Results of the search

The searches yielded 188 records, of which 154 records remained after removing 34 duplicates. We assessed that eight records were potentially eligible for inclusion after title and abstract screening. Following the screening of the full‐text articles of the eight remaining records, we excluded four of these records (Li 1996; Sachdeo 1990; Theodore 1990; Wilder 1991), and listed the reasons for exclusion in Characteristics of excluded studies. We thus judged that the remaining four records were eligible for inclusion in the review (Binelli 1999; Bourgeois 1993; Leppik 1991; Theodore 1991).

See Figure 1 for the study flow selection diagram (Moher 2009).

Figure 1

---

---

Flow chart of study selection

Included studies

See Characteristics of included studies.

Four studies met our inclusion criteria, with a total of 236 participants (Binelli 1999; Bourgeois 1993; Leppik 1991; Theodore 1991). Although all four studies were randomised, double‐blind, placebo‐controlled trials, they otherwise largely varied in study design. Two of the studies were parallel‐group trials, one was a two‐period cross‐over trial, and the other was a three‐period cross‐over trial. They utilised varying felbamate doses, varying treatment periods, differing baseline antiepileptic drugs, and differing methodology for assessment of efficacy. Participants randomised in the four studies also had differing seizure frequency during baseline.

Binelli 1999 was a parallel‐group trial including an eight‐week baseline period, followed by an undefined titration period where the dose was gradually increased to the maximum tolerated dose and was then maintained over an eight‐week maintenance phase. Participants were required to have at least eight seizures during the eight‐week baseline period. Additionally, participants' concomitant antiepileptic therapy could be made up of no more than two of the following drugs: carbamazepine, γ‐vinyl‐GABA, lamotrigine, gabapentin and benzodiazepine. However, there was no description of which drugs the felbamate group or the placebo group received. No other inclusion criteria and exclusion criteria were mentioned in the study. The study included a total of 83 participants who were randomised to one of two treatment groups, up to 3600 mg/day of felbamate or placebo.

Likewise, Bourgeois 1993 was a parallel‐group trial, which consisted of a four‐week baseline period as well as a four‐week treatment period, inclusive of a three‐day titration period. Specifically, participants underwent a routine evaluation for epilepsy surgery at the end of the four‐week baseline period. The treatment period immediately followed the surgical evaluation period and consisted of eight hospital days and 21 outpatient days. The study utilised strict diagnosis requiring video/electroencephalogram (EEG)‐confirmed focal‐onset seizures. The study publication defined both inclusion and exclusion criteria. To be eligible for inclusion, seizure frequency could not exceed an average of four complex focal‐onset seizures or feature more than one secondarily generalised seizure per day, during the last three days of the surgical evaluation. Moreover, participants were also required to have a minimum average of one seizure per day for the last three days of the surgical evaluation period. Participants had to be at least 18 years of age and have a body weight of at least 40 kg. A total of 64 participants were randomised into the study by Bourgeois 1993. Participants were randomised to receive either 3600 mg/day of felbamate (or their maximum tolerated dose) or placebo.

Importantly, during the surgical evaluation period of the Bourgeois 1993 study, standard antiepileptic drugs were reduced or discontinued. Throughout the subsequent hospitalisation period (the first eight days of the treatment period), the participants continued with the same antiepileptic drug regimen present on the last day of the surgical evaluation period. Once participants entered the second phase of the treatment period (the 21 outpatient days), participants who were on less than the full baseline dosage of one standard antiepileptic drug returned to their pre‐surgical evaluation dosage of one antiepileptic drug. If participants were instead on a reduced dosage of two antiepileptic drugs, the dosage of one was restored to its pre‐surgical evaluation dose. All participants in the felbamate group were treated at the maximum dosage. Again, there was no description of which baseline antiepileptic drugs were being taken, or which participants on which antiepileptic drugs were the best responders to felbamate.

In contrast to the two previous studies, Leppik 1991 was a two‐period cross‐over study. The study included an eight‐week baseline period, an 8‐ to 10‐day titration period and 10‐week treatment period with an additional three‐week wash‐out period. Diagnosis of epilepsy was based on the observation of at least one ictal event by trained personnel and was supported by EEG. The inclusion criteria (participants required to have four or more focal seizures per month with no more than 20 subsequent seizure‐free days despite stable plasma concentrations of both phenytoin and carbamazepine) and exclusion criteria (people with medical conditions other than epilepsy, non‐compliant or unable to accurately report seizures) were well described. The concomitant antiepileptic drugs were phenytoin and carbamazepine. The trial included a total of 59 participants who were randomised to receive either add‐on felbamate (up to a maximum dose of 2600 mg/day) or placebo. Seizure data from the final eight weeks of the first and second treatment periods were used in the efficacy analyses.

Theodore 1991 was also a cross‐over study but instead consisted of three treatment periods. Specifically, there were four treatment sequences to which the 30 included participants could be allocated: felbamate‐placebo‐felbamate; felbamate‐placebo‐placebo; placebo‐felbamate‐placebo; placebo‐felbamate‐felbamate. The treatments were administered across alternating titration and analysis periods, each lasting two weeks, and were preceded by a three‐week baseline period. During the felbamate treatment periods, participants were titrated up to a target dose of 3000 mg/day of felbamate.The diagnosis of focal epilepsy was based on clinical (the observation of focal seizures with or without secondary generalisation), EEG (onset in one cortical region), and imaging (either focal imaging abnormality or normal scan) criteria. People who had at least six seizures during the baseline period, with at least one seizure occurring per individual week, were eligible for inclusion in the trial. The exclusion criteria (acquired by contacting the first study author) specified that people with treatable causes of seizures, metastatic tumours except skin cancer, progressive neurological disorders, other serious medical or psychiatric disorders, or a history of generalised tonic‐clonic status epilepticus were to be excluded from the trial.

Of note, Theodore 1991 also included a stabilisation period, prior to the baseline period. During this period, participants' antiepileptic drug regimens were altered or discontinued, such that participants were only receiving carbamazepine for the duration of the study. There was no description of mean baseline seizure frequencies. The mean seizure frequencies during baseline were not reported and were unavailable through contacting the first author of the study.

Excluded studies

See Characteristics of excluded studies.

All four excluded studies investigated the long‐term use of felbamate but were not placebo‐controlled, and, therefore, did not meet the inclusion criteria (Li 1996; Sachdeo 1990; Theodore 1990; Wilder 1991). Notably, two of the studies also did not investigate felbamate as an add‐on therapy but instead administered it as a monotherapy (Sachdeo 1990; Wilder 1991). Another of the excluded studies was an open‐label, long‐term extension study (Theodore 1990), which we recognised was a continuation of the cross‐over trial by Theodore 1991.

Allocation

We awarded two of the included studies, Binelli 1999 and Leppik 1991, unclear risk of selection bias for both random sequence generation and allocation concealment. The two studies failed to describe the method used for either randomisation or allocation concealment. In contrast, we awarded the studies by Bourgeois 1993 and Theodore 1991 low risk of selection bias for both domains. The study by Bourgeois 1993 most likely used a permuted block randomisation sequence and participants were allocated to their respective treatment groups by personnel who were separate from the clinical sites. In Theodore 1991, a National Institutes of Health (NIH) statistician generated randomisation schedules and the pharmacy was then responsible for subsequent treatment allocation, rather than study personnel.

Blinding

Binelli 1999 was described as being double‐blind; however, the study publication failed to provide specific details about how the blinding was achieved. We thus assessed that this study was at unclear risk of performance and detection bias. In contrast, correspondence with the authors of Bourgeois 1993 confirmed that participants and study personnel were blinded by using identical packaging for both treatment groups. Similarly, Leppik 1991 used matching placebo which would ensure the effective blinding of participants and study personnel. However, the study publication for Leppik 1991 also stated that all study medication was prepared under the supervision of an unblinded pharmacist. Although this suggests that the blinding was broken, it is necessary for the pharmacist to be unblinded to ensure that study kits are correctly packaged. We therefore agreed that blinding had most likely been effectively imposed and maintained. Finally, Theodore 1991 clarified through personal correspondence that none of the participants, physicians, nurses, social workers, or other study personnel knew what treatment participants were being given, at any time. We therefore judged these three studies to be at low risk of performance and detection bias (Bourgeois 1993; Leppik 1991; Theodore 1991).

Incomplete outcome data

Twelve out of the 83 (14.5%) randomised participants failed to complete the study by Binelli 1999. The study authors did not conduct an intention‐to‐treat analysis to compensate for the loss of data. We therefore judged that the study was at high risk of attrition bias.

In Leppik 1991, three of the 59 (5.1%) randomised participants did not complete the trial. Theodore 1991 featured a similar attrition rate with only two of the 30 (6.7%) randomised participants failing to complete the trial. Again, we included only the participants who completed the trial in the analysis, thus indicating that intention‐to‐treat analysis was not performed. We are, however, uncertain about whether the data from a small proportion participants would affect the overall conclusions of the trial. As a result, we have awarded both studies unclear risk of attrition bias, as opposed to high risk.

Notably, Bourgeois 1993 was the only included study to perform intention‐to‐treat analysis. In this study, three of the 64 (4.7%) randomised participants failed to complete the trial. However, their data were incorporated into subsequent statistical analyses. We thus judged that this study was associated with low risk of attrition bias.

Selective reporting

Two of the studies fully reported the results for all of the outcomes predefined in their respective methods sections (Bourgeois 1993; Leppik 1991). We thus judged that both studies were at low risk of reporting bias.

In contrast, the study publication for Binelli 1999 did not describe any intended outcomes in the methods section. As a result, we could not determine whether they had reported all intended outcomes. One of the cross‐over studies did report all outcomes (Theodore 1991). However, Theodore 1991 only reported the primary efficacy outcome, seizure frequency, for the felbamate treatment periods and failed to disclose seizure frequency during the placebo‐control treatment periods. Moreover, both Binelli 1999 and Theodore 1991 also failed to provide details of adverse effects that occurred during treatment with add‐on placebo. We consequently assessed that both studies were at unclear risk of reporting bias.

Other potential sources of bias

We were unable to determine whether there were any other potential sources of bias associated with the study by Binelli 1999 because of the poor reporting of the trial and the lack of methodological details provided. We thus suspect that there could potentially be other sources of bias. We consequently awarded the study an unclear risk of other bias.

We also judged that Bourgeois 1993 was at unclear risk of other bias. During the surgical evaluation period, participants' antiepileptic drug regimens were reduced or discontinued. Then, during the outpatients phase, the dose of one antiepileptic drug, taken as part of the participants' regimen, was subsequently increased to the dosage they were taking prior to the surgical evaluation. Altering the dosage of concomitant antiepileptic drugs would likely influence participants' seizure control but could also affect their responsiveness to the antiepileptic drug being trialled, felbamate. As a result, it is not clear whether the result being reported is due to the intervention or due to the alterations made to the concomitant antiepileptic drugs. Most studies require that participants must be on a stable drug regimen for at least one month prior to the study, making this study very unusual in design.

We did not detect any other potential sources of bias for the two remaining studies (Leppik 1991; Theodore 1991). We thus awarded these two studies low risk of other bias.

50% or greater reduction in seizure frequency

Only one study, involving 83 randomised participants, reported the primary outcome. Binelli 1999 reported that 38% of participants allocated to felbamate treatment experienced a 50% or greater reduction in seizure frequency. Moreover, 11% of these participants had complete cessation of seizures. The study authors did not, however, specify the number of placebo‐randomised participants who attained either 50% or greater seizure reduction or complete cessation of seizures.

Absolute or percentage reduction in seizure frequency

Three of the included studies (Binelli 1999; Leppik 1991; Theodore 1991), consisting of 172 participants, reported this outcome.

Binelli 1999 reported that participants randomised to add‐on felbamate, on average, experienced a 35.8% reduction in seizure frequency during the maintenance period. In contrast, participants randomised to the placebo group experienced a 3.3% average increase in seizure frequency. The study authors did not provide any information regarding variability (i.e. standard deviation or confidence intervals) but did report that the difference in percentage seizure reduction was significant (P = 0.0005).

Leppik 1991 reported that the mean seizure frequency during the final eight weeks of the treatment period for participants randomised to add‐on felbamate was 34.4 seizures per eight weeks compared to a mean seizure frequency of 40.2 seizures per eight weeks for those randomised to placebo. Both absolute reduction (felbamate: 4.95 ± 24.55, placebo: ‐0.36 ± 27.19, P = 0.046), and percentage reduction in seizure frequency (felbamate: 4.24 ± 55.61, placebo: ‐19.14 ± 79.70, P = 0.018) were significantly greater with add‐on felbamate than with add‐on placebo. Leppik 1991 performed an additional analysis using the data collected across the 10 weeks of each treatment period and reported that this analysis revealed similar results.

Theodore 1991 reported an average 14% reduction in seizure frequency during treatment with add‐on felbamate. Although the study publication did specifically state that the percentage reduction in seizure frequency was not significantly different between the two treatments, felbamate and placebo, they did not provide the mean percentage change in seizure frequency during the placebo treatment period. As a result, the study only partially reported the outcome.

Bourgeois 1993 did not report this outcome.

Treatment withdrawal

All four included studies (Binelli 1999; Bourgeois 1993; Leppik 1991; Theodore 1991), involving all 236 randomised participants, reported treatment withdrawal.

Binelli 1999 reported that four participants randomised to felbamate withdrew from treatment. In two cases, treatment discontinuation was caused by adverse effects (diplopia in one case, asthenia and collapse in the other). One participant died from the consequences of a seizure, whilst the fourth withdrew consent. Eight of the participants randomised to add‐on placebo did not complete the study; however, the study authors did not provide any reasons.

Bourgeois 1993 reported that two participants in the felbamate group withdrew from treatment due to adverse effects. One participant in the placebo group withdrew consent.

Leppik 1991 reported that three participants withdrew from treatment during the felbamate period because of diplopia, nausea and vomiting, and fever with malaise, respectively. Theodore 1991 reported that two participants left the study during the felbamate period, one owing to seizure exacerbation and the other due to hyponatraemia, which might specifically relate to the use of carbamazepine. Notably, no participants withdrew from treatment during the placebo period during either study.

Overall, only one of the four included studies reported a higher treatment withdrawal rate amongst participants receiving placebo (Binelli 1999). The other three studies all reported a higher treatment withdrawal rate for participants receiving add‐on felbamate. Of note, the treatment withdrawal rate for participants randomised to felbamate remained below 10% in all four studies. The most common reason for treatment withdrawal amongst participants receiving add‐on felbamate was due to adverse effects.

Adverse effects

All four included studies (Binelli 1999; Bourgeois 1993; Leppik 1991; Theodore 1991), involving all 236 randomised participants, reported adverse effects.

Binelli 1999 reported 26 adverse effects in the group of participants treated with felbamate. The adverse effects were central nervous system events such as headache, dizziness, ataxia, diplopia, confusion, depression, sedation, and paraesthesia, and gastrointestinal tract events such as stomach pain, nausea, vomiting and loss of appetite. Only 4.9% (2/41) discontinued the treatment due to the occurrence of adverse effects. In the group of participants treated with felbamate, significant weight loss occurred (mean reduction from 72.9 kg to 71.4 kg). In 19.5% (8/41) of participants, the decrease in weight was between 5 kg to 7 kg. This weight loss was justifiable in two cases because of gastrointestinal adverse effects. Of the participants treated with felbamate, 2.4% (1/41) suffered a modest and transient reduction in the value of leukocytes. There was no description of adverse effects in the group of participants treated with placebo.

Bourgeois 1993 reported that the most commonly occurring adverse effect in both treatment groups was headache (40% (12/30) felbamate and 12% (4/34) placebo). Other commonly occurring adverse effects in the felbamate group were insomnia (37% (11/30)), nausea (37% (11/30)), dizziness (23% (7/30)), fatigue (20% (6/30)), constipation (20% (6/30)), anorexia (20% (6/30)), dyspepsia (17% (5/30)), anxiety (13% (4/30)), and vomiting (13% (4/30)). The most common adverse effects in the placebo group were dizziness (15% (5/34)), dyspepsia (9% (3/34)), somnolence (9% (3/34)), insomnia (6% (2/34)), fatigue (6% (2/34)), anxiety (6% (2/34)), nausea (3% (1/34)), constipation (3% (1/34)), and vomiting (3% (1/34)). Only one participant in the felbamate group had a severe adverse effect: stupor and confusion (3% (1/30)). Two participants in the felbamate group (7% (2/30)) failed to complete the trial due to adverse effects.

Leppik 1991 reported that the most frequent adverse effects were in the central nervous system and gastrointestinal tract, of which headache (36% (21/59) felbamate and 3% (2/59) placebo), dizziness (36% felbamate (21/59) and 5% (3/59) placebo), diplopia (36% (21/59) felbamate and 2% (1/59) placebo), blurred vision (22% (13/59) felbamate and 5% (3/59) placebo), ataxia (32% (19/59) felbamate and 2% (1/59) placebo), nausea (39% (23/59) felbamate and 7% (4/59) placebo), and vomiting (25% (15/59) felbamate and 3% (2/59) placebo) were noted.

According to Theodore 1991, adverse effects experienced by participants included headache (87% (26/30)), nausea (57% (17/30)), dizziness (50% (15/30)), diplopia (33% (10/30)), vomiting (33% (10/30)), blurred vision (30% (9/30)), fatigue (17% (5/30)), and poor balance (10% (3/30)). Notably, the adverse effects listed account for all participants and do not differentiate between felbamate and placebo cross‐over periods. The study did, however, state that adverse effects were reported more during treatment periods when felbamate dosage was either steady or was being increased compared to periods of either no felbamate or when felbamate dosage was being tapered. Only nausea, double vision and blurred vision were significantly associated with periods during which felbamate was administered. Importantly, felbamate led to a significant decrease in both blood urea nitrogen and white blood count.

Quality of life

Bourgeois 1993 was the only study to describe quality of life. During the four‐week outpatient baseline period, the study authors obtained each participant's vital signs and administered the Short Neuropsychological Test. For those who had the Short Neuropsychological Test and completed the treatment period, motor skills and memory skills remained the same or were improved. The study did not, however, provide any detailed data for the Short Neuropsychological Test, or specify whether there was any significant difference in outcome between the two treatment groups.