Our network meta-analysis supports the conclusion from previous reviews that, in certain populations, the newest antidepressants have a somewhat better efficacy than the generically available SSRIs. In our decision-analytic model we found that using escitalopram instead of other antidepressants was associated with a reduction in time spent in depression, leading to marginally more QALYs at no additional cost.
As noted in , there are large amounts of uncertainty in the total costs and QALYs for each investigated treatment and the differences in effect between some of the investigated treatments are rather small. Still, the probability that escitalopram is cost-effective is approximately 70% at a wide range of willingness-to-pay thresholds (). A key finding of the present study is that the relatively small differences in remission rates between therapies may have relatively large implications for differences in costs, in particular from a societal perspective, whereas the corresponding differences in QALYs are rather small.
Remission of symptoms is nowadays seen as a better outcome to study than response (defined e.g. as a 50% decrease on a rating scale). However, our analysis is generally in strong agreement with a previous study using response as outcome 
. This lends some support to the fact that the choice of endpoint to measure efficacy may not be crucial. However, the present study which explicitly attempted to estimate costs and quality of life implications of remission indicates that escitalopram is cost-effective. This conclusion is different from the tentative conclusion that sertraline is the most cost-effective alternative reported in a previous study 
An interesting observation was that the odds ratio for escitalopram against citalopram is lower in the MTC compared with pooling the head to head studies, which were all sponsored by the manufacturer of these drugs. We note that in several cases, the odds ratios from pooling direct comparative studies differ from those reported in the multiple treatment comparison. This is not surprising as the latter includes a larger body of evidence, but the underlying reasons for the differences could be worthy of further investigation as this may partly reflect a “wish-effect” (or sponsor bias) in the trials. This potential bias is reduced in the multiple treatment comparison where the drugs also appear as non-sponsored comparators.
Our study has some important strengths. First, it provides a thorough investigation regarding probabilities of remission in major depression. The large number of studies included in the systematic review provides a solid evidence base for the meta-analysis, and by using Bayesian multiple treatment comparison techniques all available evidence is efficiently utilised in the evidence synthesis. Second, against the background of little previous research evaluating the cost-effectiveness of all relevant antidepressants simultaneously in a primary care setting, we provide a framework for their evaluation. The study provides a first attempt to estimate health outcomes and cost consequences of achieving remission and as such also supports health-care decision makers in their need to include cost-effectiveness considerations systematically when deciding how to best allocate scarce health-care resources in the treatment of major depressive disorder.
Our study also has limitations. The cost data used in the health economic model come from Sweden, which could reduce generalizability of the results to other countries and settings. However, there is little evidence that the costs would differ dramatically in other European countries 
and the results are also quite robust to changes in the cost parameters. Second, studies with six weeks duration or more were included in the meta-analysis, but six weeks may be too short to study remission. However, excluding studies of duration less than eight weeks did not alter the results significantly. A related limitation concerns selection bias regarding which studies were included in the meta-analysis. A genuine attempt was made to locate unpublished material, but we may of course have failed to find important studies. Also, studies were only included if remission was reported. It has been suggested that this endpoint is more likely to be reported if a positive effect is shown for the sponsor’s drug 
. A solution to this would be to calculate remission for all studies in uniform fashion as is done in Omori et al 
. Furthermore the present study is limited to pharmacologic therapy and it should be recognized that both psychological and pharmacologic therapies appear effective in the treatment of depressive disorders, each with its own merits. It is very important to be able to individualize treatment of depression given the idiosyncratic response and tolerability issues as well as the lack of patient adherence. Our multiple treatment comparison is only able to address a subset of this large issue. However, for many decision makers, the question of whether there are efficacy (or cost-effectiveness) differences between antidepressant drugs is in itself important and worthy of careful analysis. An area for further research would be to incorporate psychological and pharmacologic therapies in a joint evaluation. Such an analysis could be based on the methods outlined in this work, although the approach to the evidence synthesis and cost-effectiveness model would have to be slightly amended.
How side effects should be handled in the analysis is a difficult question. Side effects are reported in a much less consistent way than efficacy and the reporting itself thus introduces a bias into the analysis. The most reliable measure is probably drop outs due to adverse events
. This is also a clinically relevant measure which directly affects the efficacy of the treatment. It should be noted that drop-out effects on efficacy are captured in the efficacy measure in the trials and is therefore included in the cost-effectiveness evaluation. However, there is also a direct effect on patient’s quality of life from negative side effects. The QALY-weights used in the current cost-effectiveness evaluation include side effects, but represent only an average negative effect since they were measured in a naturalistic setting and the treatments of individual patients were not reported. It should be noted that escitalopram had a rather favorable side-effect profile compared to the other treatments (Table S3
) and it is unlikely that the results would have changed if the effect of adverse events on quality of life had been explicitly modeled. Rather, we would expect the present results to be further reinforced should such an approach be taken.
Finally, it should be noted that the MTC only included active therapy and placebo arms were excluded. While we recognize that this design actually leave out some relevant evidence as placebo controlled trials also include important information on active therapy, head to head trials are the gold standard in generating clinical efficacy data and including only head to head trials implies that a potential placebo effect is actually removed from the measurement of relative efficacy in the meta-analysis.
Employing a large body of randomized head-to-head evidence, escitalopram has the highest probability of remission of the investigated antidepressants and is the most effective and cost-effective pharmacological treatment strategy for moderate to severe depression in a primary care setting, when evaluated over a one year time-horizon. The difference in effect is modest but even small differences in remission rates may be important when assessing costs and cost-effectiveness of antidepressants.