• We are sorry, but NCBI web applications do not support your browser and may not function properly. More information

PubMed Health. A service of the National Library of Medicine, National Institutes of Health.

McDonagh M, Peterson K, Lee N, et al. Drug Class Review: Antiepileptic Drugs for Indications Other Than Epilepsy: Final Report Update 2 [Internet]. Portland (OR): Oregon Health & Science University; 2008 Oct.

Results

Overview

Our literature searches identified 2308 new citations for Update 2: 540 from the Cochrane Central Register of Controlled Trials, 25 from the Cochrane Database of Systematic Reviews, 1254 from Medline, 441 from PsychINFO, and 48 from hand-searching. We received no new pharmaceutical company dossier submissions for Update 2. Figure 1 shows results of our study selection process for Update 2. Appendix F lists the excluded studies.

Figure 1

Figure 1

Results of literature search (Update 2)

Summary of Findings

Bipolar Disorder

Stabilization of acute manic/mixed episodes

  • Evidence supports the use of immediate- and extended-release forms of carbamazepine and valproate for stabilization of acute manic/mixed episodes.
    • Efficacy of older forms of carbamazepine and valproate was comparable to lithium, in that 27% to 62% of all patients had 50% or greater improvements in symptoms (“response”).
    • In more-recent placebo-controlled trials, up to 61% of patients responded after treatment with the newest extended-release forms of either carbamazepine (Equetro®) or valproate (Depakote ER®) compared with only up to 34% of placebo-treated patients.
    • Consistent evidence of differences between the 2 antiepileptic drugs is lacking.
  • Available evidence largely does not support use of phenytoin, gabapentin, lamotrigine, oxcarbazepine, or topiramate for stabilization of acute manic/mixed episodes.
  • No evidence was found regarding the efficacy of other antiepileptic drugs for acute manic/mixed episodes.

Maintenance treatment of manic/mixed episodes

  • Evidence supports the use of lamotrigine and older forms of carbamazepine and valproate as maintenance treatment in patients whose most recent episode was manic/mixed.
    • Valproate reduced odds of any relapse by half compared with placebo and had comparable odds of any relapse compared with lithium.
    • Across most of numerous small trials comparing carbamazepine with lithium, differences in relapse outcomes did not reach statistical significance, but their trends generally favored lithium.
    • Compared with placebo, lamotrigine and lithium both improved median time to intervention for recurrence of any mood episode from 85 days to 141 and 292 days, respectively. Differences between lamotrigine and lithium did not reach statistical significance (P=0.46).
  • Available evidence largely does not support use of gabapentin, oxcarbazepine, or phenytoin for maintenance treatment of manic/mixed episodes.
  • No evidence was found regarding the efficacy of other antiepileptic drugs for maintenance treatment of manic/mixed episodes.

Rapid cycling

  • Both valproate and lithium prevented relapse for 20 months in just over half of 60 patients with rapid cycling bipolar disorder.
  • Lamotrigine consistently did not significantly extend median time to intervention (pharmacotherapy or electroconvulsive therapy) compared with placebo across 2 trials.
  • No evidence was found regarding the efficacy of other antiepileptic drugs for rapid cycling bipolar disorder.

Bipolar depression

  • Evidence supports a benefit of lamotrigine monotherapy in treating acute bipolar depression over 7 to 10 weeks. The mean response rate based on the MADRS was 51% for lamotrigine and 41% for placebo, with a pooled number needed to treat of 13. Benefit was also found based on the mean change in depression scale score.
  • Evidence does not support a difference in response rates with lamotrigine compared with olanzapine/fluoxetine combination, citalopram, or lithium. Across the 3 studies, response rates with lamotrigine ranged from 45% with adjunctive treatment to 67.5% with monotherapy. The combination of olanzapine/fluoxetine was found superior on some other measures, but not all.
  • Rates of switch into hypomania with lamotrigine as monotherapy or adjunctive therapy ranged from 1% to 8% (weighted mean 2.3%) compared with 1.9% with lithium, 1% with citalopram, 4% with olanzapine/fluoxetine, and 2% (weighted mean) with placebo.
  • In maintenance of response to lamotrigine in patients with bipolar depression, lamotrigine and lithium were similar in time to intervention for any mood episode (for lamotrigine, 200 days; for lithium, 170 days; P=0.915), and both were superior to placebo.
  • Valproate has mixed evidence for treatment of acute bipolar depression, in that valproate was found superior to placebo in mean change on depression scale scores but not in response or remission rates. Limited evidence suggests valproate was not superior to placebo in preventing relapse of depressive symptoms.
  • Topiramate and extended-release bupropion resulted in similar improvements in a single small study comparing the drugs as adjunctive therapies to mood stabilizers in patients with bipolar disorder whose most recent episode was depression.
  • Limited evidence finds that lamotrigine is similar to tranylcypromine, risperidone, and inositol in response or recovery rates among patients with treatment resistant bipolar depression.

Fibromyalgia

  • Using a definition of 30% or more reduction in pain, pregabalin showed statistically significantly greater rates of response compared with placebo for 300, 450 and 600 mg per day (pooled estimate of relative risk 1.39; 95% CI, 1.26 to 1.53; number needed to treat=8). Gabapentin also showed a greater response rate than placebo (51% compared with 31%; P=0.014; our calculated number needed to treat = 5). Pregabalin 600 mg/d recorded the highest responder rate compared with placebo (30% versus 15%; P=0.0010; number needed to treat=6.62), using a definition of 50% or greater reduction in pain. The mean pain scores for gabapentin 1800 mg/d and pregabalin 450/d and 600 mg/d were statistically significantly greater than placebo at endpoint..
  • In patients who responded to pregabalin during acute treatment with a 50% or more reduction in pain, the time to loss of therapeutic response, defined as <30% reduction in pain from open-label baseline or worsening of fibromyalgia requiring alternate treatment, was statistically significantly longer with pregabalin than placebo over a 6-month period.

Migraine Prophylaxis

  • There is evidence to support the use of topiramate or valproate for migraine prophylaxis, in that these drugs reduce migraine frequency compared with placebo. Direct comparisons between the drugs are insufficient to make conclusions.
  • The evidence supporting the use of carbamazepine or gabapentin for migraine prophylaxis is weaker than the evidence base for topiramate and valproate but indicates that these drugs also may be superior to placebo in reducing headache frequency.
  • There is little evidence to support the use of lamotrigine or oxcarbazepine for migraine prophylaxis. These drugs were not found to be superior to placebo in reducing headache frequency.

Chronic Pain

Harms Associated with Antiepileptic Drugs

  • Antiepileptic drugs may be associated with an increased risk of suicidal ideation or behaviors; however, the risk associated with specific drugs, patient populations, and treatment regimens is unclear.
    • FDA analysis of placebo-controlled trials indicates that lamotrigine and topiramate are associated with statistically significant increases in risk, while valproate is not. For other drugs the evidence was weaker.
    • Evidence suggests that risk of suicide death, suicide attempts resulting in hospitalization, and suicide attempts diagnosed in the emergency department are higher with valproate than lithium. Carbamazepine has a significantly increased risk of suicide attempts resulting in hospitalization compared with lithium.
  • The risk of fracture at any site is increased by use of antiepileptic drugs. Adjusted odds ratios for any fracture in patients who used antiepileptic drugs were significantly increased with exposure to carbamazepine (1.18; 95% CI, 1.10 to 1.26), oxcarbazepine (1.14; 95% CI, 1.03 to 1.26), and valproate (1.15; 95% CI, 1.05 to 1.26). The odds ratios were nonsignificant but increased for lamotrigine (1.04; 95% CI, 0.91 to 1.19), phenytoin (1.20; 95% CI, 1.00 to 1.43), tiagabine (0.75; 95% CI, 0.40 to 1.41), and topiramate (1.39; 95% CI, 0.99 to 1.96) compared with no exposure.
  • The relative risk of Stevens-Johnson syndrome or toxic epidermal necrolysis for < 8 weeks of use was 91 (26 to infinity) for phenytoin, 120 (34 to infinity) for carbamazepine, and 24 (5.9 to infinity) for valproate. The numbers for lamotrigine were too small for meaningful analysis.
  • In 2 case-control studies the risk of agranulocytosis associated with recent use of carbamazepine gave odds ratios of 11.0 (95% CI, 1.2 to 102.6) and 10.3 (95% CI, 2.0 to 101.0). In both studies the risk with phenytoin was low and not statistically significant, and in 1 study the risk associated with valproate was 18.2 (odds ratio; 95% CI, 2.5 to infinity). With both drugs the numbers of cases were too low for precise estimates.
  • Exposure to older antiepileptic drugs during the first trimester of pregnancy is associated with an increase in risk of birth defects compared with the general population, 4%–10% compared with 2%–5%. Risks associated with newer drugs are less clear.
    • Antiepileptic drug monotherapy is associated with somewhat lower risk compared with antiepileptic drug polytherapy.
    • Valproate was associated with a higher risk compared with carbamazepine, lamotrigine, or all other antiepileptic drugs combined with odds ratios of 2 to 4.
    • Risk of birth defects was similar between carbamazepine and phenytoin, oxcarbazepine, carbamazepine and phenytoin, and lamotrigine and carbamazepine.
    • Evidence suggests a dose-dependent relationship with valproate doses of 800 to 1000 mg/d associated with higher risk. Lamotrigine doses larger than 200 mg/d were associated with risk approaching that of valproate 1000 mg/d in one study, but no association was found in another study.
    • Gabapentin, levetiraetam, and topiramate have only very limited evidence available and conclusions cannot be drawn.
  • In short-term monotherapy trials, topiramate and carbamazepine resulted in significantly greater rates of overall reports of adverse events than placebo and carbamazepine. Valproate and topiramate had higher rates of withdrawal than placebo.
    • In head-to-head comparisons, only carbamazepine had significantly higher rates of adverse events than valproate. No difference was found between topiramate and valproate. Topiramate was not directly compared with carbamazepine.
  • Comparisons of rates of specific adverse effects:
    • Diarrhea: Lamotrigine (2 trials), but not valproate (1 trial), was significantly less likely than lithium to be associated with diarrhea (pooled odds ratio 0.30; 95% CI, 0.14 to 0.59).
    • Tremor: Lamotrigine (1 trial, odds ratio 0.28; 95% CI, 0.11 to 0.68) and carbamazepine (2 trials, odds ratio 0.00; 95% CI, 0.0 to 0.30), but not valproate (1 trial), were associated with significantly lower odds of tremor than lithium. Valproate (1 trial), but not lamotrigine (1 trial), was associated with significantly higher odds of tremor than placebo (odds ratio 4.76; 95% CI, 2.38 to 10.26).
    • Headache: Lamotrigine (4 trials), but not carbamazepine (1 trial) or gabapentin (1 trial), was more likely than placebo to be associated with headache (odds ratio 1.59; 95% CI, 1.14 to 2.25).
    • Nausea: Carbamazepine (2 trials), but not divalproex (1 trial) or lamotrigine (2 trials), was more likely than placebo to be associated with nausea (odds ratio 5.16; 95% CI, 2.73 to 10.30).
    • Rash: Lamotrigine (2 trials), but not carbamazepine (1 trial), was associated with significantly higher odds of rash relative to placebo (odds ratio 2.23; 95% CI, 1.06 to 5.28).
    • Somnolence: Carbamazepine (2 trials), but not gabapentin (1 trial) or lamotrigine (3 trials), was more likely than placebo to be associated with somnolence (odds ratio 2.71; 95% CI, 1.48 to 5.36).
    • Weight change: Only valproate was reported to cause weight gain as a treatment-emergent adverse event (odds ratio 3.26; 95% CI, 1.36 to 9.03). Lamotrigine was associated with weight loss (mean change from baseline to 6 weeks, –0.96 kg), while gabapentin was associated with weight gain (1.83 kg among 31 evaluable patients; calculated difference, −2.79 kg; P=0.02). There were no significant differences between lamotrigine and placebo or between gabapentin and placebo. Weight change data should be interpreted with caution, since it was not based on randomized patients.
    • Analysis of reports of depression, headache, rash, somnolence, or weight gain did not show statistically significant differences between drugs.

Subgroups

  • In bipolar disorder
    • Demographic factors were not found to be associated with response to valproate or lithium.
    • Among males, response to lamotrigine may be better in patients who had fewer trials of prior medications compared with those with many previous trials. Analysis of data on females did not find this difference to be significant.
    • Response to gabapentin appeared to be better in younger patients with lower baseline weight.
    • Carbamazepine may benefit patients experiencing manic episodes more than patients experiencing mixed episodes.
    • Valproate has better efficacy than lithium for patients experiencing mixed manic episodes. The drugs were similar in patients with mania alone.
    • Patients with bipolar I disorder, recent mania, and previous psychiatric hospitalization may have a longer time to depressive relapse with valproate than lithium.
    • Patients with rapid-cycling bipolar II disorder may have better response with lamotrigine maintenance therapy than placebo. This difference was not seen in the subgroup with bipolar I disorder.
    • Patients 55 years and older did not differ from the overall trial population in their response to lamotrigine in preventing relapse of depressive symptoms or other mood episodes.
  • In fibromyalgia, pain relief appears to be independent of improvement in anxiety or mood.
  • In chronic pain and migraine prophylaxis, insufficient evidence is available for meaningful assessment of subgroups.

Detailed Assessment

Key Question 1 For adult outpatients with bipolar disorder, fibromyalgia, migraine, or chronic pain, do antiepileptic drugs differ in effectiveness?

Bipolar disorder

We found no trials of ethotoin, levetiracetam, pregabalin, tiagabine12, 13 or zonisamide in patients with bipolar disorders. A large proportion of included trials in patients with bipolar disorder were previously evaluated in a number of prior systematic reviews.12–22 However, findings from only the most recent and comprehensive systematic reviews are discussed in detail here.16, 17, 20

Manic/mixed episodes
Carbamazepine compared with valproate

We found 1 fair-quality, head-to-head trial that compared carbamazepine with valproate in 30 patients with bipolar disorder (DSM-III-R) and YMRS scores of ≥ 20 at a single center in India.23 After 4 weeks of therapy, valproate was superior to carbamazepine in the reduction of YMRS scores (−32.8 compared with −20.8 points; P=0.023). There was no statistically significant difference in rates of response (> 50% decrease in YMRS total score from baseline to endpoint) between carbamazepine (53.3%) and valproate (73.3%). These results should be considered preliminary, however, until they are confirmed in larger-scale, multicenter trials.

Valproate

For treatment of acute manic/mixed episodes using immediate- and extended-release forms of valproate, we included 5 trials that evaluated comparisons with placebo,24–28 2 with comparisons to lithium,24, 29 and 1 with a comparison to haloperidol.30 A number of trials conducted to evaluate olanzapine as an even newer alternative for treatment of bipolar disorder used valproate as a control.31–34 However, these were considered to be outside of the scope of this review. Discussion of their results can be found in the Drug Effectiveness Review Project’s systematic review of atypical antipsychotics.

Outcomes data from trials conducted through 2002 comparing valproate with placebo, lithium, or haloperidol25, 26, 28–30 were previously analyzed in a Cochrane review by Macritchie and others.16 Their findings are summarized here: Pooled results of 3 short-term trials suggest that valproate was more effective than placebo and comparable to lithium in the treatment of acute bipolar manic and mixed episodes. To assess how valproate compared to placebo and lithium, Macritchie calculated pooled relative risks using fixed-effects models of the outcome “failure to respond by end of study,” which is the inverse of “response.” The relative risk of failure to respond for valproate compared with placebo was 0.62 (95% CI, 0.51 to 0.77), and 1.05 (95% CI, 0.74 to 1.50) compared with lithium. For the meta-analysis of the comparison of valproate to lithium, Macritchie included data from a trial of 28 children, which did not meet our criterion of only adults. However, when we repeated their analysis and excluded the study in children, a similar relative risk was found (relative risk 1.16; 95% CI, 0.77 to 1.75). Macritchie did not express any serious concerns about methods of randomization, allocation concealment, blinding, or handling of withdrawals.

In addition, another trial investigated whether valproate could be as rapidly effective as a conventional antipsychotic in the initial treatment of acute psychotic mania associated with bipolar disorder.30 In this trial, 36 patients hospitalized with bipolar disorder with psychotic features were randomized to receive 6 days of treatment with oral loading dosages of valproate 20 mg/kg/d or haloperidol 0.2 mg/kg/d. Oral loading of valproate was found to be comparable to haloperidol in reducing both manic and psychotic symptoms as measured by mean changes in scores on the YMRS (−42% compared with −35%) and the on the global and subscale SAPS.

After the original studies of valproate immediate- and extended-release (included in the Macritchie review) results from a placebo-controlled trial of the newer, once-daily, extended-release form of valproate were published.24 This fair-quality trial compared valproate extended-release 3057 mg (final mean dose) with placebo in 377 adults who were hospitalized for an acute manic or mixed episode of bipolar disorder. For the protocol-specified primary efficacy endpoint of change on the SADS-C MRS, valproate extended-release produced significantly greater improvement than placebo (−11.5 points compared with −9.0; P=0.013). In addition, significantly more patients treated with valproate extended-release experienced at least a 50% improvement from baseline on SADS-C MRS than with placebo (48% compared with 34%; P=0.012), and more patients treated with valproate extended-release were in remission at endpoint (time point the patient left the study; 48% compared with 35%; P=0.015).

Carbamazepine

Monotherapy with carbamazepine extended-release was more efficacious than placebo in the acute treatment of patients with bipolar I disorder in 2 identically designed, pivotal trials (105.301 and 417.304)35, 36 included in the New Drug Application (NDA) submitted to the FDA in 2004 and summarized in an FDA review document.37 Mean final daily doses of carbamazepine extended-release were 756 mg (8.9 μg/mL) in study 105.30135 and 643 mg (mean plasma drug level not recorded) in study 417.304.36 Both trials were 3 weeks long. Compared with the placebo groups in both trials, patients in the carbamazepine extended-release groups had significantly greater improvements in mean YMRS total scores and more patients were considered responders at endpoint (the time point at with the patient left the study, Table 2). In the FDA review of the NDA, we also found results from a third, non-pivotal, failed placebo-controlled trial of carbamazepine extended-release in lithium-resistant patients with bipolar disorder that, to our knowledge, has not yet been fully published.37 Very little information was provided about this trial, except that its design was identical to the others; it involved 59 randomized patients; and there were no significant differences between carbamazepine extended-release and placebo on the primary outcomes of mean change in YMRS total score (−8.9 compared with −8.7; P=0.97).

Table 2

Table 2

Efficacy outcomes in acute treatment of bipolar disorder with carbamazepine extended-release compared with placebo

From older evidence supporting use of acute monotherapy with carbamazepine immediate-release, we included 3 trials published between 1987 and 1991 that involved comparisons with lithium (Table 3).38–40 In all 3 trials, included patients were diagnosed with bipolar disorder based on DSM-III or DSM-III-R criteria; most exhibited manic episodes. Trials were 4 to 8 weeks long and the dosage ratio between carbamazepine and lithium was approximately 1:1 in 2 of 3 trials39, 40 and 1:1.5 in the other. Methods of outcome assessment were heterogenous across trials, but there were no significant differences between carbamazepine immediate-release and lithium regardless of how outcome was measured. However, these findings should be interpreted with caution given that (1) patients assigned to the lithium groups were significantly older in the 2 largest trials,39, 40 and this may have biased their results; and (2) the other trial of only 34 patients may not have been large enough to reliably detect differences between the 2 drugs.

Table 3

Table 3

Efficacy outcomes in acute treatment of bipolar disorder with carbamazepine compared with lithium

Other miscellaneous trials of acute therapy with carbamazepine immediate-release were included; but, because their comparisons were with antipsychotics, they have limited usefulness here.41–45 Two trials indicated that carbamazepine had antimanic efficacy comparable to chlorpromazine.41, 43 Two trials supported the use of carbamazepine in 2 combination therapy situations.42, 44 Results from 1 trial indicated that carbamazepine showed superior efficacy to placebo when combined with haloperidol in patients with “excited psychoses, ” including mania.42 Results from two other trials indicated that carbamazepine and haloperidol provided similar benefit when added to lithium or other neuroleptics.44, 45

Topiramate

In a small (N=74), 6-week, open-label trial comparing topiramate with valproate, both in combination with risperidone,46 75.8% of topiramate-treated patients had a 50% or more reduction in YMRS score, which was comparable to the response rate in the valproate group, 70.7%. In the second trial, topiramate 254.7 mg/d was compared with placebo in 287 patients with bipolar I disorder experiencing a manic or mixed episode who were already on ongoing therapy with lithium or valproate.47 After 12 weeks, there was no significant difference between topiramate and placebo on the primary efficacy measure of reduction in YRMS total score (−10.1 compared with −9.6), in overall response rate, or on any other secondary efficacy outcomes as measured by the CGI-S, BPRS, MADRS, or GAS.

Results from 4 identically designed trials of topiramate monotherapy were reported in a single publication.48 All were 3 weeks long and collectively randomized 110 patients to topiramate 200 mg, 447 to topiramate 400 mg, 102 to topiramate 600 mg, 227 to lithium, and 429 to placebo. Based on intention-to-treat analyses of mean reduction in YRMS total score, the therapeutic benefit of topiramate was significantly lower than lithium and was not significantly different from placebo. YMRS mean reductions were −5.8 points for the topiramate 200-mg group, −7.9 points for the 600-mg group, and ranged from −5.1 to −8.2 points in the 400-mg groups, from −12.9 to −13.8 in the lithium groups, and from −6.4 to −8.4 in the placebo groups. Limited information was provided in an abstract format regarding a fifth trial that compared topiramate 256 mg (N=33), topiramate 512 mg (N=33), and placebo (N=31) in patients with bipolar I disorder.49 There, too, no statistically significant differences were reported for either dose of topiramate compared with placebo in the primary efficacy analysis of the mean change from baseline in YMRS total score, but no data were provided.

Lamotrigine

In the only published trial of lamotrigine for patients with bipolar I acute mania, 15 patients each were randomized to receive lamotrigine 100 mg or lithium 800 mg for 4 weeks.50 Results indicated no significant differences between lamotrigine and lithium for rate of improvement in mean MRS scores (58% compared with 58%; P=NS). However, these results should be interpreted with caution in light of the low dose of lithium.

We also identified 2 unpublished trials51, 52 of treatment of acute manic/mixed episodes with lamotrigine as either monotherapy (SCAA2008/GW609) or adjunct therapy (SCAB2009/GW610), with comparisons to lithium and placebo The only information we found about the results of these studies comes from a review of the use of lamotrigine in bipolar disorder, which stated that there were no significant differences between lamotrigine and placebo in either trial on the primary outcome of change in the 11-item MRS score based on data on file from GlaxoSmithKline.51 In addition, Appendix B of the FDA Medical Review of the NDA materials for lamotrigine in bipolar disorders provides brief summaries of studies of lamotrigine, including SCAA2008 and SCAB2009, but these were not available to us as those pages were withheld from the online report with reasons given as ”Trade Secret/Confidential.”

Gabapentin

Two studies of gabapentin involved patients with treatment-resistant symptoms of bipolar mania.53, 54 One compared gabapentin 900 to 3600 mg/d with placebo as add-on treatment in 117 patients with persistent bipolar disorder symptoms despite ongoing therapy with standard mood stabilizers.53 After 10 weeks, improvement in YMRS scores were significantly greater in the placebo group (−9.9 compared with −6.5; P=0.03), and there were no other differences between gabapentin and placebo for the remainder of efficacy outcomes. The second trial used a crossover design to compare 6-week treatment periods with gabapentin 3987 mg, lamotrigine 274 mg, and placebo monotherapies in 31 patients refractory or intolerant to prior treatments with standard mood stabilizers.54 Patients received all 3 agents sequentially, divided by 1-week washout periods. On the basis of an overall CGI score much or very much improved, response rates for gabapentin (26%) were significantly lower than for lamotrigine (52%; P=0.011) and were not significantly different from placebo (23%; P=0.700).

Phenytoin

A single trial evaluated the acute antimanic effects of phenytoin when used for 5 weeks in combination with haloperidol in patients with either bipolar I disorder, manic type (N=12), or schizoaffective disorder, manic type (N=18).55 The results were stratified by diagnosis, allowing for isolation of phenytoin’s effect in the subset of patients with bipolar I disorder. Interpretation of findings is limited by lack of information about whether or not the comparison groups were similar at baseline. At week 5, in the subset of patients with bipolar disorder there was added improvement with phenytoin compared with placebo for scores on the BPRS (23.7 compared with 34.5; P=0.01), the CGI (2.5 compared with 4.0; P=0.001), and the YMRS (9.5 compared with 19.7, P not reported).

Hypomania
Oxcarbazepine

The outcomes of treating hypomania with oxcarbazepine 1350 mg monotherapy or adjunct therapy were compared with valproate 1167 mg in 1 small, open-label, outcome assessor-blinded trial of 30 patients.56 A variety of concomitant medications were used by 53% of patients in the oxcarbazepine group and 40% in the valproate group. Twice as many patients in the oxcarbazepine group were using concomitant antidepressants (40% compared with 20%), and patients in the oxcarbazepine group were significantly younger (30 compared with 37 years; P=0.05). After 8 weeks, mean reduction in YMRS score with oxcarbazepine (−13.3 points) was comparable to that with valproate (−12.4). However, these results should be interpreted with caution, as it is unclear how the between-groups imbalances at baseline may have biased patient outcomes.

Maintenance of response: Manic/mixed episodes
Valproate

We included 8 trials of maintenance treatment comparing valproate monotherapy with placebo57–62 or lithium57, 63, 64 in patients previously experiencing acute mania. A summary of results from all but 260, 61 of the included trials was available in a good-quality systematic review and we will summarize those findings here.20 We will separately summarize the findings of the 2 remaining trials, 60, 61 one of which was carried out in patients with comorbid alcoholism.61

Collectively, 6 trials included in a review by Soares-Weiser and colleagues randomized 347 patients to valproate, 231 to lithium, and 102 to placebo and ranged from 6 to 20 months in duration. Populations were diverse across trials. Two trials enrolled only patients with bipolar I disorder.57, 64 One trial enrolled only patients with rapid-cycling bipolar disorder.58 And another trial enrolled only women with borderline personality disorder and comorbid bipolar II depression.59 To determine the efficacy of valproate in preventing relapse in patients with bipolar disorder, Soares-Weiser and colleagues combined data across trials using a fixed-effects model. Findings from these meta-analyses are reported in Table 4. All but 1 trial64 measured relapse outcomes. Although the trials were clinically heterogenous, no statistical heterogeneity was detected. Compared with placebo, valproate significantly reduced the odds of depressive, but not manic, outcomes. The effectiveness of valproate in reducing odds of all relapses was comparable to lithium. Additionally, results of a secondary analysis from one of the individual trials indicated that the comparability of valproate and lithium did not differ based on whether initial symptoms were euphoric or dysphoric.65

Table 4

Table 4

Odds ratios for relapse of bipolar disorder treated with valproate (Soares-Weiser 2007)

One trial included in the meta-analysis reported not only relapse outcomes but quality-of-life outcomes.64 This was an open-label trial that randomized 201 adults with bipolar disorder to either valproate 1504 mg or lithium 1213 mg and measured quality of life using the SF-36. At the end of 12 months no difference in quality-of-life outcomes was found between valproate and lithium. However, these analyses appeared to be based on only the 40% of patients that actually completed the study and for that reason should be interpreted with caution.

While several trials looked at monotherapy, 1 trial has evaluated potential benefits of combining valproate with lithium.60 In a trial that was not included in the meta-analysis above, 12 patients with bipolar I disorder were randomly assigned to open-label valproate plus lithium or placebo plus lithium and followed for up to 12 months. Although significantly fewer patients assigned to valproate plus lithium experienced a relapse compared with placebo plus lithium (0% compared with 71%; P=0.014), no definitive conclusions can be drawn from these results due to small sample size, lack of blinding, and significant between-group differences in mood polarity of patients at baseline.

Prevention of bipolar depression in patients treated for manic/mixed episodes

Patients with recent mania who had previously achieved response with valproate were randomized to valproate, lithium, or placebo and were followed for 1 year.57, 62 While statistically significant differences were not found between the 3 groups in the primary outcome (time to recurrence of any mood episode), the difference between valproate and lithium reached a P value of 0.06. A difference favoring valproate over lithium was also seen for time to a depressive episode, but again statistical significance was not achieved (P=0.08). Similar results were found for discontinuations due to any mood episode (mania or depression) except that valproate was found to have significantly fewer discontinuations due to depression than placebo (6% and 16%, respectively; P=0.017).

Carbamazepine

We included trials comparing carbamazepine monotherapy with lithium66–78 or placebo79 and evaluated efficacy for prophylaxis of recurrence of symptoms in bipolar disorder. Most trials involving lithium enrolled patients diagnosed with any bipolar disorder (I, II, or unspecified) and were heterogeneous with regard to duration, sample size, quality of methods, and method of outcome assessment (Table 5). Regardless of sources of heterogeneity, however, most trials indicated no significant difference between carbamazepine and lithium in preventing relapse, with their trends generally favoring lithium. The exceptions came from 2 of the 3 shortest trials, which followed patients for only 1 year; they reported nonsignificant trends favoring carbamazepine.66, 69

Table 5

Table 5

Relapse outcomes in bipolar disorder treated prophylactically with carbamazepine or lithium

In order to more precisely estimate the comparative effectiveness of carbamazepine and lithium in preventing relapse in persons with bipolar disorder, a recent good-quality Health Technology Assessment conducted by Soares-Weiser calculated odds ratios for each of a majority of these same trials66, 68–70, 80 and, where appropriate, pooled results across trials.20 Pooled analyses were stratified by whether investigators defined relapse events as hospitalizations only or as assessed changes in symptoms. Additional subgroup analyses were conducted to evaluate the potential effects of type of bipolar disorder and inclusion of patients who were randomized during an acute episode. However, interpretation of the findings from subgroup analyses was limited by small sample sizes. In the main analyses lithium was favored as the more effective agent for preventing relapse-related psychiatric hospitalizations (odds ratio 0.63; 95% CI, 0.33 to 1.2) and relapse-related changes in symptoms assessed by investigators (odds ratio 0.48; 95% CI, 0.27 to 0.84). In contrast, in a subgroup of 40 bipolar II disorder patients from 1 trial, carbamazepine tended to be more effective in preventing relapse-related hospitalizations (odds ratio 2.50; 95% CI, 0.54 to 11.62) and had an efficacy more comparable to lithium in preventing relapses as assessed by investigators (odds ratio 0.82; 95% CI, 0.24 to 2.83).80

Phenytoin

A single trial evaluated the prophylactic effects of phenytoin added to ongoing therapy of lithium, carbamazepine, valproate, or conventional antipsychotic in 23 patients with a manic type of either bipolar I disorder or schizoaffective disorder.81 Results were not stratified by diagnosis. Interpretation of findings was limited by lack of information about whether or not the comparison groups were well balanced at baseline; this raises the question of whether between-group differences in patient outcomes were due mainly to true differences in treatment effects or to significant differences among patients. After 6 months, only 30% of patients treated with phenytoin had a relapse event, compared with 61.5% on placebo; P=0.53.

Lamotrigine

We included 1 trial comparing lamotrigine for maintenance treatment of manic or hypomanic patients with bipolar I disorder with lithium and placebo.82 Of the original 349 patients enrolled in this trial, only 184 (53%) were eligible for randomization based on their completion of the 8- to 16-week run-in phase of open-label lamotrigine 100 to 200 mg and on meeting the criterion for response, defined as a CGI-S scale score of 3 or less maintained for at least 4 continuous weeks. Results indicated that lamotrigine improved median time to intervention for recurrence of any mood episode more than placebo (using survival analysis methods, 141 days compared with 85 days; P=0.02) and about as much as lithium (141 compared with 292 days; P=0.46).

Gabapentin

Adjunct treatment with gabapentin or placebo was compared in 25 patients being treated with mood stabilizers who were in clinical remission at study entry.83 After 1 year, improvements in scores on the modified version of the CGI-BP were significantly greater for gabapentin (−2.1 compared with −0.6; P=0.0046), but gabapentin did not significantly reduce time to first new episode compared with placebo (hazard ratio 1.34; CI not reported).

Oxcarbazepine

Due to methodological limitations, insufficient evidence was provided for drawing any strong conclusions about the general efficacy of oxcarbazepine in two small prophylaxis trials (N=33) of mixed populations (bipolar depression, schizoaffective psychosis, bipolar affective disorders, unipolar mania).21, 84, 85

Acute bipolar depression
Lamotrigine

We identified 5 short-term (7 to 10 weeks) completed trials comparing lamotrigine with placebo for acute treatment of bipolar depression. Only 1 has been fully published,86 although some information on the others has been published in a combined format.87 Additional details of the results of these unpublished trials are available through the GlaxoSmithKline Trials Registry and at ClinicalTrials.gov.88–91 These studies included a total of 1138 patients, primarily with bipolar I disorder, who either were depressed or had recently experienced a depressive episode. While we rated the published study fair quality, it was difficult to rate the quality of the others without complete publication. However, based on the available information we do not think that any were poor quality.

Our pooled analysis indicates that lamotrigine was superior to placebo based on response rate, with response defined as a mean change of 50% from baseline HAM-D 17 or MADRS. The pooled relative risk for response on the HAM-D 17 was 1.20 (95% CI, 1.04 to 1.39) and on the MADRS was 1.21 (95% CI, 1.04 to 1.41) (Figure 2). This analysis was based on intention-to-treat populations and used a DerSimmonian-Laird random-effects model. The corresponding numbers needed to treat for 1 additional person to have a response when treated with lamotrigine for 7 to 10 weeks were 13 and 11, respectively.

Figure 2

Figure 2

Response of bipolar depression to lamotrigine based on 50% improvement in MADRS (random-effects model)

In 2 studies86, 91 the primary outcome measure was improvement on the HAM-D 17, and power calculations were based on finding a 5-point difference in the mean change in score on the outcome measure between groups. In the other 3 studies88–90 the primary outcome measure was the mean change in MADRS score; again, power calculations were based on finding a 5-point difference on the outcome measure between groups. While none of the studies individually found a statistically significant difference in mean change on HAM-D 17, our pooled analysis indicates a significant benefit of lamotrigine, with a weighted mean difference of −0.99 (95% CI, −1.61 to −0.36). Similarly, only 1 study86 found a statistically significant difference in the mean change on the MADRS, but our pooled analysis indicates a significant benefit of lamotrigine, with a weighted mean difference of −1.11 (95% CI, −1.49 to −0.74).

There was no difference between lamotrigine and placebo in the risk of patients switching mood from depressed to manic, hypomanic, or mixed in 3 of the 5 trials (relative risk 1.21; 95% CI, 0.40 to 3.62).86, 88, 91

Three studies evaluated the short-term efficacy of lamotrigine and non-antiepileptic drugs over 7 to 12 weeks in patients with mainly bipolar depression.92–94 The largest study (N=410) compared lamotrigine monotherapy with a combination product containing olanzapine and fluoxetine over 7 weeks in patients with bipolar I depression.92 Across the 3 studies, response rates with lamotrigine ranged from 45% in adjunct treatment93 to 68 in monotherapy94 but were not statistically significantly different from placebo or other regimens. Similarly, remission rates were not statistically significantly different between lamotrigine and citalopram, olanzapine/fluoxetine, or lithium. However, some of the differences in response or remission rates were large (for example, remission rates of 35% with lamotrigine and 60% with citalopram); and, because the sample sizes were small, type II error may explain the lack of significant findings.

While response and remission rates did not identify statistically significant differences among the compared drugs, assessment of mean change in score on various symptom scales did identify some differences. The combination product olanzapine/fluoxetine was found to be statistically significantly superior to lamotrigine on CGS-S, MADRS, and YMRS final scores.92 Differences were not found between lamotrigine and either citalopram or lithium in mean change on the HAM-D 17, MADRS, or YMRS, although numerical differences were evident; a larger study would be required to clarify the significance of these differences.93, 94 In the study of lamotrigine and lithium, subgroup analysis among patients with rapid cycling did not indicate a statistically significant difference, and patients with hypomanic symptoms improved in both groups (YMRS ratings).94

Rates of switch into hypomania were low overall. Switch into treatment-emergent hypomania (reported as an adverse event by investigators) was not statistically significantly different in any of the comparisons, with a pooled rate of 3.5% for lamotrigine, 1.9% for lithium, 1% for citalopram, and 4% for olanzapine/fluoxetine. However, differences in definitions of switching indicate that this evidence should be interpreted cautiously.

Treatment-resistance in bipolar depression

Two trials compared lamotrigine with other drugs as adjunct therapy in patients whose symptoms were resistant to or who had not tolerated previous treatments given for at least 6 to 12 weeks.95, 96 Of these, 1 was a very small study (N=20),96 and the other was part of an NIH-funded study called STEP-BD, which used an equipoise randomization to allow patient preference to be taken into account.95 Neither study found statistically significant differences between lamotrigine and tranylcypromine, risperidone, or inositol on response or recovery rates, although response rates were numerically higher with tranylcypromine in the small, underpowered study (36.4% compared with 62.5%),96 and recovery rates (response maintained for 8 weeks) were larger with lamotrigine than risperidone in the other (23.8% compared with 4.6%).95 The findings of STEP-BD suggest that patients taking lamotrigine stayed on drug longer and had statistically significantly better final depression scores than patients taking inositol and better GAF scores than patients taking risperidone.

An additional study and its related extension study included patients with refractory bipolar and unipolar affective illness, comparing lamotrigine with gabapentin or placebo in a crossover design of 6 weeks each.54, 97 In this study, 8% of the subjects had unipolar disease. Lamotrigine resulted in more patients having a response, defined as much or very much improved on CGI (lamotrigine, 45% responded; gabapentin, 26%; and placebo, 19%; P=0.031). Analysis of only the first randomized drug, in order to avoid carryover effects, showed similar results. Post hoc comparisons of lamotrigine and gabapentin gave a P of 0.011.

Valproate

Two placebo-controlled trials of valproate monotherapy in patients with acute bipolar depression found statistically significant benefits for valproate on some, but not all, efficacy outcomes.98, 99 The studies defined the primary outcome measures as the mean change on the MADRS or HAM-D 17. The mean change on these depression scales was statistically significantly greater in the valproate groups, with a final mean MADRS score of 15.3 for valproate and 22.5 for placebo (P= 0.003)99 and a mean percent change in HAM-D of −44% for valproate compared with −27% for placebo (P=0.0002).98 Both studies also reported greater improvements on HADS and CGI with valproate, but in only 1 study was statistical significance achieved. 99 In both studies, valproate and placebo groups had similar rates of early discontinuation of assigned treatment.

Only 1 study reported rate of treatment-emergent mania; the rate was higher with placebo (17% compared with 8%), but the difference was only 1 patient.98 Mania rating scales showed a statistically significant deterioration in the placebo group in 1 study,99 but no difference in the other.98 Withdrawal due to increased or continuing symptoms of depression occurred at about the same rate in valproate and placebo groups.

Response and remission were defined very similarly in the studies. Greater than 50% improvement on MADRS indicated response in 1 study,99 while in the other >50% improvement and score <9 on the HAM-D 17 indicated remission.98 Valproate was not found to be statistically significantly superior to placebo in response or remission, although the small sample sizes (N=18 and 25) may have been underpowered to find a difference. We are aware of a small unpublished study very similar to these; it found that “43% of valproate-treated patients and 27% of placebo-treated patients achieved recovery, defined as an improvement of ≥50% in score on the 16-item HAM-D in the absence of hypomania (YMRS score <10).”100

In a small 6-month study of women with bipolar II disorder and comorbid borderline personality disorder, valproate was not statistically significantly better than placebo on the Symptom Checklist-90 depression subscale.59 These patients were not experiencing acute symptoms of bipolar disorder at the time of enrollment, so we considered this to be maintenance therapy. We also identified an ongoing maintenance trial of valproate added to lamotrigine treatment for patients with bipolar depression (ClinicalTrials.gov; study code NCT00183469). The trial is expected to be completed in June 2008.

Topiramate

A single, small, 8-week trial compared topiramate with bupropion extended-release as adjunct therapy with mood stabilizers in patients with bipolar disorder whose most recent episode was depression.22, 101 Statistically significant proportions of patients achieved response and remission in both groups; differences between the drugs were not statistically significant. Other assessments, such as the mean change on HAM-D 17, also indicated significant within-group changes, but no statistically significant differences between groups. Mania ratings also improved in both groups, without a statistically significant difference between groups. Rates of discontinuation due to any reason and due to adverse events were numerically greater in the topiramate group (44% and 33%, respectively) than the bupropion extended-release group (28% and 22%, respectively), but the difference did not reach statistical significance.

Maintenance of response: Bipolar depression
Lamotrigine

Patients who had been successfully treated for acute bipolar depression with lamotrigine were subsequently randomized to lamotrigine, lithium, or placebo in an 18-month maintenance study.102 Differences were not found between lamotrigine and lithium on the primary outcome measure, time to intervention for any mood episode (lamotrigine, 200 days; lithium, 170 days; P=0.915), although both were superior to placebo87 (93 days; P=0.029 for each comparison). Similar results were found for discontinuation from study for any reason. Depressive symptoms were the reason for intervention more often than manic symptoms. While lamotrigine was superior to placebo for time to intervention for depressive episodes, and lithium was superior to placebo for time to a manic episode, differences were not found between the drugs on either measure.

Carbamazepine

We included 1 trial that compared the general efficacy of carbamazepine (immediate-release) with that of placebo in the prophylaxis of bipolar disorder.79 This was a trial conducted in Japan which enrolled 22 patients diagnosed with bipolar- or manic-type endogenous manic-depressive psychosis according to ICD-9 criteria. Patients were randomized to either carbamazepine 200–1200 mg or placebo and followed for 1 year. Results indicated that carbamazepine completely inhibited or markedly reduced manic-depressive episodes in 60% of patients (compared with 22% for placebo; P<0.10). However, our re-analysis of findings from this trial using Fisher’s exact test indicated a P value of 0.13, suggesting that the difference between carbamazepine and placebo was not statistically significant.

Maintenance of response: Rapid cycling
Lamotrigine

For maintenance treatment of rapidly cycling bipolar disorder, we identified 2 placebo-controlled trials of lamotrigine,52, 103 only 1 of which is fully published.103 In the fully published trial,103 patients entering the 26-week randomized phase consisted of only those who were initially responsive to a preliminary phase of monotherapy with open-label lamotrigine 100 to 300 mg and scored no higher than 14 on the HAM-D and 12 on the MRS (N=182 of an original 324 patients). The main finding of this trial was that lamotrigine did not significantly improve the primary outcome, median time to additional pharmacotherapy for emerging symptoms of any mood episode compared with placebo (18 weeks compared with 12 weeks). For the second, unpublished trial, assessment of methodological quality was limited due to a lack of adequate detail provided by the FDA Medical Review and another narrative review.51, 52 Our review of the limited data found that this trial did not involve an initial run-in phase; it randomized patients with rapid-cycling bipolar I or II disorder to receive lamotrigine 50 to 400 mg (N=68) or placebo (N=69) as monotherapy or adjunct therapy for 32 weeks. The trial found that lamotrigine did not significantly extend median time to intervention with pharmacotherapy or electroconvulsive therapy compared with placebo (142 days compared with 133 days; P=0.73).51, 52

Valproate

A single trial compared monotherapy with valproate or lithium in patients with rapid-cycling bipolar disorder and found that valproate was no better than lithium in preventing relapses in this difficult-to-treat population.58 The trial was designed to test the hypothesis that valproate, at minimum blood levels of 50 μg/mL, would be more effective than lithium, at minimum blood levels of 0.8 mEq/L, in patients who were initially responsive to the combination of the 2 mood stabilizers. Of the 254 patients who initially enrolled in the open-label phase of combination therapy with valproate plus lithium, only 60 patients (23%) responded and were randomly assigned to monotherapy with either agent. After 20 months, just over half of the patients had relapsed, with no significant difference in rate between valproate and lithium, regardless of episode type. Relapses into depressive episodes were more common (31.7%) than relapses into manic episodes (21.7%).

Fibromyalgia

Four placebo-controlled trials assessed short-term (8 to14 weeks) treatment of fibromyalgia, 1 for gabapentin104 and 3 for pregabalin,105–107 and 1 placebo-controlled trial of pregabalin assessed relapse over 6 months following response to 6 weeks of treatment with pregabalin.108

Patient populations were similar across trials, with the overwhelming majority of patients being white females in their late 40s. However, patients in the gabapentin study had lower mean pain scores at baseline (5.8 on an 11-point scale) than those in the pregabalin studies (6.7 to 7.1). The three pregabalin studies were larger than the gabapentin study (N=745, 748, 530 and N=150, respectively). These studies were rated fair quality; although they were double-blind studies, used an intent-to-treat-analysis, and reported attrition, they did not report methods of randomization or allocation concealment. The results and quality of all these trials are summarized in Evidence Tables 5 and 6.

Table 6

Table 6

Change in average pain score in fibromyalgia treated with an antiepileptic drug compared with placebo

Acute treatment
Response rate

Response was defined as a 30% reduction in pain score in all short-term trials. Our pooled analyses indicate that pregabalin resulted in statistically significantly greater rates of response compared with placebo at 300 mg/d (relative risk 1.31; 95% CI, 1.11 to 1.54), 450 mg/d (relative risk 1.50; 95% CI, 1.20 to 1.87), and 600 mg/d (relative risk 1.38; 95% CI, 1.13 to 1.89). Although the 450 mg/d dose may have the highest response, overlapping confidence intervals preclude making this conclusion. Pooling the 300 mg/d, 450 mg/d and 600 mg/d data indicates a relative risk of response of 1.39 (95% CI, 1.26 to 1.53) with a number needed to treat of 8 (Figure 3). The 150 mg/d dose was not found superior to placebo.105 Gabapentin showed a greater response rate than placebo (51% compared with 31%; P=0.014; our calculated number needed to treat=5).104

Figure 3

Figure 3

Response to pregabalin relative to placebo

Using a more stringent definition of 50% or greater reduction in pain, more people responded with pregabalin 450 mg/d than placebo (28.9% compared with 13.2%; P=0.003; our calculated number needed to treat=6.4), but again at the lower doses response rates were not significantly greater than for placebo (13% and 18.9%).105 Pregabalin 600 mg/d recorded the highest responder rate compared with placebo (30% versus 15%; P=0.0010; number needed to treat=6.62), and the other treatment arms, 300 mg/d and 450 mg/d, were statistically significantly superior to placebo.106 Mease and colleagues107 did not report 50% responder rate.

Quality of life

While all 3 studies of pregabalin measured health-related quality of life using the SF-36, reporting was inconsistent, such that pooled analyses could not be undertaken. Overall, pregabalin improved scores on some, but not all, SF-36 domains with some variation based on dose, relative to placebo in 2 studies.105, 106 Both studies found that social functioning and vitality scores improved statistically significantly more with pregabalin 450 mg/d than placebo. The third study did not report the change from baseline in health-related quality of life but stated that there were no statistically significant differences at endpoint in any of the pregabalin treatment groups compared with placebo.107

Pain

Mean pain score at endpoint was the primary outcome measure in all 4 short-term studies, all using an 11-point numerical rating scale where 0=no pain and 10=worst possible pain.104–107 Mean pain scores at endpoint in the gabapentin 1800 mg/d groups (3.2) was significantly lower than in the placebo group (4.6; P=0.015). Pregabalin 300, 450, and 600 mg/d resulted in statistically significantly lower scores than placebo, with one exception (see Table 6). In the Crofford study, pregabalin 300 mg/d did not result in a final score that was statistically significantly lower than with placebo.105 Differences in mean pain score at endpoint did not reach statistical significance for pregabalin 150 mg/d.

Other outcomes

Numerous secondary outcomes were reported in all trials. In general, results from these analyses found significant improvements for gabapentin and pregabalin compared with placebo. One of the exceptions was that gabapentin was not superior to placebo in improving associated depressive symptoms. On MADRS gabapentin measured 9.1 compared with placebo 13.9, P=0.067.104 On HADS neither the depression nor anxiety scores were significantly different between pregabalin and placebo groups,105, 106 with the exception of anxiety symptoms with pregabalin 600 mg/d (difference from placebo −0.79; P=0.014).106 In the other pregabalin study by Mease, it was noted that other than sleep, secondary efficacy measures did not show any statistically significant difference between any of the treatment groups compared with placebo at endpoint.107

Relapse

Crofford and colleagues108 reported the only long-term (6-month) trial that studied relapse of symptoms of fibromyalgia. The objective of the trial was to study the duration of efficacy of pregabalin in treating fibromyalgia. All patients underwent a 6-week open-label phase in which they received escalating doses of pregabalin to determine their optimal dosages. At the end of the open-label phase, responders (greater than 50% reduction in pain using a 100-mm visual analog scale and a self-rating of “much” or “very much” improved on PGIC) entered a double-blind phase in which patients in one arm received placebo and patients in the other arm received their optimal pregabalin dosage. The primary outcome was the time to loss of therapeutic response, defined as <30% reduction in pain from open-label baseline or worsening of fibromyalgia, requiring alternate treatment.

Of 1051 patients enrolled in the open-label period, 566 were responders (53.8%). The discontinuation rate in this double-blind trial was very high, with 81% of the placebo group and 62% of the pregabalin group discontinuing the study prior to 6 months. Time to loss of therapeutic response was longer for pregabalin than placebo (P<0.001). Comparing the first quartile, the median time to loss of therapeutic response was 7 days for placebo and 34 days for pregabalin. At end of the 6-month double-blind phase, 61% of placebo patients met loss of therapeutic response criteria, compared with 32% of pregabalin patients.

Because all patients who withdrew from the study were counted as not having loss of therapeutic response in the primary analysis, sensitivity analysis was done counting these patients as having had loss of therapeutic response. This sensitivity analysis resulted in similar results, with a P<0.0001, although the time to event in the first quartile was 6 days for placebo and 18 days for pregabalin, a difference of 12 days compared with a difference of 27 days in the primary analysis. Several other sensitivity analyses were conducted; all found pregabalin superior to placebo.

Migraine prophylaxis

Previous systematic review

A Cochrane review by Chronicle and colleagues2 of antiepileptic drugs for migraine prophylaxis assessed the efficacy of carbamazepine, valproate, lamotrigine, gabapentin, and topiramate compared with placebo. Patients with chronic migraine, transformed migraine, or chronic daily headache were excluded from the Chronicle review. Reasons for excluding chronic migraine included concerns with inconsistencies in classification of chronic migraine and concerns with variability of response to treatment due to severity. Further discussion of these issues can be found in the publication.

Chronicle and colleagues conducted meta-analyses by drug for migraine frequency and for the proportion of patients achieving ≥50% reduction in migraine frequency. Table 7 summarizes the results from placebo-controlled trials: Only lamotrigine was not statistically significantly superior to placebo. Before putting significant weight on the pooled estimates from their review, the authors point out that much of the included literature had several methodologic limitations. These included selective outcome reporting, misrepresentation of intention-to-treat analyses, and inadequate measures to minimize carryover effect in crossover studies. Differences across these studies make qualitative indirect comparisons unwise. Despite these caveats, however, pooled effects for antiepileptic drugs were likely more robust in their estimates than effects estimated for agents with 1 trial. Therefore, more evidence supports use of valproate or topiramate for migraine prophylaxis than carbamazepine, lamotrigine, or gabapentin (Table 7). Furthermore, results from active-control trials that compared valproate and topiramate with propranolol or flunarizine (2 agents with evidence on efficacy) provided additional support for this conclusion. Sodium valproate and divalproex sodium are reported separately in this review (see Table 7).

Table 7

Table 7

Pooled results of antiepileptic drugs compared with placebo for reduction of migraine frequency (Chronicle 2004)

Many of the included trials studied various doses of antiepileptic drugs (Table 8). Chronicle and colleagues assessed the impact of various doses for valproate and topiramate. No clear dose-response was found for the drugs, although the 50 mg/d dose of topiramate resulted in the lowest standardized mean difference in migraine frequency among topiramate doses (50, 100, or 200 mg/d). However, the number of studies in these analyses was few, and the resulting confidence intervals were wide, such that these findings should be used with caution. Also, many of the active-control trials used dose comparisons that could be considered unequal.

Table 8

Table 8

Studied doses of antiepileptic drugs for migraine prophylaxis

Additional trials

We identified 10 trials not included in the Chronicle review: 1 valproate,109 1 carbamazepine,110 7 topiramate111–117 (including 1 trial with lamotrigine and placebo comparisons),113 and 1 oxcarbazepine.118 The carbamazepine trial110 was rated as having poor internal validity due to inadequate randomization, allocation, and blinding, and lack of an intention-to-treat analysis. Four topiramate trials were conducted in patients with chronic migraine,111, 114, 116, 117 a population excluded from Chronicle, and 1 topiramate trial112 assessed cessation compared with continuation of therapy. The results from these trials will be discussed briefly here. We also independently evaluated all trials for quality-of-life information.

Direct comparisons of antiepileptic drugs

We identified 2 trials directly comparing one antiepileptic drug with another.113, 119 In 2 small (N=60 and N=64) crossover studies topiramate was compared with lamotrigine, and topiramate was compared with valproate and placebo. At the end of 20 weeks, a larger portion of the topiramate group than the placebo group achieved ≥50% reduction in migraine frequency (63% compared with 30%; 95% CI, 0.18 to 0.46). Similar to the findings of Chronicle, there was no statistically significant difference in the proportion of patients with ≥50% reduction in migraine frequency for lamotrigine compared with placebo (46% compared with 34%; 95% CI, 0.02 to 0.26). However, when topiramate was compared with lamotrigine, more patients had a response with topiramate than with lamotrigine, although the confidence interval was wide (63% compared with 46%; 95% CI, 3% to 31%).113

In the direct comparison of valproate and topiramate, analysis of the first drug assigned found no statistically significant difference in headache frequency, but topiramate was better at reducing headache intensity (mean difference on 10-point visual analog scale 2.1; 95% CI, 1.4 to 2.9) and headache duration (mean difference 8.4 hours; 95% CI, 4.5 to 12.3). In analysis of the second randomized period, topiramate was superior in reducing the number of headaches (mean difference 1.2 per month; 95% CI, 0.2 to 2.1), but no difference was found in severity or duration. Using ANOVA to analyze the first and second randomized periods combined, the authors found no statistically significant differences. However, the conflicting findings of the first and second periods raises the question of carryover effects, such that data from the first period is preferred.

Comparisons with placebo

For topiramate, 2 studies published since the Chronicle review reported conflicting findings: The larger study was unable to find a statistically significant difference compared with placebo, while the smaller study did.113, 115 Pooling these studies with the previous studies indicates a statistically significant benefit of topiramate (odds ratio 3.04; 95% CI, 1.95 to 4.74). This compares with the pooled odds ratio for topiramate compared with placebo reported by Chronicle of 3.34 (95% CI, 2.36 to 4.73). The mean change in migraine frequency was quite different in the 2 trials, and inadequate data were reported to allow pooling with the previously reported studies (see Table 9).

Table 9

Table 9

Topiramate compared with placebo for prophylaxis of migraine

Treatment with oxcarbazepine was not superior to placebo in reducing migraine frequency (change from baseline, −1.10/mo compared with −1.16/mo; P=0.82) or in achieving ≥50% reduction in frequency (27.1% compared with 23.5%; P=0.557) over 19 weeks of therapy in 170 adults.118 No additional placebo-controlled trials were identified for carbamazepine, valproate, or gabapentin.

One new active-control trial that compared valproate with subcutaneous histamine was identified.109 No significant difference in treatment effect for lowering migraine frequency or MIDAS scores was observed in 92 adults randomized to valproate or subcutaneous histamine injection at the end of 12 weeks.

Quality-of-life and disability outcomes

Among all studies from the Chronicle review and an updated search, only 3 topiramate trials120–122 and 1 oxcarbazepine trial118 assessed quality-of-life outcomes. There were no significant differences between oxcarbazepine-treated and placebo-treated patients in improvement of quality-of-life scores using the SF-36 assessment tool.118 In 2 trials of 937 adults120, 122, patients treated with topiramate 50–200 mg reported significantly better improvement in the performance of daily activities limited by migraine headaches per MSQ-RR than patients treated with placebo. Similar findings were observed with MSQ-RP scores as well.

Topiramate more greatly reduced the number of disability days due to headache than placebo.121 Baseline mean days with disability was approximately 7. At the end of 16 weeks, the reduction in disability days was significant for topiramate and not placebo (change from baseline −4.3 days compared with −1.0 days, P<0.001).

Topiramate cessation compared with continuation

One new topiramate trial112 compared cessation with continued therapy in a study that began with a 26-week open-label phase and followed with a 26-week double-blind phase. Of 818 adults enrolled in the open-label phase using topiramate, 63% remained and were randomized to continue with topiramate or switch to placebo in the double-blind phase. The primary objective was to evaluate rebound effect after discontinuation of migraine prophylaxis by comparing the number of migraine days during the last month of the double-blind phase with the number of migraine days in the last month of the open-label phase. Patients who switched from topiramate to placebo after 26 weeks experienced an increase in the number of migraine days in a month by 1.19 (95% CI, 0.71 to 1.66). In contrast, patients who continued topiramate experienced minimal change in migraine days (+0.10 days/mo; 95% CI, −0.36 to 0.56). Despite worsening control of migraines, the number of migraine days in the placebo group during the last month of the double-blind phase (5.82 days/mo) did not return to baseline (8.9 days/mo, P<0.001).

Chronic migraine

Chronicle and colleagues excluded chronic migraine studies from their review due to concerns with significant differences in disease severity relative to patients without chronic migraines. We identified 3 placebo-controlled trials of topiramate114, 116, 117 and 1 head-to-head trial111 evaluating topiramate and valproate and decided to review the results independently and assess whether the results were significantly different between those with and without chronic migraine.

A small (N=49), open-label, active-control trial comparing topiramate 75 mg/d with divalproex sodium 750 mg/d found that at the end of 12 weeks, topiramate-treated and valproex-treated patients exhibited similar reductions in headache frequency (change from baseline, −23.3 headaches/mo with topiramate compared with −23.8 headaches/mo with valproate; between-group P value, not reported).111 The results from this poor-quality study should be considered with caution since about 10% of the population were not included in the analyses. Two placebo-controlled trials showed topiramate was more effective than placebo in reducing migraine frequency and monthly migraine days.116, 117 In a study of 306 adults 116 the change in mean monthly migraine days for topiramate was −6.4 and for placebo was −4.7, P=0.01. An additional trial showed similar findings but was rated poor.121

Although the collective result shows that topiramate is more effective than placebo, it is unclear whether the findings in patients with chronic migraine headache should be combined with results in patients without chronic headaches.

Chronic pain

Very little evidence was found to support the short-term use of tiagabine, topiramate, or gabapentin for treatment of chronic pain conditions.123–125 No evidence was found for other antiepileptic drugs. Full details are given in Evidence Tables 7 and 8.

Open-label tiagabine and gabapentin were directly compared in 91 patients with various types of chronic pain despite ongoing treatment with analgesics or antidepressants.125 Most patients were diagnosed with either musculoskeletal headache or cervical pain. Most of the population was female (78%), and the mean age was 44 years. Study medications were available only at the patients’ expense and were given in addition to ongoing treatment regimens. After 3 months, tiagabine and gabapentin were associated with similar reductions in pain score (−2.3 compared with −1.2 points on an 11-point scale; not statistically significant). Tiagabine (−3.0 points; P=0.04) resulted in significantly greater improvement in ratings for sleep quality than gabapentin (−1.54 points).

Topiramate improved pain and associated difficulties significantly more than placebo in a 10-week, fair-quality, double-blind trial of 96 patients with chronic low-back pain who had never undergone back surgery.124, 126 Overall, 75% of patients were male and their mean age was 49 years. Patients were required to discontinue analgesic or anti-inflammatory medications 1 week before randomization but were allowed to continue any prestudy antidepressant medications. Compared with placebo, topiramate significantly improved pain, associated disability, anger, and quality of life based on scores on the MPQ (−0.1 compared with −1.2 points; P<0.001), STAXI, OLBPQ, and SF-36.

Gabapentin was found to have significant analgesic effect compared with placebo in a 12-week, fair-quality, double-blind trial of 50 patients with moderate to severe chronic pain of the masticatory muscles of at least 6 months’ duration.123 All patients enrolled in this trial were female, with a mean age of 34 years. Although ongoing use of muscle relaxants and/or anti-inflammatory drugs was prohibited during the trial, acetaminophen was allowed for breakthrough pain. Patients were also allowed to continue ongoing psychotropic medication regimens (for example, tricyclic antidepressants, benzodiazepines, selective serotonin reuptake inhibitors). In addition to superior reductions in pain compared with placebo (51% compared with 24% reduction based on visual analog scale; P=0.037), gabapentin also resulted in greater reduction in number of tender palpation sites from 9.5 at baseline (−6.46 compared with −1.90; P=0.002) and greater reduction in impact of pain on daily functioning as measured using a visual analog scale (−53% compared with −19%; P=0.026).

Key Question 2 For adult outpatients with bipolar disorder, fibromyalgia, migraine, or chronic pain, do antiepileptic drugs differ in safety or adverse events?

The adverse event profiles of the antiepileptic drugs vary considerably, with overlap only in adverse effects that may affect tolerability, such as somnolence.127, 128 Comparative assessments of common, overlapping adverse effects were undertaken where possible based on direct evidence from the populations of interest in this review. Emphasis was on the comparison of rates of any adverse event, withdrawals due to adverse events, and longer-term evidence in “real-life” populations (observational studies). For the purposes of this review, side effects that are unique to individual antiepileptic drugs are summarized based on existing reviews, including rare but serious adverse events such as birth defects. Because epilepsy and its treatment are complex and may affect the adverse events experienced with an antiepileptic drug, evidence relating to the population of patients with epilepsy was not reviewed other than to provide basic estimates of rates of adverse events or to provide evidence on harms with long-term effects, such as suicidal ideation.

Suicide

An FDA advisory to healthcare professionals warning of potentially increased risk of suicidality with antiepileptic drugs was published in February 2008. In May 2008 the FDA completed an initial analysis of data on suicide relating to antiepileptic drugs, in preparation for an advisory committee meeting to be held in July 2008 (http://www.fda.gov/cder/drug/InfoSheets/HCP/antiepilepticsHCP.htm). Their analysis included 11 drugs: carbamazepine, divalproex, felbamate, gabapentin, lamotrigine, levetiracetam, oxcarbazepine, pregabalin, tiagabine, topiramate, and zonisamide. The meta-analysis was based on 199 placebo-controlled trials, with reports of completed suicides or suicidal ideation/behavior as the primary outcomes. The conclusions of this report are that as a group, these drugs are associated with an increased risk of the patient experiencing a suicidal ideation or behavior; odds ratio compared with a placebo patient was 1.80 (95% CI, 1.24 to 2.66). The number of suicide deaths was small (N=4) but greater than in the placebo groups (N=0), although numbers were insufficient to show statistical significance.

Based on these results, the FDA asked for an advisory committee review to consider regulations requiring “black box” warnings be added to all antiepileptic drugs based on the fact that 8 of 11 drugs had a numerically increased odds ratio with only 2 (lamotrigine and topiramate) reaching statistical significance. Three drugs (carbamazepine, divalproex, and tiagabine) did not have odds ratios greater than 1, and the authors of the report note that carbamazepine and tiagabine have had relatively few patients studied (N=502 and 1443, respectively), such that the risk is less certain. For felbamate, no cases were found in either group, with a total of 340 patients studied.

The advisory committee voted against adding a black box warning across the class at this time (http://www.fda.gov/ohrms/dockets/ac/08/transcripts/2008-4372t1.pdf). The committee was not convinced of a class effect and wanted to see an analysis that looked at the drugs individually; assessed geographic differences, differences among indications, longer treatment periods (the analysis was limited to studies of 24 weeks or less), and use in monotherapy versus polytherapy; and used sensitivity analyses to test assumptions about zero events and ascertainment of suicidality. Much of the discussion centered on these issues, particularly how they had been handled in the previous FDA analysis of suicidality associated with newer antidepressant drugs and the impact of the black box warning added to those drugs.

A cohort study with a mean follow-up period of 2.9 years provided data on suicide risk with carbamazepine, divalproex, and lithium in patients with bipolar disorder.129 This fair-quality study used a large, computerized, prescription database to retrospectively identify a cohort of 20 638 patients with bipolar disorder. After adjustment for age, sex, health plan, year of diagnosis, comorbid medical and psychiatric conditions, and concomitant use of other psychotropic drugs, the hazard ratio for divalproex relative to lithium was 2.7 (95% CI, 1.1 to 6.3) for suicide death, indicating an almost three-fold higher risk of fatal suicide on divalproex compared with lithium. The hazard ratios for the other outcome measures for divalproex were 1.7 (95% CI, 1.2 to 2.3) for suicide attempts resulting in hospitalization and 1.8 (95% CI, 1.4 to 2.2) for emergency department–diagnosed suicide attempts. Hazard ratios for carbamazepine relative to lithium were less consistent and stable (range, 1.4 to 2.9), showing a statistically significant result only for suicide attempts leading to hospitalization (2.9; 95% CI, 1.9 to 4.4). The results for combination treatment and no treatment, each relative to lithium, were also inconsistent. Comparing the hazard ratio estimates and confidence intervals for valproate (1.7; 1.2 to 2.3) and carbamazepine (2.9; 1.9 to 4.4) for suicide attempts leading to hospitalization, one cannot conclude there is a difference between the 2 drugs for this outcome.

Data were further analyzed for possible confounding factors, such as confounding by indication (where the differences in suicide risk could have reflected differences in preexisting illness severity or other factors affecting suicide risk). An analysis for time-dependent risk differences between valproate and lithium showed consistent results for risk of suicide attempts and less consistent risk differences for suicide deaths. A subgroup analysis of patients who switched between divalproex and lithium revealed little difference in risk in switching from divalproex to lithium and vice versa. Therefore, it appeared that any medication switch was associated with a higher, roughly 2-fold risk of suicide attempt.

Bone fractures

A good-quality case-control study included 124 patients who had sustained a fracture as identified in the National Hospital Discharge Register of Denmark and 373 962 randomly selected gender- and age-matched controls.130 Adjusted odds ratios (odds ratio; 95% CI) for any fracture in patients who used antiepileptic drugs were significantly increased for carbamazepine (1.18; 1.10 to 1.26), oxcarbazepine (1.14; 1.03 to 1.26), and valproate (1.15; 1.05 to 1.26). The odds ratios were nonsignificant but increased for lamotrigine (1.04; 0.91 to 1.19), phenytoin (1.20; 1.00 to 1.43), tiagabine (0.75; 0.40 to 1.41), and topiramate (1.39; 0.99 to 1.96). Fracture risk analyzed by various skeletal sites was significant for carbamazepine at the hip (1.33; 1.13 to 1.58), lamotrigine at the spine (2.47; 1.13 to 5.39), and oxcarbazepine at the hip (1.48; 1.11 to 1.97). Risk was not significant by skeletal site for phenytoin, tiagabine, topiramate, or valproate. There was a significant dose-response relationship for carbamazepine, oxcarbazepine, and valproate, and no significant dose-response relationship for lamotrigine, phenytoin, tiagabine, or topiramate. The results suggest that the risk for any or site-specific fracture may be greater for carbamazepine, lamotrigine, oxcarbazepine, and valproate than for phenytoin, tiagabine, and topiramate; however, one cannot definitely conclude that there are differences between antiepileptic drugs, because the confidence intervals overlapped. No data were available for gabapentin and levetiracetam.

A second case-control study of 1018 cases and 1842 matched controls also found that exposure to antiepileptic drug increased risk of fracture.131 The risk increased with duration of exposure, with the strongest association at greater than 12 years of use (adjusted odds ratio 4.15; 95% CI, 2.71 to 6.34), with higher risk among women and no difference between antiepileptic drugs that do or do not induce the hepatic cytochrome P450 system. It should be noted that this study was done within a cohort study of epilepsy patients; the data may or may not translate to nonepileptic patients.

Stevens-Johnson syndrome and toxic epidermal necrolysis

Two fair-quality case-control studies provided comparative assessments of risk for Stevens-Johnson syndrome and toxic epidermal necrolysis.132, 133 The first provided comparative data for 5 antiepileptic drugs. It was conducted in hospitals in France, Germany, Italy, and Portugal.133 There were 352 cases of Stevens-Johnson syndrome or toxic epidermal necrolysis with onset before hospitalization and 1579 matched hospitalized controls. The univariate relative risk of Stevens-Johnson syndrome or toxic epidermal necrolysis for 8 or fewer weeks of use was 57 (95% CI, 16 to 360) for phenobarbital, 91 (26 to infinity) for phenytoin, 120 (34 to infinity) for carbamazepine, 25 (5.6 to infinity) for lamotrigine, and 24 (5.9 to infinity) for valproate. The multivariate relative risk for phenobarbital was 59 (12 to 302). The univariate relative risk for more than 8 weeks of use was 6.2 (2.4 to 17.0) for phenobarbital, 1.2 (0 to 5.4) for phenytoin, 0.4 (0.02 to 2.1) for carbamazepine, and 7.0 (2.4 to 21.0) for valproate. The multivariate risk for long-term use was 2.1 (0.5 to 9.3) for phenobarbital and 2.0 (0.3 to 15.0) for valproate (neither were significant). Short-term use of other antiepileptic drugs was a potential confounder for an association with valproate. Therefore, the risks of these serious skin reactions appear to be increased for short-term (≤8 weeks) use of phenobarbital, phenytoin, and carbamazepine. The numbers for lamotrigine were too small for meaningful analysis.

The second study identified 35 case subjects with Stevens-Johnson syndrome or toxic epidermal necrolysis based on hospital discharge ICD-9-CM codes and 105 randomly selected, matched controls.132 The crude relative risk (95% CI) was 33.0 (4.3 to 255.6) for carbamazepine and 9.6 (2.0 to 46.6) for phenytoin. Multivariate risks were 301.8 (13.6 to 6700.2) and 290.8 (9.2 to 9239.3), respectively. The results suggest that carbamazepine and phenytoin are similar in their risks of Stevens-Johnson syndrome or toxic epidermal necrolysis; however, confidence intervals were wide because of the small number of cases. Ascertainment of cases may have been incomplete because of misdiagnoses or missing records.

Aplastic anemia and agranulocytosis

A good-quality, population-based, case-control study of antiepileptic drug–related agranulocytosis and aplastic anemia was conducted in Barcelona, Spain, as part of a 22-year systematic, multicenter (17 hospital hematology units), collaborative surveillance study (International Agranulocytosis and Aplastic Anemia Study, IAAAS).134 A total of 177 case subjects and 586 matched controls was included. In the conditional primary analysis, 5 cases and 1 control were exposed to carbamazepine, and 2 cases and 1 control were exposed to phenytoin. The odds of drug exposure within the week before the index day of agranulocytosis were significant for carbamazepine (odds ratio 10.96; 95% CI, 1.17 to 102.64). The odds ratio was not calculated for phenytoin because of the small number of exposures. The population-attributable risk and incidence of agranulocytosis for exposure to carbamazepine within the week before the index day were 2.57 (95% CI, 0.03 to 5.04) and 0.09 (95% CI, < 0.01 to 0.17) per 1 million per year, respectively. These results suggest that the risk of agranulocytosis is greater with carbamazepine than phenytoin; however, confidence intervals were wide.

A similar study used data from the UK General Practitioners Research Database to identify 173 cases and 497 matched controls.135 The study covered the years 1987 to 2002, when carbamazepine, phenytoin, and valproate were the most commonly used antiepileptic drugs, and lamotrigine saw only limited use. Only 16 of the 173 cases were using an antiepileptic drug prior to the event, although use of any antiepileptic drug was statistically significantly associated with aplastic anemia (odds ratio 9.5; 95% CI, 3.0 to 39.7). The odds ratios for individual drugs were carbamazepine 10.3 (95% CI, 2.0 to 101), phenytoin 3.5 (95% CI, 0.4 to 44), and valproate 18.2 (95% CI, 2.5 to infinity). The broad confidence intervals reflect the small number of cases.

Birth defects

We found 19 studies reporting the risk of birth defects among women treated with antiepileptic drugs during pregnancy.131, 136–153 Of these, 9 are studies of only women with epilepsy and are not considered here due to the complex nature of both the disease and use of multiple antiepileptic drugs concurrently, potentially resulting in drug interactions and drug-disease interactions that may have complex adverse impact on fetal development.131, 136–138, 140, 145, 147, 148, 152 The subject of whether epilepsy is associated with birth defects has been debated and reviewed elsewhere.154 Five studies138, 139, 142–144, 146 examined the relationship between birth defects and exposure to antiepileptic drugs among broader populations of patients, 2 of which combined data for all antiepileptic drugs.142, 143 These studies are reviewed here.

In 2005, a review of the relationship between birth defects and exposure to antiepileptic drug during pregnancy (for any reason) found that exposure to older antiepileptic drugs during the first trimester is associated with an increased risk compared with the general population, 4%–10% compared with 2%–5%.155 The review also confirms the belief that antiepileptic drug monotherapy is associated with somewhat lower risk of birth defects than antiepileptic drug polytherapy. While specific rates vary among studies, differences in rates of birth defects among infants exposed in utero to carbamazepine, phenytoin, and phenobarbital were not found. However, valproate was associated with a higher risk, with odds ratios of 2 to 4, than carbamazepine, lamotrigine, and all other antiepileptic drugs combined. Some studies indicate a dose-dependent relationship, with valproate doses of 800 to 1000 mg/d associated with higher risk. A more recent case-control study found an increased risk of cleft palate among infants exposed to phenytoin during the second and third month of pregnancy and increased risk of posterior cleft palate among infants exposed to carbamazepine during the third and fourth months of pregnancy.144

Of the newer antiepileptic drugs, only lamotrigine has been well studied, through 2 registries. In the review conducted in 2005, analysis of data from one of these registries indicated a potential dose-response association for lamotrigine, with doses of > 200 mg/d associated with risk approaching that of valproate 1000 mg/d.155 However, in an analysis of the manufacturer’s registry a dose-effect was not seen in doses up to 400 mg/d. Data on doses above 400 mg/d were too limited for meaningful analysis.138 Studies did not indicate a significant difference in risk between lamotrigine and carbamazepine. Oxcarbazepine had a risk similar to carbamazepine and phenytoin in a single retrospective study; the risk for valproate was higher. Studies of topiramate exposure during pregnancy are limited to 2 small registry studies, one including only women with epilepsy152 and the other a very small study in women taking topiramate for unspecified reasons.153 This study found the rate of nongenetic major malformations to be 4.9% with topiramate, compared to 3.4% in a control group not exposed to topiramate. This difference was not statistically significant. Gabapentin and levetiraetam have only very limited evidence, such that conclusions cannot be drawn.

Polycystic ovary syndrome

Polycystic ovary syndrome is an endocrine disorder with increased androgen production, abnormal gonadotropin secretion, anovulation, and menstrual dysregulation. Valproate has been identified as a drug that is potentially associated with polycystic ovary syndrome, although there is debate about the relationship between polycystic ovary syndrome and the underlying disease states, such as epilepsy or bipolar disorder. In a study that enrolled women taking valproate for bipolar disorder, with no preexisting polycystic ovary syndrome, new-onset oligomenorrhea that could not be explained by other reasons was identified and compared with a group of women being treated with another mood stabilizer, including other antiepileptic drugs. The resulting sample size was small, N = 230. The rate of new-onset oligomenorrhea with hyperandrogenism was 10.5% in the valproate group and 1.4% in the control group (P=0.002).

While we found 3 other studies examining the effects of valproate in women with bipolar disorder, concerns over study design limits their usefulness in this report.156–158 One is a cross-sectional study using interviews to obtain menstrual histories; another is a related study with a cross-sectional component and a 17 month follow-up of 56% of the original cohort. The third is an extension of the cohort study discussed above, but this one reports only on women who developed polycystic ovary syndrome while on study or were considered at risk. This is also a very small study, with only 14 women participating.

Delirium

Valproate was not found to be associated with a statistically significant increase in diagnosis of delirium compared with lithium among older patients (age > 65 years) being treated for mood disorders.159 Using 4 databases, the study found that the hazard ratio of a diagnosis of delirium during a hospitalization was 1.07 (95% CI, 0.67 to 1.70) for valproate compared with lithium.

Overall adverse event rates

Seven head-to-head trials compared topiramate with sodium valproate for migraine prophylaxis; 1 compared topiramate with divalproex for acute mania; 1 compared topiramate with lamotrigine for migraine prophylaxis; 1 compared lamotrigine with gabapentin for refractory mood disorders; 1 compared gabapentin with tiagabine for chronic pain; and 1 compared carbamazepine with sodium valproate for acute mania.23, 46, 54, 111, 113, 119, 125 Rates of any adverse event and withdrawals due to adverse events were reported in most of these trials, and those data provided the basis for evaluation of direct comparative safety among the antiepileptic drugs (Table 10).

Table 10

Table 10

Risk of any adverse event in head-to-head trials of antiepileptic drugs

In the trial of carbamazepine and divalproex, a larger number of patients reported an adverse event with carbamazepine than divalproex, with no difference in withdrawals. None of the other trials individually showed statistically significant differences in rate of overall adverse events or withdrawals due to adverse events. Two studies compared sodium valproate and topiramate; again, the pooled analysis did not indicate a significant difference between the drugs. However, these were small fair- to poor-quality studies, with the largest enrolling only 91 patients; it is unlikely that these studies would find statistically significant differences.

Adverse events were reported in 51 placebo-controlled trials. Of these, 31 reported the number of patients with any adverse event, and 47 reported the number of withdrawals due to adverse events, allowing comparisons to be made. Specific adverse event rates are discussed elsewhere, but because the antiepileptic drugs differ so greatly in their adverse effects, it may be more useful to compare the overall rates of adverse events and rates of withdrawal from study drug due to adverse events. Withdrawal due to adverse events is an appealing measure, because it incorporates the severity of the events. Pooled estimates (Table 11) show that in nonepileptic populations, only topiramate and carbamazepine result in significantly greater rates of adverse event reports than placebo. Withdrawal from study drug due to adverse events is also statistically significantly greater with carbamazepine and topiramate, and also with valproate (immediate- release). These results apply when these drugs are used primarily as short-term monotherapy.

Table 11

Table 11

Risk of adverse events with antiepileptic drugs compared with placebo

Results of all studies are shown by drug in the forest plot in Figure 4. For studies in which zero events occurred in one group, a correction factor was used in the analysis and the resulting confidence intervals extend to infinity. For studies in which zero events occurred in both groups, no point estimate could be calculated. For rates of any adverse events, there is 1 outlier, a small study of valproate (N=74) in which no adverse events were reported in the placebo group and adverse events were experienced by 54% of patients in the valproate group.

Figure 4

Figure 4

Relative risk of any adverse event: Antiepileptic drug compared with placebo

Based on crude indirect comparisons of the antiepileptic drugs, no difference in the overall rate of adverse events is apparent, although the rate relative to placebo is higher for carbamazepine and topiramate. In Figure 5, withdrawals across all studies show that although most point estimates indicate higher rates (relative to placebo) with the study drug, the differences are not statistically significant. The graph supports the pooled estimates: Carbamazepine, valproate, and topiramate have higher rates of withdrawal than placebo.

Figure 5

Figure 5

Risk of discontinuing treatment due to adverse effects relative to placebo

In the analysis of placebo-controlled trials (above) the 3 drugs with higher rates of adverse events or withdrawals due to adverse events were carbamazepine, divalproex, and topiramate. In head-to-head comparisons, only carbamazepine had significantly higher rates of adverse events than divalproex. Topiramate was not found different from divalproex or sodium valproate, but was not directly compared with carbamazepine.

We found 1 good-quality systematic review providing comparative data on adverse events with carbamazepine and valproate relative to lithium. Based on 2 randomized controlled trials of acute (4-week) treatment of mania,160 no statistically significant difference was seen in the risk of adverse events between carbamazepine (relative risk compared with lithium 0.71; 95% CI, 0.49 to 1.02; N=139) and valproate (relative risk 1.09; 95% CI, 0.95 to 1.26; N=105). These findings indirectly suggest that carbamazepine and valproate have similar risks of adverse events, since neither was statistically different from a common comparison treatment, lithium.

Specific adverse events

In an analysis of adverse events we included 14 trials and evaluated 8 specific adverse events (diarrhea, dizziness, headache, nausea, rash, somnolence, tremor, and weight gain). There were no reports of hepatotoxicity, thrombocytopenia, or hyperammonemia in any of the placebo-controlled trials.

The results of our meta-analysis are shown in Appendix E. The only consistent finding was a higher likelihood of tremor with valproate than lamotrigine, based on data from lithium- and placebo-controlled trials. However, the 95% confidence intervals overlapped in both analyses (0.61 to 1.77 for valproate compared with lithium and 0.11 to 0.68 for lamotrigine compared with lithium; 2.38 to 10.26 for valproate compared with placebo and 0.33 to 3.79 for lamotrigine compared with placebo). Therefore, we cannot conclude that valproate and lamotrigine definitely differ in their association with tremor. One of the limitations of the evaluation of specific adverse events and pooled analyses of adverse events is inconsistency among trials in the definition of common adverse event. That is, common was defined as occurring in at least 5%, 8%, or 10% of patients in different trials. This variation in reporting of common adverse events may influence indirect comparisons of antiepileptic drugs.

Our statistical analysis of the 1 small trial that compared carbamazepine with valproate found that carbamazepine was significantly more likely than valproate to be associated with dizziness, with an odds ratio of 15.50; however, the 95% confidence interval was wide, 1.53 to 826.43.23 The incidence of rash was not found to be different and was low in both groups.

We analyzed data for carbamazepine, valproate, and lamotrigine relative to lithium. The numbers of trials and patients were small, and the 95% confidence intervals were wide. However, 2 findings reached statistical significance. Lamotrigine (2 trials),82, 102 but not valproate (1 trial),57 was significantly less likely than lithium to be associated with diarrhea (pooled odds ratio 0.30; 95 % CI, 0.14 to 0.59). Lamotrigine (1 trial; odds ratio 0.28; 95% CI, 0.11 to 0.68)102 and carbamazepine (2 trials; odds ratio 0.00; 95% CI, 0.0 to 0.30),38, 161 but not valproate (1 trial),57 was also associated with significantly lower odds of tremor than lithium. Analysis of reports of depression, headache, rash, somnolence, or weight gain did not result in statistically significant differences. (See Appendix E for complete results).

Similarly, we pooled data for carbamazepine, valproate, gabapentin, and lamotrigine compared with placebo. Again, the numbers of trials and patients were small, and the 95% confidence intervals were wide. Lamotrigine (4 trials),82, 86, 102, 162 and not carbamazepine (1 trial)35 or gabapentin (1 trial),53 was more likely than placebo to be associated with headache (odds ratio 1.59; 95% CI, 1.14 to 2.25). Carbamazepine (2 trials),35, 36 and not valproate (1 trial)57 or lamotrigine (2 trials),82, 86 was more likely than placebo to be associated with nausea (odds ratio 5.16; 95% CI, 2.73 to 10.30). Lamotrigine (2 trials),82, 86 and not carbamazepine (1 trial),36 was associated with a significantly higher odds of rash relative to placebo (odds ratio 2.23; 95% CI, 1.06 to 5.28). Carbamazepine (2 trials),35, 36 and not gabapentin (1 trial)53 or lamotrigine (3 trials),82, 86, 163 was more likely than placebo to be associated with somnolence (odds ratio 2.71; 95% CI, 1.48 to 5.36). Valproate (1 trial),57 and not lamotrigine (1 trial),102 was associated with significantly higher odds of tremor compared with placebo (odds ratio 4.76; 95% CI, 2.38 to 10.26). Only valproate was reported to cause weight gain as an adverse event (odds ratio 3.26; 95% CI, 1.36 to 9.03).57

In 31 evaluable patients, lamotrigine was associated with weight loss (mean change from baseline at 6 weeks, −0.96 kg), while gabapentin was associated with weight gain (+1.83 kg; calculated difference, −2.79 kg; P=0.02).54 There was no significant difference between lamotrigine and placebo (−0.40 kg) or between gabapentin and placebo. The findings should be interpreted with caution, since they were not based on randomized patients.

Key Question 3 Are there subgroups of patients based on demographics (age, racial groups, and gender), other medications, or comorbidities for which one antiepileptic drug is more effective or associated with fewer adverse events?

Bipolar disorder

Patient characteristics
Subtype

A fair-quality trial in a hospitalized inpatient population evaluated possible predictors of clinical response to lamotrigine and gabapentin in 45 patients with bipolar or unipolar mood disorder.97 Responder rates were higher for lamotrigine (51%) than gabapentin (28%) or placebo (21%). Univariate analyses and linear regression showed that response to lamotrigine may be better in male patients with fewer trials of prior medications. A better response to gabapentin appeared to occur in younger patients with lower baseline weight; however, there was no statistically significant difference in response between gabapentin and placebo. These results should be considered preliminary because of the post hoc subgroup analyses, the small and selective (treatment-refractory) study population, and the heterogeneous patient diagnoses. Another trial showed no demographic factors to be predictors of a differential response between valproate and lithium.62 However, for patients with bipolar I disorder with recent mania and previous psychiatric hospitalization, valproate was associated with a longer time to depressive relapse than lithium.62

Two placebo-controlled trials evaluated the impact of bipolar subtype, 1 with carbamazepine and 1 with lamotrigine. The trial evaluating carbamazepine showed no differential effect of bipolar subtype by YMRS total score. However, when depressive symptoms were measured on HAM-D, patients with manic episodes appeared to benefit more greatly from carbamazepine than patients with mixed episode; improved symptoms were not consistently of the same type(s). Similarly, valproate was found to have superior efficacy compared with lithium for patients experiencing mixed manic episodes, while in a systematic review of valproate in bipolar disorder, response to the drugs was similar in patients with mania alone.15 These authors also found that irritability was more responsive to valproate than lithium or carbamazepine.

Subgroup analyses by bipolar subtype were performed in a trial that compared lamotrigine with placebo maintenance therapy in patients who had bipolar I or II disorder with rapid cycling. The bipolar II subgroup consistently responded better to lamotrigine than placebo on time to premature discontinuation for any reason, proportion of patients who were stable without relapse for 6 months, and GAS score.162 However, while time to relapse (the primary efficacy measure) was also longer with lamotrigine than placebo in the bipolar II subgroup (17 weeks compared with 7 weeks), this difference between treatments was not statistically significant (P=0.073). The bipolar I subgroup showed no significant difference between lamotrigine and placebo for any outcome. According to the authors, this finding was unexpected, since lamotrigine had previously been shown to be effective in bipolar I disorder. A high rate of response to placebo was observed in bipolar I patients and may be a confounder or an indication of other possible confounders. The factors accounting for different responses between the 2 bipolar subtypes need further clarification.

Age

We found 2 reports on the effect of antiepileptic drugs on symptoms of bipolar disorder in older patients.14, 164 A pooled analysis evaluated data on 98 patients ≥55 years who had been randomized to lamotrigine, lithium, or placebo in these 2 studies of lamotrigine maintenance therapy for which the primary outcome measure was time to intervention for a mood episode.164 Because the subgroups were small and not stratified at randomization, differences at baseline were present, such as mean lifetime hospitalizations, which were 11.2 in the lamotrigine group, 4.8 for lithium, and 6.3 for placebo. Similar to the findings of the results across all ages, compared with placebo lamotrigine delayed the time to intervention for any mood disorder (manic, mixed manic, or depressive episode), while lithium delayed time to intervention for manic and mixed episodes only. The mean age in these subgroups was 61 years, older than the typical bipolar population but not elderly. Because these are post hoc subgroup analyses, they should be interpreted with caution.

In a 2006 systematic review of evidence on antiepileptic drugs for bipolar disorder in patients > 60 years, Aziz and colleagues reported that there were no “published, controlled studies with these medications that focus on late-life bipolar disorder.”14 The authors went on to report results of observational reports, primarily case series, in older patients with bipolar disorder.

Comorbidity

A small placebo-controlled trial in patients with both bipolar disorder and alcoholism found that valproate as adjunct treatment to lithium was no different from placebo in treating manic or depressive symptoms. Valproate did reduce the number of heavy drinking days; the number of drinks per day on heavy drinking days was about the same as with placebo.61

Fibromyalgia

Typically, trial populations were about 90% women. Pregabalin at 450 mg/d was statistically more efficacious than placebo in the primary analyses that included both men and women, as well as a secondary analysis including only women.105

In a companion paper165 Arnold and colleagues studied the effect of anxiety and depression on improvement in pain in the pregabalin trial. Significantly more patients reported symptoms of anxiety (71%) than depression (56%; P<0.0001). The pain treatment did not depend on baseline HAM-D score, suggesting that pregabalin improves pain in patients with or without symptoms of depression and anxiety. These analyses indicate that much (75%) of the pain reduction appears to be independent of improvements in anxiety or mood symptoms.

The results of the gabapentin trial104 might not apply to patients with some comorbid psychiatric disorders, such as psychosis or bipolar disorder; rheumatologic or other musculoskeletal disorders; or unstable psychiatric or medical disorders, because patients with these conditions were excluded from the trial. Similarly the pregabalin trial105 might have excluded the most severely affected patients and patients with psychiatric comorbidity.

Migraine prophylaxis

Included trials did not provide sufficient evidence to determine comparative efficacy or safety in patients with migraine.

Chronic pain

Included trials did not provide sufficient evidence to determine comparative efficacy or safety in patients with chronic pain.

Copyright © 2008, Oregon Health & Science University, Portland, Oregon.
Cover of Drug Class Review: Antiepileptic Drugs for Indications Other Than Epilepsy
Drug Class Review: Antiepileptic Drugs for Indications Other Than Epilepsy: Final Report Update 2 [Internet].
McDonagh M, Peterson K, Lee N, et al.
Portland (OR): Oregon Health & Science University; 2008 Oct.

PubMed Health Blog...

read all...

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...