure types can occur in a single participant): 269 (82%) had generalised tonic‐clonic seizures, 18 had tonic seizures, 61 had focal seizures and three had myoclonic seizures. Participants were randomised by a random‐number table to 0.5 mg/kg of rectal diazepam or buccal midazolam. A placebo which was identical in volume and similar in colour was simultaneously given with the study drug. The participants were well‐balanced by age, sex, and type of seizure between the two groups. The primary outcome measure was cessation of visible seizure activity within 10 minutes, without recurrence in the subsequent hour. If the convulsion lasted longer than 10 minutes or recurred within one hour, this was considered a treatment failure and the child was given intravenous diazepam. Secondary outcome measures were the proportion with cessation of convulsions within 10 minutes, the proportion with recurrence in the next hour and within 24 hours of initial control, and time to recurrence within these periods.

Shah 2005 was a prospective controlled quasi‐randomised study of the treatment of acute seizures in children in a tertiary general hospital in Mumbai, India, including children presenting to the emergency department and those who were already admitted to the ward or intensive care unit (ICU). The study enrolled 115 children with an acute seizure (definition unclear) over a one‐year period in a single centre. Those who had already had treatment for the seizure were excluded. Participants who already had intravenous access were treated with 0.2 mg/kg of intravenous diazepam. Those without were randomised to treatment with 0.2 mg/kg of intramuscular midazolam or to the establishment of intravenous access and treatment with intravenous diazepam. Sixty‐three children had generalised tonic‐clonic seizures, and were equally divided between the two treatment groups. Of the remaining participants, 47 had focal, four had tonic and one had clonic seizures. Outcome measures were mean time to cessation of seizures and the presence of adverse effects. Those who did not respond after five minutes were treated with other "anticonvulsants" (not specified).

Sreenath 2010 was a randomised controlled study of the management of convulsive status epilepticus (defined as continuous convulsive activity lasting for five minutes or more) in children presenting to a single centre in North India. Exclusion criteria were treatment with any anti‐epileptic medication in the preceding four weeks, acute head trauma, history of poisoning and jaundice, suspected renal failure, or diarrhoea presenting with seizures. The study randomised 178 children aged one to 12 years by computer‐generated table to receive lorazepam 0.1 mg/kg or diazepam 0.2 mg/kg. If intravenous access was not present, the drug was given rectally at the same dose. If the seizures recurred within an undefined time frame, a second dose of the same drug was given. The diazepam group were given a loading dose of 18 mg/kg of phenytoin after 15 to 30 minutes, regardless of whether the seizure had recurred. Sixty‐three per cent had generalised tonic‐clonic seizures, while the rest were tonic (10%), clonic (10%), myoclonic (0.5%), simple partial (2%), complex partial (10%), and partial with secondary generalisation (4%). The majority were therefore generalised convulsive or motor seizures. The primary outcome measure was cessation of seizure activity, with treatment considered successful if this occurred within 10 minutes of the first intervention and without recurrence over the next 18 hours. Secondary outcomes were time taken for initial (presenting) convulsion to stop after administration of the first drug, the number of doses of study drug required to treat the initial convulsion, the use of an additional anti‐epileptic drug, the total number of seizures occurring in the first 18 hours following administration of the study drug, the development of respiratory depression, the number of participants requiring transfer to ICU for mechanical ventilation, and the number of participants requiring cross‐over to an alternative regimen (i.e. from diazepam to lorazepam and vice versa).

Talukdar 2009 was a prospective randomised study of 120 children who attended the paediatric emergency department of a Delhi hospital with a seizure, the length of which was not defined. Those with myoclonic, absence and atonic seizures were excluded. The mean age of the participants was 3.2 years, with 73.3% under five years of age and 53.3% under one year. Seventy‐four per cent of the children presented with GTC, while the rest were complex partial or tonic seizures. The groups were not significantly different in age, sex, seizure type or underlying aetiology. Participants were allocated by a random‐number table to 0.2 mg/kg of buccal midazolam or 0.3 mg/kg of intravenous diazepam. The primary outcome measure was cessation of all motor activity within or by five minutes of administration of the drug. The drug response was further analysed as treatment initiation time (time from noting seizure to drug administration), drug effect time (time from drug administration to effect) and total controlling time, a combination of the previous two.

Excluded studies

We excluded 20 studies from the review for the following reasons (see Characteristics of excluded studies for further information):

• The study evaluated refractory status epilepticus (i.e. children who have failed a first or second treatment for status epilepticus) rather than acute status epilepticus (Agarwal 2007; Arpita 2014; Mahmoudian 2006; Mehta 2007; Mittal 2014; Rosati 2016; Singhi 2002).

• The study was not randomised or did not have a control group, or both (Kutlu 2003; Morton 2007; Qureshi 2002).

• Fewer than 70% of included participants had GTCs (or it was unclear how many participants had GTCs) (Bhattacharyya 2006; Scott 1999; Tonekaboni 2012).

• The study examined drug management for the long‐term prevention of recurring febrile seizures (Camfield 1980; Heckmatt 1976; Strengell 2009) or clusters of seizures (Cereghino 1998), rather than management of acute convulsions.

• The study was not conducted in a hospital setting (Holsti 2010; Silbergleit 2012).

• The study was published only as a conference abstract and we could not contact the authors for additional information to assess eligibility (McCormick 1999).

Allocation

Fourteen studies reported adequate methods for random sequence generation (e.g. computer‐generated randomisation, block randomisation, etc.) and we judged them to be at low risk of bias. Four studies (Appleton 1995; Baysun 2005; Fişgin 2002; Shah 2005) reported inadequate methods for random sequence generation and we judged these to be at high risk of bias. Three of these studies (Appleton 1995; Baysun 2005; Fişgin 2002) used an alternating ('odd' and 'even') day approach to randomisation and one study (Shah 2005) was partially randomised, including patients attending the emergency department and inpatients in the paediatric ward or intensive care unit. Those who already had intravenous access were selected to receive intravenous diazepam, and those without were further randomised to intramuscular midazolam or intravenous diazepam. For those who were randomised, it was unclear how randomisation was performed.

Regarding allocation concealment, nine studies described adequate methods of concealment such as centralised allocation or sealed opaque envelopes and we judged them to be at low risk of bias. Six studies (Appleton 1995; Baysun 2005; Fişgin 2002; McIntyre 2005; Momen 2015; Shah 2005) did not conceal allocation and we judged them to be at high risk of bias. The remaining three studies did not mention allocation concealment and we judged them to be at unclear risk of bias (Ashrafi 2010; Chamberlain 1997; Talukdar 2009).

Blinding

One study was double‐blinded (Chamberlain 2014) and one trial had single‐blinded participants (Mpimbaza 2008). The remaining sixteen studies were unblinded, and for many of them blinding would have been impractical, due to different routes of intervention. However, given the objective nature of the main outcomes of these studies (e.g. seizure cessation), it is unlikely that the lack of blinding would affect results, so we rated all studies at low risk of bias.

Incomplete outcome data

We judged 15 studies to be at low risk of bias, since all recruited participants were included and analysed on an intention‐to‐treat basis. We judged one study (Appleton 1995) to be at high risk of bias. In this study there were a relatively large number of protocol violators (16 of 102 children, or 16% of the total study population) and these violators were excluded from the analyses. The analysis was therefore not an intention‐to‐treat analysis.

We judged two studies (Chamberlain 1997; Gathwala 2012) as being at unclear risk of bias. In Chamberlain 1997, three children who were randomised to receive diazepam were subsequently excluded, as their seizures did not persist for 10 minutes. There was also a protocol violator who was randomised to receive intravenous diazepam but received intramuscular midazolam after 25 minutes, due to unsuccessful intravenous access. This participant was excluded from the analysis and obviously would have skewed the results significantly if he/she had been included. It may have been helpful to know the response time of this child once treatment was administered, as this is an important example of the disadvantages of the intravenous route. In Gathwala 2012, three children were excluded due to difficulties obtaining intravenous access, and this may have introduced a source of bias. It could be argued that the data cannot be considered to have been analysed on an intention‐to‐treat basis, as these participants were excluded from the analysis. However, given that all routes were intravenous the effect of this is likely to have been small.

Selective reporting

Fourteen studies reported all expected and prespecified outcomes and we judged them to be at low risk of bias. We rated two studies at high risk of bias; Javadzadeh 2012 did not report seizure cessation, which we would expect to be reported, and in Mahmoudian 2004 the authors did not report the time taken to insert intravenous cannulae in the intravenous diazepam group. This would have a significant effect on the time from arrival to seizure cessation. Other studies that compared intravenous with other routes have included this information.

We judged two studies to be at unclear risk of bias. Fişgin 2002 stated that information about previous convulsions and history of anti‐epileptic medication were collected according to the Methods but not reported in the Results section. It is unlikely that this information influenced outcomes, but we are unclear why the information was not reported. Lahat 2000 defined seizure cessation in the Methods section as "successful" if seizures stopped in less than five minutes, "successful but delayed" if seizures stopped after five to 10 minutes, and "failure" if seizures had not stopped after 10 minutes. However, results seem to be presented only in terms of treatment success and failure. It is unclear if this is selective reporting of results.

Other potential sources of bias

We identified additional high risks of bias in five studies. In two studies, a high proportion of the children recruited had either cerebral malaria or meningitis, which may have impacted upon the results (Ahmad 2006; Mpimbaza 2008). In Appleton 1995 there was a large discrepancy in the two routes of administration used in the study, probably due to clinician uncertainty about the use of rectal lorazepam. This discrepancy is likely to have impacted upon results. In Chamberlain 1997, one child was enrolled in the study twice, and is represented in both groups. Due to the small numbers of children included in the study, this double‐enrolment may have impacted on the results. In Gathwala 2012, the definition of the 'seizure cessation' outcome used is different from all the other included studies, and is not an appropriate criterion for judging seizure cessation. This definition is likely to have impacted upon results.

In four studies, it was unclear whether additional bias was present. In three studies (Ashrafi 2010; Mahmoudian 2004; Sreenath 2010), one or both treatment arms showed a 100% seizure cessation rate, which is higher than expected. However, it was unclear whether these unexpected results were due to a particular element of the trial design. In Fişgin 2002, the description of the seizure type and aetiology of participating children was unclear, so we cannot be sure that the population of this study is generalisable.

We found no other biases in the remaining nine studies (Arya 2011; Baysun 2005; Chamberlain 2014; Javadzadeh 2012; Lahat 2000; McIntyre 2005; Momen 2015; Shah 2005; Talukdar 2009).

1. Lorazepam versus diazepam

Three trials recruiting 455 participants compared lorazepam to diazepam (Appleton 1995; Chamberlain 2014; Gathwala 2012). All participants in Chamberlain 2014 and Gathwala 2012 received drugs intravenously; in Appleton 1995, children received the drugs either intravenously or rectally (where intravenous access was not possible). As the route of administration in this trial was not randomised, we test the route of administration for lorazepam and diazepam by subgroup analyses rather than by separate comparisons.

2. Intranasal lorazepam versus intramuscular paraldehyde

One trial (Ahmad 2006), recruiting 160 participants, compared intranasal lorazepam to intramuscular paraldehyde.

3. Intravenous lorazepam versus intravenous diazepam/intravenous phenytoin combination

One trial (Sreenath 2010), recruiting 178 participants, compared intravenous lorazepam to intravenous diazepam/intravenous phenytoin combination.

4. Intravenous lorazepam versus intranasal lorazepam

One trial (Arya 2011), recruiting 141 participants, compared intravenous lorazepam to intranasal lorazepam. Results are presented for the subgroup of 58 participants with GTC.

5. Buccal midazolam versus rectal diazepam

Four trials, recruiting 648 participants, compared buccal midazolam to rectal diazepam (Ashrafi 2010; Baysun 2005; McIntyre 2005; Mpimbaza 2008). One trial (177 participants; McIntyre 2005) reported on 219 seizure episodes; in other words, the same child was randomised and treated for multiple seizures in the same trial. Results are not available at the participant level so results reported for McIntyre 2005 are by episode. This is a limitation, as meta‐analysis assumes independence between measurements, and more than one treated seizure per child would not be statistically independent. A result of ignoring this unit‐of‐analysis issue could be overoptimistic confidence intervals.

6. Buccal midazolam versus intravenous diazepam

One trial (Talukdar 2009), recruiting 120 participants, compared buccal midazolam to intravenous diazepam.

7. Intranasal midazolam versus intravenous diazepam

Three trials, recruiting 174 participants, compared intranasal midazolam to intravenous diazepam (Javadzadeh 2012; Lahat 2000; Mahmoudian 2004).

8. Intranasal midazolam and rectal diazepam

One trial (Fişgin 2002), recruiting 45 participants, compared intranasal midazolam and rectal diazepam.

9. Intramuscular midazolam versus intravenous diazepam

Two trials, recruiting 138 participants, compared intramuscular midazolam to intravenous diazepam (Chamberlain 1997; Shah 2005). Shah 2005 included some non‐randomised participants who received intravenous diazepam as they already had intravenous access; only randomised children are reported in this review. Chamberlain 1997 reported that one child was enrolled in the study twice, so is represented in both groups. It was not possible to identify this child in the reported results so we note that results presented in this section must be interpreted with caution, due to the representation of this child in both treatment groups.

10. Intramuscular midazolam versus rectal diazepam

One trial (Momen 2015), recruiting 100 participants, compared intramuscular midazolam to rectal diazepam.

11. Intravenous midazolam versus intravenous diazepam

One trial (Gathwala 2012), recruiting 80 participants, compared intravenous midazolam to intravenous diazepam.

12. Intravenous midazolam versus intravenous lorazepam

One trial (Gathwala 2012), recruiting 80 participants, compared Intravenous midazolam to intravenous lorazepam.