NPC Archive Item: Reboxetine — a striking example of publication bias

NOTE – This is an archive post from the NPC and has not been updated since first publication. Therefore, some hyperlinks may no longer be working.
MeReC Rapid Review NPC Logo

1 November 2010

A meta-analysis of published and unpublished data on reboxetine for the acute treatment of major depression found it, overall, to be an ineffective and potentially harmful antidepressant. The published evidence overestimated the benefits of reboxetine while underestimating the harms. This study raises important questions about access to, and publication of, clinical trial data.

Level of evidence:
Level 1 (good quality patient-oriented evidence) according to the SORT criteria.

Prescribers should continue to follow the NICE clinical guideline on the management of depression. When it is appropriate for an antidepressant to be prescribed, NICE recommends that it should normally be a selective serotonin reuptake inhibitor (SSRI) in a generic form. The choice of agent largely depends on its safety and tolerability profile and its propensity to cause discontinuation symptoms as well as cost. Patient preference and previous experience of treatments is also important.

Health professionals should consider the implications of these new data, both in terms of reboxetine prescribing, and the potential impact of publication bias in clinical trials. Patients taking reboxetine for the treatment of depression should be identified and reviewed, and clinicians should carefully weigh the balance of risks and benefits on an individual patient basis.

What is the background to this?
Reboxetine is a selective inhibitor of noradrenaline, and is licensed for the treatment of major depression in many European countries (including the UK). It was rejected by the FDA for use in the United States in 2001. In the 2009 clinical guideline, NICE concluded that reboxetine was superior to placebo and as effective as other antidepressants in the treatment of depression. There was insufficient evidence on the tolerability of reboxetine compared with placebo or alternative antidepressants. These findings were based on efficacy and tolerability data from up to 1,068 trial participants, from six published trials of reboxetine. Furthermore, the NICE health economic analysis ranked the cost effectiveness of antidepressants for moderate and severe depression. Reboxetine was the least cost-effective of all antidepressants for both moderate and severe depression.

There have been few large, high quality, independently sponsored comparative trials of antidepressants. In addition, publication bias has been previously found in trials of antidepressants. The aim of this new systematic review and meta-analysis of RCTs was to assess the benefits and harms of reboxetine, compared with placebo or SSRIs in the acute treatment of depression, and to measure the impact of potential publication bias in trials of reboxetine.

What does this study claim?
The study was conducted by IQWIG, the German equivalent of NICE. Their preliminary health technology assessment (HTA) suggested that reboxetine had been tested in about 4,600 patients. However, Pfizer, the manufacturer of reboxetine, did not provide a complete list of unpublished trials as requested and data on almost two-thirds of patients were not accessible. A meaningful assessment of reboxetine was not initially considered possible due to a high risk of publication bias. The manufacturer provided the missing data after publication of the preliminary HTA report.

Data on 74% of patients analysed in the study were previously unpublished. For benefit outcomes (remission rates; 50% response rates), published data showed that reboxetine was superior to placebo. However, when the unpublished data were added, the full dataset showed no significant difference (see Table 1 in study details below). Also, whereas the published data showed no significant difference between reboxetine and SSRIs, the full data set showed that reboxetine was significantly inferior to SSRIs (see Table 2 below).

For harm outcomes (adverse events [AEs]; withdrawal due to AEs), a similar picture emerged. Published data showed no significant difference between reboxetine and placebo but when the unpublished data were added, the full dataset showed that reboxetine was inferior to placebo (see Table 1). Also, for rates of withdrawals due to AEs, whereas the published data showed no significant difference between reboxetine and SSRIs, the full dataset showed that reboxetine was significantly inferior to SSRIs (see Table 2). For patients with at least one AE, there was no significant impact of unpublished data in the comparisons between reboxetine and SSRIs.

How does this relate to other studies?
A meta-analysis of 117 RCTs by Cipriani and colleagues compared the efficacy of 12 second generation antidepressants. For more details on the results of this meta-analysis, and its limitations, see MeReC Rapid Review 283. In relation to reboxetine, the study found that it was significantly less efficacious and less well tolerated than all of the other 11 antidepressants. They concluded that reboxetine should not be used as a routine first-line acute treatment for major depression.

So what?
Overall, the authors of this meta-analysis concluded that reboxetine was ineffective as an antidepressant because it showed no benefit over placebo and was inferior to SSRIs for remission and 50% response rates. Published data on reboxetine overestimated the beneficial effect of reboxetine compared with placebo by 99–115% and of reboxetine compared with SSRIs by 19–23%. These findings have a number of important implications.

Richard Lehman recently commented that ‘reboxetine is an antidepressant drug which has never really caught on, and does not deserve to’. Although prescribing rates of reboxetine are low compared to other antidepressants (approximately 13,000 items per quarter for reboxetine vs. over 2.5 million items for citalopram)a, this could represent an estimated 4,000 patients taking reboxetine in England alone. This significant new evidence suggests an unfavourable risk:benefit profile for reboxetine. Clinicians should identify and review any patients taking reboxetine and, in discussion with patients, consider whether reboxetine is still an appropriate option for them.

The results of this meta-analysis are in contrast to the findings of other meta-analyses (including the NICE meta-analysis), which only considered published studies. An alarming 74% of the analysed clinical trial data on reboxetine were previously unpublished. In seven out of the eight comparisons the findings of published research were reversed when data from previously unpublished studies were included. The published evidence overestimated the benefits of reboxetine while underestimating the harms. This is a striking example of how publication bias can affect research.

As pointed out in a related BMJ editorial, estimates of treatment benefits are not always altered when previously unpublished clinical trial data become available. However, it is unclear as to the extent to which integration of missing data would support or refute key portions of the existing evidence-base. Meta-analyses of RCTs are usually considered the ‘gold standard’ of evidence — but it is difficult to form accurate conclusions if a significant proportion of available data is either unpublished or inaccessible.

This meta-analysis also raises questions about the role of drug manufacturers in providing full access to all data from clinical trials, to show a complete picture of their drugs. The authors highlighted the difficulties they encountered in obtaining all the reboxetine data from the manufacturer. This is not the first time this issue has made headlines. The two most notable examples are perhaps relating to rofecoxib and the omission of important cardiovascular safety data in the VIGOR study, and more recently the reliability of data on rosiglitazone from the RECORD study. Rofecoxib was withdrawn worldwide in 2004 and rosiglitazone was recently withdrawn across Europe. The authors of this new meta-analysis conclude that mandatory publication of clinical trial results is needed.

a Prescribing data supplied by NHSBSA for the quarter to December 2009 in general practice in England.

Study details
Eyding D, et al. Reboxetine for acute treatment of major depression: systematic review and meta-analysis of published and unpublished placebo and selective serotonin reuptake inhibitor controlled trials. BMJ 2010;341:c4737

Systematic review and meta-analysis of 13 double-blind RCTs for the acute treatment (6 weeks or more) of major depression, including unpublished data. Effect sizes in the published, unpublished, and overall dataset were compared.

4,098 adults with major depressive disorder as their primary diagnosis according to specified diagnostic criteria e.g. ICD

Intervention and comparison
Comparison of reboxetine with placebo or SSRI. 2,256 patients in the reboxetine vs. placebo comparisons, and 2,641 in the reboxetine vs. SSRI comparisons (citalopram, fluoxetine or paroxetine).

Benefit outcomes:

1) response rates (reduction in score on Hamilton depression rating scale of 50% or more from baseline to end of study)

2) remission rates (reduction in score on Hamilton depression rating scale to below an absolute threshold at end of study [score <10 in all except one trial where score threshold was <8]).

Harm outcomes:

1) rates of patients with at least one AE (any AE according to the definitions used in the primary trials)

2) withdrawals due to AEs (any AE according to the definitions used in the primary trials).


Table 1.

Reboxetine vs. placebo: Published studiesOdds ratio [OR] (95%CI) All studiesbOR (95%CI)
Remission 2.51 (1.49 to 4.25) 1.17 (0.91 to 1.51)
Response 2.47 (1.49 to 4.11) 1.24 (0.98 to 1.56)
Patients with AEs 2.67 (0.52 to 13.79) 2.14 (1.59 to 2.88)
Withdrawal due to AEs 0.95 (0.45 to 1.99) 2.21 (1.45 to 3.37)

Table 2.

Reboxetine vs. SSRI: Published studiesOR (95%CI) ALL studiesbOR (95%CI)
Remission 0.98 (0.68 to 1.40) 0.80 (0.67 to 0.96)
Response 0.95 (0.66 to 1.38) 0.80 (0.67 to 0.95)
Patients with AEs 1.07 (0.72 to 1.61) 1.06 (0.82 to 1.36)
Withdrawal due to AEs 1.58 (0.81 to 3.08) 1.79 (1.06 to 3.05)

b All studies includes published and previously unpublished studies

Sponsorship: The German Institute for Quality and Efficiency in Health Care (IQWiG).
More information on depression can be found on the depression section of NPC.

Please comment on this blog in the NPC discussion rooms, or using our feedback form.

Make sure you are signed up to NPC Email updates — the free email alerting system that keeps you up to date with the NPC news and outputs relevant to you