Bipolar disorder is among the top causes of worldwide disability and is characterised by both depressive and manic episodes. Reference Murray and Lopez1 The depressive symptoms are now recognised to be the predominant cause of disability in the long term for most people with bipolar disorder. Reference Judd, Akiskal, Schettler, Coryell, Endicott and Maser2–Reference Kupfer, Frank, Grochocinski, Luther, Houck and Swartz4 Prior to 1999, the treatment of bipolar depression had been little studied and there was uncertainty about the treatment of this phase of the disorder. There have long been concerns about the risk of mood destabilisation with antidepressant drugs and there remains uncertainty about efficacy. A meta-analysis found some evidence for efficacy Reference Gijsman, Geddes, Rendell, Nolen and Goodwin5 but a subsequent large trial conducted as part of the National Institute of Mental Health funded Systematic Treatment Enhancement Program for Bipolar Disorder (NIMH STEP–BD) found no evidence of benefit for adjunctive therapy with bupropion or paroxetine. Reference Sachs, Nierenberg, Calabrese, Marangell, Wisniewski and Gyulai6
Recent guidelines have suggested a role for lamotrigine, an inhibitor of voltage-sensitive sodium channels in the acute treatment of bipolar depression. Lamotrigine is only licensed in the USA by the Food and Drug Administration and in some European countries for prevention of relapse in bipolar disorder. 7 None the less, it is already in common clinical use for bipolar disorder, particularly in the USA. Reference Marangell, Martinez, Ketter, Bowden, Goldberg and Calabrese8 Although the evidence for the long-term efficacy of lamotrigine is reasonably robust, the five pivotal trials in acute phase therapy have been reported as individually neutral, with no statistically significant benefit from lamotrigine. Reference Calabrese, Huffman, White, Edwards, Thompson and Ascher9 This apparent lack of acute efficacy sits rather uncomfortably beside evidence for efficacy for relapse prevention. Although lamotrigine may indeed be ineffective in the acute phase, it is possible that the therapeutic effect size of lamotrigine may be smaller than predicted and that the acute trials were consequently underpowered. It is also possible that the need to increase the dose of lamotrigine gradually may have made detection of the acute therapeutic effect more difficult or that any positive effects are confined to a subgroup of individuals and not observable in the full trial samples.
Meta-analysis can increase the statistical power for analysis of the available randomised data. Data can be combined in a variety of ways: pooling individual participant data from several trials is the most informative for meta-analysis because it allows investigation of potential differential effects across participant subgroups. Reference Stewart and Clarke10,Reference Clarke and Stewart11 We report a meta-analysis of the individual participant data from the five trials conducted by GlaxoSmithKline investigating the efficacy of lamotrigine in acute bipolar depression.
We included all the randomised controlled trials conducted by GlaxoSmithKline comparing lamotrigine with placebo in bipolar depression (online Table DS1 and Table 1).
|Study 1 GW602/SCAB2001||Study 2 GW603/SCAA2010||Study 3 SCA40910||Study 4 SCA10022||Study 5 SCA30924|
|Age, years: mean (s.d.)|
|Lamotrigine||42.2 (11.5)||40.5 (11.3)||37.6 (12.6)||38.1 (11.5)||40.5 (12.5)|
|Placebo||42.4 (12.7)||40.9 (11.2)||37.3 (11.5)||36.5 (11.9)||38.2 (12.1)|
|White ethnicity, n (%)|
|Lamotrigine||57 (90)||90 (87)||113 (88)||70 (64)||94 (74)|
|Placebo||62 (94)||89 (86)||99 (84)||82 (75)||84 (69)|
|Female, n (%)|
|Lamotrigine||35 (56)||66 (64)||74 (57)||70 (64)||69 (54)|
|Placebo||39 (59)||61 (59)||62 (53)||69 (63)||66 (54)|
|Duration of current episode, weeks: n (%)|
|≤ 24||49 (78)||79 (77)||104 (81)||76 (70)||83 (65)|
|> 24||14 (22)||24 (23)||25 (19)||33 (30)||43 (34)|
|≤ 24||47 (71)||79 (77)||91 (77)||83 (76)||70 (58)|
|> 24||19 (29)||24 (23)||27 (23)||26 (24)||52 (43)|
|Intensity of depression based on SCID, n (%)|
|Mild||2 (3)||8 (8)||0||0||0|
|Moderate||34 (54)||66 (64)||91 (71)||75 (69)||58 (46)|
|Severe||27 (43)||29 (28)||38 (29)||34 (31)||69 (54)|
|Moderate||40 (61)||75 (73)||80 (68)||74 (68)||55 (45)|
|Severe||26 (39)||22 (21)||38 (32)||35 (32)||67 (55)|
|Suicide ever attempted, n (%)|
|Lamotrigine||20 (32)||33 (32)||50 (39)||33 (30)||50 (39)|
|Placebo||24 (36)||38 (37)||32 (27)||25 (23)||50 (41)|
To identify any additional randomised trials, we conducted a search of electronic databases including MEDLINE, EMBASE, CINAHL, PsycINFO, CENTRAL (online Appendix DS1).
Individual patient data-sets were compiled from the five GlaxoSmithKline-sponsored trials comparing lamotrigine with placebo. Analyses were of the full intention-to-treat trial populations. The last available observation was used for participants who withdrew from the trial before the end of the study. The a priori data analysis plan included analyses of both categorical and continuous outcomes. Trial-specific estimates of the relative risks of response (>50% reduction in baseline score on Hamilton Rating Scale for Depression Reference Hamilton13 (HRSD) and Montgomery–Åsberg Depression Rating Scale Reference Montgomery and Åsberg14 (MADRS)) and remission (<8 on HRSD and <12 on MADRS) were calculated and pooled using the Mantel–Haenszel fixed effect approach in metan in STATA version 9. The number needed to treat (NNT) was estimated from the inverse of the weighted mean absolute difference in event rates. For continuous measures, separate ANCOVA analyses were conducted of final score (adjusted for baseline) and then pooled using metan.
A planned subgroup analysis was conducted to investigate whether the treatment effect differed between (a) individuals with bipolar I and bipolar II disorder and (b) individuals with severe depressive illness and moderate illness at randomisation. As there is no universally accepted cut point for the HRSD, 15 we dichotomised the sample around the mean baseline score in the trials. We used meta-regression for the subgroup analysis which is a method that allows the investigation of whether any particular covariates are related to the observed study-specific treatment effects. Reference Thompson and Higgins16 In STATA, the metareg command uses a random effects iterative method to provide a restricted maximum likelihood estimate of the regression parameters (with their asymptotic variances) and the residual heterogeneity variance. Reference Sharp17 Meta-regression can be a powerful technique but its use in aggregated data is limited to the investigation of study-level variables. The use of individual patient data allows more informative meta-regression analyses of trial data because precisely defined subgroup analyses can be conducted applying consistent definitions across trials.
As a measure of overall acceptability, discontinuation rates from the study arms were compared in each trial using relative risks and pooled using the Mantel–Haenszel fixed effect approach in metan in STATA version 9.
There were five randomised controlled trials (total 1072 participants) conducted by GlaxoSmithKline to compare lamotrigine with placebo in bipolar disorder. Summary details of the trials are presented in online Table DS1 and Table 1, and are reported in detail elsewhere. Reference Calabrese, Huffman, White, Edwards, Thompson and Ascher9 All five trials compared lamotrigine monotherapy with placebo. Individuals were only included if they had discontinued any other psychoactive drug at least five elimination half-lives before trial entry. Prior treatment with lamotrigine was an exclusion criterion in all trials. Three trials (GW602/SCAB2001, SCA40910, SCA30924) included only people with bipolar I disorder, one trial (SCA10022) included only people with bipolar II disorder and one trial (GW603/SCAA2010) included people with both bipolar I and II disorder. Duration of the trial varied from 7 to 10 weeks. Lamotrigine dose was 50 mg or 200 mg in GW602/SCAB2001, flexible 100–400 mg in GW603/SCAA2010 and 200 mg in SCA40910, SCA30924 and SCA10022. After the initial result from GW602/SCAB2001, the 50 mg dose was considered too low and so was not included in the other trials. We did not therefore include the 50 mg arm in the meta-analysis.
Two non-GlaxoSmithKline sponsored trials were also identified. Reference Frye, Ketter, Kimbrell, Dunn, Speer and Osuch18,Reference van der Loos, Nolen and Vieta19 As individual patient-level data were not available from these trials, and the protocols differed substantially from the GlaxoSmithKline trials, these were not included in the main analysis. Both trials, however, reported substantial and statistically significant benefits with lamotrigine compared with placebo; in one case, in combination with lithium. Reference Sharp17
Individuals treated with lamotrigine were more likely to respond to treatment than those treated with placebo on both HRSD (Fig. 1(a)) (pooled RR=1.27, 95% CI 1.09–1.47, heterogeneity χ2=1.80, d.f.=4 P=0.772, test of RR=1, z=3.13, P=0.002) and MADRS (Fig. 1(b)) (pooled RR=1.22, 95% CI 1.06–1.41, heterogeneity χ2=3.12, d.f.=4, P=0.538, test of RR=1, z=2.80, P=0.005). The NNT to achieve one more response than would have been observed on placebo was 11 (95% CI 7–25) on HRSD and 13 (95% CI 7–33) on MADRS. Remission rates were not statistically significantly higher for lamotrigine on HRSD (pooled RR=1.10, 95% CI 0.90–1.36, heterogeneity χ2=9.03, d.f.=4, P=0.060, test of RR=1, z=0.93, P=0.351) but were on MADRS (pooled RR=1.21, 95% CI 1.03–1.42, heterogeneity χ2=5.86, d.f.=4, P=0.210, test of RR=1, z=2.30, P=0.021).
On continuous symptom measures, the weighted mean difference on MADRS was −1.43 (95% CI −2.80 to −0.06, P=0.04) and −1.01 (95% CI −2.17 to 0.14, P=0.084) on HRSD. The standardised mean difference on MADRS was −0.12 (95% CI −0.24 to −0.00, P=0.04) and −0.11 (95% CI −0.23 to 0.01, P=0.084) on HRSD.
Discontinuation rates for each study are shown in online Table DS1. Overall, there was no difference between lamotrigine and placebo (RR=1.02, 95% CI 0.93–1.11, P=0.731, heterogeneity χ2=4.95, d.f.=4, P=0.292).
There was no statistically significant interaction (regression coefficient −0.06, 95% CI −0.35 to 0.24, P=0.705) between diagnostic subgroup and treatment effect (bipolar type I RR=1.24, 95% CI 1.04–1.46; bipolar type II RR=1.15, 95% CI 0.90–1.47).
Baseline severity of depression
The mean HRSD score at randomisation was 24.37 (s.d.=3.83, range 18–37). We therefore dichotomised the HRSD at ≤24 and >24. There was a significant interaction by severity of depressive symptoms at randomisation (regression coefficient=0.30, 95% CI 0.14–0.60, P=0.04). Lamotrigine was superior to placebo in individuals with severe depressive symptoms at randomisation (baseline HRSD score >24, RR=1.47, 95% CI 1.16–1.87, P=0.001, NNT=7, 95% CI 4–17) but not in people with moderate symptom severity (RR=1.07, 95% CI 0.90–1.27, P=0.445) (Fig. 2). This interaction was replicated in a secondary meta-regression using the continuous MADRS score as the independent variable (regression coefficient=3.99, 95% CI 1.02–6.94, P=0.008). Lamotrigine was superior to placebo in individuals with severe depressive symptoms at randomisation (baseline HRSD score >24, standardised mean difference=–0.24, 95% CI −0.42 to −00.06, P=0.011) but not in those with moderate symptom severity (standardised mean difference=–0.02, 95% CI −0.19 to 0.14, P=0.442).
In the severe group, the response to lamotrigine rate was 110/242 (45.5%) compared with 71/236 (30.1%) in the placebo group. In the moderate group, the response to lamotrigine rate was 142/299 (47.5%) compared with 131/294 (44.6%) in the placebo group. Thus, the interaction by severity was because of a higher response rate in the moderately ill placebo-treated group, rather than, for example, a higher response rate in the severely ill lamotrigine-treated group.
Meta-analysis of the individual patient data from five manufacturer-sponsored randomised trials found consistent evidence of an overall modest benefit for lamotrigine. About 11 people would need to be treated to achieve one more response than would be achieved on placebo: this is at the margins of being clinically worthwhile. The magnitude of the treatment effect in the main analysis was reduced by the high placebo response rate in the moderately ill subgroup of participants. Consequently, the treatment effect appears to be more substantial in participants who are more severely ill (NNT=7) but the actual response rate on lamotrigine is very similar in both subgroups.
As with all quantitative reviews, this review is subject to a number of limitations. Publication bias, or the tendency for trials with negative or neutral findings not to be published, can seriously limit the reliability of meta-analysis. Moreover, studies from which individual data can be obtained may not represent an unbiased sample of all the trials. In this analysis, the bipolar depression trials included in the main analysis are the total of the acute studies conducted by GlaxoSmithKline, the manufacturer of lamotrigine. There is some evidence that trials conducted by the manufacturer of a drug may be more likely to detect and report results that favour their drug, Reference Montaner, O'Shaughnessy and Schechter20 but this is not the case here. Indeed, our search revealed two trials Reference Frye, Ketter, Kimbrell, Dunn, Speer and Osuch18,Reference van der Loos, Nolen and Vieta19 conducted independently of GlaxoSmithKline that both reported substantial benefits with lamotrigine compared with placebo in acute bipolar depression. This probably increases the confidence that we can have in the overall results.
This review was highly focused on the effect of lamotrigine on depressive symptoms and we did not request access to data on specific adverse events. Withdrawal rates were similar for participants allocated to lamotrigine and placebo. A comprehensive analysis of the rates of adverse events in these trials has been recently published. Reference Calabrese, Huffman, White, Edwards, Thompson and Ascher9 The most common adverse events were headache and nausea. Reference Calabrese, Huffman, White, Edwards, Thompson and Ascher9 The incidence of non-serious rash was low and there were no reports of serious rash in any of the five trials.
The finding of an interaction between the severity of depressive symptoms at randomisation and the size of the treatment effect appeared to be because of a larger placebo response in individuals who are less severely ill. This suggests that the finding should be interpreted cautiously clinically as it probably does not mean that only people who are severely ill are likely to respond to lamotrigine. The finding is, however, of considerable methodological importance, especially for the design of placebo-controlled monotherapy trials. Although there is still some consensus that placebo-controlled trials are required for regulatory purposes, Reference Charney, Nemeroff, Lewis, Laden, Gorman and Laska21 a number of artefacts and biases (such as inflation of baseline scores to meet eligibility criteria) can result from the difficulties of conducting placebo-controlled trials when existing standard treatments are available. 15,Reference Geddes and Cipriani22 These problems can inflate placebo response making it difficult to detect drug effects but they are often overlooked in the interpretation of the results of such trials. We believe that our results suggest that it is essential to ensure a reliable and sufficient severity of illness at baseline in participants in clinical trials to minimise such artefacts. This methodological challenge is likely to increase in importance because the increasing number of drugs for which an indication for bipolar disorder is being sought has made recruitment of more participants who are severely ill into placebo-controlled trials in bipolar disorder more difficult. Reference Calabrese, Rapport, Shelton and Kimmel23 Although placebo control remains a regulatory requirement for the development of new antidepressants, an adequate severity of symptoms at the point of randomisation will remain a challenging prerequisite for success.
There was no difference in response to lamotrigine between bipolar disorder type I and II subgroups. This parallels the findings in the BOLDER (BipOLar DepRession) trials of quetiapine, Reference Calabrese24,Reference Thase, Macfadden, Weisler, Chang, Paulsson and Khan25 and supports the idea that depressive episodes arising in an illness course characterised by mania or hypomania are probably very similar in terms of treatment response.
The effect of lamotrigine compared with placebo only became statistically significant when data from five randomised controlled trials were pooled by meta-analysis. This might mean that the true effect of lamotrigine is too small to be reliably detected by individual studies and, possibly, too small to be clinically important. However, the interaction by baseline severity suggests that the trial may actually have underestimated the true efficacy of lamotrigine. It is worth noting that a simple ‘vote counting’ analysis (as is often used by regulatory authorities) would fail to quantify the treatment effect at all. None of the individual trials reported a statistically significant effect on the primary outcome: Reference Calabrese, Huffman, White, Edwards, Thompson and Ascher9 this meta-analysis suggests that there is a modest but consistent effect across the trials. The current regulatory approach of requiring two ‘successful’ pivotal trials may encourage the conduct of multiple, underpowered trials. Conversely, there may be an unjustified focus on unrepresentatively large effects from one or two trials that meet criteria for statistical significance, ignoring (and failing to publish) any trials that do not produce a statistically significant result. Reference Chalmers and Lau26,Reference Easterbrook, Berlin, Gopalan and Matthews27 The vote-counting approach does not make the most efficient or reliable use of the total randomised data and routinely risks bias in the estimation of true treatment effects. In this case, vote counting would have failed to detect a positive treatment effect of uncertain clinical significance.
Finally, in addition to their clear methodological importance, the results of this meta-analysis are potentially of clinical importance for people with bipolar disorder and their doctors because they provide some solid empirical support for a commonly used drug treatment in a disorder with few proven therapies. It suggests that lamotrigine may be an effective treatment for acute bipolar depression arising in a bipolar disorder of type I or II as well as for prevention of relapse. Further trials are warranted to clarify the size of the treatment effect of lamotrigine both as monotherapy and in combination: the ongoing CEQUEL (Comparative Evaluation of QUEtiapine–Lamotrigine combination v. quetiapine monotherapy (and folic acid v. placebo) in people with bipolar depression) trial will provide further knowledge of the effectiveness of lamotrigine as add-on therapy to quetiapine in ‘real world’ people with bipolar disorder ().
The randomised trials included in this analysis were conducted and funded by GlaxoSmithKline, who provided raw data-sets for independent analysis by the authors. J.R.G. had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. This analysis was initiated entirely by the authors. No additional support, either scientific or pecuniary, was provided by GlaxoSmithKline. We are grateful to Walter Paska and Gary Evoniuk of GlaxoSmithKline for their assistance in accessing the individual patient data and for their comments on a draft of the manuscript. We are grateful to Ly-mee Yu and Will Stevens of the Oxford Clinical Trial Unit for Mental Illness for help with the analyses of the continuous data and data processing, and Sarah Stockton of the Centre for Evidence-Based Mental Health for designing and implementing the search strategy.