The British Journal of Psychiatry

Abstract

Background In individual studies and limited meta-analyses venlafaxine has been reported to be more effective than comparator antidepressants, particularly selective serotonin reuptake inhibitors (SSRIs).

Aims To perform a systematic review of all such studies.

Method We conducted a systematic review of double-blind, randomised trials comparing venlafaxine with alternative antidepressants in the treatment of depression. The primary outcome was the difference in final depression rating scale value, expressed as a standardised effect size. Secondary outcomes were response rate, remission rate and tolerability.

Results A total of 32 randomised trials were included. Venlafaxine was more effective than other antidepressants (standardised effect size was -0.14, 95% Cl -0.07 to -0.22). A similar significant advantage was found against SSRIs (20 studies) but not tricyclic antidepressants (7 studies).

Conclusions Venlafaxine has greater efficacy than SSRIs although there is uncertainty in comparison with other antidepressants. Further studies are required to determine the clinical importance of this finding.

Depressive disorders are the second most important cause of disability in developed countries (Murray & Lopez, 1997) but a substantial minority of depressed patients fail to respond to antidepressant treatment (Anderson et al, 2000). Although newer antidepressants have tolerability and safety benefits over older tricyclic antidepressants (TCAs), similar efficacy generally is reported (Edwards, 1992; Gross & Huber, 1999; Anderson, 2000; Geddes et al, 2001). It is potentially of great clinical importance if an antidepressant were to be more effective than comparators, and understanding why may shed light on how antidepressants work. It has been proposed (Nelson et al, 1991; Seth et al, 1992; Heninger et al, 1996) that antidepressants with a dual action of inhibiting the reuptake of both noradrenalin and serotonin (5-hydroxytryptamine, 5-HT) may be more effective than drugs acting on a single monoamine (e.g. selective serotonin reuptake inhibitors, SSRIs). Venlafaxine is the first drug to be marketed that inhibits both noradrenalin and 5-HT reuptake without actions at other receptors (Holliday & Benfield, 1995). We present a systematic review investigating the relative efficacy and tolerability of venlafaxine compared with other antidepressants.

METHOD

Relevant trials were identified from our existing database (Eccles et al, 2000) and from systematic searches of electronic databases. The search terms were VENLAFAXINE, EFEXOR or EFFEXOR. Databases searched included Medline, Embase, Biosis, PsychLit, National Research Register, Healthstar, SIGLE, Cochrane Database of Systematic Reviews, DARE, Cochrane Controlled Trials Register and Current Controlled Trials. A total of 2349 trials were identified from our electronic search strategy. We carried out a manual search of reference lists of included studies and requested unpublished data from authors and study sponsors.

Inclusion criteria

Trials were included if they were double-blind, randomised studies comparing venlafaxine with an alternative antidepressant for the treatment of depression. The definition of depression was intentionally broad and included explicit clinical or research criteria for major depression (such as DSM-IV; American Psychiatric Association, 1994) or if the clinician considered the patient to be depressed and eligible for antidepressant treatment. Two of the researchers (D.S. and C.D.) made an independent assessment of each potentially eligible study and disagreements were resolved through discussion within the team.

Data abstraction

Design characteristics and quality assessment

We abstracted data on the inclusion and exclusion criteria for each study, the dose and regimen of venlafaxine and alternative antidepressants, the adequacy of randomisation and concealment of allocation (as reported in the paper), number of patients randomised, loss to follow-up, form of analysis (completer analysis or last observation carried forward), relevant clinical outcomes reported, age and gender of participants and length of follow-up. When specific variables were not reported within a given trial, the authors of the paper were contacted to obtain the missing data. If this was unsuccessful, we contacted the sponsors. Data were abstracted on all available patients randomised in the trials and patients were analysed on the basis of initial random allocation to treatment group (intention to treat) whenever possible.

Clinical outcomes

The primary outcome was the mean depression severity measure assessed by the final (end of trial) Hamilton Rating Scale for Depression (HRSD; Hamilton, 1960), the Montgomery and Åsberg Depression Rating Scale (Montgomery & Åsberg, 1979) or the Clinical Global Impression (Guy, 1976), with preference given in that order if more than one scale was reported. Secondary outcome variables were response rate (typically 50% or greater drop in depression rating scale from baseline) and remission rate (depression rating scale below a certain score, e.g. HRSD <8). Data on tolerability were abstracted by collecting ‘all cause’ withdrawals from each treatment group and also the attributed reason for withdrawal from therapy (lack of efficacy and adverse effects).

Statistical analysis

The primary efficacy outcome was the pooled standardised difference in mean treatment effect. For this measure, standardised effect sizes (difference in final rating scale means divided by the within-study standard deviation) were estimated from the efficacy data for each treatment group. Where an estimate of study variance was not available, this was imputed by taking the average for the studies using the same outcome measure. Secondary binary outcomes of response and remission, as well as tolerability data, were calculated as the odds ratio and absolute risk difference.

A simulation method was used to estimate pooled treatment effects using Gibbs sampling in BUGS software (Smith et al, 1995; Freemantle et al, 1999). This method is analogous to standard methods but does not require large sample assumptions, making it superior in meta-analysis where these assumptions frequently are not met. It has the additional advantage that the predictive value of different factors, such as patient severity or dose, may be examined using meta-regression approaches (Freemantle et al, 1999). Absolute risk differences were calculated using standard methods (DerSimonian & Laird, 1986) and interpreted as ‘number needed to treat’ (NNT). Negative NNTs are often described as ‘number needed to harm’.

Fixed effects approaches to meta-analysis assume that each trial contributes an estimate of a constant population effect for a treatment, whereas random effects approaches assume that there is no single population effect but a distribution (range) of effects. Random effects models were used where venlafaxine was compared with a variety of agents (e.g. in comparison with SSRIs) but fixed effects models were used where venlafaxine was compared with individual agents.

Meta-regression was used to examine the predictive value of potentially important explanatory factors on the primary efficacy outcome measure (Freemantle et al, 1999). This hierarchical approach to data modelling enables examination of the effect of trial characteristics while preserving the structure of individual trials. The factors that we identified a priori were: size of trial; in-patient v. out-patient status; design criteria (last observation carried forward v. completer analysis). The analysis on size of trial is a particularly helpful method of identifying potential publication bias and is analogous to using a funnel plot. Other factors also investigated were age and gender, comparator drug class, length of follow-up, rating scale used (e.g. HRSD or Montgomery and Å sberg Depression Rating Scale), dose of venlafaxine and if the variance was imputed.

RESULTS

Included trials

A total of 32 studies met the inclusion criteria (Table 1), with comparisons of venlafaxine with TCAs (clomipramine, imipramine, dothiepin (dosulepin) and amitriptyline), SSRIs (fluoxetine, fluvoxamine, paroxetine and sertraline) and other drugs (trazodone and mirtazapine). There were 5562 patients in total1: 3844 in the twenty trials comparing venlafaxine with SSRIs (SSRI n=1857); 1356 in the nine trials comparing venlafaxine with TCAs (TCA n=579); and 418 in the three trials comparing venlafaxine with other drugs (other n=212). The average trial size was 179 patients (range 28-382). The average length of follow-up was 10 weeks (range 4-48). Most trials used the last observation carried forward for the primary analysis (see Table 1). For three of the trials, we imputed the measure of variance because the data were not available and could not be obtained from the authors or sponsors. None of the trials indicated whether concealment of allocation was conducted appropriately.

View this table:
Table 1

Description of included trials1

Primary outcome

There were 29 comparisons in the effect size analysis of clinical efficacy (Table 2). The overall effect size estimate was -0.14 (95% CI -0.22 to -0.07) in favour of venlafaxine. The size of effect (given a pooled standard deviation of 8.3) is equivalent to the final HRSD score, being about 1.2 points lower on venlafaxine. For the SSRIs, the effect size estimate was -0.17 (95% CI -0.27 to -0.08). Effect sizes for the TCAs and the ‘other drug’ categories were similar but not significantly different from venlafaxine (Table 2, Fig. 1).

View this table:
Table 2

Effect size analysis

Fig. 1

Plot of pooled efficacy of venlafaxine compared with other antidepressants. The bars show the effect size (difference in final rating scale score divided by pooled final standard deviation) and the 95% CI. Results falling to the left of the line of no effect (zero) indicate an advantage to venlafaxine.

The results appeared consistent across the SSRIs but there were differences between the TCA studies, notably imipramine: the effect size was -0.38 (95% CI -0.57 to -0.19), favouring venlafaxine, whereas there was no benefit in studies against other TCAs (Table 2, Fig. 1).

Response rates

Table 3 shows the estimated response rates. The overall odds ratio for response was 1.27 (95% CI 1.07-1.52). The risk difference was 0.05 (95% CI 0.02-0.09), with an NNT of 19 (95% CI 11-63). The pooled results for different drug classes were similar to this overall effect (Fig. 2).

View this table:
Table 3

Response analysis

Fig. 2

Plot of pooled response rate to venlafaxine compared with other antidepressants. The bars show the odds ratio and the 95% CI. Results falling to the right of the line of no effect (I) indicate an advantage to venlafaxine.

Remission rates

Table 4 and Fig. 3 display the pooled remission results. The overall odds ratio for remission rate was 1.36 (95% CI 1.14-1.61), favouring venlafaxine. The overall risk difference was 0.07 (95% CI 0.03-0.11), giving an NNT of 14 (95% CI 9-29).

View this table:
Table 4

Remission analysis

Fig. 3

Plot of pooled remission rate on venlafaxine compared with other antidepressants. The bars show the odds ratio and the 95% CI. Results falling to the right of the line of no effect (I) indicate an advantage to venlafaxine.

Remission rates were measured in only 18 of the trials and, of these, 16 used an SSRI agent as the comparator. The result for the pooled SSRI comparison was similar to the overall effect.

None of the factors that were hypothesised to influence the estimate of primary outcome were significantly predictive of greater efficacy in meta-regression analyses (analysis not shown).

Meta-regression analysis and visual inspection of funnel plots provided no evidence of publication bias, although did not exclude the possibility of the existence of such bias.

Treatment discontinuation

Table 5 shows an analysis of drop-outs by reason and comparator drug class. The overall risk difference of -0.004 (95% CI -0.029 to 0.020) indicates that there are 0.4% fewer drop-outs overall in the venlafaxine group, and the difference is not statistically or clinically significant. The only statistically significant drop-out comparison exists for drop-outs due to side-effects compared with the ‘other drug’ category, where there is a risk difference of 0.221 (95% CI 0.065-0.376), giving an NNT of 5 (95% CI 3-15) in favour of other drugs. However, because the overall difference in drop-out is equivalent, this result is countered by drop-out for all other causes.

View this table:
Table 5

Drop-out analysis by cause and drug class

DISCUSSION

Efficacy

This meta-analysis provides evidence that in the treatment of depressive disorders all antidepressants are not equal. Pooling data from all currently available studies reveals that venlafaxine carries an advantage of about 1.2 HRSD points over other antidepressants. The majority of comparisons were with SSRIs, where the effect appeared consistent across the different drugs. In contrast, it is less clear that the advantage is consistent across other antidepressants such as TCAs, where imipramine is the only individual drug that clearly demonstrates lesser efficacy. This does not, however, reduce the importance of the findings for the primary outcome measure — venlafaxine v. any other antidepressent in reducing symptoms of depression — in which a clear advantage was identified.

The results are of probable clinical significance, with an NNT of 19 (95% CI 11-63) for response and 14 (95% CI 9-29) for remission. The two data-sets do not include all of the same studies and are not as comprehensive as the data used in primary analysis of effect sizes, so the absolute figures must be viewed as approximate. However, this magnitude of advantage for venlafaxine over other antidepressants is potentially of considerable importance, given the often prolonged or even chronic nature of depressive episodes. It is increasingly recognised that improvement of depression on antidepressants is often incomplete or partial so that remission rates are relatively low (Ferrier, 1999) and only 42% of patients in the studies that we included achieved remission by the end of the study. Patients who fail to reach remission have significantly greater continuing morbidity and higher relapse rates than those who do experience remission (Cornwall & Scott, 1997). If only one extra person reaches remission when treated with venlafaxine instead of an SSRI for every 14 patients treated, then this is a potentially important health benefit. It suggests that even if not used first line, venlafaxine should be considered for patients having an inadequate response to other antidepressants.

Our study confirms the more limited meta-analysis recently reported by Thase et al (2001), which only included a small subset (eight) of studies against SSRIs and therefore cannot be considered systematic. It only assessed efficacy using remission rates with an odds ratio of 1.5 (95% CI 1.3-1.9) in favour of venlafaxine. The NNT was not calculated formally but appears to be about 10 from the difference in remission rates (45% v. 35%); this is a greater advantage to venlafaxine than we found with a larger data-set.

Our analysis of the tolerability of venlafaxine as measured by total treatment drop-outs and those due to side-effects did not suggest that greater efficacy was offset by poorer tolerability overall or against SSRIs or TCAs. More patients dropped out of treatment owing to side-effects on venlafaxine than trazodone or mirtazapine, suggesting poorer tolerability than these drugs, but the small number of studies makes it difficult to draw conclusions.

Mechanism underlying venlafaxine's greater efficacy

We have reported previously being unable to identify a relationship between pharmacology and efficacy using a meta-regression analysis of a variety of antidepressants compared with SSRIs (Freemantle et al, 2000). There were, however, considerable problems in that analysis, relating to being able to identify accurately the acute pharmacology of many antidepressants in vivo. In this study some of these problems are overcome through using a single agent and it appears that the most plausible mechanism by which venlafaxine may exert increased efficacy in comparison with SSRIs is its ability to inhibit not only 5-HT reuptake but also the reuptake of noradrenalin (Holliday & Benfield, 1995). Whether this is the mechanism in the case of venlafaxine has yet to be confirmed, however. The profile of its binding to human monoamine transporters suggests a weak affinity for the noradrenalin transporter compared with the 5-HT transporter (Owens et al, 1997; Tatsumi et al, 1997). At lower doses, venlafaxine appears to act as an SSRI and it is unclear at what dose significant noradrenalin effects occur. Preliminary evidence suggests that, at least outside the brain, this is somewhere between 75 and 225 mg, with one study suggesting that it may occur by 150 mg (Abdelmawla et al, 1999). It is of interest that previous meta-analyses have suggested superior efficacy for amitriptyline against other antidepressants, particularly SSRIs (Anderson, 2000; Barbui & Hotopf, 2001), which adds some support to dual action conferring greater efficacy than occurs when blocking the reuptake of a single transmitter.

We did not find an effect of dose on the size of the advantage to venlafaxine over SSRIs, raising some question as to the mechanism underlying its greater efficacy. However, the studies in this meta-analysis were not designed to detect dose—response effects, most employing flexible dosing. The lack of an association between efficacy and a venlafaxine dose below or above 150 mg is probably against a strong linear dose—response over the range used but cannot rule out a non-linear relationship. Two fixed-dose studies of venlafaxine against placebo have suggested a dose—response over the range 60-225 mg (Kelsey, 1996; Rudolph et al, 1998), but the differentiation between doses has not been statistically significant and the dose at which any possible greater efficacy may arise is not clear.

Methodological considerations

The major methodological challenge to all systematic overviews is publication bias — the selective availability of trials with positive results. The comprehensive search strategies used to identify trials, the systematic attempts to identify unpublished trials and unpublished data and examination of the distribution of the results from included trials all mediate against the importance of this threat to the validity of the results of this overview. However, it has to be acknowledged that the majority of studies were sponsored by the company that markets venlafaxine and sponsorship has been suggested as a potential factor influencing the outcome of the trials (Stewart & Parmar, 1996; Freemantle et al, 2000).

Although over 5000 patients were included in the trials identified for this meta-analysis, this number is small against other clinical areas where this number of patients commonly may be included in a single trial. Further randomised trials, including those of a naturalistic design, involving larger numbers of patients in different clinical settings (particularly primary care, where the majority of treatment for major depressive disorder is conducted) are required to find out how generalisable this result is to different settings and whether venlafaxine has increased effectiveness in usual clinical practice.

Clinical Implications and Limitations

CLINICAL IMPLICATIONS

  • The findings from this systematic review and meta-analysis provide some confidence that venlafaxine is more effective than selective serotonin reuptake inhibitors (SSRIs) with comparable tolerability.

  • The size of this advantage is of probable clinical importance when the prolonged or chronic nature of depression is taken into account.

  • Venlafaxine should be considered for patients in whom efficacy needs to be maximised and in those failing to respond to an SSRI.

LIMITATIONS

  • Apart from the comparison with fluoxetine, there are insufficient comparisons between venlafaxine and individual SSRIs and other antidepressants to draw strong conclusions with regard to specific comparisons.

  • Meta-analysis is dependent on the quality of individual studies included in the analysis.

  • Drop-outs are a relatively crude proxy for tolerability.

Footnotes

  • * The Editor, Greg Wilkinson, is in receipt of a consultancy fee from Neurolink (sponsored by Wyeth) so took no part in, and was kept blind to, the assessment of this paper. After enquiries by Professor Wilkinson, neither the Editorial Board member who acted as Editor in respect of this paper, nor those assessors who took part in the peer-review process, declared any interest relevant to the publication of this paper.

  • 1 Multiple comparisons were made in a number of trials. As a quality criterion, and a rule of thumb, we cite the total number of patients in the trials, rather than the comparisons included, because there is good empirical evidence that the quality of studies is affected directly by the overall size.

  • Received July 8, 2001.
  • Revision received January 18, 2002.
  • Accepted January 31, 2002.

References

APPENDIX: Studies included in the meta-analysis

View Abstract