10.2.1. Introduction

This section reviews the relative efficacy of individual antidepressants in the treatment of depression. Where there were sufficient data, the effect of patient setting (inpatient, outpatient or primary care) on choice of drug was also examined. It covers the following drugs:

• Tricyclic antidepressants (TCAs) (Section 10.3)

  –

  Amitriptyline**

  –

  An overview of TCAs used as comparator treatments in trials reviewed elsewhere⁸⁹**

• Selective serotonin reuptake inhibitors (SSRIs) except escitalopram (Section 10.4)

  –

  Citalopram**

  –

  Fluoxetine**

  –

  Fluvoxamine**

  –

  Paroxetine**

  –

  Sertraline**

• Escitalopram (Section 10.5)
• Monoamine oxidase inhibitors (Section 10.7)

  –

  Moclobemide**

  –

  Phenelzine**

• ‘Third-generation’ drugs (Section 10.8)

  –

  Duloxetine

  –

  Mirtazapine**

  –

  Reboxetine**

  –

  Venlafaxine**

• Other preparations

  –

  St John’s wort** (Section 10.9)

10.3.1. Introduction

**TCAs have been used to treat depression for over 40 years. Currently, nine TCAs are available in the UK. They are thought to exert their therapeutic effect by inhibiting the re-uptake of monoamine neurotransmitters into the presynaptic neurone, thus enhancing noradrenergic and serotonergic neurotransmission. Although all TCAs block the reuptake of both amines, they vary in their selectivity with, for example, clomipramine being primarily serotonergic and imipramine noradrenergic.

All TCAs cause, to varying degrees, anticholinergic side effects (dry mouth, blurred vision, constipation, urinary retention, and sweating), sedation and postural hypotension. These side effects necessitate starting with a low dose and increasing slowly. In many patients a ‘therapeutic dose’ is never reached either because the patient cannot tolerate it or because the prescriber does not titrate the dose upwards.

All TCAs except lofepramine are toxic in overdose, with seizures and arrhythmias being a particular concern (see Chapter 11, Section 11.9). This toxicity and the perceived poor tolerability of these drugs in general have led to a decline in their use in the UK over the last decade.

10.3.2. Amitriptyline

Although amitriptyline was not the first TCA and is not the best tolerated or the most widely prescribed, it is the standard drug against which new antidepressants are compared with respect to both efficacy and tolerability. Amitriptyline may be marginally more effective than other antidepressants, a potential benefit that is offset by its poorer tolerability (Barbui & Hotopf, 2001). Efficacy benefits may be more marked in hospitalised patients (Anderson et al., 2000).

10.3.3. Tricyclic antidepressants – an overview of selected data

**This section combines data from other reviews where a TCA was used as a comparator treatment. It is, therefore, not a systematic review since a systematic search for all trials of TCAs was not conducted. It specifically does not include comparisons of TCAs with other TCAs.

10.4.1. Introduction

**The selective serotonin reuptake inhibitors (SSRIs) inhibit the reuptake of serotonin into the presynaptic neurone thus increasing neurotransmission. Although they ‘selectively’ inhibit serotonin reuptake, they are not serotonin specific. Some of the drugs in this class also inhibit the reuptake of noradrenaline and/or dopamine to a lesser extent.

As a class, they are associated with less anticholinergic side effects and are less likely to cause postural hypotension or sedation. Dosage titration is not routinely required so subtherapeutic doses are less likely to be prescribed. They are also less cardiotoxic and much safer in overdose than the TCAs or MAOIs. These advantages have led to their widespread use as better-tolerated first-line antidepressants.

The most problematic side effects of this class of drugs are nausea, diarrhoea and headache. Fluvoxamine, fluoxetine and paroxetine are potent inhibitors of various hepatic cytochrome metabolising enzymes (Mitchell, 1997) precipitating many significant drug interactions. Sertraline is less problematic although enzyme inhibition is dose-related, while both citalopram and escitalopram are relatively safe in this regard.

There are other important differences among the SSRIs (Anderson & Edwards, 2001), as outlined below.

10.4.2. Studies considered for review of selective serotonin reuptake inhibitors apart from escitalopram^105,106

The GDG used an existing review (Geddes et al., 2002) as the basis of this section, for which the authors made their data available to the NCCMH team. Since this review did not cover escitalopram which achieved its UK licence in late 2001, a separate review¹⁰⁷ of this drug was undertaken.

The Geddes and colleagues’ (2002) review included 126 studies of which 72 did not meet the inclusion criteria set by the GDG. In addition, one trial (Peselow et al., 1989) included in the original review was considered to be part of a multicentre trial (Feighner92) rather than a separate trial. Another trial (Feighner1989) excluded by Geddes and colleagues (2002) was included in this review because it contained tolerability data (which Geddes and colleagues [2002] did not include). A further two trials excluded by Geddes and colleagues (2002) were also considered part of the Feighner92 multicentre trial (Dunbar et al., 1991; Feighner & Boyer, 1989).

Since the Geddes and colleagues’ (2002) review compared SSRIs with TCAs only, 59 additional studies were identified from other reviews undertaken for this guideline, including two identified from hand searching reference lists. Thirty-three of these were included and 26 excluded. Thus 107 trials are included in this review providing data from up to 11,442 participants. A total of 97 trials were excluded.

All included studies were published between 1983 and 2003 and were between 4 and 24 weeks’ long (mean = 6.5 weeks). Twenty-four studies were of inpatients, 51 of outpatients and six undertaken in primary care. In the remaining 26, it was either not clear from where participants were sourced, or they were from mixed sources. In 11 studies, more than 80% of participants were aged 65 years and over (although only eight of these reported extractable efficacy outcomes). In two studies participants had depression with additional atypical features.

In addition to the standard diagnostic criteria, most studies required a minimum baseline HRSD score of between 10 and 22 on the 17-item version (61 studies) or between 18 and 22 on the 21-item version (28 studies). The ten studies reporting MADRS scores required minimum baseline scores of between 18 and 30.

Data were available to compare SSRIs (citalopram, fluoxetine, fluvoxamine, paroxetine and sertraline) with amineptine, amitriptyline, clomipramine, desipramine, dothiepin/dosulepin, doxepin, imipramine, lofepramine, nortriptyline, maprotiline, mianserin, trazodone, phenelzine, moclobemide, mirtazapine, venlafaxine and reboxetine.

The Geddes and colleagues’ (2002) review, on which this review is based and for which the data were made available to the GDG, included only one outcome measure (mean endpoint scores) and did not include tolerability data. Tolerability data, but not additional efficacy outcomes, have been extracted by the NCCMH team.

10.4.3. Clinical evidence statements for selective serotonin reuptake inhibitors apart from escitalopram¹⁰⁸

10.4.4. Clinical summary of selective serotonin reuptake inhibitors apart from escitalopram

SSRIs are relatively well-tolerated drugs with equal efficacy compared with alternative antidepressants. They may have an advantage for those with suicidal intent, due to their safety in overdose (see Section 11.10).**

10.5.1. Introduction

Escitalopram was reviewed in the previous guideline, but a relatively large number of studies (compared with the number previously available) have been published since then and so the review has been updated for this guideline. For the present review, both published and unpublished double-blind RCTs were sought that compared escitalopram either with placebo or with another antidepressant. The marketing authorisation holder, Lundbeck, was also contacted for data.

10.5.2. Databases searched and the inclusion/exclusion criteria

Information about the databases searched for published trials and the inclusion/exclusion criteria used are presented in Table 68. Details of the search strings used are in Appendix 8.

Table 68

Databases searched and inclusion/exclusion criteria for clinical effectiveness of pharmacological treatments.

10.5.3. Studies considered¹¹¹

A total of six trials were included in the review in the previous guideline and these were supplemented by another 18 trials. Some of the studies used in the previous review that had been unpublished, have since been published with different first authors; therefore, the study identifier has changed for some studies. Five studies in the current review are unpublished and supplied by the drug’s manufacturer.

Data were available to compare escitalopram with placebo, and with a range of other antidepressants. Sub-analyses were undertaken to assess the effect of the severity of depression at baseline and by dose, and to ascertain effectiveness against individual drugs (in particular, citalopram), other SSRIs and non-SSRI antidepressants.

Summary study characteristics of the included studies are presented in Table 69 with full details in Appendix 17c, which also includes details of excluded studies.

Table 69

Summary study characteristics of escitalopram.

10.5.4. Clinical evidence

Escitalopram versus placebo

Eleven studies were found that compared escitalopram with placebo. Those that used a fixed dose of 10 or 20 mg were included in sub-analyses by dose. The summary evidence profile can be found in Table 70. The full evidence profiles and associated forest plots can be found in Appendix 16c and Appendix 19c.

Table 70

Summary evidence profile for escitalopram versus placebo.

Escitalopram was effective when compared with placebo, although overall effect sizes were small and the quality of evidence graded moderate (largely because of heterogeneity). Sub-analyses by dose indicated that both 10 and 20 mg doses were effective, although effect sizes were greater and graded moderate with the larger dose. However, more people left treatment early for any reason and because of side effects, and more people taking 20 mg reported side effects compared with those taking 10 mg.

Escitalopram versus all other antidepressants

Twenty one studies were found that compared escitalopram with other antidepressants. Table 71 gives the summary evidence table for escitalopram compared with all other antidepressants together. (Separate analyses follow for escitalopram compared with SSRIs, citalopram and other antidepressants are below). The full evidence profiles and associated forest plots can be found in Appendix 16c and Appendix 19c, respectively.

Table 71

Summary evidence profile for escitalopram versus all other antidepressants.

Compared with all antidepressants for which there are data, escitalopram was more effective although effect sizes were small. Fewer participants taking escitalopram left treatment early for any reason or because of side effects compared with those taking other antidepressants, although the numbers reporting side effects were roughly equal.

Escitalopram versus selective serotonin reuptake inhibitors

Eight studies were found that compared escitalopram with SSRIs. Escitalopram is also compared with citalopram separately. The summary evidence profile can be found in Table 72. The full evidence profiles and associated forest plots can be found in Appendix 16c and Appendix 19c, respectively.

Table 72

Summary evidence profile for escitalopram versus SSRIs.

Compared with all SSRIs together, escitalopram is more effective although the effect sizes are small. Compared with individual SSRIs, there were no clinically important differences on efficacy outcomes other than compared with citalopram, where escitalopram was more effective with a small effect size. Escitalopram was also more acceptable and tolerable than SSRIs, apart from sertraline, although differences were again small.

Escitalopram versus non-selective serotonin reuptake inhibitors

Seven studies were found that compared escitalopram with non-SSRI antidepressants. The summary evidence profile can be found in Table 73. The full evidence profiles and associated forest plots can be found in Appendix 16c and Appendix 19c, respectively.

Table 73

Summary evidence profile for escitalopram versus non-SSRIs.

There were no clinically important differences between escitalopram and duloxetine, venlafaxine or bupropion on efficacy measures, although all effect sizes favoured escitalopram. Escitalopram was mostly more acceptable and tolerable, although differences were small.

10.5.5. Clinical summary

Escitalopram is superior to placebo in the treatment of depression. There is some evidence that 20 mg may be more effective than 10 mg, but at the expense of increased side effects. Escitalopram is more effective than citalopram although the effect size is small. It is at least as effective as other SSRIs and marginally better tolerated, except against sertraline.

Escitalopram was more effective than other antidepressants, with statistically significant differences versus SSRIs (although effect sizes are small and unlikely to be clinically important), but not against other antidepressants (duloxetine, venlafaxine and bupropion). Effect sizes compared with citalopram were largest, although these were still relatively small. This was particularly the case for escitalopram at 20 mg. It was also marginally more acceptable and tolerable, apart from compared with sertraline. However, differences were again small and unlikely to be clinically important.

Several more detailed comparisons were considered by the GDG, in addition to those presented above, which helped inform interpretation of the data. These can be found in Appendix 19c (forest plot numbers Pharm Esc 05 to Esc 11).

Overall, the quality of the evidence tended to be downgraded because of heterogeneity between trials. Since escitalopram is still in patent its acquisition costs are relatively high compared with antidepressants available in generic form.

10.7.1. Introduction

**MAOIs exert their therapeutic effect by binding irreversibly to monoamine oxidase, the enzyme responsible for the degeneration of monoamine neurotransmitters such as NA and serotonin. This results in increased monoamine neurotransmission. The first antidepressant drug synthesised was an irreversible MAOI and drugs in this class have been available in the UK for nearly 50 years.

All MAOIs have the potential to induce hypertensive crisis if foods containing tyramine (which is also metabolised by monoamine oxidase) are eaten (Merriman, 1999) or drugs that increase monoamine neurotransmission are co-prescribed (Livingstone & Livingstone, 1996). These foods and drugs must be avoided for at least 14 days after discontinuing MAOIs. Reversible inhibitors of monoamine oxidase (RIMAs) have a much lower likelihood of causing a hypertensive crisis and dietary restrictions are usually not required. Moclobemide is the only RIMA licensed in the UK.

Dietary restrictions, potentially serious drug interactions and the availability of safer antidepressants have led to the irreversible MAOIs being infrequently prescribed in the UK, even in hospitalised patients. However, MAOIs are still widely cited as being the most effective antidepressants for the treatment of atypical depression (see Section 11.5).

For this class of drugs, the GDG chose to review phenelzine and moclobemide.

10.7.2. Moclobemide

10.7.3. Phenelzine

10.8.1. Introduction

**This diverse group of antidepressants was marketed after the SSRIs. The aim was to broaden the mechanism of action beyond serotonin in order to improve efficacy without incurring the side effects or toxicity in overdose associated with the TCAs.**

The following drugs are reviewed in this section: duloxetine (a new review for this updated guideline), mirtazapine, reboxetine and venlafaxine.

10.8.2. Duloxetine

Databases searched and the inclusion/exclusion criteria

For the present review, both published and unpublished double-blind RCTs were sought that compared duloxetine either with placebo or with another antidepressant. The marketing authorisation holder, Eli Lilly, was also contacted for data. Information about the databases searched for published trials and the inclusion/exclusion criteria used are presented in Table 74. Details of the search strings used are in Appendix 8.

Table 74

Databases searched and inclusion/exclusion criteria for clinical effectiveness of pharmacological treatments.

Studies considered¹²³

In total, 27 acute-phase trials were sourced from searches of electronic databases and from the website of the drug’s manufacturer, Eli Lilly, which included links to the clinical trials website www.clinicaltrialresults.org from where full trial reports were downloaded. In all, 18 trials (four unpublished) were included with nine excluded (seven unpublished). (One trial is also included in the review of treatment-resistant depression [see Chapter 12, Section 12.3] because it re-randomised patients who did not respond to acute phase treatment.) Only data from patients given at least the licensed dose (60 mg) were included in the analyses, apart from in trials that used a variable dose and in trials where comparisons with the licensed dose were possible.

Data were available to compare duloxetine with placebo, with duloxetine at different doses, and with other antidepressants (SSRIs or venlafaxine). In addition, three trials continued treatment for those with at least a partial response (>30% improvement in baseline depression scores). Summary study characteristics of the included studies are presented in Table 75 with full details in Appendix 17c, which also includes details of excluded studies.

Table 75

Summary study characteristics of studies of duloxetine.

Clinical evidence

Duloxetine versus placebo

Although the effect sizes for all three efficacy outcomes for duloxetine (dose at least as large as the licensed dose of 60 mg) versus placebo were statistically significant and favoured duloxetine, with only that for non-response approaching clinical importance, there were similar effect sizes for duloxetine at different doses when these data were looked at separately, although the effect sizes for duloxetine at 120 mg versus placebo was larger than those for lower does (WMD = −2.57, −3.77 to −1.37). The data for duloxetine at different doses can be seen in the full evidence profiles and forest plots (Appendix 16c and Appendix 19c, respectively).

Two trials specifically examined depression-related pain using the self-report Brief Pain Inventory (BPI) scale. There was an average reduction of three-quarters of a point (on an 11-point Likert scale) for the ‘average pain in last 24 hours’ item.

There was little difference between the number of people receiving duloxetine who left treatment early for any reason and those receiving placebo on this measure. However, of those leaving treatment early, twice as many taking duloxetine as those taking placebo left specifically because of side effects while twice as many taking placebo left because of lack of efficacy. The numbers reporting side effects were high in both groups, with more among those taking duloxetine. Those taking duloxetine also experienced a small average weight loss compared with those on placebo, although these data were of low quality largely because of heterogeneity. The quality of the evidence was moderate or low, largely because of the selective population included in the studies.

Evidence from the important outcomes and overall quality of evidence are presented in Table 76. The full evidence profiles and associated forest plots can be found in Appendix 16c and Appendix 19c, respectively.

Table 76

Summary evidence profile for duloxetine versus placebo (acute phase).

Three studies continued patients who achieved at least partial response to acute-phase treatment (defined as >= 30% decrease in baseline HAMD scores) (DETKE2004, ELI LILLY HMAQ, PERAHIA2006B), although there were no extractable data in ELI LILLY HMAQ. There was no difference in symptoms of depression or on acceptability and tolerability measures between duloxetine at either 80 or 120 mg and placebo.

Evidence from the important outcomes and overall quality of evidence are presented in Table 77. The full evidence profiles and associated forest plots can be found in Appendix 16c and Appendix 19c, respectively.

Table 77

Summary evidence profile for duloxetine versus placebo (continuation phase for partial responders).

Duloxetine comparing different doses

Data were available to compare duloxetine at 40 mg (less than the licensed dose) with 80 mg, 30 mg with 60 mg, and 80 mg with 120 mg. There were no statistically or clinically important differences between the doses on either efficacy or acceptability and tolerability outcomes, although there were few trials. Evidence from the important outcomes and overall quality of evidence are presented in Table 78. The full evidence profiles and associated forest plots can be found in Appendix 16c and Appendix 19c, respectively.

Table 78

Summary evidence profile for duloxetine comparing different doses (acute phase).

One study comparing duloxetine at different doses included a continuation phase for those who achieved at least partial response to acute-phase treatment (defined as >= 30% decrease in baseline HAMD scores) (PERAHIA2006B). This showed no difference between the doses. The quality of the evidence was low or very low. Evidence from the important outcomes and overall quality of evidence are presented in Table 79. The full evidence profiles and associated forest plots can be found in Appendix 16c and Appendix 19c, respectively.

Table 79

Summary evidence profile for duloxetine comparing different doses (continuation phase for partial responders).

Duloxetine versus other antidepressants

Data were available to compare duloxetine with paroxetine, fluoxetine, escitalopram and venlafaxine. There was no difference between duloxetine and other antidepressants, except venlafaxine which was more effective on mean change scores at endpoint (although the effect size was small and not quite statistically significant). Duloxetine was less acceptable to patients, as measured by the number leaving treatment early, and more people taking duloxetine left specifically because of adverse reactions. However, there was no difference between duloxetine and other antidepressants on numbers leaving treatment early because of lack of efficacy, on the number of people reporting side effects or on weight change. The quality of the evidence was moderate, low or very low, largely because of the selective population included in the studies.

Evidence from the important outcomes and overall quality of evidence are presented in Table 80. The full evidence profiles and associated forest plots can be found in Appendix 16c and Appendix 19c, respectively.

Table 80

Summary evidence profile for duloxetine versus other antidepressants (acute phase).

Two studies comparing duloxetine with other antidepressants included a continuation phase for those who achieved at least partial response to acute-phase treatment (defined as >= 30% decrease in baseline HAMD scores) (DETKE2004, PERAHIA2006B). Both studies compared duloxetine with paroxetine. Only one outcome was reported by both studies. This showed no difference between the doses. The quality of the evidence was low. Evidence from the important outcomes and overall quality of evidence are presented in Table 81. The full evidence profiles and associated forest plots can be found in Appendix 16c and Appendix 19c, respectively.

Table 81

Summary evidence profile for duloxetine versus other antidepressants (continuation phase for partial responders).

One study comparing duloxetine with other antidepressants included a continuation phase for all those entering the study regardless of response during the acute phase of the study (WADE2007). This compared duloxetine with escitalopram. There was a small difference in favour of escitalopram in efficacy measures, which was not clinically important, and the number of patients leaving treatment early specifically because of side effects favoured escitalopram. Evidence from the important outcomes and overall quality of evidence are presented in Table 82. The full evidence profiles and associated forest plots can be found in Appendix 16c and Appendix 19c, respectively.

Table 82

Summary evidence profile for duloxetine versus other antidepressants (continuation phase for all).

10.8.3. Mirtazapine

10.8.4. Reboxetine

10.8.5. Venlafaxine

10.9.1. Introduction

**St John’s wort, an extract of the plant Hypericum perforatum, has been used for centuries for medicinal purposes including the treatment of depression. It is not licensed as a medicine in the UK but can be bought ‘over the counter’ from health food shops, herbalists and community pharmacies. Many different branded preparations are available. St John’s wort is licensed in Germany for the treatment of depression.

St John’s wort is known to contain at least ten constituents or groups of components that may contribute to its pharmacological effects (Linde & Mulrow, 2004), but its exact mode of action is unknown. These include naphthodianthrons, flavonoids, xanthons and biflavonoids (Wagner & Bladt, 1994). In common with all herbal preparations, the quantity and proportions of each constituent varies among batches (Wang et al., 2004). Most commercial products are standardised with respect to hypericin content, but it is not known if this is the only active component. Individual brands or batches of the same brand may, therefore, not be therapeutically equivalent. Many clinically important drug interactions have been reported (Committee on Safety of Medicines, 2000). St John’s wort may also cause photosensitivity.

10.9.2. Studies considered^135,136

Forty studies were found in a search of electronic databases, with 19 being included and 21 being excluded by the GDG.

Ten studies were available for a comparison with placebo (Davidson02, Hansgen1996, Kalb2001, Laakmann98, Lecrubier02, Philipp99, Schrader98, Shelton2001, Volz2000, Witte1995); four studies for a comparison with TCAs (Bergmann93, Philipp99, Wheatley97, Woelk2000); one for a comparison with TCA-related antidepressants (Harrer94); and six studies for a comparison with SSRIs (Behnke2002, Brenner00, Davidson02, Harrer99, Schrader00, VanGurp02)¹³⁷. Data from up to 1520 participants were available from studies comparing St John’s wort with placebo, and data from up to 1629 participants were available from comparison with antidepressants.

All included studies were published between 1993 and 2002 and were between 4 and 12 weeks’ long (mean = 6.47 weeks). In 16 studies participants were described as outpatients and in the other three it was either not clear from where participants were sourced or they were from mixed sources. In one study (Harrer99), all participants were aged 60 years and over. All participants had either moderate or severe depression.

It is very difficult to assess the exact content of the preparation of St John’s wort used in included studies so no study was excluded on grounds of inadequate dose. Included studies described the following range of preparations:

• 2 × 150 mg (300 mg) at 0.450 to 0.495 mg total hypericin per tablet
• 900 mg LI 160
• 4 × 200 mg (800 mg) LoHyp-57: drug extract ratio 5–7:1
• 3 × 300 mg (900 mg) WS5572: drug extract ratio 2.5–5:1, 5% hyperforin
• 3 × 300 mg (900 mg) WS557= : 0.5% hyperforin
• 3 × 300 mg (900 mg) WS5570: 0.12 to 0.28% hypericin
• 3 × 350 mg (1050 mg) STEI 300: 0.2 to 0.3% hypericin, 2 to 3% hyperforin
• 2 × 200 mg (500 mg) ZE117: 0.5 mg hypericin
• 3 to 6 × 300 mg (900 mg to 1800 mg) at 0.3% hypericum
• 3 × 300 mg (900 mg) LI 160 = 720 to 960 mcg hypericin
• 2 × 250 mg (500 mg) ZE117: 0.2% hypericin
• 900 mg to 1500 mg LI 160: standardised to 0.12 to 0.28% hypericin
• 4 × 125 mg (500 mg) Neuroplant
• 200–240 mg Psychotonin forte
• 3 × 30 drops Psychotonin (500 mg)
• 3 × 30 drops Hyperforat: 0.6 mg hypericin.

In addition, six studies with low doses of standard antidepressants were also included.

10.9.3. Clinical evidence statements for St John’s wort compared with placebo¹³⁸

10.9.4. Clinical evidence statements for St John’s wort compared with antidepressants¹⁴³

10.9.5. Clinical summary

St John’s wort is more effective than placebo on achieving response in both moderate and severe depression, and on reducing symptoms of depression in moderate depression.

There appears to be no difference between St John’s wort and other antidepressants, other than in moderate depression where it is better at achieving response and in severe depression where it is less effective than low-dose antidepressants in achieving response.

However, St John’s wort appears as acceptable as placebo and more acceptable than antidepressants, particularly TCAs, with fewer people leaving treatment early due to side effects and reporting adverse events.

10.9.6. Recommendation

10.9.6.1.

Although there is evidence that St John’s wort may be of benefit in mild or moderate depression, practitioners should:

• not prescribe or advise its use by people with depression because of uncertainty about appropriate doses, persistence of effect, variation in the nature of preparations and potential serious interactions with other drugs (including oral contraceptives, anticoagulants and anticonvulsants)
• advise people with depression of the different potencies of the preparations available and of the potential serious interactions of St John’s wort with other drugs¹⁴⁴.

10.10.1. Systematic literature review and economic considerations

The systematic search of the economic literature undertaken for this guideline update identified nine studies. Two unpublished evaluations submitted by pharmaceutical companies were also included. Pharmacological companies producing the drugs under review were identified and contacted to provide/recommend unpublished or soon-to-be published studies in order to ensure up-to-date evidence was included in the evidence base for the guideline.

10.10.2. Escitalopram and duloxetine

Five industry-funded studies that assessed the cost effectiveness of escitalopram and duloxetine against various antidepressant comparators in the UK were included in the systematic review of economic literature (Benedicte et al., 2010; Fernandez et al., 2005; Wade et al., 2005a; Wade et al., 2005b; Wade et al., 2008).

Wade and colleagues (2005a) investigated the cost effectiveness of escitalopram at a dose of 20 mg per day compared with citalopram at 40 mg per day in those with severe depression (MADRS = > 30) in primary and secondary care in the UK. This cost-effective analysis was reported to be an adaptation of models described in other studies such as Borghi and Guest (2000). A decision tree with a 6-month time horizon was developed. It incorporated effectiveness data derived from a study review and expert opinion. Data for response rates and other relevant inputs such as remission and discontinuation rates were derived from a 506-sample meta-analysis reporting at week 8; these were then extrapolated to 6 months. Costs were calculated from the societal perspective as well as from that of the NHS and reported in 2003 pound sterling. Conventional resource use directly related to treatment as well as treatment-emergent adverse events and attempted suicide were also included. Lost productivity costs due to absenteeism from work were calculated using the human capital approach, based on mean market wages for 2003. Cost estimates for the majority of the resources used were derived from national published studies. The primary outcome measure was patient treated successfully, defined as a patient in remission (MADRS < = 12 at week 24), while the secondary outcome measure was first-line success (that is, remission without switching drug treatment). Univariate sensitivity analysis and Monte Carlo simulations were conducted to evaluate the effect of uncertainty.

From the NHS perspective, the expected total cost per patient was £422 (£404 to £441) for escitalopram and £454 (£436 to £471) for citalopram. Escitalopram also fared better in terms of the effectiveness outcomes. For example, overall success was 53.7% (50.3 to 57.5%) compared with 48.7% (45.8 to 51.7%) for citalopram. Escitalopram was demonstrated to be more effective and less costly, and therefore escitalopram dominated citalopram.

Wade and colleagues (2005a) concluded that escitalopram was a cost saving alternative to citalopram for the treatment of people with severe depression in the UK despite the price of escitalopram being higher than other generic drugs. Cost savings were shown from both perspectives. Multivariate sensitivity analysis further demonstrated that escitalopram was dominant at all ranges of probabilities tested in more than 99% of simulations. This study is deemed to be of good quality; however, depression is a chronic illness and a 6-month time horizon may well be too short to capture all costs and benefits. There are many commonly used drugs for depression and other comparators from other drug classes may have been relevant for analysis and their inclusion would possibly have been more informative.

Another study by Wade and colleagues (2005b) was reviewed, which examined the cost effectiveness of three drug therapies for the treatment of depression in primary care. Escitalopram (10 to 20 mg daily) was compared with venlafaxine-XR (75 to 150 mg daily) and then generic citalopram (20 to 40 mg daily) over a 6-month time horizon from the perspective of the NHS and society. Because of an absence of relevant head-to-head studies, two separate analyses were run. An Austrian cost-effectiveness model (Hemels et al., 2004) was adapted for the UK by Wade and colleagues (2005b). The model encompassed remission, treatment failure, referral to secondary care, dosage titration and switching of antidepressants as required. A decision tree representation was developed. The clinical evidence came from a meta-analysis of four studies (n = 1472) and head-to-head clinical trials. The summary benefit measure was the overall success rate and this was estimated using the decision model. The direct health service costs included in the economic evaluation were drugs, GP visits and psychiatrist visits. The General Practice Research Database (GPRD) was searched for treatment pattern data; expert opinion was also sought and unit costs were taken from published cost data for the UK. The price year was 2003.

When escitalopram was compared with citalopram from the NHS perspective the cost per successfully treated patient was £732 (95% CI 665, 807) for escitalopram and £933 (95% CI 850, 1023) for citalopram. In the comparison between escitalopram and venlafaxine, the cost per successfully treated patient was £546 (95% CI 481, 618) for escitalopram and £607 (95% CI 542, 677) for citalopram. ICERs were not calculated because escitalopram was found to always dominate both citalopram and venlafaxine, which were more expensive and less effective.

Sensitivity analysis showed robust findings for the analysis between escitalopram and citalopram. However, the comparison with venlafaxine was sensitive to changes in parameters such as remission rates and relapse rates used in the model.

Quality of life (QoL) is an important dimension in the depression spectrum and the impact of the interventions under review on QoL may have been informative. An indirect comparison analysis could have been conducted had there been relevant head-to-head trials published. However, the authors argue that an indirect comparison would not have changed the conclusions of the analysis.

Fernandez and colleagues (2005) aimed to assess the cost effectiveness of escitalopram (10 to 20 mg/day) compared with venlafaxine-XR (75 to 150 mg/day) in UK primary care patients with depression. The effectiveness data were derived from a double-blind, multinational¹⁴⁵ RCT with 8-week follow-up (n = 293). Costing was undertaken prospectively on the same patient sample. The perspectives of the NHS and society were adopted. The direct costs for the average patient were reported to be 40% higher for venlafaxine-XR than for escitalopram. The analysis of efficacy data was based on the basis of treatment completers only. The primary health outcome was quality of life measured on the Quality of Life Depression Scale (QLDS). Mean QLDS scores decreased in both groups: from 18.6 to 12.4 for escitalopram-treated patients (p < 0.01) and from 18.8 to 12.1 for venlafaxine-treated patients (p < 0.01). No statistically significant differences were observed between the groups. CEACs were not produced because there were no significant differences in efficacy. The results showed escitalopram to be less costly and equally effective as venlafaxine-XR. The authors concluded that escitalopram is as effective as venlafaxine-XR on the treatment of depression and may be associated with lower costs from both perspectives. Limited details of the effectiveness study were reported making it difficult to assess the study quality or validity. An 8-week follow-up is quite short for a depression-related study and, as a result, long-term costs and benefits may not have been captured. Fernandez and colleagues (2005) acknowledged that larger sample sizes are required to increase the power of performed tests and to enable the detection of differences in costs between escitalopram and venlafaxine-XR.

The study by Benedicte and colleagues (2010)¹⁴⁶ was also reviewed. It described an economic evaluation of duloxetine in comparison with SSRIs, venlafaxine-XR and mirtazapine in primary and secondary care settings in Scotland. Two analyses were conducted; in the first duloxetine was compared with SSRIs, venlafaxine and mirtazapine in patients with moderate to severe depression (HAMD-17 = > 19) in primary care. The second analysis set in secondary care compared duloxetine with venlafaxine and mirtazapine in patients with severe depression (HAMD-17 = > 25). Efficacy data, drug dosages and resource utilisation differed in both. The perspective adopted was that of the NHS. The clinical effectiveness parameters were from published and unpublished RCT data, other clinical study data and expert interviews. Resource use estimates were sourced from the Scottish Psychiatrists’ Panel, literature and UK practicing GPs. Direct medical costs consisted of all outpatient and inpatient visits and drug costs. The main outcome of the model was QALYs.

In the primary care setting, when compared with SSRIs and mirtazapine, duloxetine produced additional benefits at higher costs leading to ICERs of £6,300/QALY and £2,400/QALY gained. It dominated venlafaxine in this setting. Duloxetine also dominated venlafaxine and mirtazapine in the secondary care setting. The cost effectiveness results in the primary care setting were sensitive to changes in efficacy parameters (that is, duloxetine relapse, remission and response rates). The secondary care scenario was less sensitive to changes. The study limitations considered that efficacy data for SSRIs had been collected from other duloxetine trials and for mirtazapine from a single old meta-analysis of limited quality. The authors acknowledged the risk of bias given the problems of comparability of trial populations. Resource use data were collected from a small physician panel that is not considered to be a good source of such evidence.

Wade and colleagues (2008) evaluated the cost effectiveness of escitalopram and duloxetine in the treatment of patients with depression in an outpatient setting. This analysis was carried out alongside a double-blind, multisite randomised study. The study time horizon was 24 weeks. The primary effectiveness outcome of the analysis was the Sheehan Disability Scale (SDS) score. Resource use estimates over this time were sourced from the health economics assessment questionnaires taken alongside the trial. The societal perspective was adopted and results were reported in 2006 UK pound sterling.

The results showed that over the study period escitalopram was associated with significant cost savings compared with duloxetine (£1127 versus £2,001 total/patient cost respectively). Escitalopram also resulted in significantly lower sick leave duration compared with duloxetine (31 versus 62 days). Escitalopram dominated duloxetine in the primary analysis (that is, when assessed with the SDS scale). Indirect costs because of sick leave accounted for two-thirds of the total costs. This study was conducted in several countries in addition to the UK, which limits the generalisability of the results to the UK. Because of the marked differences in healthcare systems there would be differences in healthcare resource use costs and the relative economic burden of sick leave. The perspective adopted in this study is not that of the health services and is therefore less useful for those making decisions on behalf of health services. The short time horizon modelled may not capture all the costs and benefits of the drugs for the treatment of depression.

10.10.3. Selective serotonin reuptake inhibitors, tricyclic antidepressants and lofepramine

One study that assessed the cost effectiveness of SSRIs, TCAs and lofepramine (a newer TCA which is safer in overdose) in the treatment of depression in adult patients in the UK was included in the systematic review of economic literature (Kendrick et al., 2006b). The study was carried out alongside a prospective, randomised, open-label, clinical trial in primary care from the perspective of the health service. This trial provided effectiveness and costing data. The costing was carried out prospectively on the same sample (n = 327) of patients. The length of follow-up was 12 months.

The primary clinical measure was the number of weeks free from depression (HADS-D <8). No statistically significant differences between the groups were observed in this measure. The differences in the total costs did not reach statistical significance either. Cost-effectiveness planes and CEACs were computed to illustrate the uncertainty around the estimates. The cost-effectiveness planes for each comparison included points in all four quadrants reflecting statistically non-significant differences in outcomes and costs. The CEACs suggested that, for values placed on an additional QALY of over £5,000, SSRIs were likely to be most cost effective, although the probability of this did not rise above 0.6. This analysis was based on a trial that was well described and reflected usual practice. It also drew from a population from several centres across the UK, which was representative of the wider UK population. A limitation of the study was the failure to recruit the desired number of patients thereby reducing the study’s power to detect differences in effectiveness and costs. Loss to follow-up approaching 50% over 12 months further limited the power.

10.10.4. Mirtazapine and venlafaxine

Two industry funded UK based studies compared mirtazapine to older agents such as TCAs and SSRIs (Borghi & Guest, 2000; Romeo et al., 2004).

Borghi and Guest (2000) aimed to determine the cost effectiveness of mirtazapine compared with amitriptyline and fluoxetine in the treatment of moderate and severe depression in the UK, as well as the costs related to antidepressant discontinuation. Effectiveness data were derived from a literature review and also from a panel of GPs and psychiatrists. Direct costs included costs of hospitalisation, visits to GPs and psychiatrists, antidepressant and concomitant medication, community psychiatric nurse and community mental health team visits, and attendance at day wards. The study adopted the perspective of the health service. The estimation of quantities and costs was based on actual data, a panel of ten GPs and three psychiatrists, and literature. The price year was 1997/1998. The measure of benefit used was the proportion of successfully treated patients, determined by the HAMD-17 score (7 or less). Mirtazapine was observed to be cheaper and more effective than amitriptyline and therefore dominated amitriptyline. Six months’ treatment with mirtazapine compared with fluoxetine increased the proportion of successfully treated patients by 22% at a net additional cost to the NHS of £27 per patient. Mirtazapine’s cost effectiveness relative to amitriptyline was sensitive to the cost of managing adverse events. Mirtazapine’s cost effectiveness relative to fluoxetine was sensitive to the cost of managing patients who discontinue antidepressant treatment, the number of psychiatric consultations with GPs and the percentage of patients who completed 6 weeks’ treatment with mirtazapine and achieved a 50% reduction in the HAMD-17 score. A significant limitation of this study was that 6-week data comparing mirtazapine with fluoxetine was extrapolated to 6 months using assumptions derived from published literature due to the lack of available comparison data at the time of the study. The authors recommend an update of the model when longer-term data are available. Another limitation was that resource use data were obtained from interviews with a panel of experts; this is not considered to be ideal.

Romeo and colleagues (2004) compared the cost effectiveness of 30 to 45 mg/day mirtazapine with 20 to 30 mg/day paroxetine for patients with depression treated in primary care. The model data were obtained from an RCT. The effectiveness data and costing, which was conducted prospectively, were obtained from a subgroup of patients participating in the trial (treatment completers only). The study was conducted in general practices in Scotland and had a 24-week follow-up.

Costs were reported from the NHS and societal perspectives. Effectiveness outcomes were reported in the form of number of HAMD responders (that is, patients with a 50% decrease in the HAMD-17 score) and the change in QLDS score (from baseline) at the 24-week end point to capture change in quality of life. Both antidepressants were efficacious for 24 weeks of treatment in depressed primary care patients. Compared with paroxetine, mirtazapine was associated with greater improvements in quality of life. The primary measure of cost effectiveness was the incremental cost per responder. There were no significant differences in costs and effects on the primary outcome measure; therefore, they were not combined in the form of ICERs. In addition, there were no significant differences in the benefits between the two groups when the number of HAMD responders was the outcome considered. However, improvement in quality of life was shown to be significantly higher with mirtazapine than with paroxetine. These results were robust under all scenarios examined in the sensitivity analysis.

Sensitivity analysis revealed that if society were willing to pay nothing for a point improvement in depressive syndromes, there was an 80% probability that mirtazapine would be more cost effective than paroxetine. If the willingness-to-pay increased to £1000, this probability rose to 89%. Romeo and colleagues (2004) concluded that compared with paroxetine, mirtazapine might be a cost-effective treatment choice for depression in a primary care setting. However, when considering improvements in quality of life following the administration of these two agents, it can be inferred that mirtazapine should be considered the treatment of choice. The potential limitations are that the analysis may be subject to potential selection bias. The subgroup used consisted of treatment completers only. Nevertheless, it was reported that patients excluded from the subgroup did not differ from the patients included in terms of baseline characteristics. No further statistical analyses, to account for potential biases and confounding factors, were undertaken.

Doyle and colleagues (2001) described a multinational pharmacoeconomic evaluation which compared the cost effectiveness of venlafaxine, SSRIs and TCAs in acute depression. A decision analytic model with a 6-month time horizon was developed. This model was adapted with country specific estimates from a clinical management analysis, meta-analytic rates and two published meta-analyses and a resource valuation of treatment costs by local health economists in each country. Cost effectiveness was determined using the expected values for both a successful outcome and a composite measure of outcome termed ‘symptom-free days’. Venlafaxine dominated the other two options since its expected total health service costs were the lowest and it was more effective in terms of both success rate and symptom free days. These findings were explored with sensitivity analysis. This study was conducted in several countries in addition to the UK, which limits the generalisability of the results to the UK. Because of the marked differences in healthcare systems there would be differences in healthcare resource use patterns and patient variations. The short time horizon modelled may not capture all the costs and benefits of the drugs for the treatment of depression.

10.10.5. Summary of health economic evidence

The pharmacoeconomic evidence (much of it industry funded) presented above suggests that escitalopram is better in terms of costs and benefits compared with some of the antidepressants. There is also a weak trend that reflects that SSRIs may be more cost effective than TCAs. (In the previous guideline, pharmacoeconomic evidence suggested that SSRIs were more cost effective than TCAs for the first-line treatment of depression.)

In the previous guideline, pharmacoeconomic evidence suggested that venlafaxine was more cost effective than SSRIs; however, the clinical evidence review at the time highlighted that the clinical estimates used in the economic studies of the drugs compared were inconsistent with the results of the NCCMH clinical evidence review. Therefore an opportunity cost approach was adopted and primary care costs of the different antidepressants were considered alongside the clinical evidence. It is evident that the nature of the current pharmacoeconomic data is piecemeal – no study compares all the relevant antidepressants drugs in a single evaluation. Such an evaluation could inform future guideline recommendations.

The updated meta-analyses of clinical evidence in this guideline points to similar levels of effectiveness across the antidepressants reviewed; that is, they show no robust clinically important superiority in terms of effectiveness. The guideline update recommends that normally an SSRI should be prescribed because they are as effective as other antidepressants, are better tolerated and are less likely to be discontinued because of side effects. Most SSRIs are off patent and available in generic form. In the case of newer drugs, the lack of any greater effect than older drugs makes the added cost potentially not worthwhile (see Table 83). Additionally, a better tolerated drug may also result in cost savings because of the potential decrease in adverse event related healthcare resource use. Therefore, when making a treatment decision regarding the use of an antidepressant, many factors should be taken into consideration for example, clinical history, side effect profile, cost of drug and patient choice.

Table 83

Drug acquisition costs.

The findings from the health economic evidence highlighted the need for de novo economic modelling for this guideline (see Section 10.12).

No new pharmacoeconomic evidence on relapse prevention, maintenance therapy or switching and sequencing patterns were identified in the UK setting.

10.12.1. Introduction

As described in Section 10.10, the systematic search of economic literature identified a number of studies on pharmacological treatments for the management of depression in the UK. The studies were characterised by varying quality in the methods employed. The number of antidepressants assessed in this literature was limited and did not include the whole range of drugs available in the UK for the treatment of people with depression. These findings highlighted the need for de novo economic modelling for this guideline. The objective of economic modelling was to explore the relative cost effectiveness of antidepressants for people with depression in the current UK clinical setting, incorporating the results of a recently published network meta-analysis (Cipriani et al., 2009), as described in Section 10.11.

10.12.2. Methods

Model structure

A pragmatic decision analytical model was constructed using Microsoft Excel XP. The model constructed for the economic analysis of combination therapy versus antidepressant treatment in Section 8.9 was adapted for this analysis. Within the antidepressant model, patients were initiated on a specific antidepressant and either continued or discontinued treatment. Patients continuing their initial antidepressant treatment either responded or did not respond. Patients who responded to initial treatment received 6 months of maintenance therapy and then were assumed to either relapse or enter remission. People who discontinued from initial antidepressant treatment were assumed to receive various levels of care for their depression, including no care. Some of these people were assumed to clinically improve, and then either relapse or enter remission. The time horizon of the analysis was 14 months; this consisted of 2 months of treatment, reflecting the time point at which the clinical efficacy and acceptability parameters reported in Cipriani and colleagues (2009) were measured, plus 12-month follow-up, for which relapse data were available. Switching to second-line antidepressants was not considered for those patients who discontinued their first-line antidepressant treatment or who did not respond to treatment. Two separate analyses were conducted for hypothetical cohorts of 100 patients with either moderate or severe depression, each assessing the relative cost effectiveness of the ten antidepressants assessed. A schematic diagram of the economic model is presented in Figure 9.

Figure 9

Schematic diagram of the economic model structure.

Estimation of response and discontinuation rates in the economic model

The efficacy and acceptability results from the network meta-analysis by Cipriani and colleagues (2009) are summarised in Table 84. The odds ratios reported for fluoxetine versus each of the other antidepressants were converted into probabilities (response and dropouts) for each antidepressant considered in the economic model using the following formulae:

Table 84

Efficacy (expressed as response rate) and acceptability (reflected in dropout rates) of antidepressants, expressed as odds ratios (OR) of fluoxetine versus each of the antidepressants assessed (taken from Cipriani et al., 2009).

$${\text{Odds}}_{\text{FL}}={\text{P}}_{\text{FL}}/(1-{\text{P}}_{\text{FL}})$$

1

$${\text{OR}}_{(\text{FL},\text{AD})}={\text{Odds}}_{\text{FL}}/{\text{Odds}}_{\text{AD}}=>{\text{Odds}}_{\text{AD}}={\text{Odds}}_{\text{FL}}/{\text{OR}}_{(\text{FL},\text{AD})}$$

2

$${\text{P}}_{\text{AD}}={\text{Odds}}_{\text{AD}}/(1+{\text{Odds}}_{\text{AD}})$$

3

Odds_FL and P_FL are the odds and probability (of relapse or dropping out) for fluoxetine at 8 weeks; Odds_AD and P_AD are the odds and probability (of relapse or dropping out) for each of the other antidepressants considered at 8 weeks; and OR_(FL,AD) is the odds ratio of fluoxetine versus each antidepressant (of relapse or dropping out) at 8 weeks.

The probabilities for fluoxetine were estimated based on data reported for 54 RCTs considered in the network meta-analysis that included fluoxetine in one of their arms. Two of the trials had three arms and compared fluoxetine with paroxetine and sertraline. The data on fluoxetine from these two trials were reported twice, and therefore have been double-counted at the estimation of probabilities on response and dropping out for fluoxetine because it was not possible to identify and isolate respective data coming from these two RCTs. Because both response and dropout rates referred to an 8-week period, the probabilities for discontinuation and response were applied over a period of 2 months in the economic model. The probabilities for response and discontinuation for each antidepressant over 8 weeks, along with their 95% credible intervals, are presented in Table 85.

Table 85

Probabilities estimated for use in the economic model (adapted from Cipriani et al., 2009).

Estimation of quality-adjusted life-years

To calculate QALYs, QoL weights estimated in a study of patients with depression were used (Sapin et al., 2004) (see Section 8.9 for further details). Utility weights used in the economic analysis are presented in Table 86.

Table 86

Quality-of-life weights utilised in the economic model.

Drug acquisition costs

Drug acquisition costs were taken from BNF 56 (British Medical Association and the Royal Pharmaceutical Society of Great Britain, 2008), with the exception of the cost of venlafaxine which was obtained from the Electronic Drug Tariff (NHS, Business Services Authority, 2009) because this antidepressant has recently become available in generic form but BNF 56 has not captured this information. The daily dosage of all ten antidepressant drugs was informed by the midpoint of the range of daily dosages presented in Cipriani and colleagues (2009) and by the BNF. It was assumed that patients with moderate or severe depression would both receive the same average daily dosage. For all patients, the total costs of antidepressants were calculated over the 8 weeks of initial therapy. It was assumed that all patients who did not discontinue and responded to their initial treatment after 8 weeks would continue to receive maintenance antidepressant treatment at the same dose over a further 6 months in the model. The average daily dosages and the drug acquisition costs are presented in Table 87.

Table 87

Average daily dosages and acquisition costs of antidepressant drugs included in the economic model.

Monitoring costs

All patients receiving antidepressant treatment were assumed to be actively monitored either in primary or secondary care during both the initial treatment period and the maintenance treatment period. Based on the same assumptions used in the combination therapy versus antidepressant treatment model (see Section 8.9), all patients with moderate depression and 50% of patients with severe depression would receive standard GP care while the remaining 50% of patients with severe depression would receive specialist mental health outpatient care. According to the expert opinion of the GDG, it was estimated that patient monitoring in both primary and secondary care consists of two fortnightly visits in the first month followed by one visit in the second month; the maintenance therapy period consists of one GP/specialist visit every 2 months. The unit costs of a GP consultation (£36) and a mental health outpatient consultation (£130) were both taken from the latest PSSRU estimates (Curtis, 2009). The total antidepressant treatment costs including patient monitoring are presented in Table 88.

Table 88

Total healthcare costs applied in the economic model.

10.12.3. Data analysis and presentation of the results

A deterministic analysis was undertaken, where data are analysed as point estimates; results are presented as mean total costs and QALYs associated with each treatment option assessed. Relative cost effectiveness between alternative treatment options is estimated using incremental analysis: all options are first ranked from the most to the least effective; any options that are more expensive than options that are higher in ranking are dominated (because they are also less effective) and excluded from further analysis. Subsequently, ICERs are calculated for all pairs of consecutive options, starting from the most to the least effective. ICERs express the additional cost per additional unit of benefit associated with one treatment option compared with another. Estimation of such a ratio allows consideration of whether the additional benefit is worth the additional cost when choosing one treatment option over another. If the ICER for a given option is higher than the ICER calculated for the previous intervention in ranking, then this strategy is also excluded from further analysis on the basis of extended dominance. After having excluded cases of dominance and extended dominance, ICERs are recalculated. The treatment option with the highest ICER below the cost-effectiveness threshold is the most cost effective option.

10.12.4. Results

Mirtazapine appears to be the most cost-effective option among those assessed for both moderate and severe depression, producing the highest number of QALYs and the lowest costs among all drugs assessed (dominant option). Full results of the deterministic analysis for both moderate and severe depression are presented in Table 89, where the antidepressant drugs have been ranked from the most to the least effective in terms of number of QALYs gained.

Table 89

Mean costs and QALYs associated with each antidepressant assessed for patients with depression.

If mirtazapine is not a suitable treatment option for patients with moderate or severe depression, the next option would be escitalopram or sertraline because venlafaxine is dominated by escitalopram and the remaining antidepressants are dominated by sertraline. The ICERs of escitalopram versus sertraline are £32,987 per QALY for moderate depression and £27,172 per QALY for severe depression. Both ICERs are above the current cost-effectiveness threshold of £20,000 per QALY recommended by NICE (NICE, 2008b). Therefore, based on the results of the deterministic analysis, for patients with either moderate or severe depression, sertraline is the second most cost-effective option after mirtazapine and escitalopram is the third most cost-effective option. By repeating this process in steps, and excluding from each new incremental analysis all options found to be cost effective in previous analyses, it is possible to rank all antidepressants in terms of cost effectiveness.

The rankings of antidepressants in terms of QALYs in Table 89 were identical for both moderate and severe depression. Reboxetine was ranked last in both cases, resulting in the lowest number of QALYs and the highest costs. Overall, the rankings of antidepressants in terms of cost-effectiveness are very similar to the ranking of antidepressants in terms of efficacy, based on the ORs of fluoxetine versus each anti-depressant as reported by Cipriani and colleagues (2009). In their analysis, mirtazapine, followed by escitalopram, venlafaxine, sertraline and citalopram were ranked as the five best antidepressants in terms of efficacy (measured by ORs versus fluoxetine), with results being statistically significant for the first four of them. In the economic analysis, mirtazapine, followed by sertraline, escitalopram, citalopram and venlafaxine were ranked as the five best antidepressants in terms of cost effectiveness for both moderate and severe depression. Escitalopram and venlafaxine both fell slightly in the cost-effective rankings because escitalopram remains under patent and venlafaxine has only recently become available in generic form and its price remains high (although it may be expected to fall substantially). The other three antidepres-sants are available in generic form and hence much cheaper. Table 90 presents the rankings of each antidepressant in terms of both their efficacy and cost effectiveness.

Table 90

Rankings of each antidepressant in terms of efficacy and cost effectiveness.

Sensitivity analysis was undertaken to explore uncertainty around the ORs for efficacy and acceptability estimated in the network meta-analysis by using the upper and lower limits of the 95% credible intervals. The analysis demonstrated that overall results were robust with mirtazapine remaining the dominant option for both moderate and severe depression.

10.12.5. Discussion – limitations

Given the time constraints involved, a preliminary economic analysis was undertaken based on the results of the network meta-analysis by Cipriani and colleagues (2009). The model used to compare the cost effectiveness of combination therapy and antidepressant treatment was adapted for this analysis. The network meta-analysis examined 12 new-generation antidepressants, of which two (bupropion and milnacipran) were excluded from the economic analysis in this guideline. The study did not analyse older antidepressants including TCAs and MAOIs, which is a limitation in terms of the comprehensiveness of the economic analysis presented here. The study evaluated efficacy regarding response and acceptability in terms of dropouts over the acute phase of depression (8 weeks). As Cipriani and colleagues (2009) acknowledge, other important outcomes such as side effects, toxic effects, discontinuation symptoms and social functioning were not investigated in the meta-analyses. Other possible limitations of the study have been highlighted in the clinical review in Section 10.11. A more comprehensive economic analysis would be able to consider costs and outcomes over a longer time horizon, consider issues of drug sequencing or switching (for patients who discontinue initial antidepressant treatment), and give more explicit consideration (captured in estimation of QALYs) of the side effects of different antidepressants as well as impacts on patient mortality (because of side effects or increased suicide risk).

The economic analysis did not consider the possibility of switching to second-line antidepressants for patients who discontinue their first-line antidepressant, which is another possible limitation. In clinical practice, if a patient discontinues their initial antidepressant because of adverse side affects or other factors, another second-line antidepressant would almost certainly be offered. Another issue relates to the current and future costs of the antidepressants analysed. Venlafaxine has recently been available in generic form and, although the current price listed in the NHS Drug Tariff (NHS, Business Services Authority, 2009) remains high, it is anticipated that this price will fall further to non-proprietary levels. Similarly, it is anticipated that escitalopram will shortly be available in generic form. As the prices of both antidepressants in generic form are likely to be lower in the future, their relative cost effectiveness compared with other antidepressants is likely to be further improved.

Another major limitation of this economic model is the inadequate exploration of uncertainty around the results in terms of the assumptions and the clinical efficacy and acceptability data used. Given the considerable uncertainty around some of the input parameters used in the model, and the underlying assumptions behind them, comprehensive deterministic sensitivity analyses are required. Ideally, probabilistic sensitivity analysis, which demonstrates the joint uncertainty between all of the different parameters used in the model, is also required. However, this would have necessitated full access to the results (posterior estimates of every iteration) of the network meta-analysis by Cipriani and colleagues (2009).

10.12.6. Conclusions

The findings of this preliminary economic analysis suggest that mirtazapine might be more cost effective than other antidepressants in the treatment of people with moderate and severe depression and support the findings of Cipriani and colleagues (2009) regarding the clinical superiority of mirtazapine. However, these economic findings are subject to considerable uncertainty arising from the limitations of the current model and lack of incorporation of data on the relative adverse effects of the drugs in the model. Addressing these issues may alter the outcome of the model.

10.14.1. Choice of antidepressant¹⁴⁷

10.14.1.1.

Discuss antidepressant treatment options with the person with depression, covering:

• the choice of antidepressant, including any anticipated adverse events, for example, side effects and discontinuation symptoms (see Section 11.8.7.2) and potential interactions with concomitant medication or physical health problems¹⁴⁸
• their perception of the efficacy and tolerability of any antidepressants they have previously taken.

10.14.1.2.

When an antidepressant is to be prescribed, it should normally be an SSRI in a generic form because SSRIs are equally effective as other antidepressants and have a favourable risk–benefit ratio. Also take the following into account:

• SSRIs are associated with an increased risk of bleeding, especially in older people or in people taking other drugs that have the potential to damage the gastrointestinal mucosa or interfere with clotting. In particular, consider prescribing a gastroprotective drug in older people who are taking non-steroidal anti-inflammatory drugs (NSAIDs) or aspirin.
• Fluoxetine, fluvoxamine and paroxetine are associated with a higher propensity for drug interactions than other SSRIs¹⁴⁹.
• Paroxetine is associated with a higher incidence of discontinuation symptoms than other SSRIs¹⁵⁰.

10.14.1.3.

When prescribing drugs other than SSRIs, take the following into account:

• The increased likelihood of the person stopping treatment because of side effects (and the consequent need to increase the dose gradually) with venlafaxine, duloxetine and TCAs.
• The specific cautions, contraindications and monitoring requirements for some drugs. For example:

  –

  the potential for higher doses of venlafaxine to exacerbate cardiac arrhythmias and the need to monitor the person’s blood pressure

  –

  the possible exacerbation of hypertension with venlafaxine and duloxetine

  –

  the potential for postural hypotension and arrhythmias with TCAs

  –

  the need for haematological monitoring with mianserin in elderly people¹⁵¹.

• Non-reversible monoamine oxidase inhibitors (MAOIs), such as phenelzine, should normally be prescribed only by specialist mental health professionals.
• Dosulepin should not be prescribed.

10.15.1. Introduction

Received wisdom has been that antidepressants have a delayed onset of action and that it takes 2 to 4 weeks for them to begin to work. This is now recognised as incorrect and it has been shown from data from clinical trials that improvement can start immediately, with the greatest degree of improvement occurring in the first week; the curve begins to flatten off thereafter, with a smaller degree of improvement as time goes on. Posternak and Zimmerman (2005), in a meta-analysis of 47 placebo-controlled studies followed up at 6 weeks, found that 35% of the improvement occurred between weeks 0 and 1, and 25% between weeks 1 and 2. However, it is important to recognise that although the curve flattens, some people continue to improve after this and the assessment of the literature is influenced by the duration of follow-up. For example, in the large naturalistic STAR*D study (Trivedi et al., 2006), which enrolled nearly 2,876 patients followed up to 12 weeks, the mean time to response with citalopram (defined by at least 50% reduction in the Quick Inventory of Depressive Symptomatology-Self Report [QIDS-SR]) was 5.7 weeks with about 65% of patients responding by 6 weeks, but some patients continuing to respond at 12 weeks and beyond. Malt and colleagues (1999) undertook an RCT of 372 primary care patients randomised to sertraline, mianserin or placebo, and treated for 24 weeks. Response was defined as at least a 50% improvement on the HAMD plus at least much improved and not more than mildly ill on the CGI. Depending on the treatment arm, 58 to 91% of those responding by 24 weeks had done so by 6 weeks and 79 to 98% responded by 12 weeks.

The rate and degree of improvement also appears to be influenced by the frequency of follow up. Posternak and Zimmerman (2007), in a systematic review of 41 RCTs, found that weekly assessment between weeks 2 and 6 led to a greater reduction in HAMD score than less frequent assessment in a dose-related manner. This applied to both placebo and drug treatment arms and they estimated that follow-up frequency accounted for about 40% of the placebo response.

These studies emphasise the importance of the early stages of treatment in response to antidepressants and highlight the role of frequency of monitoring. A key issue related to this is the optimum time to change treatment. Switching treatment too early could lead to rejection of an effective treatment, which in the long run will be unhelpful when future treatment options are considered and could lead to a merry-go-round of treatment changes. Increasing the dose too early could lead to patients unnecessarily being maintained on higher than needed doses of antidepressants over a prolonged period of time with associated increased side effects or treatment discontinuation (Bollini et al., 1999; Furukawa et al., 2002b). Delaying change in treatment too long could prolong the period of depression if symptoms are not going to respond to current drug/dose, lead to a patient’s loss of faith in treatment, and increase depression-related morbidity and even mortality.

There is limited but consistent evidence, mostly from studies with SSRIs, that increasing the dose after 3 weeks treatment in those not responding (<50% decrease in rating scale) or remitting (HAMD <9) at this early stage does not improve outcome at 6 weeks (Adli et al., 2005). However, these are stringent criteria and do not allow clinicians to judge whether altering treatment is beneficial in those not improving at all. Stassen and colleagues (1993) found that the natural variation in assessment makes the minimum reliably detectable improvement in a rating scale in the range of 15 to 25% and most subsequent studies have examined the predictive value of non-improvement using a criterion of 20% or less (these are referred to here as 20% improvers).

10.15.2. Early prediction of eventual response

Most studies have found that early improvement in the first 2 weeks (20% or greater improvement) is a good predictor of response by the end of the study (Nierenberg et al., 1995; Nierenberg et al., 2000; Szegedi et al., 2003; Szegedi et al., 2009). This is consistent with usual clinical practice.

The outcome of concern is the number of non-improvers at each time point who subsequently respond or remit by the end of a certain time frame because this provides some guide as to when changing treatment is likely to improve outcome. This can be assessed using the negative predictive value (NPV), which is the proportion of non-improvers not going on to achieve response/remission at the last evaluation. Where this is low, non-improvement at that time point is not a useful predictor of outcome at endpoint. A matter for debate is: what is a reasonable value for the NPV that should trigger change of treatment? It is suggested that it lies somewhere between 70 and 80%; in other words, if a non-improver still has more than a 20 to 30% chance of responding then it is probably reasonable to persist longer with treatment before adding to the potential side-effect burden by increasing the dose, adding another drug or changing the drug. To put this into context, in the large naturalistic STAR*D study (Rush et al., 2006), changing treatment in non-responders to the first treatment only resulted in an average increase in response of about 30%.

The strongest recent case for changing treatment as early as 2 weeks in non-improvers is Szegedi and colleagues (2009). They pooled data from 41 RCTs in which mirtazapine was compared with active comparators or placebo. Most studies were only 6 weeks in duration. They found that 60 to 76% of patients on antidepres-sants compared with 52% on placebo were 20% improvers at 2 weeks. Using a definition of stable response (response at both 4 weeks and subsequently, usually 6 weeks), the overall NPV for those not having a 20% improvement at 2 weeks was 89%, in other words only 11% would have a stable response as defined. The limitations are the short time frame of most of the studies and the requirement to have responded by 4 weeks.

It is useful to consider other studies according to the length of follow up. A 5-week study found an NPV for 20% improvement on response at 5 weeks of 48 to 54% at 2 weeks, 74 to 83% at 3 weeks and 96 to 99% at 4 weeks (Stassen et al., 1993). A 6-week study found an NPV defined in the same way as 65 to 72% at 2 weeks, 77 to 94% at 3 weeks and 82 to 94% at 4 weeks (Szegedi et al., 2003). Two 8-week studies of fluoxetine (defining improvement as 20% reduction in one and 30% in another) (Nierenberg et al., 1995; Nierenberg et al., 2000) and a pooled analysis of 14 escitalopram studies (20% improvement) (Baldwin et al., 2009) found NPVs of 55 to 64% at 2 weeks, 80 to 82% at 4 weeks and 90 to 93% at 6 weeks. In contrast, an open 12-week study of fluoxetine (Quitkin et al., 2003) using 25% improvement to predict remission (HAMD <8) found an NPV of only 49% at 4 weeks, 59 to 69% at 6 weeks and 77% at 8 weeks. Finally, a naturalistic study of 795 inpatients (Henkel et al., 2009) with a variable follow-up (discharge, mean = 60 days), using 20% improvement found only a 37% NPV at 2 weeks for response and 43% at 4 weeks. NPVs for remission (HAMD <8) were higher at 69% and 72% respectively.

It is possible to draw only tentative conclusions from these studies. Higher early NPVs are associated with shorter studies and RCTs and lower NPVs with longer, open studies and possibly more severe patients. Taking the middle ground with an assessment period of 8 weeks and an NPV based on less than 20% improvement predicting lack of response at 8 weeks, a reasonable time to consider a change of treatment in these patients would be at 3 to 4 weeks. In patients who have failed previous trials of treatment, and in more severely ill patients, longer trials of treatment may be warranted before making changes.

10.15.3. From evidence to recommendations

Antidepressant studies examining the onset of improvement in relation to response or remission at the end of the study vary in their findings according to the exact methodology used. Taking studies evaluating response over an 8-week time frame, which was thought by the GDG to present a realistic clinical situation, and using less than a 20% improvement on the HAMD score to indicate patients with a lack of, or barely detectable, response, when evaluated at 2 weeks these patients had about a 40% chance of achieving a response at 8 weeks falling to a 20% chance if they had failed to improve by 4 weeks. The rate of improvement after 6 to 8 weeks of treatment is slower and only a minority of non-responders at this stage will go on to have an adequate response over the next 6 to 18 weeks.

In addition, the GDG noted that there is some evidence that a higher frequency of assessment between weeks 2 and 6 is associated with a better outcome. Taken together these led to the recommendation that if there is no, or barely any detectable improvement at 2 to 4 weeks, patients should be followed weekly and consideration given to changing treatment at 3 to 4 weeks. Patients who are improving should have their improvement monitored and if there has been insufficient response at 6 weeks in the absence of a continuing trajectory of improvement, consideration should be given to changing treatment at that stage.

10.15.4. Clinical practice recommendations

10.15.4.1.

For people started on antidepressants who are not considered to be at increased risk of suicide, normally see them after 2 weeks. See them regularly thereafter; for example, at intervals of 2 to 4 weeks in the first 3 months, and then at longer intervals if response is good.¹⁵²

10.15.4.2.

If a person with depression develops side effects early in antidepressant treatment, provide appropriate information and consider one of the following strategies:

• monitor symptoms closely where side effects are mild and acceptable to the person or
• stop the antidepressant or change to a different antidepressant if the person prefers or
• in discussion with the person, consider short-term concomitant treatment with a benzodiazepine if anxiety, agitation and/or insomnia are problematic (except in people with chronic symptoms of anxiety); this should usually be for no longer than 2 weeks in order to prevent the development of dependence.

10.15.4.3.

If the person’s depression shows no improvement after 2 to 4 weeks with the first antidepressant, check that the drug has been taken regularly and in the prescribed dose.

10.15.4.4.

If response is absent or minimal after 3 to 4 weeks of treatment with a therapeutic dose of an antidepressant, increase the level of support (for example, by weekly face-to-face or telephone contact) and consider:

• increasing the dose in line with the Summary of Product Characteristics if there are no significant side effects or
• switching to another antidepressant as described in Section 12.3.16 if there are side effects or if the person prefers.

10.15.4.5.

If the person’s depression shows some improvement by 4 weeks, continue treatment for another 2 to 4 weeks. Consider switching to another antidepressant as described in 12.3.16 if:

• response is still not adequate or
• there are side effects or
• the person prefers to change treatment.