OUP user menu

A critical review of pharmacotherapy for major depressive disorder

Jamie M. Dupuy, Michael J. Ostacher, Jeffrey Huffman, Roy H. Perlis, Andrew A. Nierenberg
DOI: http://dx.doi.org/10.1017/S1461145711000083 1417-1431 First published online: 1 November 2011


Newer generation antidepressant drugs, with improvements in safety and tolerability, have replaced tricyclic antidepressants as first-line treatment of depressive illness. However, no single antidepressant drug from any class has distinguished itself as the obvious first-line treatment of major depression. The choice of therapy is driven primarily by patient choice, with informed consent for the risks of adverse effects. Cost has become an additional factor in this decision as several of the newer antidepressant drugs are now available in generic form. Several augmentation and drug-switching strategies have demonstrated benefit in refractory illness. While no single strategy distinguished itself as superior to the others, some have been more rigorously tested. Ongoing efforts at improving effectiveness, time to response, and tolerability have led to novel drug therapies. Efforts at characterizing predictors of treatment outcomes now include pharmacogenetic studies.

Key words
  • Antidepressant
  • major depressive disorder
  • outcomes
  • pharmacological treatment


Major depressive disorder (MDD) is estimated by the World Health Organization to be a leading cause for loss of disability-adjusted life years. In the USA, the 2001 National Comorbidity Survey showed overall lifetime prevalence estimates for MDD of 16.2% (Kessler et al. 2003). Over 50% of respondents with MDD received some type of treatment in the 12 months before their interview. However, only about one-fifth of patients received ‘adequate’ treatment during that time, indicating that more efforts are required at improving access to and utilization of care. Even for patients who are treated, pharmacotherapy is not uniformly effective. Many episodes do not completely respond to initial antidepressant, response may be delayed for weeks or months, and residual symptoms may cause significant morbidity. In this paper, we review the evidence for first-line treatment of MDD, and strategies for patients with treatment-resistant depression.


For the purposes of this review, we searched Medline and PsycINFO for controlled trials published in English between January 1981 and October 2010, in which adults with MDD were randomly assigned to receive medication, placebo, or active comparator drugs, and for all meta-analyses of pharmacotherapy for MDD. The number of published randomized placebo-controlled trials (RCTs) of antidepressants for the acute treatment of MDD is too vast to allow discussion of each study. In order to recommend first-line treatment for MDD, we examined meta-analyses of RCTs of antidepressant drugs for MDD as a statistical means of weighing the relative efficacy of antidepressants that have not been compared directly. Because placebo response rates in antidepressant trials tend to be high, many positive trials have small effect sizes, and meta-analysis of multiple trials may be used to determine whether drug–placebo differences are meaningful.

First-line treatment for MDD

Although an array of options exists for first-line treatment of MDD, no single agent has distinguished itself as clearly superior. Selective serotonin reuptake inhibitors (SSRIs) are still the most widely prescribed antidepressants, due to safety, tolerability, and cost. However, newer agents have gained in market share, as patients and clinicians search for treatment options while struggling to contain costs. Many trials have investigated the efficacy of SSRIs compared to other antidepressants.

SSRIs vs. tricyclic antidepressants (TCAs)

A Cochrane Collaboration meta-analysis identified 98 trials comparing SSRIs to other antidepressants, and failed to detect any clinically significant difference in efficacy between SSRIs and TCAs (Geddes et al. 2000). Another Cochrane Collaboration meta-analysis investigated the tolerability and efficacy of the TCA amitriptyline in comparison with other antidepressants and SSRIs, and found no difference in overall efficacy between amitriptyline and other TCAs or the SSRI comparators, but tolerability and acceptability measures favoured SSRIs (Guaiana et al. 2007). SSRIs also demonstrate efficacy for depression, without clear evidence of superiority over older drugs, when studied in particular patient subgroups. A smaller meta-analysis including 365 SSRI-treated geriatric depressed patients found SSRIs and TCAs to be equally efficacious (Wilson et al. 2003).

SSRIs demonstrate benefit over older medications in tolerability and acceptability. A Cochrane Collaboration review identified 136 randomized trials in which SSRIs and TCAs were compared among depressed patients, and found a modest but significant difference favouring SSRIs in terms of discontinuation of treatment (Barbui et al. 2003). SSRIs have the advantage of safety and tolerability over TCAs, and monoamine oxidase inhibitors (MAOIs), and remain the treatment choice for MDD.

SSRIs vs. newer-generation antidepressants

Antidepressants with dual-action and triple-action mechanisms have proved effective for depression in both outpatient and in-patient settings, and in placebo-controlled and comparator trials. Data suggest that these newer-generation antidepressants improve response over SSRIs. For example, a meta-analysis of 93 trials of treatment for MDD with antidepressants that have serotonergic plus noradrenergic action showed that these drugs were slightly more efficacious in achieving clinical response than were SSRIs (response rate of 63% vs. 59%) (Papakostas et al. 2007). However, the number needed to treat (NNT) statistic was 24. In addition, some studies have shown that mirtazapine offers an earlier onset of response than SSRIs (Versiani et al. 2005), but use of this medication may be limited by its side-effect profile (Papakostas et al. 2008).

Overall, efforts at distinguishing efficacy and acceptability between SSRIs and newer antidepressants have been limited by the lack of RCTs examining direct, head-to-head comparisons between medications. In 2009, Cipriani and colleagues conducted a systematic review of 117 RCTs comparing 12 new-generation antidepressants, in terms of efficacy and acceptability, as monotherapy for the acute-phase treatment of MDD, with outcome measures at 8 wk (Cipriani et al. 2009). Data from 25 928 individual subjects were included in the study, of which approximately two-thirds were women. In analysing the data, the authors used number of patients who responded as a measure of ‘efficacy’, and number of patients who dropped out as a measure of ‘acceptability’. Direct pair-wise comparisons of efficacy and acceptability were performed, as well as a multiple-treatments meta-analysis. The authors found that, in terms of response, mirtazapine, escitalopram, venlafaxine, and sertraline were more efficacious than duloxetine, fluoxetine, fluvoxamine, paroxetine, and reboxetine [odds ratio (OR) 1.22–2.03]. In terms of acceptability, the better-tolerated drugs were escitalopram, sertraline, citalopram, and bupropion. In a complex statistical analysis combining and comparing both response rates and discontinuation rates, the authors conclude that sertraline and escitalopram, followed by mirtazapine and venlafaxine, have better risk-benefit profiles than other antidepressants. Considering the additional element of cost, the authors suggest that sertraline might be the best first choice for treating moderate to severe depression.

This study was limited by the lack of data from direct comparisons between the medications. In the absence of head-to-head comparisons, the authors analysed comparative efficacy by using fluoxetine as a reference drug. This statistical method has the advantage of using available data to draw inferences about drugs that have never been directly compared. However, the results must be interpreted with caution, and must not be viewed as having the same validity that analysis of results from RCTs would provide. In addition, these authors interpreted data from studies of different patient populations, and compared outcomes measured by different rating scales. They used response rate as a dichotomous variable, rather than change in score on symptom rating scales, which could have inflated differences between treatments. The use of discontinuation rate as a proxy for tolerability and safety is problematic as well, since different medications can cause widely varying side-effects or adverse events.

To illustrate the difficulty in interpreting the data, we next consider another large-scale meta-analysis comparing second-generation antidepressants, by Gartlehner et al. (2008), who used similar methods but came to a somewhat different conclusion. Their analysis of 203 studies produced a conclusion that there are no substantial differences in comparative efficacy and effectiveness of second-generation antidepressants for the treatment of MDD. However, there were differences identified among individual drugs. For example, bupropion had fewer sexual side-effects, and mirtazapine had a faster onset of action, compared to some of the SSRIs. In subgroup analyses, no patient factors were identified that could allow prediction of response or non-response to an individual drug.

Subpopulations and predictors of response

Symptom severity at baseline

In studies of antidepressants, baseline symptom severity may affect treatment outcome, such that more severely depressed patients have a more robust response (Khan et al. 2002; Kirsch et al. 2008). However, the generalizability of these data has been limited by exclusion of patients with ‘minor’ depression, or less severe major depressive symptomatology. In a 2010 meta-analysis, Fournier and colleagues combined data from six RCTs that included patients with the lower range of baseline symptom severity [Hamilton Depression Rating Scale (HAMD) scores from the low teens] (Fournier et al. 2010). The authors analysed individual patient data, allowing for a ‘mega-analysis’. The authors employed a severity x treatment interaction statistic, and concluded that the efficacy of antidepressant medication for depression varies considerably as a function of symptom severity. The advantage over placebo was non-existent or negligible among patients with mild, moderate, and even severe baseline symptoms, while the advantage over placebo was large for patients with very severe symptoms. This result is consistent with the findings of Khan et al. and Kirsch et al. summarized above. A limitation was that Fournier's analysis interpreted data from small sub-populations of patients with minor depression or mild-moderate depression, and the results may not be generalizable to this population in clinical practice. Additionally, it should be noted that of the thousands of clinical trials of antidepressants, very few have included patients with mild or moderate depression. This lack of evidence for antidepressant effect should not be misinterpreted as negative evidence. As nearly half of outpatients meeting diagnostic criteria for MDD have symptoms in the mild-moderate range (Zimmerman et al. 2002), it is clear that this question has significant clinical implications, and requires further investigation.

For those patients who do have severe and melancholic forms of depression, limited data are available to determine whether one antidepressant or class is preferable to another. In a study of depressed in-patients, one large meta-analysis found TCAs to have a small amount of benefit over SSRIs (Geddes et al. 2002). The authors caution that this difference was attributed to a one-point difference on the HAMD, and may be due to chance. Meta-analysis showed that while depressed in-patients responded better to TCAs vs. SSRIs, the SSRIs were moderately more tolerable (Anderson, 1998, 2000). In the STAR*D study, nearly a quarter of subjects had melancholic features at baseline, and these patients had higher severity scores, higher current suicidal risk, higher rates of previous suicide attempts, and more psychiatric comorbidity at enrolment (McGrath et al. 2008). After initial treatment with citalopram, patients with melancholic features showed a 24% decrease in remission rate relative to patients without melancholic features. Together, the findings above suggest that the presence of melancholic features may be a negative prognostic factor for SSRI treatment, particularly among in-patients, but, in light of the STAR*D findings, perhaps in outpatients as well. However, it is not clear whether the STAR*D Level 1 (citalopram treatment) finding can be generalized to include response to other SSRIs. For example, as patients with melancholic features participated in subsequent levels of treatment in STAR*D, it did not appear that they responded differentially to sertraline vs. bupropion-SR or venlafaxine-SR (Rush et al. 2008). The limitations of the STAR*D study included lack of placebo control, and lack of structured diagnostic algorithms to ascertain and track melancholic features.

Patients at risk for suicide

Patients with MDD are at higher risk for suicide, and guidelines indicate that patients should be assessed for suicide at the start of treatment and regularly over the course of treatment (Fochtman & Gelenberg, 2005). In 2005, Fergusson and colleagues conducted a large meta-analysis of RCTs comparing SSRIs with placebo or an active comparator, to investigate this association (Fergusson et al. 2005). The authors examined 702 clinical trials, comprised of n=87 650 patients, and included trials of SSRIs for MDD, but nearly 60% of included trials had investigated SSRIs for other clinical conditions. In meta-analysis, the authors found a significant increase in the odds of suicide attempts [OR 2.28, 95% confidence interval (CI) 1.14–4.55, NNT to harm 684] for patients receiving SSRIs compared to placebo. In the pooled analysis of SSRIs vs. TCAs, there was no difference in the odds ratio of suicide attempts (OR 0.88, 95% CI 0.54–1.42). Notably, the authors reported, ‘Estimates for patients with major depression favored a decrease in suicides with SSRIs, whereas patients with depression and other clinical indications may have as much as an eightfold increase in the rates of suicide.’ This meta-analysis had several limitations, including that only about half of the trials included directly reported rates of suicide attempts, and most trials included small numbers of patients followed for short durations (average of 10 wk), making it difficult to analyse longer-term outcomes. And, as Mulder and colleagues point out, much of the evidence used to assess risk of suicidality is generated by industry-sponsored antidepressant trials, which typically exclude patients with high-risk factors including current suicidal ideation, substance abuse, or comorbid personality disorders (Mulder et al. 2008). This practice essentially produces studies of depressed patients without suicidal ideation or recent attempts, and biases trials to detect emerging suicidal ideation and behaviour, as well as making the data difficult to generalize to outpatients seen in clinical practice.

In a prospective trial, Mulder and colleagues followed a group of unselected outpatients who were started on antidepressants, for a 6-month study period (Mulder et al. 2008). The study population included patients with comorbid Axis I and II disorders, and suicidal ideation at study entry. Patients were randomly assigned to fluoxetine or nortriptyline, and clinicians were allowed to titrate doses or make augmentation or switching decisions based on clinical response. One-hundred seventy-six patients completed follow-up at 6 months. The proportion of patients with significant suicidal ideation [Montgomery–Asberg Depression Rating Scale (MADRS) suicide item score ⩾3] fell from 47% to 14% during the first 3 wk of the trial. Around 10% of patients did have ‘emergent’ suicidality, but of these, it was found that nearly three-quarters did have suicidal ideation during the 6 months prior to study entry. Excluding those with prior suicidal ideation, only 3% of patients had emergent suicidal ideation for the first time after starting antidepressants. The authors firmly conclude that antidepressant use is associated with a reduction in suicidal ideation and attempts.

The Mulder study supports the findings of other investigators, who, in epidemiological studies, found decreased rates of suicide associated with prescriptions of antidepressants (Gibbons et al. 2005, 2006; Nakagawa et al. 2007). An interesting study by Gibbons amd colleagues examined epidemiological data for suicide rates among adolescents in 2003–2005, after US and European regulators issued warnings about a possible association between antidepressants and suicide risk (Gibbons et al. 2007). Subsequent to the warning, antidepressant prescriptions decreased by about 20% in the USA and The Netherlands. During that period, youth suicide rates increased strikingly in both countries. Again, the epidemiological data appear to support the utility of SSRIs and newer-generation antidepressants in reducing risk for suicide across age groups. However, much debate continues about the risk for suicide among child and adolescent patients who are prescribed antidepressants, and the FDA black-box warning was extended in 2006 to include young adults (Leon, 2007). While the data may be conflicting, it is without question that any patient at risk for suicide requires careful monitoring and ongoing assessment of risk. (See below for a discussion of the antisuicidal properties of lithium.)

Timing of clinical improvement

The delayed onset of antidepressant therapeutic effect is an obstacle in the treatment of depression. The full therapeutic effect of antidepressants may take many weeks to develop, leaving patients to endure long periods of depressive symptomatology before making subsequent treatment decisions. Recently, data have emerged that re-examine the necessity of waiting 8–12 wk for a full antidepressant trial, and challenge the notion that early responders are experiencing a placebo effect. For example, in a meta-analysis of eight studies (n=7121 patients), Papakostas and colleagues compared early sustained response rates between antidepressant- and placebo-treated adults with MDD (Papakostas et al. 2006). Sustained early response, defined as clinical improvement not followed by worsening of symptoms, was more likely among antidepressant-treated patients than among patients receiving placebo at weeks 1 or 2 (OR 1.50 and 2.06, respectively). The authors conclude that a ‘true’ antidepressant response can occur within the first 1 or 2 wk of treatment. An earlier study of response to fluoxetine showed that more than half of eventual responders will show response by week 2, and over 75% respond by week 4; conversely, if patients had not experienced onset of response by weeks 4–6, they were unlikely to respond by the end of the 8-wk study period (Nierenberg et al. 2000). In addition, a study of 182 outpatients showed that earlier clinical improvement with fluoxetine predicted greater symptom resolution at 8 wk (Papakostas et al. 2007b). This study was limited by lack of a placebo comparator group, and lack of follow-up for a longer duration.

Wade et al. (2009) performed a pooled analysis of four RCTs of escitalopram vs. a comparator antidepressant in MDD. Onset of improvement at week 2 was correlated with response at week 8: 80% of patients with onset at week 2 responded by week 8, but among those without onset of improvement at week 2, only 43% responded by week 8. Of those who did not show improvement at week 4, only 13% had responded at week 8. In this meta-analysis the authors also showed that response at week 8 predicted a greater probability of achieving remission, and completing 6 months of treatment.

Thus the emerging data suggest that early response is predictive of longer-term treatment outcomes. The application of this data to clinical practice is not as clear, as the clinical trials have been limited by lack of placebo control, and lack of longer-term follow-up. More studies in this line of inquiry could prove very valuable to patients, should there come a time when the data support evolving the treatment strategy at the earliest possible opportunity.

Continuation and maintenance treatment

After resolution of an acute episode, the period of time to continue antidepressant treatment has not been definitively determined. Longitudinal, naturalistic follow-up data of patients who recover from an index episode of MDD found that 85% of subjects had a recurrence over the 15-yr study (Mueller et al. 1999). After remission of an acute episode, the most robust predictor of relapse is having had a prior episode (Coryell et al. 1991; Keller et al. 1983; Maj et al. 1992; Mueller et al. 1999; Roy-Byrne et al. 1985; Simpson et al. 1997). A 10-yr follow-up study of 318 subjects after an index episode of depression found that the risk of recurrence increases with each subsequent episode (Solomon et al. 2000). Another strong predictor of recurrence is severity of illness in the acute episode. Angst et al.'s (2003) 40-yr follow-up of patients initially hospitalized for unipolar or bipolar depression found steady recurrence rates over the study for this more severely ill population, and suggests that maintenance treatment is warranted for more severely ill patients.

Residual symptoms after remission of an acute episode also predict recurrence (Judd et al. 1998; Nierenberg et al. 2010). Kennedy & Paykel (2004) followed-up with 60 severely and recurrently depressed patients, at 8–10 yr after remission from an index episode of depression. The presence of significant residual symptoms predicted more time with depressive symptoms, as well as impaired social functioning and reduced quality of life. In the STAR*D trial, at 1-yr follow-up, over 90% of patients who had remitted with citalopram had at least one residual symptom, and, concordant with other findings, patients with more residual symptom domains had a higher probability of relapse (Nierenberg et al. 2010). In contrast, those patients with complete resolution of their symptoms have a longer time to next episode than those patients who do have residual depressive symptoms (Pintor et al. 2003).

Most antidepressants have been studied in continuation after resolution of an acute depressive episode, and appear superior to placebo for preventing relapse (Weihs et al. 2002). A thoughtful and rigorous pooled analysis of 31 trials of antidepressant continuation up to 36 months found that the benefit of antidepressants over placebo was sustained no matter what the length of continuation (Geddes et al. 2000). Most of the studies examined were of 12 months duration, so the authors suggest that the data for longer continuation be confirmed. Patients with shorter (1–2 months) and longer (4–6 months) treatment before randomization to continued antidepressant, and from 6 months to 36 months of follow-up after randomized treatment all appeared to have similar reductions in proportional risk. The results for different classes of antidepressants were similar, suggesting that all antidepressants that are effective in treating depression will be effective for maintenance treatment. To examine the efficacy and effectiveness of second-generation antidepressants in preventing relapse and recurrence during continuation and maintenance phases of treatment, Hansen et al. (2008) assembled a large meta-analysis of 23 placebo-controlled, and four active comparator trials. Pooled data for the class of second-generation antidepressants compared to placebo suggested a large effect size, and the NNT to prevent both relapse and recurrence was five patients, suggesting the overall benefit of these medications for continuation and maintenance.

Treatment-resistant depression

A significant proportion of patients do not respond to initial antidepressant treatment. Before switching or augmenting antidepressants, initial treatment must be optimized; under-utilization and under-dosing of antidepressants for MDD have been documented repeatedly. ‘Pseudoresistance’, as this phenomenon has been termed, is the result of inadequate dosing or inadequate treatment duration, and must be differentiated from true treatment resistance (Nierenberg & Amsterdam, 1990). Under-prescribing is as ineffective over the long term as no pharmacological treatment at all (Leon et al. 2003).

Antidepressant switching

A review of the evidence for switching pharmacotherapy after a first trial with a SSRI examined eight RCTs and 23 open studies (Ruhe et al. 2006). A limitation of this review, and indeed of most studies of this patient population, is that definitions of response, remission rate, and treatment-resistant depression differed. Response rates and remission rates varied widely between the included studies, from 12–86% and 5–39%, respectively. The authors concluded that no unequivocal evidence is available to prove differences in outcome between switches within classes or between classes of antidepressants.

In STAR*D, n=727 patients were randomly assigned to the switch strategy option in Level 2, after failing to remit with Level 1 treatment (citalopram) (Rush et al. 2008). The overall remission rate in Level 2 was about 31% (Rush et al. 2006a). About one-quarter of patients who switched to another SSRI (sertraline), a serotonin norepinephrine reuptake inhibitor (SNRI) (venlafaxine-XR), or a norepinephrine dopamine reuptake inhibitor (NDRI) (bupropion-SR) achieved remission with the second antidepressant, and the overall side-effect burden and rate of serious adverse events did not differ (Rush et al. 2006b). Clinical symptom patterns or demographic measures were not useful for predicting which patients would respond to any particular medication (Rush et al. 2008). In Level 3, about 14% of participants achieved remission (Rush et al. 2006a). Patients entering this level could be considered treatment-resistant, having failed to achieve remission with two different antidepressant regimens. As a strategy, switching to a third antidepressant monotherapy resulted in lower remission rates than in the first two levels (<20%) (Fava et al. 2006). There were no statistically significant differences in remission between a second monotherapy switch to mirtazapine (12%) or nortriptyline (20%), and both were equally tolerable (Fava et al. 2006). Patients entering Level 4 had failed three consecutive antidepressant trials, and were considered to have highly treatment-resistant illness. About 13% of subjects achieved remission in this level (Rush et al. 2006b), and there was no significant difference in remission between those who switched to the MAOI tranylcypromine (7%), or those who switched to venlafaxine-XR plus mirtazapine (14%) (McGrath et al. 2006). However, interpretation of the efficacy of tranylcypromine is limited by its poor tolerability; nearly half of patients receiving the MAOI discontinued the medication before 6 wk. There are many valid criticisms of STAR*D. These include use of lower-than-recommended medication dosages in later levels, broadly inclusive enrolment criteria that may have dampened effect sizes, and underpowered trials in the later levels of the study.

Antidepressant augmentation

Additional antidepressants

A double-blind, placebo-controlled trial of mirtazapine augmentation of patients who remained depressed on antidepressant monotherapy was performed (Carpenter et al. 2002). Twenty-six patients were randomized to receive mirtazapine or placebo for 4 wk. Response to mirtazapine was marked, with a response rate of 64% (7/11) for those on drug compared to 20% for placebo (3/15) (p=0.043). The remission rate was 45.4% for drug and 13.3% for placebo. Aside from its small size, a limitation of this study is the wide range of doses of a broad array of antidepressants that were augmented with mirtazapine. It is difficult to state whether one antidepressant or class of antidepressant is more likely to benefit from mirtazapine augmentation.

STAR*D allowed an RCT of adding a second antidepressant to an ineffective first antidepressant. The Level 2 switching strategy randomly assigned patients to augmentation of citalopram (a SSRI) with bupropion-SR (a NDRI) or buspirone (a partial agonist at the 5-HT1A receptor) (Trivedi et al. 2006). Both augmentation strategies led to similar rates of remission, 30% each, as measured by reduction in HAMD scores. However, bupropion-SR was associated with a greater reduction in the Quick Inventory of Depressive Symptomatology – Self Report (QIDS-SR), a patient self-report of depression symptoms, and also was somewhat better tolerated by patients over buspirone.

Atypical antipsychotics

The 1990s saw the introduction of second-generation antipsychotic drugs, with an advantage over typical antipsychotics in the lower likelihood for extrapyramidal side-effects (Correll et al. 2004). These medications block central dopamine receptors, but have significant heterogeneity in their mechanisms of action, a number of which suggest antidepressant properties; e.g. serotonin 5-HT2 receptor antagonist and 5-HT1A and dopamine receptor partial agonist activity (DeBattista & Hawkins, 2009).

A meta-analysis by Nelson & Papakostas (2009) reviewed RCTs of acute-phase treatment with atypical antipsychotic augmentation in patients with MDD, in addition to other unpublished data. In meta-analysis of the data, the authors found that augmentation with atypical antipsychotics was significantly more effective than placebo, for outcome measures of response (OR 1.69) and remission (OR 2.00). No significant differences in efficacy were found between the four atypical agents studied (olanzapine, risperidone, quetiapine, aripiprazole). The atypical antipsychotics as a class did show significantly higher rates of discontinuation for adverse effects, and this result was seen in studies of short duration (4–12 wk). Longer-term effects of atypical antipsychotics can include weight gain and metabolic syndrome, and extrapyramidal side-effects including tardive dyskinesia, or rare events including neuroleptic malignant syndrome. Thus safety and tolerability, as well as the significant economic cost should be considered when evaluating augmenting strategies for MDD. Finally, the efficacy of these agents for continuation and maintenance therapy is yet to be studied.


Until the advent of atypical antipsychotics, lithium had been the most-studied medication for antidepressant augmentation (Nelson & Papakostas, 2009). Early studies of lithium augmentation demonstrated benefit compared to placebo when added to TCAs for partial response, but there had until recently been no studies of lithium augmentation in patients with a history of non-response to multiple antidepressants. In a 2003 study, Nierenberg and colleagues treated 35 treatment-resistant subjects prospectively with nortriptyline for 6 wk (Nierenberg et al. 2003). Those who tolerated the medication but did not respond were then randomized to receive lithium or placebo augmentation for another 6 wk. At that time, 12.5% of lithium-augmented patients showed a response, vs. 20% of placebo-augmented patients, a non-significant difference between lithium and placebo. The authors concluded that the utility of lithium augmentation may be restricted to patients with depression refractory to a single medication. A meta-analysis of lithium augmentation for unipolar and bipolar patients with major depression was conducted by Crossley & Bauer (2007). In the meta-analysis of 10 augmentation studies, the data suggest that lithium is an effective augmenting strategy, with an OR of 3.11. A limitation of this meta-analysis was that included studies were not limited to patients with treatment-resistant depression, and combined patients with unipolar and bipolar depression.

Lithium has long been postulated to have antisuicidal properties. A meta-analysis of the antisuicidal effect of lithium in patients with recurrent MDD concludes that the overall risk for suicide or attempt was 88.5% lower in patients taking lithium vs. those without lithium (Guzzetta et al. 2007). The effect was observed even in two studies that selected patients with high risk for suicide. Limitations of the study were relatively small number of trials included, with a small number of patients contributing data (n=78), and that in some studies, suicide was observed incidentally, rather than as an outcome measure. However, the data are consistent with earlier findings that lithium reduces suicide rates in patients with bipolar disorder or other psychiatric illness. An RCT investigating the effect of adjunctive lithium treatment in the prevention of suicidal behaviour was published in 2008 (Lauterbach et al. 2008). In this study, 167 patients with a suicide attempt in the past 3 months, in the context of a depressed episode, were treated with either lithium or placebo, and followed for 12 months. Seventy-six percent of patients had MDD, and the others had adjustment disorders or dysthymia; nearly three-quarters of patients had comorbid psychiatric or substance use disorders as well. Lithium and placebo were studied as add-on treatments, and patients received usual clinical care in community settings. The composite primary outcome measure was the occurrence of a suicide attempt or completed suicide during the follow-up period. Over the study period, there were 17 suicide attempts or completions, seven in the lithium group and 10 in the placebo group. Survival analysis showed no significant difference of suicidal acts between the lithium- and placebo-treated groups. However, in post-hoc analysis, it was revealed that all completed suicides occurred in the placebo group, accounting for a significant difference in the incidence rates. The authors conclude that lithium may be effective in reducing the risk of completed suicide in adults with depressive disorders. This was the first prospective RCT examining suicide as an outcome measure in a sample of depressed patients at high risk for suicide. Although the findings were limited by small sample size, this study adds to the body of evidence suggesting an antisuicidal effect of lithium, and should encourage further investigation.


A few small studies have shown some efficacy for lamotrigine as an augmenting agent for antidepressants in patients with treatment-resistant depression. In one study, 23 patients with treatment failure for MDD were selected to receive treatment with fluoxetine, and randomized to concomitant treatment with lamotrigine or placebo (Barbosa et al. 2003). Lamotrigine at 100 mg was statistically superior to placebo on the outcome measure of Clinical Global Impression (CGI) scale scores, but failed to separate from placebo on the primary outcome measure, HAMD scores. The study was limited by small sample size and inclusion of bipolar II and MDD patients.

In an open-label trial of lithium vs. lamotrigine for treatment-resistant depression, 34 patients were randomized to receive one or the other as antidepressant augmentation (Schindler & Anghelescu, 2007). In both groups, clinically significant reductions in HAMD scores were achieved, and there was no statistically significant difference between the groups. Twenty-three percent of the lamotrigine group and 18% of the lithium group achieved remission. This study was limited by small sample size and non-blinded observations, but the results suggest that further investigation of lamotrigine augmentation is warranted.

Triiodothyronine (T3)

T3 has been well-studied in antidepressant non-responders. A meta-analysis by Altshuler and colleagues found that the pooled, weighted effect size of the addition of T3 to TCAs in non-refractory patients was 0.58 (Altshuler et al. 2001). Five of the six randomized trials of T3 found benefit in augmentation. All of the trials, however, took place between 1969 and 1974, in patients prescribed relatively low doses of TCAs. Whether the effect would have been apparent in subjects with higher TCA dosing is unclear. While most of the published data support the use of T3 as augmentation for TCAs, some more recent studies have shown moderate efficacy of T3 augmentation for SSRIs (Abraham et al. 2006; Iosifescu et al. 2005).

A recent prospective, placebo-controlled pilot study was performed to evaluate the utility of T3 in accelerating antidepressant response under naturalistic conditions (Posternak et al. 2008). Fifty consecutive outpatients who were diagnosed with MDD and started on antidepressants were concurrently randomized to receive T3 or placebo, and followed for 6 wk. Response rates, defined as a >50% reduction in MADRS scores, were higher for the adjunctive cohort after weeks 1 and 2, and the authors conclude that T3 may be used to accelerate the antidepressant response; further investigation is needed.

S-adenosyl methionine (SAMe)

SAMe has been studied previously for its antidepressant properties, and has been prescribed in Europe since the 1970s. SAMe was initially studied in its parenteral forms; these early studies demonstrated efficacy vs. placebo, and also suggested that SAMe was similar in efficacy to TCAs (see Papakostas, 2009, for a review). When a stable oral form of SAMe became available for study, three RCTs demonstrated its comparable efficacy to TCAs, and two studies demonstrated efficacy superior to placebo (Papakostas, 2009). In a 2002 meta-analysis by the Agency for Healthcare Research and Quality, results of studies in which SAMe was administered orally, intramuscularly, and intravenously were combined, and supported the conclusion that SAMe was similarly effective to TCAs (effect size 0.08, 95% CI −0.17 to 0.32), and superior to placebo (effect size 0.65) for the treatment of depression (Hardy et al. 2002).

More recent data support the use of SAMe as an antidepressant augmentation strategy. Papakostas and colleagues presented a RCT focusing on the use of oral SAMe for antidepressant augmentation in patients with MDD who were SRI non-responders (Papakostas et al. 2010). Seventy-three patients were randomized to adjunctive placebo or 800–1600 mg/d SAMe. The response rate (the primary outcome measure) for SAMe-treated patients was 36.1%, which differed significantly from the response rate for placebo-treated patients, 17.6%. Discontinuation rates for side-effects were similar in both groups. Replication and further studies are needed, but this first randomized and placebo-controlled study demonstrates a promising finding for the use of SAMe as an antidepressant augmentation strategy.

In summary, after an initial trial of antidepressant monotherapy, both switching and augmenting are reasonable next steps for patients who fail to achieve remission. One advantage to switching medication is to avoid polypharmacy and its inherent risk for side-effects or adverse events. An advantage to augmenting is to continue any partial response that might have occurred as a therapeutic effect of the initial antidepressant agent. Evidence-supported augmenting strategies include addition of atypical antipsychotics, an antidepressant with a different mechanism of action, lithium, T3, or SAMe.

Emerging evidence


The development of agomelatine, a melatonin receptor agonist and a 5-HT2C receptor antagonist, has added a new approach to treating depression. The drug increases monoamingeric transmission, and through its effect on melatonin receptors, contributes to improvements in sleep onset and quality. Two recent RCTs found significant improvements in HAMD scores over the 8-wk study periods (Stahl et al. 2010; Zajecka et al. 2010). The studies conflicted in determining the effective dose of agomelatine, with Stahl et al. reporting 25 mg was more efficacious, and Zajecka et al. reporting efficacy with only a 50-mg dose. Both studies observed transient aminotransferase elevations in the 50-mg groups. In a longer, 24-wk study of agomelatine responders, subjects were randomized to continue treatment with agomelatine or to placebo; over the study period, the relapse rate for agomelatine-treated patients was 21.7%, compared to 46.6% for placebo-treated patients (Goodwin et al. 2009). Previous studies with agomelatine reported a superior sexual side-effect profile with agomelatine compared to other antidepressants, specifically venlafaxine XR (Kennedy et al. 2008), and paroxetine (Montejo et al. 2010).

Glutamatergic agents

Converging lines of evidence suggest that the glutamatergic system plays a role in the pathophysiology of MDD, and treatments targeting this system are already being studied for their effect on depressive symptomatology (Hashimoto, 2009; Sanacora, 2009). Several known mood-modulating medications act on the glutamatergic system: lamotrigine affects glutamate release, ketamine and memantine antagonize the NMDA receptor, and the antioxidant N-acetyl-l-cysteine (NAC) activates the cystine-glutamate antiporter. Early studies of ketamine by Zarate et al. (2006a) demonstrated a robust and very rapid antidepressant response to an intravenous infusion of ketamine in patients with treatment-resistant depression. Patients were randomized to an infusion of ketamine or placebo on two days, 1 wk apart. Fifty percent of patients met response criteria within 2 h, and 71% within 24 h; the effect was significantly different than outcome measures of the placebo group, and was sustained at 7 d post-treatment. A study of memantine, a NMDA receptor antagonist, failed to separate from placebo on MADRS scores at 8 wk (Zarate et al. 2006b). Finally, riluzole, a drug used for the treatment of amyotrophic lateral sclerosis, affects both glutamate release and reuptake, and has shown some effect on depressive symptoms in treatment-resistant depression, both when used as augmentation and as monotherapy (Sanacora, 2009). Agents that affect glutamatergic neurotransmission will continue to be an active topic for study in the treatment of MDD.


Pramipexole is a pre-synaptic dopamine agonist, with a preference for the D3 receptor subtype. In addition to potentiating dopamine receptors, pramipexole has a suppressive effect on REM sleep, and has demonstrated neurotrophic effects as well. A meta-analysis by Aiken in 2007 reviewed the literature supporting its use in psychiatric conditions (Aiken, 2007). Aiken identified three placebo-controlled trials, but pooled data for all treated subjects in both prospective and retrospective studies, and found large effect sizes in unipolar and bipolar depression. In the only controlled trial of patients with unipolar MDD, Corrigan et al. (2000) compared pramipexole as monotherapy to 20 mg fluoxetine, and placebo, in 174 patients. In this industry-sponsored study, pramipexole improved scores on HAMD, MADRS, and CGI, in a statistically significant finding at 8 wk. The most improvement was seen in patients taking the 5-mg dose, although more patients at this dose range dropped out. Other studies of mood symptoms in bipolar depression and in Parkinson's disease show improvement in pramipexole-treated patients (Aiken, 2007). More study is needed to show efficacy, and the authors raise concern over previously reported adverse effects including sleep attacks, compulsive behaviour, mania, and psychosis in other patient populations.


Modafinil is an analeptic (wake-promoting) agent that has FDA approval for narcolepsy and shift-work sleep disorder, and as adjunctive treatment for obstructive sleep apnoea/hypopnoea syndrome. Its novel mechanism of action includes increased release of monoamines, as well as increased localized histamine release in the hypothalamus; investigators have hoped to see a benefit of modafinil on residual depressive symptoms, in particular in measures of fatigue. Small studies of open-label treatment, and a small RCT, have had positive results for augmentation of antidepressants (DeBattista, 2003, 2004; Rasmussen et al. 2005). In 2005, Fava and colleagues reported an RCT of 311 outpatients who had partial response to SSRI monotherapy, and had persistent fatigue and sleepiness (Fava et al. 2005). The dual treatment group reported improved scores on the Brief Fatigue Inventory that separated from placebo by the end of the 8-wk trial. In addition, patients in the dual-treatment group had greater improvements on CGI scores, and a (non-significant) trend towards greater improvement on HAMD scores. Finally, in a meta-analysis of 33 double-blind, placebo-controlled trials of modafinil for any clinical indication, the authors concluded that there is some evidence for modafinil's off-label use as an adjunct to antidepressants in depressive disorders, for treatment of depressed mood and depression-related fatigue and sleepiness (Ballon & Feifel, 2006).


Galantamine is an acetylcholinesterase inhibitor that modulates activity at nicotine receptor sites. It is approved for treatment of Alzheimer disease, but as findings emerged showing improvement in mood symptoms among galantamine-treated patients with mild cognitive impairment, interest developed in studying the medication as an adjunct therapy for depression (Elgamal & MacQueen, 2008). Two small RCTs have been published to date. In 2008, Elgamal and colleagues reported on a trial of 20 patients with MDD, not in an acute episode, who were randomized to placebo or galantamine (up to 16 mg) as augmentation to their current antidepressants (Elgamal & MacQueen, 2008). Although the antidepressant effect did not separate from placebo overall, certain patients with higher symptom severity at study entry did show some response. The galantamine-treated group did show improvement in cognitive scales, but the results were not statistically significant. This study was limited by the small sample size, and the study of patients without acute mood symptoms at study entry. Another small study investigated galantamine augmentation of antidepressant treatment in older adults with MDD (Holtzheimer et al. 2008). Thirty-eight non-demented older adults were randomized to galantamine or placebo augmentation of standard antidepressant therapy (citalopram or venlafaxine XR). The study period was 24 wk, and failed to demonstrate a benefit of galantamine that separated from placebo. Galantamine-treated patients were more likely to withdraw early from the study. However, there was some evidence that galantamine-treated patients had lower depression scores at week 2 of treatment. These small pilot studies suggest that further investigation is warranted.


Much attention has been focused on testing individual patients for genetic markers that may predict antidepressant response, and the hope has been to more closely match patients with suitable and effective treatments. Several recent studies have suggested that response to antidepressants may have a genetic basis (see McMahon et al. 2006; Perlis et al. 2008; Serretti et al. 2007; and recently published results of the Genome-based Therapeutic Drugs for Depression study; Uher et al. 2009). However, replication of the findings has been difficult to achieve. Genetic studies require large, well-characterized populations, as the effects of a single gene are very small and difficult to detect. Thus it has been difficult to generalize results to various minority racial-ethnic groups, as most association studies have used Caucasian samples. In addition, any single gene is likely to exert only a small effect, and useful pharmacogenetic testing would require incorporating and interpreting results from multiple loci. Finally, considerations of cost and access will become relevant as testing improves and becomes commercially available (Perlis et al. 2009). However, pharmacogenetic testing has already become a reality in other fields of medicine, and it seems inevitable that we will soon use genetic markers to predict risk and outcomes in psychiatric conditions as well.


SSRIs and newer agents have proven to be safe and effective alternatives to the older antidepressants; these agents are all suitable for first-line use in patients with MDD. Each medication's unique side-effect profile should guide drug selection for individual patients. In this economic age, cost should be considered as well, both from an individual patient perspective, and from an aggregate perspective. For depressive episodes not responding to initial therapy, no single strategy – whether switching to a different drug of the same class, switching to a different class of drug, or augmenting with an additional agent – has been clearly established as superior. As augmenting agents, atypical antipsychotics, lithium, and T3 have been studied the most extensively, and shown to have benefit. However, their risks and side-effect profiles may make them less attractive to patients, and patient preference and safety should determine treatment decisions for refractory or chronic MDD.

To have advantages over existing drugs, novel antidepressants must have a more rapid onset of action, higher rates of response and remission, or marked improvements in tolerability and ease of use. Some of the novel agents, targeting other neurotransmitter systems such as glutamate, show promise in producing early response. The use of biomarkers, including pharmacogenetic testing, may one day provide more accurate predictors of response or adverse outcomes, allowing targeted treatments and the promise of personalized medicine.


Supported in part by the National Institute on Alcoholism and Alcohol Abuse (Grant K23AA016340-01A1), and the Scholars in Clinical Science Program of Harvard Catalyst/The Harvard Clinical and Translational Science Center (Award #UL1 RR 025758 and financial contributions from Harvard University and its affiliated academic health care centres) (Dr Ostacher). The content is solely the responsibility of the authors and does not necessarily represent the official views of Harvard Catalyst, Harvard University and its affiliated academic health care centres, the National Center for Research Resources or the National Institutes of Health.

Statement of Interest

Dr Ostacher has received research support from Pfizer, and has served on the advisory/consulting boards of Pfizer, Schering-Plough, and Concordant Rater Systems. He has received speaker's fees from AstraZeneca, Bristol–Myers Squibb, Eli Lilly & Company, GlaxoSmithKline, Janssen Pharmaceutica, Pfizer, and Massachusetts General Psychiatry Academy.

Dr Perlis has received honoraria, consulting, or speaker's fees from AstraZeneca, Bristol–Myers Squibb, Eli Lilly & Co., GlaxoSmithKline, Pfizer, Proteus Biomedical, and Concordant Rater Systems. He holds patents and receives royalties with Concordant Rater Systems.

Dr Nierenberg is a full-time employee of the Massachusetts General Hospital (MGH). In the past 12 months (as of 3 March 2009), he has served as a consultant to the American Psychiatric Association (only travel expenses paid), Appliance Computing Inc. (Mindsite), Brandeis University. Through the MGH Clinical Trials Network and Institute, he has consulted for Brain Cells, Inc., Dianippon Sumitomo/Sepracor Novartis, PGx Health, Shire, Schering-Plough, Targacept, and Takeda/Lundbeck Pharmaceuticals. He received grant/research support through MGH from NIMH, PamLabs, Pfizer Pharmaceuticals, and Shire. He received honoraria from Belvior Publishing, University of Texas Southwestern Dallas, Hillside Hospital, American Drug Utilization Review, American Society for Clinical Psychopharmacology, Baystate Medical Center, Columbia University, IMEDEX, MJ Consulting, New York State, MBL Publishing, Physicians Postgraduate Press, SUNY Buffalo, University of Wisconsin, and the University of Pisa. Dr Nierenberg is a presenter for the Massachusetts General Hospital Psychiatry Academy (MGHPA). The education programs conducted by the MGHPA were supported through Independent Medical Education (IME) grants from the following pharmaceutical companies in 2008: Astra Zeneca, Eli Lilly, and Janssen Pharmaceuticals; in 2009 Astra Zeneca, Eli Lilly, and Bristol–Myers Squibb. No speakers' bureaus or boards since 2003. He is on the advisory boards of Appliance Computing, Inc., Brain Cells, Inc., Eli Lilly and Company, and Takeda/Lundbeck and Targacept. Dr. Nierenberg owns stock options in Appliance Computing, Inc. and Brain Cells, Inc. Through MGH, he is named for copyrights to the Clinical Positive Affect Scale and the MGH Structured Clinical Interview for the Montgomery–Asberg Depression Scale exclusively licensed to the MGH Clinical Trials Network and Institute (CTNI). Also, through MGH, Dr Nierenberg has a patent extension application for the combination of buspirone, bupropion, and melatonin for the treatment of depression.


View Abstract