Lithium or an atypical antipsychotic drug in the management of treatment resistant depression: a systematic review and economic evaluation

Pathophysiology

The aetiology of depression is not fully understood, although there is evidence to suggest depression is a complex interaction among biological, genetic, psychosocial and environmental factors. 2 The highest rates of depression typically occur in people between 25 and 44 years old, and females are twice as likely as males to experience depression,2 although how these figures relate to the subgroup of patients with TRD is difficult to know owing to the lack of epidemiological data and the lack of a consistent definition for TRD. Family history of depression is also a risk factor for depression2 and a previous history of MDD increases the risk of future episodes (i.e. relapses). 6 In addition, it has been reported that patients with depression have increased morbidity and mortality. 2 For example, they are more likely to die from cardiovascular disease7 or suicide. 8

Diagnosis and assessment of response to treatment

People presenting with depression may complain of depressed mood, loss of interest or pleasure, feelings of guilt or low self-worth, suicidal ideation, disturbed sleep or appetite, low energy and poor concentration. Depression can be diagnosed clinically using different criteria. The most commonly used criteria are the DSM-IV (Diagnostic and Statistical Manual of Mental Disorders-Fourth Edition published by the American Psychiatric Association)9 and the ICD-10 (International Classification of Diseases, Tenth Revision) criteria developed by the World Health Organization (WHO). 10 The DSM-IV system requires at least five out of nine symptoms for a diagnosis of major depression, including at least one of the following two symptoms: low mood; or loss of interest and pleasure. 9 A diagnosis of moderately severe depressive episode using the ICD-10 classification system requires the presence of at least three out of ten depressive symptoms, including at least two of the following three symptoms: low mood; loss of interest and pleasure; or loss of energy. 10 In both cases, symptoms should be present for at least 2 weeks and each symptom should be present at sufficient severity for most of every day (Box 1 shows the full diagnostic criteria).

The recommended treatment goal in depression is to reach remission, which is defined as the relative absence of clinical symptomatology and is usually determined by reaching a certain score on a treatment response rating scale. 11,12 Response to treatment in TRD is commonly measured by a reduction of at least 50% on either the Hamilton Depression Rating Scale (HAMD)13 or Montgomery–Åsberg Depression Rating Scale (MADRS). 14 Both scales are designed to be administered via a clinical interview and consist of a list of symptoms of depression which the clinician must assess the patient for during the interview. The clinician then rates the patient's symptoms on the scales provided for each symptom and adds up the individual scores to provide the overall score. In both cases, the lower the score, the less severely depressed the patient. The definition of remission on the HAMD is typically defined as a score of ≤  7 on the 17-item version of the HAMD. 15 However, there is currently no consensus on a definition of remission for the MADRS,15 although clinical expert advisors reported that scores of ≤  10 on the MADRS are commonly used.

Relevant national guidelines

The key clinical guideline for depression in adults in the UK is the NICE clinical guideline on the treatment and management of depression in adults (CG90; extracts from this guideline have been reproduced here with permission);16 this guideline is the updated version of CG23. 3 It should be noted that in CG9016 a decision was taken to no longer use the term ‘treatment-resistant depression’, as there were concerns that the term implies there is a ‘natural cut-off between people who respond to one or two antidepressants compared with those who do not’, and that this ‘is not supported by the evidence, and the term may be taken by both doctors and patients as a pejorative label’. 16 As a result, in CG9016 it was decided to approach the problem of inadequate response by considering sequenced treatment options rather than by a category of patient. This is reflected throughout CG9016 by use of the label ‘inadequate response to initial interventions’.

The recommendations in CG9016 for the sequencing of drug treatments in patients with an initial inadequate response are presented in Box 2, together with the additional recommendations for monitoring when treatment with lithium or atypical antipsychotic drugs (AAPs) is chosen.

In addition to the NICE clinical guideline (CG90) for the treatment and management of depression in adults,16 there are guidelines published by the British Association of Psychopharmacology (BAP). 20 The BAP 2008 guidelines20 provide similar advice to CG9016 on the use of augmentation therapy in TRD, although they do not specifically mention a definition of how many treatment failures are required for a diagnosis of TRD (Box 3).

A key problem highlighted in both of the guidelines16,20 is that there is limited randomised controlled trial (RCT) evidence comparing the different potential augmentation treatments and thus there is currently much uncertainty as to which augmentation therapy is the most clinically effective and/or cost-effective in the management of TRD.

Selective serotonin reuptake inhibitors

The SSRIs licensed for use in the UK are citalopram, escitalopram, fluoxetine, fluvoxamine, paroxetine and sertraline. They are all available in England and Wales as both generic and branded drugs: citalopram (Cipramil^®, Lundbeck); escitalopram (Cipralex^®, Lundbeck); fluoxetine (Prozac^®, Lilly); fluvoxamine (Faverin^®, Abbott Healthcare); paroxetine (Seroxat^®, GlaxoSmithKline); and sertraline (Lustral^®, Pfizer). SSRIs work by selectively inhibiting the reuptake of serotonin [5-hydroxytryptamine (5-HT)]; hence, they are termed SSRIs. SSRIs are commonly used first-line for treating depression as they are better tolerated and are safer in overdose than other classes of antidepressants. 19 In particular, the SSRIs are less sedating and have fewer antimuscarinic and cardiotoxic effects than tricyclic antidepressants. 19 Side effects of SSRIs include gastrointestinal effects (e.g. nausea, vomiting, abdominal pain, diarrhoea, constipation), anorexia, rash, dry mouth, anxiety, headache, insomnia, tremor, dizziness, asthenia, drowsiness, convulsions, sexual dysfunction, urinary retention, sweating and hyponatraemia. 19

SSRIs are recommended for use with caution in patients with epilepsy, cardiac disease, diabetes mellitus, susceptibility to angle-closure glaucoma, a history of mania or bleeding disorders, or taking drugs that increase the risk of bleeding. 19 SSRIs are associated with a risk of seizures and should therefore be used with caution in those receiving concurrent ECT. 19

Lithium

Lithium is used in the UK to augment antidepressants in patients with TRD, although this is an unlicensed indication. 19 Lithium is more commonly used for its licensed indication as a mood-stabilising agent, although the precise mechanism of action of lithium remains unknown.

Lithium is available in both generic and branded formulations in England and Wales. These are as follows: lithium carbonate (Camcolit^®, Norgine; Lithonate^®, Teva UK; Liskonum^®, GlaxoSmithKline; Priadel^® tablets, Sanofi-aventis); and lithium citrate (Li-Liquid^®, Rosemont; Priadel^® liquid, Sanofi-aventis).

Lithium salts have a narrow therapeutic–toxic ratio and therefore serum lithium concentrations should be monitored regularly during treatment. 19 Levels should be measured 12 hours after the dose with the aim of achieving a serum-lithium concentration of 0.4–1  mmol/l. 19 The BNF recommends that routine ‘serum-lithium monitoring should be performed weekly after initiation and after each dose change until concentrations are stable, then every 3 months thereafter’. 19 It should also be noted that different lithium preparations have different bioavailability and so caution is required when changing the lithium preparation. 19

Renal function should also be monitored at baseline and every 6 months thereafter as lithium is excreted renally, and so renal impairment could cause lithium levels to build up, leading to toxicity. 19 Serum lithium levels are also affected by a patient's sodium or fluid intake, with the risk of lithium toxicity increasing if there is sodium depletion or dehydration. Long-term use of lithium has been associated with thyroid disorders and mild cognitive and memory impairment. 19 Therefore, long-term treatment requires careful assessment of risk and benefit, and monitoring of thyroid function every 6 months. 19

Side effects of lithium therapy include gastrointestinal disturbances, fine tremor, renal impairment, polydipsia, leucocytosis, weight gain, cognitive dulling, hyperparathyroidism, hyperthyroidism, hyperglycaemia, hypermagnesaemia and hypercalcaemia. 19 The signs of lithium toxicity include blurred vision, anorexia, vomiting, diarrhoea, muscle weakness, polyuria and increasing drowsiness eventually leading to coma. 19

It is recommended that lithium is avoided if possible in patients with renal impairment and used with caution in patients with cardiac disease; QT-interval prolongation; conditions with sodium imbalance (e.g. Addison's disease); diarrhoea; vomiting; intercurrent infection; concurrent ECT treatment; psoriasis; and myasthenia gravis. 19 In addition, caution is recommended in the use of lithium in surgical patients, patients on diuretics and the elderly. 19

Atypical antipsychotic drugs

As discussed above for lithium, AAPs can similarly be used as adjunctive therapies to antidepressants in the treatment of patients with TRD. AAP drugs are also known as the second-generation antipsychotic drugs and act on a range of receptors in comparison with the first-generation antipsychotic drugs that predominantly act on only one type of receptor. The only AAP licensed for use in the UK as an adjunctive treatment in MDD is quetiapine. 19 However, there are several other AAPs that are used as unlicensed treatments in patients with unipolar TRD, including aripiprazole, olanzapine and risperidone. In addition, there are other AAPs classed as second-generation antipsychotic drugs that could also potentially be used, albeit unlicensed.

Most of the AAPs are still patented and thus are available only as branded drugs. The AAP drugs available for use in England and Wales are as follows: amisulpride (Solian^®, Sanofi-aventis); aripiprazole (Abilify^®, Bristol–Myers Squibb); clozapine (Clozaril^®, Novartis; Denzapine^®, Merz; Zaponex^®, Teva UK); olanzapine (Zyprexa^®, Lilly); paliperidone (Invega^®, AstraZeneca); quetiapine [Seroquel^®, AstraZeneca; Seroquel XL^® (modified release), AstraZeneca]; risperidone (Risperdal^®, Janssen-Cilag). In addition to these AAPs, ziprasidone (Geodon^®/Zeldox^®, Pfizer) is used elsewhere in Europe, but is not licensed or used routinely for any indication in the UK. As this report is focused on treatments available for use in the NHS, ziprasidone will not be discussed further in this section.

The choice of AAP medication is usually influenced by the patient's medication history, and consideration of individual patient factors, for example, the risk of particular side effects such as weight gain or impaired glucose tolerance. As previously discussed, AAPs each act on different receptors. These are summarised in Table 1.

Full blood count, urea and electrolytes, and liver function test monitoring are required at the start of therapy with antipsychotic drugs, and then annually thereafter. 19 In addition, clozapine requires differential white blood cell monitoring weekly for 18 weeks then fortnightly for up to 1 year, and then monthly as part of the clozapine patient monitoring service. 19 Blood lipids and weight should also be measured at baseline, at 3 months, and then yearly, and fasting blood glucose should be measured at baseline, at 4–6 months, and then yearly. 19 Patients taking clozapine or olanzapine should have fasting blood glucose tested at baseline, after 1 month's treatment, and then every 4–6 months. 19 It is also advisable to monitor prolactin concentration regularly. Blood pressure monitoring is also advised before starting therapy and frequently during dose titration of antipsychotic drugs, and ECG monitoring may also be required if the patient has cardiovascular risk factors. 19

There are numerous side effects associated with AAP drugs and the side effects contribute significantly to the reasons for non-adherence to therapy. 19

Most antipsychotic drugs increase prolactin concentration because dopamine inhibits prolactin release, but aripiprazole reduces prolactin because it is a dopamine-receptor partial agonist. 19 Risperidone and amisulpride are most likely to cause symptomatic hyperprolactinaemia. 19 The clinical symptoms of hyperprolactinaemia include sexual dysfunction, reduced bone mineral density, menstrual disturbances, breast enlargement and galactorrhoea. 19

Other side effects associated with AAPs include cardiovascular side effects, such as tachycardia, arrhythmias and hypotension. 19 Hyperglycaemia and sometimes diabetes can occur, particularly with clozapine, olanzapine, quetiapine and risperidone. 19 All antipsychotic drugs may cause weight gain, although the risk and extent varies, with clozapine and olanzapine being the most commonly associated with weight gain. 19 Clozapine and quetiapine can cause postural hypotension (especially during initial dose titration), which may be associated with syncope or reflex tachycardia in some patients. 19 Hypersalivation is also associated with clozapine therapy. 19 In addition, other possible side effects include drowsiness, agitation, restlessness, increased appetite, insomnia, dizziness, headache, confusion, gastrointestinal disturbances, venous thromboembolism, and antimuscarinic symptoms (e.g. dry mouth, constipation, difficulty with micturition, blurred vision and also, very rarely, precipitation of angle-closure glaucoma). 19 Neuroleptic malignant syndrome (NMS) is a rare, but potentially fatal, side effect of all antipsychotic drugs and requires discontinuation of the antipsychotic drug. 19 NMS is characterised by hyperthermia, a fluctuating level of consciousness, muscle rigidity, pallor, irregular pulse, tachycardia, sweating and urinary incontinence.

It is recommended that AAPs are used with caution in patients with cardiovascular disease, a history of epilepsy or those on concomitant drugs that increase the QT interval (on an ECG). 19 In addition, caution is required in the elderly owing to an increased risk of mortality associated with antipsychotic drugs and an increased risk of other serious side effects. 19

Anticipated costs associated with intervention

The direct costs associated with the interventions under review (SSRIs, AAPs and lithium) are limited to the price of the individual tablet or liquid formulations, as there is no requirement for them to be administered in a specialised setting. In addition, there are some costs associated with the monitoring requirements of each therapy, although these costs vary between lithium and AAPs, as well as among the individual AAPs. The costs of the interventions along with the wider costs associated with each intervention are discussed in detail in the cost-effectiveness section of this report (see Chapter 5, Drug costs).

Subgroup analyses

The a priori subgroup analyses deemed to be most important were as follows:

• different durations of depression (i.e. time since first onset of current episode of depression)

• class of previous antidepressants (e.g. SSRI or tricyclic antidepressant)

• sex (i.e. male and female)

• age (i.e. those   <  75 years and those ≥  75 years)

• people with different severities of depression (i.e. based on trial entry HAMD score13).

The reason for selecting these subgroups is that they were highlighted by clinical experts to be the most clinically important subgroups. This is because:

• People who have had TRD for longer periods of time are likely to be more difficult to treat and, thus, could be less likely to respond to augmentation therapy.

• Previous class of antidepressant therapy may have an impact on the response to future treatments (i.e. if two SSRIs have been failed in the current episode rather than two different classes of antidepressants).

• It is unknown whether or not sex has an effect on response to treatment in TRD, but more females tend to be treated for depression than males and thus RCTs may have a higher female–male ratio.

• People of <  75 years of age are known to have different pathophysiologies for their depression and also to respond differently to antidepressants than people aged ≥  75 years.

• People with more severe depression at baseline (i.e. higher HAMD13 score) require a greater improvement to enter remission and so could potentially be less likely to enter remission.

It is thus considered that these subgroups of patients could respond differently to augmentation therapy and so each subgroup will be analysed for the primary outcome in this review if sufficient data are identified to enable such comparison.

Search strategy

The search strategy comprised the following elements:

1. searching of electronic bibliographic databases

2. contact with clinical experts in the field

3. review of the reference lists of retrieved papers

4. searching of the Cochrane Collaboration Depression, Anxiety and Neurosis Review Group (CCDAN) Controlled Trials Register (CTR) databases (CCDANCTR-Studies and CCDANCTR-References).

The following electronic databases were searched:

1. (a) EMBASE (searched from 1974 to August 2011)

2. (b) MEDLINE (searched from inception to August 2011)

3. (c) PsycINFO (searched from inception to August 2011)

4. (d) Cochrane Central Register of Controlled Trials (CENTRAL) (searched from inception to August 2011).

Full details of the search strategies used are provided in Appendix 2.

In addition, to assist the drawing up of final recommendations, the website ClinicalTrials.gov was searched to identify relevant ongoing clinical trials. Trials considered relevant were those that when completed may have an impact on the results of this review. Clinical experts in the relevant therapy areas were contacted for details of trials (published and unpublished) of which they may be aware.

The references from any relevant review papers or RCTs uncovered in the search were also examined for additional references potentially relevant to the review.

The CCDANCTR-Studies and CCDANCTR-References were searched using terms consistent with the search terms used in the other electronic bibliographic databases as a validation exercise of the searches. The searches on the CCDANCTR-Studies and CCDANCTR-References registers were conducted on 7 December 2011.

Abstract appraisal

Titles and abstracts of studies identified by the search process were assessed independently by two reviewers (VH and SB) for inclusion. For cases in which the reviewers were unable to reach a consensus on whether or not the full text should be obtained for further appraisal, the full text was obtained.

When potentially relevant data were available in only an abstract format then attempts were made to contact the corresponding author and drug manufacturer to obtain the full publication or additional information if possible.

The a priori inclusion criteria applied to the review:

• for the review of clinical effectiveness, only RCTs were included

• adults ≥  18 years

• people with unipolar depression

• TRD defined as failure to respond to at least two previous antidepressants in the current episode of depression

• SSRI given as baseline treatment and patient randomised to either lithium or an AAP

• comparator treatment of SSRI plus either lithium, AAP, placebo or no treatment

• minimum duration of 4 weeks' treatment with study medication for the current episode of depression

• studies reporting on one or more of the following outcomes:

  • response

  • QoL

  • adverse events

  • adherence to medication or withdrawals (all cause)

  • relapse rate

  • mortality

  • cost-effectiveness.

The a priori exclusion criteria applied to the review:

• non-randomised studies

• narrative reviews, editorials, opinions

• studies performed in animals

• studies not focusing on the treatment of the acute phase of depression (i.e. those focusing solely on maintenance therapy)

• bipolar depression or bipolar disorder diagnosis prior to study entry

• underlying medical condition or another substantial comorbid psychiatric condition (e.g. psychosis)

• trials reporting only post-crossover results

• trials using non-SSRI antidepressants as the baseline treatment for augmentation with lithium or an AAP.

Study inclusion assessment

Two reviewers (VH and SB) independently assessed the full-text papers of the trials identified during the abstract assessment stage for inclusion and any differences in opinion were arbitrated by a third reviewer (SJE).

Data extraction strategy

A sample of five papers was fully independently data extracted by two reviewers (VH and SB) using a standardised data extraction form (for a copy of the data collection form, please see Appendix 3) and then validated by one reviewer (SB). Agreement between the two reviewers was high and so, owing to time constraints, the remaining papers were independently extracted by one reviewer (VH) and validated by a second reviewer (SB). Discrepancies in the data extracted by the two reviewers were resolved through discussion, with involvement of a third reviewer (SJE) if necessary.

Data from intention-to-treat (ITT) analyses were extracted and it was planned that per-protocol (PP) data would also be extracted for use in a sensitivity analysis, although PP data were not reported in any of the included papers. For the purpose of this review, ITT was defined as patients being analysed in the treatment group to which they were allocated at randomisation regardless of whether they received the wrong intervention, withdrew or were lost to follow-up. Should a trial not report ITT data then missing data were treated as treatment failures to allow the analysis to conform to an ITT analysis.

Study authors and drug manufacturers were contacted to supply any additional information not included in published sources (including relevant subgroup data and additional methodological data required for the quality assessment).

Quality assessment strategy

Outcomes from the studies that met the inclusion criteria were assessed using the updated risk of bias tool developed by the Cochrane Collaboration (March 2011). 27

These criteria assess the following areas:

1. random sequence generation

2. allocation concealment

3. blinding of participants and personnel

4. blinding of outcomes assessment

5. incomplete outcome data

6. selective reporting

7. ‘other bias’.

Based on these criteria, an assessment for each outcome reported in the trial was allocated based on the identified risk of bias. The three bias assessment categories used were low risk, high risk and unclear risk. Only trials that were deemed to be at low or unclear risk of bias were included in the main analysis, with plans to include the trials rated as high risk in a sensitivity analysis; no trial was rated as being at high risk of bias.

Two reviewers (VH and SB) independently rated the trial outcomes for inclusion and any differences in opinion were arbitrated by a third reviewer (SJE). Outcomes reported by each RCT were considered appropriate for inclusion unless the trial demonstrated a high risk of bias across several of the seven risk-of-bias domains assessed for that outcome. No trial was excluded from any of the outcomes analysed based on the risk of bias assessments.

Methods of analysis/synthesis

Data have been tabulated and, where appropriate, meta-analysis undertaken to estimate a summary measure of effect on relevant outcomes based on ITT analyses. Standard pairwise meta-analysis was conducted when more than one trial was identified for inclusion for any pair of treatments under investigation. This was carried out using a fixed-effects model with the Mantel–Haenszel method. 28 Sensitivity analyses were conducted using a random-effects model with the DerSimonian and Laird method. 29

Only one direct head-to-head trial was identified comparing augmentation with AAP with lithium (available as two abstracts and one poster). 30–32 The main analyses of this trial consisted of people with resistance to either one or two antidepressants in their current episode of depression. This trial also included a mixture of different SSRIs and venlafaxine, a serotonin–norepinephrine reuptake inhibitor (SNRI), and was excluded from the primary analyses. The decision to exclude it from the primary analyses for this review was a result of being unable to obtain subgroup data in the subgroup of patients meeting the inclusion criteria for this review (i.e. patients with resistance to two or more antidepressants and taking SSRIs, with the data reported separately for each SSRI). As a result, it was necessary to carry out an indirect comparison to estimate the efficacy of SSRI  +  AAP compared with SSRI  +  lithium. A mixed-treatment comparison (MTC; also called a multiple-treatment meta-analysis and network meta-analysis) was chosen as the method to estimate the effects of SSRI  +  AAP compared with SSRI  +  lithium. A MTC can be seen as an extension of traditional pairwise meta-analysis. 29,33–35

The MTC was conducted using a fixed- and random-effects model, with the most appropriate model chosen for the reporting of the results. This was determined by the model with the lowest deviance information criterion (DIC). 36 DIC measures the fit of the model while penalising for the number of effective parameters. 34,37 For the chosen model, the consistency of the evidence was assessed using the posterior mean residual deviance, which should approximate the number of unconstrained data points in a good-fitting model.

For dichotomous outcomes, the odds ratio (OR) is reported as the summary statistic, and, for continuous outcomes, the mean difference (MD).

The primary analysis is:

• response (measured by a reduction of at least 50% in HAMD13 or MADRS14 score).

The secondary analyses are:

• QoL as reported using a validated QoL rating scale25 (e.g. SF-36)

• adverse events (total number of events, individual adverse events for which comparable data were available for both augmentation with AAP and with lithium, and withdrawal rates due to an adverse event)

• withdrawals (all cause) as a surrogate outcome for adherence to medication

• relapse rate

• mortality (all cause).

In addition, remission rates and mean change from baseline MADRS score14 were also chosen a priori to further assess response to treatment as these were additional clinical parameters that were required for the economic model. Eight-week outcome data were collected where reported. If 8-week data were not available, outcome data from the nearest available time point were collected.

Subgroup analyses were planned in the following populations on only the primary outcome (response), subject to the availability of data:

• different durations of depression (i.e. time since first onset of current episode of depression, short term <  6 months, long term   >  6 months)

• class of previous antidepressants (e.g. SSRI or tricyclic antidepressant)

• sex (i.e. male and female)

• age (i.e. those of ≥  75 years and those of <  75 years old)

• people with different severities of depression, that is, based on trial entry HAMD13 rating using the following categories:13

  • 8–13  =  mild depression

  • 14–18  =  moderate depression

  • 19–22  =  severe depression

  • ≥  23  =  very severe depression.

In the absence of suitable data to perform a meta-analysis, the available data have been tabulated where possible and discussed in a narrative review.

Heterogeneity

Heterogeneity in pairwise meta-analysis has been explored through consideration of the study populations, methods and interventions, by visualisation of results and, in statistical terms, by the chi-squared test for homogeneity and the I²-statistic. Statistically significant heterogeneity has been defined as p  <  0.05. Levels of inconsistency have been assessed using I² and defined as follows. I² of: 0–25%  =  low level of inconsistency; 26–50%  =  moderate level of inconsistency; and >  50%  =  high level of inconsistency. 38

When statistically significant heterogeneity was detected in any of the primary or secondary analyses, hypothesis-generating subgroup analysis was conducted, although the results from such analyses are highlighted in the text and should be treated with caution. Meta-regression was planned if significant statistical heterogeneity was identified among trials analysed and there were 10 or more trials in the review. However, there were insufficient trials in the review to consider any meta-regression for the pairwise meta-analyses.

For the MTC, where a random-effects model was deemed the best fit, the degree of heterogeneity has been investigated by evaluating the posterior mean tau-squared statistic. 39

Sensitivity analysis

The following sensitivity analyses were specified a priori on the primary analysis:

• assuming a ‘class’ effect with SSRIs and AAPs

• different number of prior antidepressants for the current episode of depression

• changing the quality assessment to include the trial outcomes excluded on grounds of methodological quality, i.e. those categorised as being of high risk of bias

• changing the analysis from using ITT data to PP data.

In addition, the following post hoc sensitivity analysis was conducted:

• limiting the primary analysis to trials reporting response measured by a ≥  50% reduction on the MADRS scale. 14

None of the trials included in this review was rated as ‘high risk of bias’ and so the sensitivity analysis including such trials was not required. In addition, no trial reported PP data and so this sensitivity analysis could not be performed.

Publication bias

For each of the primary pairwise meta-analyses, a funnel plot was used to assess publication bias. A regression of normalised effect compared with precision was calculated as a test for small study effects (using p  <  0.10 to indicate a significant result). 40,41

Quantity of research available

The search of electronic databases identified 3717 potentially relevant articles, which, after initial screening, resulted in the identification of 61 potentially relevant full-text articles that were ordered for further screening. An additional three RCTs42–44 were identified from the reference list of a systematic review by Wang et al. ,45 and a further RCT46 was identified from a systematic review by Nelson et al. 11 Both systematic reviews were from the 61 full-text articles assessed. In addition, an unpublished poster32 was provided in response to a request for further information on one RCT. 30

Following the full assessment of all 66 full-text papers, a total of 11 studies30,43,46–53 reported in 15 publications were identified by both reviewers (VH and SB) as meeting the criteria for inclusion: one publication53 included data from two studies, one RCT30 was published as two abstracts30,31 with additional data available from an unpublished poster,32 and one RCT51 was published in four publications51,54–56 (one full-text paper and three abstracts). From here on, each RCT will be referred to by only the primary source of the data included in this review, that is, Franco et al. 30 and Shelton et al. 51 It should also be noted that two trials49,52 included some patients who had failed to respond to only one antidepressant in their current episode of depression and had a historical failure to a second antidepressant in a previous episode of depression. Another RCT43 may also have included such patients, although the numbers are not reported in the paper. All three RCTs were included in the primary analyses, although a post hoc decision was taken to perform a sensitivity analysis to assess the impact of excluding these trials from the analysis. In addition, it should be noted that one additional study57 that was potentially suitable for inclusion was excluded following appraisal of the full-text papers and after discussion with clinical experts because it was deemed to involve unusually high doses of AAP and the AAP used was one that is not licensed or routinely used in the NHS. The RCT was three armed and compared two different doses of the AAP ziprasidone with placebo augmentation of SSRI. The ziprasidone doses used were 80  mg twice daily (b.i.d.) and 160  mg b.i.d., and the trial was titled a ‘pilot study’. In addition, it is noted that there is currently a clinical trial in progress in patients with MDD and failure to respond to an SSRI in their current episode of depression that is assessing the efficacy of augmentation of SSRIs with ziprasidone at doses of 20–80  mg/day compared with augmentation with placebo. This trial is listed on ClinicalTrials.gov as a Phase II trial with an anticipated completion date of March 2013 (ClinicalTrials.gov identifier: NCT00633399). 58

Of the 11 RCTs agreed by both reviewers (VH and SB) as suitable for inclusion, 10 RCTs were for a comparison of SSRI  +  AAP with SSRI  +  placebo/no treatment. 43,46–53 The remaining study30 compared SSRI or SNRI  +  AAP with SSRI or SNRI  +  lithium. Baseline antidepressants in this study were venlafaxine and mixed SSRIs and thus it was agreed among the reviewers (VH, SB and SE) that it should not be included in the primary analysis as the trial author was unable to supply suitable subgroup data. No studies were identified that compared SSRI  +  lithium with SSRI  +  placebo/no treatment.

Owing to the absence of suitable trials for the primary analysis including lithium as a comparator, a pragmatic decision was taken to review all the previously screened full-text papers evaluating lithium (n  =  20), with a view to identifying trials that most closely matched the inclusion criteria. Following this review of previously excluded papers, a single study59 that met all of the inclusion criteria, with the exception of the population criterion, was identified. The trial reported in Katona et al. 59 considered the comparative effectiveness of SSRI  +  lithium with SSRI  +  placebo in patients who had failed one or more antidepressant regimens. Furthermore, in light of the new NICE guideline for depression in adults (CG90; extracts from this guideline have been reproduced here with permission),16 it was considered that this trial would suffice as a proxy for a lithium trial in the required population; CG9016 states that a ‘natural cut-off between people who respond to one or two antidepressants compared with those who do not .  .  . is not supported by the evidence’.

The decision to include this trial in the review was validated by a third reviewer (SJE). In addition, CG9016 and a systematic review of placebo controlled trials of lithium augmentation therapy in TRD60 was used to validate that all of the other potentially relevant SSRI  +  lithium compared with SSRI  +  placebo trials had been identified and excluded appropriately. However, the patient population of the surrogate trial for lithium augmentation could be less treatment resistant than the patients in the trials informing treatment augmentation with AAP. The potential impact of this difference in the trial populations is discussed further in the discussion section (see Assessment of effectiveness, below).

The search of ClinicalTrials.gov identified no clinical trials for the comparison of SSRI  +  AAP with SSRI  +  lithium in the population of interest (i.e. people with TRD and a failure to respond to two or more antidepressants in their current episode of depression) that were completed within the past 12 months or registered as still recruiting patients or ongoing. In addition, clinical experts for this review were not aware of any additional published or unpublished relevant trials.

For a full breakdown of studies included and excluded at each stage of the search and appraisal process, see Figure 1 (PRISMA diagram). For details of the full-text studies excluded and the individual reasons for exclusion, see Appendix 4.

The total number of trials agreed for inclusion in this review was 12 RCTs in 16 publications30,43,46–53,59 (two RCTs are reported in one publication53). CG9016 reported that the DSM-IV tool was used to diagnose depression in most of the evidence reviewed in the guideline and, thus, DSM-IV was the preferred diagnostic measure in the guideline. All but one of the RCTs included in this review required patients to have a diagnosis of MDD based on the DSM-III (one RCT) or DSM-IV (nine RCTs) criteria.

The one RCT43 identified that did not implement DSM criteria to diagnose depression used the Chinese Classification of Mental Disorders, Version 3 (CCMD-3) measure. A sensitivity analysis excluding this study is reported in the results section of the report [see Quality assessment (sensitivity analysis 3)].

For a summary of the characteristics of each of the studies included in this review, see Table 2.

FIGURE 1.

Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow diagram for studies included and excluded from the clinical effectiveness review.

The search of the CCDANCTR-Studies and CCDANCTR-References resulted in the identification of 1487 articles. Initial screening identified three additional potentially relevant papers. All three of these articles represented additional conference abstracts for a study already included following the primary electronic database searches (CENTRAL, EMBASE, MEDLINE and PsycINFO), and the decision was taken not to include them in the results as none of them provided any additional information to that available in the full-text publication.

Quality assessment

All 12 of the included RCTs were assessed for quality using the Cochrane risk-of-bias tool. 27 In the overall assessments for each study, as well as the majority of the assessments for the individual outcomes of interest, all of the trials were rated as unclear risk of bias. This was generally the result of a lack of information being reported in the methods, and that, despite contacting study authors, the additional information could not be obtained. This reason for rating studies as having unclear risk of bias is not unusual as it has been reported elsewhere that unclear risk is likely to be assigned owing to poor reporting of how a trial was conducted rather than a poorly conducted trial. 61 For full details of the risk-of-bias assessments for each study see Appendix 5.

Assessment of effectiveness

The RCTs meeting the inclusion criteria for the primary analyses in the clinical effectiveness review comprise trials comparing SSRI  +  AAP with SSRI  +  placebo/no treatment, and SSRI  +  lithium with SSRI  +  placebo. These trials were used to create a network for the MTC to address the review question regarding comparison of the clinical effectiveness of SSRI  +  AAP with that of SSRI  +  lithium. The individual clinical effectiveness results are presented separately below for each of the following comparisons:

1. SSRI  +  AAP vs. SSRI  +  placebo/no treatment

2. SSRI  +  lithium vs. SSRI  +  placebo

3. SSRI  +  AAP vs. SSRI  +  lithium.

Discussion and summary findings

The available clinical effectiveness data informing the comparison of SSRI  +  AAP with SSRI  +  lithium in the primary analysis were based on fluoxetine  +  olanzapine43,49,51–53 compared with fluoxetine  +  lithium. 59 The results from the MTC of the star-shaped network demonstrate no significant differences between treatment regimens for any of the outcomes assessed. A non-significant trend in benefit was observed for the lithium-based augmentation strategy compared with the olanzapine-based augmentation strategy for response, mean change in MADRS from baseline, and fewer discontinuations. The results of the MTC also demonstrated a non-significant trend in favour of the olanzapine-based augmentation strategy compared with the lithium-based augmentation strategy for fewer adverse events. However, care should be taken when interpreting non-significant results.

When the results of the MTC are compared with the individual results for the pairwise meta-analyses there is general agreement with the results obtained when SSRI is used as the baseline. This suggests a reasonable fit of the modelling approach used within the MTC. The radiating star shape of the network means that only the trials providing the results from the pairwise meta-analyses are providing the results within the MTC. The results for the lithium-based augmentation strategy compared with SSRI alone in the MTC tend to have wider 95% CrIs than the 95% CIs provided from the single trial informing this comparison. 59 This is probably owing to the random-effects model tending to be the preferred model for the MTC outcomes assessed.

The results for both the pairwise meta-analysis and MTC estimates of SSRI  +  AAP compared with SSRI alone showed a statistically significant benefit in favour of augmentation with AAP for the outcomes of response and mean change in MADRS score. The equivalent results for SSRI  +  lithium compared with SSRI alone showed a statistically non-significant trend in favour of augmentation with lithium. The results for lithium augmentation could be considered inconclusive because they do not reach statistical significance, although it should be noted that they are based on data from only a small subgroup of patients in one RCT. 59 In addition, other publications have reported results that suggest lithium is an effective augmentation strategy. 60,65,66 A recent meta-analysis by Crossley et al. 60 included 10 RCTs and demonstrated a statistically significant benefit in terms of response rate with lithium augmentation compared with placebo (OR 3.11; 95% CI 1.80 to 5.37). This meta-analysis is not entirely comparable with the patient population under review here because it included patients with bipolar disorder, patients on various antidepressants (including tricyclic antidepressants), and patients with a minimum of one previous antidepressant failure. However, the Crossley et al. 60 meta-analysis does provide evidence to suggest that lithium is an effective augmentation agent in TRD and thus is supportive of the results from the MTC. In addition, the results from the full trial population for the Katona et al. 59 RCT included in the MTC demonstrate a statistically significant benefit in terms of response with lithium augmentation compared with placebo (OR 3.21; 95% CI 1.09 to 9.48).

Relapse rates and mortality were prespecified as outcomes of interest for this review. However, none of the identified trials reported comparable mortality or relapse rate data. In addition, extremely limited subgroup data, if any, were reported for the trials included in the clinical effectiveness review and no trial reported suitable subgroup data for the prespecified subgroup analyses.

The trials included in this review were validated against the trials included in CG9016 and those included in two separate systematic reviews; one for the comparison of AAP augmentation with placebo in MDD11 and the other for lithium augmentation compared with placebo in TRD. 60 All three publications included additional trials compared with this review, but none of the additional trials was found to be suitable for inclusion. Moreover, each trial was excluded for multiple reasons based on the inclusion criteria for this review. Mostly, the additional RCTs did not meet the following criteria: the augmentation of SSRI as baseline therapy;67–82 a 4-week minimum duration of treatment;74–80,81 or two or more failures of antidepressant therapy in the current episode of depression67–73 (it is unclear how many trials in Crossley et al. 60 are affected by this criterion). The results of both the Crossley et al. 60 meta-analysis evaluating lithium compared with placebo augmentation and the Nelson et al. 11 review of AAP compared with placebo augmentation demonstrated that the respective augmentation agents were statistically significantly more effective than placebo at achieving treatment response in people with MDD or TRD.

Overview of included economic evaluations

The four economic evaluations identified in the systematic literature review were carried out in various countries: Scotland,86 Singapore,87 Thailand88 and the USA. 89 The study by Benedict et al. 86 was from the perspective of the Scottish NHS. Therefore, the costs and resources considered in the evaluation by Benedict et al. 86 were the most relevant to the decision problem that is the focus of this review. However, it is well established that TRD is associated with a substantial societal burden. 16,22 Benedict et al. 86 considered the impact of the societal perspective in sensitivity analysis. The US study by Simpson et al. 89 took a similar approach to that of Benedict et al. 86 (payer perspective as the base case and societal perspective in sensitivity analysis), whereas the studies carried out in Thailand88 and Singapore87 were conducted from a payer and societal perspective, respectively. A summary of the four included economic evaluations is given in Table 7; the quality of each evaluation (individually discussed below; see Narrative review of included studies) was assessed using the Philips checklist90 (the checklist for each study can be found in Appendix 8).

All of the retrieved studies were recent, being published in 2009 or 2010, and are therefore likely to reflect current clinical practice in relation to TRD. Each study included a population of patients who were treatment resistant (either the whole study population or a subgroup); TRD is defined as per the population criteria for this review.

The health states used in each of the identified studies informed the number and type of health states included in the de novo model (see Chapter 5, Model overview and Model structure). However, the methodological approach to the simulation of patients varied across the identified studies. Two of the studies used a decision tree model,87,88 one study used a Markov model86 and the last89 used a hybrid model, combining a decision tree to simulate the acute phase of treatment and a Markov component to simulate the maintenance phase of treatment. The hybrid approach used by Simpson et al. 89 accounted for both the short (duration of acute therapy) and medium (1-year time horizon) term implications of treatment. Therefore, the study by Simpson et al. 89 was considered to be the most relevant to the current decision problem. Furthermore, none of the identified studies considered the longer-term (>  1 year after acute treatment) implications of treatment and the time horizons evaluated ranged from 6 weeks88 to 1 year. 86,89 This reflects the paucity of long-term follow-up data available for TRD patients. 16

Only one study88 evaluated augmentation of antidepressant therapy [augmented with an AAP (aripiprazole)]; the three remaining studies86,87,89 evaluated antidepressant monotherapy regimens compared with either each other or transcranial magnetic stimulation (TMS; Neuronetics, Inc., Malvern, PA, USA). Incremental cost per quality-adjusted life-year (QALY) gained was assessed in all studies except that by Xie et al. ,87 which limited the outcomes assessed to the percentage of primary and secondary care patients achieving remission and the associated costs.

Narrative review of included studies

Introduction

This literature review was carried out to identify utility values associated with depression health states in a patient population with TRD. As in the review of the economic literature, it was expected a priori that, if available, QoL literature on TRD would be limited. In addition, it was considered that the health-related quality of life (HRQoL) associated with TRD was unlikely to be different to that associated with depression in general. Therefore, a decision was made to expand the remit of the search to consider studies of depression in general rather than limiting to studies in TRD. Hence, the primary objectives of this review were to identify health-state utility values (HSUVs) associated with different levels of depression. The secondary objective of this review was to identify any issues that positively or negatively impact upon the QoL in depressed patients.

The review of the QoL literature identified five HSUV studies that addressed the primary objective of the review: to retrieve utility values associated with different levels of depression. A further 12 studies30,43,46–53,59 were identified that addressed the secondary objective: to understand the scope of issues that affect quality of life in the TRD patient population. The sections that follow provide an overview and quality assessment of the HSUV studies identified as addressing the primary objective of this review; the rationale for the utility values used in the economic model; and an overview of the QoL studies identified as addressing the secondary objective of this review. The quality assessment of HSUV studies is based on quality assessment criteria outlined by NICE's decision support unit: Technical Support Document 9 2011 (TSD 9 2011). 101 As recommended by the decision support unit, factors considered when evaluating quality included selection and recruitment of respondents, inclusion and exclusion criteria and whether or not the study included a description of the baseline characteristics of the population from which values were derived. Response rates of the measures used to derive the HSUVs and loss to follow-up were also evaluated.

Systematic literature search and selection process

As stated above, a systematic literature search was carried out to identify utility values associated with depression health states and to identify any issues that positively or negatively impact on the QoL of depressed patients.

Literature search terms and strategies used to identify cost-effectiveness studies relevant to the decision problem were based on validated search strategies developed by Haynes et al. 84 and Dickersin et al. 85 Multiple databases encompassing medical and economic literature were searched to maximise the potential of capturing relevant studies. Searches were carried out in the following databases:

• EMBASE (for the period 1988 to August 2011)

• MEDLINE, including MEDLINE In-Process & Other Non-Indexed Citations (for the period 1950 to August 2011)

• CENTRAL (from inception to August 2011)

• HTA database (from inception to August 2011)

• NHS EED accessed via Wiley Online Library (for the period 1999 to August 2011)

• PsycINFO (from inception to August 2011).

Details of the search strategy are provided in Appendix 7. The search terms used covered condition, population and intervention; no country or language restrictions were applied. All references were exported to the Reference Manager bibliographic database and deduplicated. In addition, the reference lists of identified systematic reviews were hand-searched for additional references. Inclusion and exclusion criteria specific to the QoL literature review were developed (Table 8).

The initial search identified 352 papers, of which 114 were duplicates. One reviewer (LN) carried out the first appraisal of the title and abstract of the 238 studies (level one screening) for potential inclusion. At this stage, 205 papers were excluded.

Further assessment of included abstracts (level two screening) was carried out by a second reviewer (NT) and 13 abstracts were excluded. The full-text publications of 29 studies identified as potentially relevant by both reviewers were ordered (including nine studies identified through bibliographic hand-searching). The papers were independently assessed for final inclusion by two reviewers (LN and NT) using the criteria presented in Table 8: 12 studies were excluded at this stage. Figure 12 summarises the selection process.

FIGURE 12.

Flow diagram of studies in systematic review to identify HSUVs.

Overview of included health-state utility valuation studies

The five HSUV studies identified in the QoL literature review reported utility values for depression severity and/or treatment response. Two studies102,103 reported utility by depression severity, one study104 reported utility associated with treatment response and two studies91,105 considered both depression severity and treatment response. Most studies were carried out in North America (three in Canada91,102,103 and one in the USA;104 the remaining study was carried out in France105). The studies were published between 1998 and 2004 and sample sizes ranged from 58 to 250. The studies each used a different tool to evaluate health-state utility: the tools included the EQ-5D questionnaire, the self-administered Quality of Well-Being Scale (QWB-SA), the SF-36, and the McSad instrument. The mean age of the study participants was <  50 years and more than two-thirds of study participants were women. Table 9 below summarises the five included studies. 91,102–105

Narrative review of included studies (critique of identified quality-of-life studies)

Utility data used in the model

There is ongoing debate around the optimum approach for measuring patient outcomes for use in economic evaluation and decision making. 101 Areas of uncertainty include which health status instrument to use [EQ-5D, Health Utilities Index (HUI) or Short Form questionnaire-6 Dimensions (SF-6D)]; which valuation technique [e.g. TTO, SG or visual analogue scale (VAS)] to apply; and whose preferences (patients, clinicians or the general public) to consider. However, NICE's methods guide108 recommends that:

1. Health status should be reported by the patients experiencing the condition.

2. The values placed on changes in health should come from the UK general population using a choice-based method.

3. EQ-5D is the preferred measure of HRQoL in adults.

In line with these recommendations, one of the HSUV studies identified in the review of the QoL literature, Sapin et al. (2004),105 met the above criteria and was considered as the source of utility for the model. Utility was reported by level of treatment response (remission, response and non-response) and by severity of depression (mild, moderate and severe). The study was carried out in French patients but used UK general public valuation scores. The model assumes that at baseline everyone had the utility of severe depression, as all patients in this model have TRD.

Overview of included quality-of-life studies

The inclusion criteria for this review were broad, and studies examining HRQoL in depression of any severity (not limited to TRD) were considered in addition to HSUV studies. This was to identify any issues that positively or negatively impact on the QoL of depressed patients (these issues were considered likely to be similar in depressed patients, regardless of level of treatment resistance). Twelve studies considering HRQoL were identified that covered a wide range of issues relevant to depression, including the impact on HRQoL of treatment (drug dosage and duration of treatment), physical illness, exercise and depression severity. Ten of the 12 studies were conducted in the USA. Of these 10 studies, four measured health status using the SF-36,109–112 three used the Quality of Life Enjoyment and Satisfaction Questionnaire (Q-LES-Q),67,69,113 one used both SF-36 and Q-LES-Q,114 and one used the World Health Organization Quality of Life Assessment (WHOQOL-BREF). 115 Finally, one study used both the SF-36 and the Late-Life Function and Disability Instrument (LLFDI). 116 A further study using the WHOQOL-BREF was carried out in Brazil117 and a study using the Sickness Impact Profile (SIP) was carried out in the Netherlands. 118

The patient populations examined varied across studies; however, two studies69,110 included patients with TRD as defined in this review (failed on at least two adequate antidepressant treatments in the current episode of depression). Two studies109,117 assessed outcomes in patients with unipolar depression. The remaining studies67,111–118 included patients meeting DSM-IV criteria for MDD. A range of interventions were considered in the identified studies: eight studies69,110,111,113,115–118 considered antidepressant treatment (SSRI/SNRI/tricyclic antidepressant); two studies67,112 considered augmentation of antidepressant therapy with risperidone; one study114 considered the augmentation of antidepressant therapy with physical exercise; and the remaining study109 considered any therapeutic regimen that had been agreed by patients and their psychiatrist. All studies reported an improvement in QoL with treatment.

There is some evidence that the impact of treatment on QoL is sustained and a longer duration of treatment results in further improvement. A 2-year maintenance study by Trivedi et al. 113 assessed the psychosocial outcomes in patients with recurrent MDD. The patients were responders to ER venlafaxine during acute and continuation phase and at 1- and 2-year maintenance periods. The authors found that the longer a patient continues treatment, the higher their QoL. The study found that people who responded to treatment and switched to placebo had worse QoL than those who remained on treatment. The authors also found that patients with greater severity of depression have worse QoL across multiple domains and concluded that improvement in depressive symptoms would be reflected in assessment of functioning.

Another study by Karp et al. 116 assessed the correlation between disability depression severity and QoL in patients with or without depression. The study used the LLFDI, which is defined by the authors as a measure of instrumental activity of daily living, personal role and social role function. The authors found that there was a correlation between disability and depression severity measured on the HAMD-17 scale and that antidepressant treatment improved functional ability. The study also demonstrated that those patients who stayed longer on treatment continued to improve.

In addition, a prospective multicentre observational study by Dunner et al. 110 in people with TRD who were severely depressed and receiving treatment as usual [any therapeutic regimen agreed to by the treating physician and the patient (i.e. drugs or ECT)] found that, after 12 months of treatment, 48% of the patients reported that their QoL had not changed. After 2 years of treatment, the proportion of patients reporting no change in QoL fell slightly to 42%. The proportion of patients reporting an improvement in QoL was 30% and 36% at 12 months and 24 months' follow-up, respectively. Therefore, based on the results from the studies by Trivedi et al. 113 and Dunner et al. 110 it could be inferred that patients should be kept on treatment as long as possible to maximise the HRQoL gains. There is, however, some evidence from relapse prevention studies in depression that indicates a higher rate of relapse in patients who cease therapy following response. 15 It may be that patients in Trivedi et al. 113 and Dunner et al. 110 who ceased therapy or switched to placebo experienced a ‘rebound’ effect that resulted in a lower QoL than in patients who remained on therapy.

About half of the studies reported that the physical domain of QoL for different instruments (SF-36, SIP, Q-LES-Q and WHOQOL-BREF) did not improve with treatment (antidepressants or antidepressant plus risperidone). 109,110,112,113,118 However, one study by Carta et al. 115 assessing physical activity as an adjunctive therapy in women with MDD found that the physical component of the WHOQOL-BREF improved significantly in women who did physical activity compared with women who did not. There were no differences in other domains (relationships, environment and psychological) between those who did physical activities and those who did not. These findings suggest that physical activity may be used as adjunctive therapy to antidepressants to improve the overall QoL of people suffering from MDD.

A study by Small et al. 109 in elderly patients (age >  60 years) with unipolar depression assessed the impact of physical illness on QoL. The indicators for physical illness were number of current chronic and historical illnesses and these indicators were found to influence QoL measures. QoL was measured using the SF-36, and depression levels were the same between patients with physical illness and those who did not have physical illness. The commonly reported chronic illnesses were joint diseases, cardiovascular disease, allergies and gastrointestinal disease. Frequently reported historical illnesses were surgical procedures, gastrointestinal disorder, accidental injuries and cardiovascular disease. The authors concluded that current and previous physical illness is associated with poor QoL, especially physical and mental functioning in older patients with depression. Therefore, based on the findings of the studies by Carta et al. 115 and Small et al. ,109 physical and psychological health may be considered as relatively independent factors of QoL.

In conclusion, there is evidence to suggest that treatments for depression are effective in improving the QoL of patients, particularly if treatment is sustained. In addition, the QoL evidence base indicates that pharmacological treatment of depression does not result in any improvement of physical outcomes, and physical therapies do not result in any improvement of psychological symptoms. This, in turn, suggests that the physical and psychological domains of QoL are independent of each other in determining the overall QoL of a depressed patient.

Acute treatment phase

Patient outcomes following the decision tree component of the model are dictated by the clinical efficacy of and the proportion of patients completing their acute augmentation therapy. In this section, discussion focuses on the:

• acute efficacy associated with each augmentation therapy

• rate of discontinuation associated with each therapy

• likelihood of clinical improvement of symptoms after discontinuation from acute therapy.

Maintenance phase

The maintenance phase of the model includes 6 months of maintenance treatment and 4 months of follow-up (where patients may remain on or cease therapy depending on the level of response achieved). An absence of comparative long-term data for maintenance augmentation treatment led to the decision to assume that the maintenance phase of the model is predominantly treatment independent. Therefore, the parameters used in the Markov component of the model, representing the maintenance phase, are non-drug specific.

Table 12 summarises the probability associated with each possible transition for patients in the Markov model.

Drug costs

As discussed above (see Acute treatment phase), a class effect was assumed for the SSRIs and AAPs. Therefore, the cost of SSRI used in the model is a weighted average of the commonly prescribed SSRIs. The weights are based on clinical opinion rather than prescription cost analysis, as prescription cost analysis encompasses depression of any severity rather than focusing on TRD. The four most commonly prescribed SSRIs are citalopram (20%), escitalopram (20%), fluoxetine (30%) and sertraline (30%). However, clinical experts acknowledge that citalopram use is decreasing because of concerns over side effects of QT interval prolongation, particularly with respect to augmentation. The most commonly prescribed AAPs are aripiprazole, olanzapine, quetiapine and risperidone; experts estimated that the proportion of use for each of these would be 30%, 30%, 20% and 20%, respectively. Analogous to the assumptions made for SSRIs, a weighted average cost was calculated for AAPs. All drug costs were taken from the BNF 63. 19 Table 13 summarises the weighted drug costs used in the model. The maximum licensed dose of each SSRI is used in the cost calculations, as it is assumed that patients eligible for augmentation will be receiving the maximum dose of their current SSRI therapy. The dose assumed in the AAP cost calculation is the usual maintenance dose stated in the BNF. 19 Lithium was assumed to be given in tablet form as the cost of tablets was lower than that of lithium injections (Priadel^®, 800  mg per day). A threshold analysis around the cost of AAPs is carried out as part of the sensitivity analysis (see Sensitivity analysis, below). The cost of fluoxetine was also varied in sensitivity analysis using 3  ×  20-mg tablets, which cost £4 per month, as the base case assumed the use of 1  ×  60-mg fluoxetine tablet, which cost £53 per month.

Health-care professional costs

For the purpose of costing the patient pathway, the analysis assumes that during the acute phase of the disease a proportion of patients will be managed in the community and the remainder will be managed in hospital. For those patients managed in the community, a fraction will be seen by CRHTTs, who provide intensive home-based support, whereas the remainder will receive usual care from their GP and CMHT. Clinical experts estimate that 70% of patients with TRD will receive usual care and 20% will be seen by the CRHHT. The remaining 10% of patients would receive inpatient care. These estimates are in line with those cited in the NICE Depression Guideline,16 as well as the King's Fund Report. 22 Following acute treatment, all patients are assumed to receive care in the community, with the amount of care received depending on the level of response achieved.

Costs associated with each health state

Probabilistic sensitivity analyses

To assess the simultaneous effect of parameter uncertainty on outcomes of cost-effectiveness, the model was built to be fully probabilistic. This was achieved by assigning appropriate distributions to each parameter from which estimated values were repeatedly sampled. Samples of 1000, 2000 and 5000 were used in the base-case model and the stability of the model was assessed with respect to non-linearity between the deterministic and probabilistic results. A sample size of 2000 was chosen, as this gave a more stable probabilistic estimate than a sample size of 1000 and was more efficient than a sample size of 5000. Table 21 displays the distributions assigned to each model parameter.

One-way sensitivity analyses

A range of OWSA was undertaken on key model inputs to investigate the impact the independent variation of each input had on the incremental cost, incremental benefit and ICERs. Each key parameter was alternately assigned a low and high value and the deterministic cost-effectiveness results using this value recorded. The deterministic model was used for the OWSA as a pragmatic measure because the probabilistic model was computationally intensive (average run time ∼10 hours).

In the OWSA, it was assumed that there is no uncertainty with regards to current drug costs in the NHS; therefore, drug costs were not varied in these analyses. The interquartile ranges reported in the National Schedule of Reference Costs were used as the low and high values of HCP costs. However, the variation associated with the cost of monitoring was not reported; therefore, the values used in the OWSA were assumed to be ±  25% of the model's base-case value.

To assess the impact of univariate changes in acute treatment efficacy, 95% CIs were calculated from the means and SEs reported in the MTC (used to inform the normal distribution from which treatment effect is sampled in the base case) for each treatment. Then for each treatment, in turn, the upper and lower thresholds of the CI associated with that treatment were alternatively used to represent the mean change in MADRS score. The mean change in MADRS score was then used to inform the normal distribution from which acute treatment efficacy is sampled. The upper and lower values used in the assessment of the model's sensitivity to acute discontinuation were derived from the 95% CI associated with the log-normal distribution assigned to acute discontinuation (see discussion of PSA).

The uncertainty surrounding the assumption that the probabilities of relapse and of improvement (for patients in response) are independent of treatment status was assessed using assumed variance of ±  25%.

Table 22 summarises the upper and lower values used in each univariate sensitivity analysis. The results of this and other sensitivity analyses are presented below (see Results). In addition, any parameters identified as being significant in the OWSA based on the deterministic model were assessed in the probabilistic model.

Threshold analysis on the cost of atypical antipsychotic drugs

At present, there is significant uncertainty surrounding the future cost of AAPs to the NHS. This is because at the time of writing this report the patents held by Eli Lilly for olanzapine and AstraZeneca for quetiapine were close to expiration. (Note that by the time of publication of this report these patents had expired.) In addition, the patent for aripiprazole held by Otsuka will expire in 2015–16. 128 Therefore, the price of AAPs is likely to decrease as the manufacturers face competition from the generic market. However, there is uncertainty over the extent of the price reduction expected. A threshold analysis on the price of aripiprazole, olanzapine and quetiapine used in the base case was carried out to explore the impact of various price reductions on the base-case model results. The results of this are displayed below (see Results).

Scenario analyses

In the base case, the model assumes a class effect for the SSRIs and the AAPs. This assumption was based on an absence of evidence of a difference of effect, rather than evidence of no difference in effect. Therefore, there is a degree of uncertainty surrounding this assumption. Moreover, data were not available to relax this assumption. However, a scenario analysis was carried out to assess the sensitivity of the model to the use of additional data to inform the class effect of AAP augmentation. In the scenario analysis, data were used from a MTC carried out as part of the clinical effectiveness sensitivity analysis (see Chapter 3, Assessment of effectiveness). The MTC included trials in which lithium, olanzapine, aripiprazole or quetiapine were used to augment a patient population treated with a mixture of SSRI or venlafaxine therapy. Table 23 summarises the acute efficacy and discontinuation derived from the MTC and used in this scenario analysis. The results of this analysis are presented below (see Results).

Base-case results

The probabilistic and deterministic base-case results of the comparison between augmentation of SSRI therapy with lithium and augmentation with an AAP are displayed in Table 24. Deterministic results are reported to facilitate understanding on the OWSA results. The cost-effectiveness plane and cost-effectiveness acceptability curve (CEAC) associated with the probabilistic base case are displayed in Figures 16 and 17, respectively.

FIGURE 16.

Cost-effectiveness plane associated with base-case probabilistic analysis of augmentation of SSRI therapy with lithium vs. augmentation with an AAP.

FIGURE 17.

Cost-effectiveness acceptability curve associated with the probabilistic base case of augmentation of SSRI therapy with lithium vs. augmentation with an AAP.

The base-case results indicate that augmentation of SSRI therapy with an AAP is dominated by augmentation of a SSRI with lithium. The cost-effectiveness plane (see Figure 16) represents the incremental costs and QALYs associated with AAP augmentation (vs. lithium augmentation) as points on the graph. This scatter plot suggests that there is no uncertainty about the dominance result and it is important to note that AAP augmentation did not provide more benefit than lithium augmentation in any of the probabilistic runs.

Conclusion

The economic evaluation presented as part of this review indicated that, in patients with TRD, augmentation of SSRI therapy with an AAP may be dominated by augmentation with lithium. However, the cost-effectiveness results are predominantly driven by estimates of acute efficacy and discontinuation derived from the assessment of clinical effectiveness. The results of the clinical effectiveness analyses indicate that there is no statistically significant difference between the two augmentation strategies. In addition, the general paucity of RCTs available for augmentation therapy using SSRI as background treatment in patients with TRD results in a high level of uncertainty in both the clinical and economic analyses.

Description of the identified clinical studies

Initially, 11 RCTs30,43,46–53 were identified that considered the augmentation of SSRI therapy in the relevant population (patients who had failed to respond to two or more antidepressant regimens). However, 10 of these RCTs considered the augmentation of SSRI therapy with an AAP. 43,46–53 The remaining RCT30 considered the augmentation of SSRI or SNRI (venlafaxine) therapy with lithium compared with AAP. None of the identified studies considered augmentation of SSRI therapy with lithium. The primary analysis did not assume a class effect for either the SSRIs or AAPs. Therefore, the single trial identified that considered lithium was ineligible for inclusion in the primary analysis. Consequently, to enable a primary analysis of the relative treatment effect of augmentation with lithium compared with augmentation with an AAP, a surrogate lithium trial was included. The patient population of this trial had failed on one or more antidepressant therapy.

Summary of clinical findings

Primary and secondary analyses of the relative clinical effectiveness were carried out based on the RCTs identified as part of the clinical effectiveness review for the following comparisons:

• SSRI  +  AAP vs. SSRI alone

• SSRI  +  lithium vs. SSRI alone

• SSRI  +  AAP vs. SSRI  +  lithium.

The primary analyses assumed that the SSRIs and AAPs were associated with different efficacy profiles, whereas the secondary analyses assumed a class effect for both classes of drug.

Discussion of clinical findings

The available clinical effectiveness data informing the comparison of SSRI  +  AAP with SSRI  +  lithium in the primary analysis was based on fluoxetine  +  olanzapine43,49,51–53 compared with fluoxetine  +  lithium. 59 The results from the MTC of the star-shaped network demonstrate no significant differences between treatment regimens for any of the outcomes assessed. A non-significant trend in benefit was observed for the lithium-based augmentation strategy compared with fluoxetine  +  olanzapine for response, mean change in MADRS from baseline, and fewer discontinuations. The results of the MTC also demonstrated a non-significant trend in favour of the olanzapine-based augmentation strategy compared with fluoxetine  +  lithium for fewer adverse events. However, care should be taken when interpreting non-significant results.

When the results of the MTC are compared with the individual results for the pairwise meta-analyses, there is general agreement with the results obtained when SSRI is used as the baseline. This suggests a reasonable fit of the modelling approach used within the MTC. The radiating star shape of the network means that only the trials providing the results from the pair-wise meta-analyses are providing the results within the MTC. The results for the lithium-based augmentation strategy compared with SSRI alone in the MTC tend to have wider 95% CrIs than the 95% CIs provided from the single trial informing this comparison. 59 This is likely to be due to the random-effects model tending to be the preferred model for the MTC outcomes assessed.

The results for both the pair-wise meta-analysis and MTC estimates of SSRI  +  AAP compared with SSRI alone showed a statistically significant benefit in favour of augmentation with AAP for the outcomes of response and mean change in MADRS score. The equivalent results for SSRI  +  lithium compared with SSRI alone showed a statistically non-significant trend in favour of augmentation with lithium. The results for lithium augmentation could be considered inconclusive, although it should be noted that they are based on data from only a small subgroup of patients in one RCT59 and other publications have reported results demonstrating lithium to be an effective augmentation strategy. 60,65,66 A recent meta-analysis by Crossley et al. 60 included 10 RCTs and demonstrated a statistically significant benefit in terms of response rate with lithium augmentation compared with placebo (OR 3.11; 95% CI 1.80 to 5.37). This meta-analysis is not entirely comparable with the patient population under review in the current research because it included patients with bipolar disorder, patients on various antidepressants and patients with a minimum of one previous antidepressant failure. However, the Crossley et al. 60 meta-analysis does provide strong evidence to suggest that lithium is an effective augmentation agent in TRD and thus is supportive of the results from the MTC.

Relapse rates and mortality were prespecified outcomes of interest for this review; however, none of the trials included reported comparable mortality or relapse rate data. In addition, extremely limited subgroup data, if any, were reported for the trials included in the clinical effectiveness review, and no trial reported suitable subgroup data for the prespecified subgroup analyses.

The trials included in this review were validated against the trials included in the NICE clinical guideline for depression in adults (CG90)16 and those included in two separate systematic reviews; one for the comparison of AAP augmentation with placebo in MDD11 and the other for lithium augmentation compared with placebo in TRD. 60 All three publications included additional trials compared with this review, although none of the additional trials were found to be suitable for inclusion. Moreover, each trial was excluded for multiple reasons based on the inclusion criteria for this review. Mostly, the additional RCTs did not meet the following criteria: the augmentation of SSRI as baseline therapy;67–81 a 4-week minimum duration of treatment;74–80 or two or more failures of antidepressant therapy in the current episode of depression67–73 (it is unclear how many trials in Crossley et al. 60 are affected by this criterion). The results of both the Crossley et al. 60 meta-analysis evaluating lithium compared with placebo augmentation and the Nelson et al. 11 review of AAP compared with placebo augmentation demonstrated that the respective augmentation agents were statistically significantly more effective than placebo at achieving treatment response in people with MDD or TRD.

The sensitivity analysis based on a class effect (sensitivity analysis 1) allowed the inclusion of two additional AAPs (aripiprazole and quetiapine) in the MTC. For the outcomes assessed (response) the non-significant trend favouring lithium augmentation observed in the primary analysis is diminished (mean OR changes from 4.15 to 1.07) and remains non-significant. This could be supportive evidence for no clinically meaningful difference between the two augmentation strategies, a reflection of the lack of information available for the comparison, or an indication that the assumption of a class effect is flawed and that there is a difference between the individual treatments within a class. With regards to the last concern, the trial providing information on quetiapine in the MTC30 demonstrated a non-significant trend in favour of a quetiapine-based augmentation strategy compared with lithium-based augmentation (OR 1.25; 95% CI 0.74 to 2.12). The result of this trial is thus in contrast with the result of the MTC and therefore suggests a difference in treatment effect between the different AAPs.

Summary of the cost-effectiveness review

No economic evaluations were identified that considered the relative cost-effectiveness of augmentation of SSRI therapy with lithium or with an AAP. However, four economic evaluations were identified that included a population of people with depression who were treatment resistant. Each study highlighted issues surrounding the economic evaluation of TRD, particularly around the long-term consequences of treatment and the paucity of data available in this area. A variety of approaches were taken to assess the cost-effectiveness of various interventions (mostly SSRI therapy). The hybrid approach used by Simpson et al. 89 was considered to be most relevant to the current decision problem and was used to inform the structure of the de novo economic model.

Summary of quality-of-life review

A systematic review of the QoL literature was carried out to identify:

• HSUV to inform the economic model

• any issues that positively or negatively impact on the QoL of depressed patients.

Five studies reporting utility values91,102–105 by severity of depression or level of response were identified. Of these, only one study by Sapin et al. 105 met the criteria for elicitation and valuation outlined in the NICE methods guide. 108 The study was carried out in France but used UK general public valuation scores. Sapin et al. 105 reported utility by level of treatment response (remission, response and non-response) and by severity of depression (mild, moderate and severe). Therefore, it was these utility values that were used to inform the QALY calculations in the de novo economic model.

A further 12 studies were identified that considered the overall QoL in depressed patients. 30,43,46–53,59 There is evidence from these studies to suggest that treatments for depression are effective in improving the QoL of patients, particularly if treatment is sustained. In addition, the QoL evidence base indicates that pharmacological treatment of depression does not result in any improvement of physical outcomes, and physical therapies do not result in any improvement of psychological symptoms. This, in turn, suggests that the physical and psychological domains of QoL are independent of each other in determining the overall QoL of a depressed patient.

Summary of economic evaluation

A hybrid economic model was constructed to simulate the clinical and economic consequences of augmenting an SSRI with either lithium or an AAP in the treatment of TRD. The model considered outcomes from the perspective of the NHS over a 1-year time horizon. The model consists of two distinct components: a decision tree (8-week time horizon) and a Markov component (10-month time horizon). A hybrid model was chosen, as this facilitated capturing the granularity of the acute treatment phase and also accounted for the patient progression within 1 year.

The model was constructed around the level of response to acute treatment (assessed by changes in MADRS score) and the extent to which this response (or otherwise) is maintained. The acute efficacy and discontinuation associated with each treatment was obtained from a MTC carried out as part of the clinical effectiveness review. Discontinuation was a key driver of the model, as patients who discontinued their acute therapy were far less likely to experience response or remission.

Furthermore, as a result of the paucity of long-term data, progression within the Markov component of the model was assumed to be independent of treatment regimen. The progression of patients who did not respond to acute treatment (and patients who relapsed during the Markov component of the model) was simplified such that these patients were assumed not to benefit from further augmentation treatment and received a standard package of care. It is acknowledged that, in reality, patients who do not respond to (or relapse following) acute augmentation therapy are likely to experience some benefit from further treatment (e.g. change in augmentation treatment, ECT, etc.). However, these follow-on patient pathways are diverse and there is a paucity of data to inform them, and, therefore, the decision was made to simplify this aspect of patient care. In addition, while non-responding patients were exposed to the probability of ‘spontaneous’ remission or response, in accordance with expert clinical opinion, patients who relapsed were assumed to enter an absorbing health state.

Background

Depression is a common mental disorder affecting about 121 million people worldwide and is among the leading causes of disability. 1 People presenting with depression may complain of depressed mood, loss of interest or pleasure, feelings of guilt or low self-worth, disturbed sleep or appetite, low energy and poor concentration.

Depression can be diagnosed clinically using different criteria, the most commonly used of which are the DSM IV criteria as published by the American Psychiatric Association and the ICD 10 criteria developed by the World Health Organisation. 3^,4

Up to two thirds of patients with major depression will either not respond to or will have a sub-optimal response to first-line treatment with antidepressants (i.e. they may respond but not enter remission which is the relative absence of clinical symptomatology). There are several potential pharmacological treatment options for patients not achieving sufficient response with antidepressants, one of which is to augment the antidepressant with an agent not approved for use as monotherapy in major depressive disorder. 5

Current NICE guidance2 for the sequencing of treatments in depression after an inadequate response to at least one antidepressant recommends that people who are informed about and prepared to tolerate the increased side-effect burden, should be considered for treatment with the combination or augmentation of an antidepressant with lithium or an antipsychotic such as aripiprazole, olanzapine, quetiapine or risperidone or another antidepressant such as mirtazapine or mianserin.

Objective

This report aims to determine the clinical effectiveness and cost-effectiveness of SSRI antidepressant therapy with either lithium or an atypical antipsychotic in the management of people with treatment resistant unipolar depression.

For this review, treatment resistant depression will be defined as failure to respond to at least two previous antidepressant medications. We will not impose restrictions on the maximum number of previous antidepressant drugs allowed so as not to reduce the amount of data available for analysis as we aware that there will be limited relevant SSRI RCT data available. This assumes that there is a consistent relative treatment effect independent of line of therapy, i.e. addition of an atypical or lithium has the same relative benefit whether given with third-line SSRI or fourth line SSRI, etc. However, a sensitivity analysis will be conducted to assess the impact of this assumption.

PICO criteria

The planned PICO is as follows:

• Population: Adults with treatment resistant unipolar depression defined as failure to respond to at least two previous antidepressants in the current episode of depression only.

• Intervention:

  • An SSRI (selective serotonin reuptake inhibitor) (defined as either Citalopram (Cipramil), Escitalopram (Cipralex), Fluoxetine (Prozac, Felicium, Prozep, Prozit), Fluvoxamine (Faverin), Paroxetine (Seroxat) or Sertraline (Lustral)), PLUS

  • An atypical antipsychotic drug (defined as either Amisulpride (Solian), Aripiprazole (Abilify), Clozapine (Clozaril, Denzapine, Zaponex), Olanzapine (Zyprexa, Zypadhera), Paliperidone (Invega), Quetiapine (Seroquel), Risperidone (Risperdal) or Ziprasidone (Geodon))

• Comparator:

  • An SSRI (defined as either Citalopram (Cipramil), Escitalopram (Cipralex), Fluoxetine (Prozac, Felicium, Prozep, Prozit), Fluvoxamine (Faverin), Paroxetine (Seroxat) or Sertraline (Lustral)) PLUS

  • Lithium (Lithium carbonate (Camcolit, Liskonum, Priadel) or Lithium citrate (Li-Liquid, Priadel) or Lithium (Litarex, Lithonate, Phasal))

• Outcomes:

  • Disease severity

  • Quality of life

  • Adverse effects

  • Withdrawals (all cause) as a surrogate outcome for adherence to medication

  • Relapse rate

  • Mortality

  • Cost-effectiveness

Subgroup analyses

The planned subgroup analyses are as follows:

• Different durations of depression (i.e. time since first onset of current episode of depression)

• Classes of previous antidepressants (e.g. SSRI or tricyclic antidepressant)

• Sex (i.e. males and females)

• Age (i.e. those <  75 years and those ≥  75 years old)

• People with different severities of depression (i.e. based on trial entry Hamilton Depression Rating Scale rating)

Objectives

The key areas that we plan to address in this report are:

• To identify and review the existing evidence relating to the clinical outcomes as pre-specified above

• To report the cost-effectiveness of these treatments

• To identify what the potential areas for future research might be

Search strategy

The search strategy will comprise the following main elements:

1. Searching of electronic bibliographic databases

2. Contact with clinical experts in the field

3. Review of the reference lists of retrieved papers

1. The electronic databases that will be searched are EMBASE, MEDLINE, PsycINFO and the Cochrane Controlled Trials Register.

We will also search the ClinicalTrials.gov website to identify relevant ongoing clinical trials that when completed may have an impact on the results of this review, to assist us in drawing up our final recommendations.

2. We will contact clinical experts in the relevant therapy areas to request details of trials (published and unpublished) of which they may be aware. We will allow the experts 1 calendar month to provide an initial response, with any additional time allowed being dependent on whether we have reached the data analysis stage of the review.

3. The references from any relevant review papers or randomised controlled trials (RCTs) uncovered in the search will also be examined for additional references potentially relevant to the review.

Abstract appraisal

Titles and abstracts of studies identified by the search process will be assessed independently by two reviewers (VH and SB) for inclusion. In cases where the reviewers are unable to reach a consensus as to whether the full text should be obtained for further appraisal, the full text will be obtained.

When potentially relevant data are available in only an abstract format then we will attempt to contact the corresponding author in order to obtain the full publication; however, there will be a pre-specified deadline of 1 calendar month by which they will need to have contacted us, but we may allow additional time for them to supply the data requested depending on where we are in the review process. Any information supplied after the deadline will be included in only the discussion section of the review report.

Inclusion criteria

• For the review of clinical effectiveness, only RCTs will be included

• Adults ≥  18 years

• People with unipolar depression only

• Treatment resistant depression defined as failure to respond to at least two previous antidepressants in the current episode of depression only

• SSRI (selective serotonin reuptake inhibitor) given as baseline treatment and patient randomised to either lithium or an atypical antipsychotic

• Minimum duration of 4 weeks treatment with study medication for the current episode of depression

• Studies reporting on one or more of the following outcomes:

  • Disease severity

  • Quality of life

  • Adverse effects

  • Adherence to medication or withdrawals (all cause)

  • Relapse rate

  • Mortality

  • Cost-effectiveness

Exclusion criteria

• Non-randomised studies

• Narrative reviews, editorials, opinions

• Studies performed in animals

• Studies not focusing on the treatment of the acute phase of depression (i.e. those only focusing solely on maintenance therapy)

• Bipolar depression or bipolar disorder diagnosis prior to study entry

• Underlying medical condition or another substantial co-morbid psychiatric condition

• Trials reporting only post-crossover results

Study inclusion assessment

Two reviewers (VH and SB) will independently assess for inclusion the full text of the trials identified during the abstract assessment stage and any differences in opinion will be arbitrated by a third reviewer (SJE).

Data extraction strategy

Data will be extracted by one reviewer (VH) using a standardised data extraction form (for draft copy of data collection form, please see appendix 10.2) and validated by second reviewer (SB).

A pragmatic decision for data validation will be made depending on the number of trials identified owing to the time constraints for completing this review. If a large number of trials are identified then all data will be validated (checked) by a second reviewer, with a sample being fully independently data extracted. This sample will be 25% or a minimum of 5 papers (whichever is larger).

The Data Extraction Form will be pilot tested on a sample of three papers by the reviewers and a final version agreed.

Discrepancies in the data extracted by the two reviewers will be resolved through discussion, with involvement of a third reviewer (SJE) if necessary.

Data from intention-to-treat (ITT) analyses will be extracted (per protocol (PP) data will also be extracted for use in a sensitivity analysis). Should a trial not report ITT data, we will treat missing data as treatment failures to allow our analysis to conform to an ITT analysis. For the purpose of this review, ITT will be defined as patients being analysed in the treatment group they were allocated to at randomisation regardless of whether they received the wrong intervention, withdrew or were lost to follow-up.

Study authors will be contacted to supply any additional information not included in published sources (including pre-crossover results in those trials reporting only post-crossover results) and there will be a pre-specified deadline by which we would require a response. The deadline will be 1 calendar month from the date of sending the request by which time they must have contacted us with at least an initial response acknowledging their intent to supply some of the information required. We may allow additional time for them to supply the data requested depending on where we are in the review process, however any information received after the deadline will be included in only the discussion section of the review.

Quality assessment strategy

Outcomes from the studies that meet the inclusion criteria will be assessed using the updated risk of bias tool developed by the Cochrane Collaboration (March 2011). 7

These criteria assess the following areas:

1. Random sequence generation

2. Allocation concealment

3. Blinding of participants and personnel

4. Blinding of outcomes assessment

5. Incomplete outcome data

6. Selective reporting

7. ‘Other bias’

Based on these criteria, an assessment for each outcome reported in the trial will be allocated based on the identified risk of bias. The three bias assessment categories used will be: low risk, high risk and unclear risk. Unclear risk is likely to be assigned owing to poor reporting of how the trial was conducted rather than a poorly conducted trial. 8 Trials that are deemed to be at low or unclear risk of bias will be included in the main analysis; however, the trials rated high risk will be included in a sensitivity analysis.

Two reviewers (VH and SB) will independently rate the trial outcomes for inclusion and any differences in opinion will be arbitrated by a third reviewer (SJE). An outcome from an RCT will be considered appropriate for inclusion unless the trial demonstrates some feature that necessitates the exclusion of that outcome.

Methods of analysis/synthesis

Data will be tabulated and, where appropriate, meta-analysis will be employed to estimate a summary measure of effect on relevant outcomes based on ITT analyses (with a sensitivity analysis based on per protocol data).

We will not be assuming there is a class effect for any of the drugs included and so each individual drug will be considered separately in the review, i.e. each SSRI and atypical antipsychotic or lithium combination will form separate analyses.

Standard pairwise meta-analysis will be conducted when more than one trial is identified for inclusion for any pair of treatments under investigation. This will be carried out using a fixed effects model with the Mantel-Haenszel method. 9 Sensitivity analysis will be conducted using a random effects model with the DerSimonian & Laird method. 10

It is anticipated that a mixed treatment comparison (MTC; also called a multiple treatment meta-analysis and network meta-analysis) will need to be conducted to estimate the effects of the different treatments included in the research. A MTC can be seen as an extension of traditional pairwise meta-analysis. 11^–13 It has advantages over standard pairwise meta-analysis as it is based on a network of connected trials where a new trial may enter the network if it is in a clinically comparable patient population, has a similar design to other trials incorporated in the network and contains at least one treatment that already exists within the network. It has been argued that this underlying assumption of exchangeability of data is no different from the practice within standard pairwise meta-analysis of combining similar trials.

The MTC will be conducted based on a fixed effects and a random effects model with the most appropriate model identified as the one with the lowest deviance information criterion (DIC). 14 DIC measures the fit of the model while penalising for the number of effective parameters. 12^,15 For the chosen model, consistency of the evidence will be assessed using the posterior mean residual deviance, which should approximate the number of unconstrained data points in a good-fitting model.

For dichotomous outcomes we will use odds ratio as the summary statistic, and for continuous outcomes we will use the weighted mean difference as the summary statistic.

• Primary analysis will be:

  • Disease severity (measured by a reduction of at least 50% on Hamilton Depression Rating Scale (HDRS)16 or Montgomery-Åsberg Depression Rating Scale (MADRS). 17 Where a study reports both we will use only the HDRS data).

• Secondary analyses will be:

  • Quality of life (QoL) as reported using a validated QoL rating scale18, e.g. EQ-5D, SF-36, HUI.

  • Adverse effects (data will be collected on those adverse effects most burdensome to patients such as agitation, akathisia, anxiety, cognitive dulling, constipation, diarrhoea, dry mouth, dyspepsia, extrapyramidal symptoms, kidney and thyroid dysfunction, lipid disturbance, gastrointestinal bleeding, (orthostatic) headache, hyperglycaemia, hypotension, nausea, polyuria, restlessness, sedation, sexual dysfunction, sleep disturbance, thirst, tremor, visual problems, and weight gain).

  • Withdrawals (all cause) as a surrogate outcome for adherence to medication

  • Relapse rate

  • Mortality (all cause)

• 8-week outcome data will be collected where reported. If 8-week data are not available, we will use outcome data reported from the nearest available time point

• Subgroup analyses will be performed in the following populations on only the primary outcome (disease severity), subject to the availability of data:

  • Different durations of depression (i.e. time since first onset of current episode of depression, short term <  6 months, long term >  6 months)

  • Classes of previous antidepressants (e.g. SSRI or tricyclic antidepressant)

  • Sex (i.e. males and females)

  • Age (i.e. those ≥  75 years and those <  75 years old)

  • People with different severity's of depression, i.e. based on trial entry HDRS rating using the following categories: 16

    • 0  –  7  =  Normal

    • 8  –  13  =  Mild Depression

    • 14  –  18  =  Moderate Depression

    • 19  –  22  =  Severe Depression

    • ≥  23  =  Very Severe Depression

In the absence of suitable data to perform a meta-analysis, the available data will be tabulated where possible and discussed in a narrative review.

Heterogeneity

In addition to the existing pre-specified subgroups, other potential sources of clinical heterogeneity could be a result of combining different preparations of drugs.

For pairwise meta-analysis, heterogeneity will be explored through consideration of the study populations, methods and interventions, by visualisation of results and, in statistical terms, by the χ² test for homogeneity and the I² statistic. Statistically significant heterogeneity will be defined as p  <  0.10. Levels of inconsistency will be assessed using I² and will be defined as follows: I² of: 0%–25%  =  low level of inconsistency; 26%–50%  =  moderate level of inconsistency; and >  50%  =  high level of inconsistency. 19

If statistically significant heterogeneity is detected in any of the primary or secondary analyses, hypothesis-generating subgroup analysis will be conducted, but the results from such analyses will be treated with caution. Meta-regression will be attempted if significant statistical heterogeneity is identified among trials analysed and there are 10 or more trials in the review.

For the MTC, where a random effects model is deemed the best fit, the degree of heterogeneity will be investigated by evaluating the posterior mean tau-squared. Where possible, any closed loops formed by the network of trials will be assessed separately to determine if the results from the ‘direct’ evidence is coherent with the ‘indirect’ evidence when the wider network is introduced. Any incoherence identified will be investigated.

Sensitivity analysis

Sensitivity analyses are planned on the primary analysis and consist of:

• different number of prior antidepressants for the current episode of depression;

• assuming a ‘class’ effect with SSRIs and atypical antipsychotics;

• changing the quality assessment to include the trial outcomes excluded on grounds of methodological quality; i.e. those categorised as of high risk of bias;

• changing the analysis from using ITT (intention to treat) data to per protocol data.

Publication bias

For each of the primary pairwise meta-analyses, a funnel plot will be used to assess publication bias. A regression of normalised effect compared with precision will also be calculated as a test for small study effects (using a p  <  0.10 as an indicator of a significant result). 20

Identifying and systematically reviewing published cost-effectiveness studies

The following databases will be used to identify studies of the cost-effectiveness. MEDLINE, EMBASE, PsycINFO, CINAHL NHS Economic Evaluation Database, Health Technology Assessment Database and Office of Health Economics Health Economic evaluation database. We will apply a cost search filter to the comprehensive clinical search strategy described in Section 5.

In order to express clinical outcomes in the form of QALYs, utility weights for health states relating to treatment resistant depression are required. Utility weights represent the health related quality of life (HRQOL) associated with specific health states; they are estimated based on people's preferences and perceptions of quality of life characterising the health states under consideration. We will undertake a systematic quality of life search where health economics and quality-of-life search filters will be used in MEDLINE, EMBASE, PsycINFO and CINAHL to identify relevant studies.

The inclusion and exclusion criteria for economic evaluations will be the same as those for the systematic review of clinical effectiveness and in addition the health economic evaluation will also include:

• non-randomised studies will be included (e.g. decision-model based analysis or analysis of person-level cost and effectiveness data alongside observational studies)

• full cost-effectiveness analyses, cost–utility analyses, cost–benefit analyses and cost–consequence analyses will be included

• stand-alone UK cost analysis will also be sought and appraised

Titles and abstracts returned by the search strategy will be assessed independently by two health economists (LN and NT) and screened for possible inclusion. Any disagreements will be resolved by a third health economist (SJE).

Evaluation of costs and cost-effectiveness (may include development of a de novo economic model)

The methodological quality of economic evaluations will be assessed according to internationally accepted criteria such as the Consensus on Health Economic Criteria list questions developed by Evers et al. (2005). 21 Any studies based on decision models will be assessed using the checklist developed by Phillips et al. (2004). 18

In addition, a new economic evaluation will be carried out from the perspective of the UK NHS using a probabilistic decision-analytic (Markov) modelling approach to estimate the costs and QALYs of SSRI with an atypical antipsychotic compared with SSRI with lithium in the management of treatment resistant unipolar depression. An annual discount rate of 3.5% will be used for both costs and QALYs in accordance with NICE guidance. 22 Model structure, data inputs and modelling assumptions will be determined in consultation with clinical experts to ensure they reflect the best current clinical practice and evidence. Uncertainty in the data used to populate the model will be characterised using appropriate methods, such as probabilistic sensitivity analysis. The time horizon of our analysis will preferably be a patient's lifetime in order to reflect the chronic nature of the disease. However time horizon may be dictated by the availability of data in which case shorter time horizons will be modelled.

Ideally, evidence on the impact of these therapies on HRQoL will be available directly from the trials included within the review. In the absence of such evidence, the mathematical model may use indirect evidence on quality of life from alternative sources, such as related technology appraisals or clinical guidelines. Quality of life data will be reviewed and used to generate the quality adjustment weights required for the model. We will also adjust utility for age using data from the Health Survey of England. 23

Results will be presented as incremental cost-effectiveness ratios (ideally cost per quality adjusted life year) and cost-effectiveness acceptability curves, which quantify the degree of uncertainty.

TAR Centre

The BMJ Evidence Centre comprises over 30 specialists with a wealth of experience in diverse health-related areas and includes clinicians, pharmacists, information specialists, health informatics specialists, project managers, systematic reviewers, clinical guideline developers and health economists.

The BMJ-TAG core team consists of 5 members. Together, we have an array of experience amongst us in producing focussed reports in a short timescale for policy customers such as NICE. Please see below for further details of each team member's experience.

• Dr Steven J Edwards DPhil MSc BSc (Hons), Head of Health Technology Assessment: Over the past 12 years, Steve has conducted over 40 systematic reviews and health economic evaluations in a range of therapeutic areas including cardiovascular, CNS, gastroenterology, infection, oncology and respiratory medicine. His interests are in the use of the best available evidence for decision making with an emphasis on the design and conduct of clinical trials, systematic reviews, meta-analyses, adjusted indirect comparisons and their subsequent use in economic evaluations. His postgraduate research in this area at the University of Oxford resulted in him being awarded the first doctorate of evidence based health care. In addition, Steve is an honorary senior lecturer in health economics at the London School of Hygiene & Tropical Medicine, a member of the Cochrane Statistical Methods Group, the Campbell & Cochrane Economics Methods Group, and an Editorial Board member of the International Journal of Clinical Practice.

• Dr Samantha Barton PhD BSc (Hons), Health Technology Assessment Analyst: Sam has extensive experience in the critical appraisal of studies. During the past 4 years, she has contributed to the publication of over 50 systematic reviews on prevention and treatment of various clinical conditions. She has worked on reviews in the areas of mental health, sexual health, infectious diseases, cardiovascular disorders, respiratory disorders and oncology.

• Dr Victoria Hamilton MBChB, Health Technology Assessment Analyst: Vicky has a clinical background with relevant experience in the fields of general surgery, general medicine, general practice, paediatrics and orthopaedic surgery. Vicky also has experience in the critical appraisal of clinical studies and over the last year has contributed to the publication of systematic reviews in a variety of clinical areas. She also has experience in the process and use of clinical audit to review current clinical practice within both primary and secondary care settings.

• Mr Leo Nherera MSc BSc (Hons), Health Economist: Over the past 6 years, Leo has been working for the NICE clinical guideline programme and has successfully worked in eight published clinical guidelines and one Public Health guideline. His work involved appraising economic evaluations as well as doing original economic analysis for various guideline questions to assist in guideline recommendations. Leo was involved in organising and teaching the Health Economics module at Queen Mary University of London. He has also peer-reviewed papers for the International Journal of Clinical Practice. His interests are in the use of the best available evidence for decision making with an emphasis on systematic reviews and meta-analyses and their subsequent use in economic evaluations.

• Ms Nicola Trevor MSc BSc (Hons), Health Economist: Nicola has a strong mathematical background, with a Masters in analytical, numerical and statistical modelling techniques, which over the past 2 years she has applied in the field of health economics, conducting economic evaluations and statistical analysis for systematic review in disease areas such as multiple sclerosis, cardiovascular disease, Gaucher's disease and oncology. Her interests are in the use of the best available techniques for decision making with an emphasis on survival analysis, meta-analysis, modelling approaches and the use of Bayesian methods in economic evaluations.

Recent publications from the team members include:

Nherera L, Marks D, Minhas R, et al. Probabilistic cost-effectiveness analysis of cascade screening for Familial Hypercholesterolaemia using alternative diagnostic and identification strategies. Heart 2011;97:1175–81.

National Collaborating Centre for Women's and Children Health. Multiple pregnancy: The management of twin and triplet pregnancy in the antenatal period London, Royal College of Obstetricians and Gynaecologists, 2011; (in press).

Edwards SJ, Wordsworth S, Clarke MJ. Treating pneumonia in critical care in the UK following failure of initial antibiotic: a cost-effectiveness analysis comparing meropenem with piperacillin/tazobactam. European Journal of Health Economics 2011;12: (available online first at: www.springerlink.com/content/q044j5t32601vt4l/).

Halpin DMG, Gray J, Edwards SJ, et al. Budesonide/formoterol versus salmeterol/fluticasone in COPD: a systematic review and adjusted indirect comparison of pneumonia in randomized controlled trials. International Journal of Clinical Practice 2011;65:764–74.

Edwards SJ, Borrill J. Network meta-analysis: importance of appropriate trial selection. Value in Health 2010; 13:681–2.

Edwards SJ, von Maltzahn R, Naya IP, et al. Budesonide/formoterol for maintenance and reliever therapy: a meta-analysis of randomised controlled trials. International Journal of Clinical Practice 2010; 64:619–27.

Gray J, Edwards SJ, Lip GYH. Comparison of sequential rosuvastatin doses in hypercholesterolaemia: a meta-analysis of randomized controlled trials. Current Medical Research and Opinion 2010;26:537–47.

Trevor NC, Alnwick K. How can the use of predictive biomarkers lead to positive HTA recommendations? Value in Health 2010;13:A423–4.

Trevor NC, Tang M, Samuels ER. Investigating the impact of R&D investment and policy on innovative performance in Europe. Value in Health 2010;13:A414.

Edwards SJ, Gray J. Budesonide/formoterol plus tiotropium (BUD/FORM+TIO) vs salmeterol/fluticasone plus tiotropium (SALM/FLU+TIO): a systematic review and adjusted indirect comparison between two alternative triple treatments in chronic obstructive pulmonary disease (COPD). Value in Health 2010;13:A319.

Edwards SJ, Welton NJ, Borrill J. Gefitinib compared with doublet chemotherapy for first-line treatment non-small-cell lung cancer (NSCLC) a systematic review and adjusted indirect comparison. Value in Health 2010;13:A252–3.

Edwards SJ, Welton NJ, Borrill J. Tolerability of first-line treatments of locally advanced or metastatic non-small-cell lung cancer (NSCLC) a systematic review and adjusted indirect comparison. Value in Health 2010;13:A250.

Nherera L, Calvert NW, DeMott K, et al. Cost effectiveness analysis of the use of a high intensity statin compared to a low intensity statin in the management of patients with familial hypercholesterolaemia. Current Medical Research and Opinion 2010;26:529–36.

Visintin C, Mugglestone MA, Almerie MQ, et al. on behalf of the Guideline Development Group. Management of hypertensive disorders during pregnancy: summary of NICE guidance. British Medical Journal 2010;341:c2207.

National Collaborating Centre for Women's and Children Health. Hypertension in pregnancy: The management of hypertensive disorders during pregnancy. London, Royal College of Obstetricians and Gynaecologists 2010.

External Clinical Expert Advisors

Professor Philip J. Cowen – MRC Clinical Scientist and Professor of Psychopharmacology; Specialist in Psychopharmacology of Mood Disorders

Neurosciences Building,

Warneford Hospital,

Oxford OX3 7JX,

United Kingdom

phil.cowen@psych.ox.ac.uk

Recent publications include:

• McCabe C, Mishor Z, Cowen PJ, et al. Diminished neural processing of aversive and rewarding stimuli during selective serotonin reuptake inhibitor treatment. Biological Psychiatry 2010;67:439–45.

• Harmer CJ, O'Sullivan U, Favaron E, et al. Effect of acute antidepressant administration on negative affective bias in depressed patients. American Journal of Psychiatry 2009;166:1178–84.

• Harmer CJ, Goodwin GM, Cowen PJ. Why do antidepressants take so long to work? A cognitive neuropsychological model of antidepressant drug action. British Journal of Psychiatry 2008;195:102–8.

• Gelder M, Cowen P, Harrison P. Shorter Oxford Textbook of Psychiatry. Oxford University Press, Oxford, 1995, 2001, 2006.

Dr Luiz Dratcu – Consultant Psychiatrist and Specialist in Psychopharmacology, Treatment Resistant Mental Illness, Schizophrenia and Affective Disorders

Maudsley Hospital

South London & Maudsley NHS Foundation Trust

Denmark Hill,

London SE5 8AZ,

United Kingdom

luiz.dratcu@slam.nhs.uk

Recent publications include:

• Dratcu L. The quest for the pharmacological treatment of schizophrenia: from conventional neuroleptics to atypical anti-psychotics and beyond. Vertex 2010;21:385–93.

• Dratcu L. The future of depression: a complex neuroendocrine, inflammatory and neurodegenerative systemic illness. Vertex 2009:20:329–41.

• Dratcu L, Grandison A, McKay G, et al. Clozapine-resistant psychosis, smoking, and caffeine: managing the neglected effects of substances that our patients consume every day. American Journal of Therapeutics 2007;14:314–8.

• Dratcu L, Olowu P, Hawramy M, et al. Aripiprazole in the acute treatment of male patients with schizophrenia: effectiveness, acceptability, and risks in the inner-city hospital setting. Neuropsychiatric Disease and Treatment 2006;2:191–7.

Appendix 10.1.1 Draft MEDLINE search strategy

<  1948 to June Week 1 2011  >

Search Strategy:

1. Randomized Controlled Trials as Topic/ (73,451)

2. randomized controlled trial/ (308,386)

3. Random Allocation/ (71,692)

4. Double Blind Method/ (110,600)

5. Single Blind Method/ (15,044)

6. clinical trial/ (463,236)

7. clinical trial, phase i.pt. (11,244)

8. clinical trial, phase ii.pt. (17,834)

9. clinical trial, phase iii.pt. (6176)

10. clinical trial, phase iv.pt. (614)

11. controlled clinical trial.pt. (82,578)

12. randomized controlled trial.pt. (308,386)

13. multicenter study.pt. (131,287)

14. clinical trial.pt. (463,236)

15. exp Clinical Trials as topic/ (242,013)

16. or/1-15 (859,148)

17. (clinical adj trial$).tw. (155,138)

18. ((singl$ or doubl$ or treb$ or tripl$) adj (blind$3 or mask$3)).tw. (107,934)

19. PLACEBOS/ (29,733)

20. placebo$.tw. (129,547)

21. randomly allocated.tw. (12,594)

22. (allocated adj2 random$).tw. (14,831)

23. or/17-22 (326,697)

24. 16 or 23 (954,423)

25. case report.tw. (158,367)

26. letter/ (716,157)

27. historical article/ (275,084)

28. or/25-27 (1,139,766)

29. 24 not 28 (928,335)

30. exp Depression/ or exp Depressive Disorder/ (127,115)

31. (depress* or adjustment disorder* or mood disorder* or affective disorder* or affective symptom* or dysthymi* or dysphori*).mp. (344,159)

32. 30 or 31 (344,159)

33. 29 and 32 (39,592)

34. Serotonin Uptake Inhibitors/ or Antidepressive Agents, Second-Generation/ or ssri*.mp. or exp Serotonin

35. Antagonists/ (59,672)

36. citalopram.mp. or exp Citalopram/ (4058)

37. escitalopram.mp. (779)

38. fluoxetine.mp. or exp Fluoxetine/ (9218)

39. fluvoxamine.mp. or exp Fluvoxamine/ (2286)

40. paroxetine.mp. or exp Paroxetine/ (4469)

41. sertraline.mp. or exp Sertraline/ (2966)

42. 36 or 37 or 38 or 39 or 40 (16,313)

43. 35 and 41 (1722)

44. 35 or 37 or 38 or 39 or 40 (18,583)

45. 36 and 43 (713)

46. 35 or 36 or 38 or 39 or 40 (11,526)

47. 37 and 45 (2095)

48. 35 or 36 or 37 or 39 or 40 (17,294)

49. 38 and 47 (931)

50. 35 or 36 or 37 or 38 or 40 (15,845)

51. 39 and 49 (1665)

52. 35 or 36 or 37 or 38 or 39 (16,994)

53. 40 and 51 (1311)

54. 42 or 44 or 46 or 48 or 50 or 52 (3310)

55. 33 and 53 (799)

56. lithium.mp. or exp Lithium Carbonate/ or exp Lithium/ or exp Lithium Compounds/ or exp Lithium Chloride/ (29,746)

57. (antipsychotic* or anti?psychotic* or anti-psychotic*).mp. (41,739)

58. amisulpride.mp. (571)

59. aripiprazole.mp. (1454)

60. clozapine.mp. (8530)

61. olanzapine.mp. (5275)

62. paliperidone.mp. (153)

63. quetiapine.mp. (2428)

64. risperidone.mp. (5910)

65. or/34-40 (66,927)

66. or/55-63 (72,543)

67. 33 and 64 and 65 (713)

68. 54 or 66 (1455)

10.1.2 Drat MEDLINE search strategy (Health Economics and Quality of life)

PART ONE: REVIEW, REVIEWER AND STUDY INFORMATION

Study ID:

Reviewer name:

Date of completion of this form:

Title of paper/abstract:

Source (journal, year, volume, pages):

Authors:

Language of publication:

Type of paper (e.g. full paper/abstract/poster):

PART TWO: VERIFICATION OF STUDY ELIGIBILITY

PART THREE: INFORMATION ABOUT THE STUDY

PART FOUR: CLINICAL TRIAL QUALITY

Please describe the method of randomisation and allocation concealment used in the clinical trial:

Please describe the method of blinding and who was blinded in the clinical trial:

Please describe the number of patients lost to follow up (the overall number and number by treatment group, give reasons for loss to follow up):

How would you describe the trials design to minimise bias for (please tick):

How would you rate the trials overall risk of bias? Low risk Unclear High risk

Do you have any additional comments you would like to make about this clinical trial?

Ideally, would you like further information about the clinical trial from the authors (If so, please give details)?

Appendix 10.3 Team members' contributions

Steve Edwards, Head of HTA, will develop the protocol, act as the third reviewer for assessment of trials and cost-effectiveness studies, validate data extraction and any data analysis required, validate the economic model, contribute to writing/editing of the report, be overall director of the project and act as guarantor of the report.

Sam Barton, HTA Analyst, will act as co-reviewer for assessing trials for inclusion and data extraction, and contribute to the writing/editing of the report.

Vicky Hamilton, HTA Analyst, will provide overall project management, develop the protocol, write and run the search strategy, act as co-reviewer for assessing trials for inclusion and data extraction (and perform data analysis as required), and contribute to the writing/editing of the report.

Leo Nherera, Health Economist, will develop the protocol, act as co-reviewer of the cost-effectiveness studies, develop the economic model, and contribute to the writing/editing of the report.

Nicola Trevor, Health Economist, will act as co-reviewer of the cost-effectiveness studies, validate the economic model, and contribute to the writing/editing of the report.

Professor Cowen and Dr Dratcu, Clinical Expert Advisors, will provide clinical advice as required through out the protocol development and review processes.

Appendix 10.4 References

Search strategy

1. Depression/ or Depressive Disorder/ or Depressive Disorder, Major/ (126,194)

2. depress$.tw. (267,648)

3. 1 or 2 (300,405)

4. Serotonin Uptake Inhibitors/ or Antidepressive Agents, Second-Generation/ or ssri$.tw. or (serotonin adj2 inhibitor$).tw. or Serotonin Antagonists/ (35,140)

5. (citalopram or citalopramum or celexa or cipramil).tw. or Citalopram/ (4259)

6. (escitalopram or S-citalopram or lexapro or cipralex).tw. (939)

7. (fluoxetin$ or fluokset$ or felicium or prozac or prozep or prozit or sarafem or selfemra or symbyax).tw. or Fluoxetine/ (9738)

8. (fluvoxamin$ or fluvoksami$ or faverin or luvox).tw. or Fluvoxamine/ (2372)

9. (paroxetin$ or parokset$ or seroxat or paxil or pexeva).tw. or Paroxetine/ (4698)

10. (sertralin$ or lustral or zoloft).tw. or Sertraline/ (3149)

11. 4 or 5 or 6 or 7 or 8 or 9 or 10 (42,763)

12. (refract$ or resistan$ or nonrespon$ or unrespon$ or fail$ or (incomplet$ adj respon$) or (no$ adj2 respon$)).tw. (1,393,254)

13. treatment failure/ (22,171)

14. (inadequat$ respon$ or (sub$ adj2 respon$) or (poor$ adj respon$)).tw. (36,866)

15. 12 or 13 or 14 (1,427,769)

16. randomized controlled trial.pt. (314,314)

17. (random$ or rct or placebo$).tw. (623,289)

18. ((singl$ or doubl$ or tripl$ or treb$) adj (blind$3 or mask$3 or sham$ or dummy)).tw. (112,536)

19. (case reports or comment or editorial or in vitro or letter).pt. (2,881,916)

20. 19 not (randomized controlled trial or controlled clinical trial).pt. (2,874,149)

21. (16 or 17 or 18) not 20 (693,670)

22. exp animals/ not humans.sh. (3,650,161)

23. 21 not 22 (621,736)

24. (anti-depress$ or antidepress$).tw. (41,423)

25. Antidepressive Agents/ (28,927)

26. 24 or 25 (54,177)

27. (augment$ or adjunct$ or combin$ or add$ or potentiation).tw. (2,822,627)

28. Drug Therapy, Combination/ (125,552)

29. 27 or 28 (2,888,000)

30. Lithium/ or exp Lithium Compounds/ (23,585)

31. antipsychotic agents/ or clozapine/ or risperidone/ (38,767)

32. (Lithium or Camcolit or Liskonum or Priadel or Li-Liquid or Litarex or Lithonate or Phasal or Lithny or Licio or Lithii or Litio or Litium or Lityum or dilithium or litu or Cibalith-S or Eskalith or Lithane or Lithobid or Lithonate or Lithotabs).tw. (25,357)

33. ((atypical adj antipsychotic?) or (atypical adj anti-psychotic?) or Amisulprid$ or Solian or Aripiprazol$ or Abilify or Clozapin$ or Clozaril or Klotsapiini or Klozapin$ or FazaClo or Fazalco or Denzapine or Zaponex or Olanzapin$ or Olantsapiini or Zyprexa or Zypadhera or Paliperidon$ or Invega or Xeplion or Quetiapin$ or Seroquel or Risperidon$ or Riszperidon or Rysperydon or Risperdal or Ziprasidon$ or Geodon).tw. (19,696)

34. or/30-33 (74,184)

35. (26 and 29 and 34) or 11 (44,571)

36. 3 and 15 and 23 and 35 (938)

Search strategy

1. exp Major Depression/ (75,749)

2. depress$.tw. (180,819)

3. 1 or 2 (182,464)

4. Serotonin Antagonists/ or serotonin reuptake inhibitors/ or (ssri$ or (serotonin adj2 inhibitor$)).tw. (9804)

5. (citalopram or citalopramum or celexa or cipramil).tw. or Citalopram/ (1781)

6. (escitalopram or S-citalopram or lexapro or cipralex).tw. (615)

7. (fluoxetin$ or fluokset$ or felicium or prozac or prozep or prozit or sarafem or selfemra or symbyax).tw. or Fluoxetine/ (5066)

8. (fluvoxamin$ or fluvoksami$ or faverin or luvox).tw. or Fluvoxamine/ (1376)

9. (paroxetin$ or parokset$ or seroxat or paxil or pexeva).tw. or Paroxetine/ (2593)

10. (sertralin$ or lustral or zoloft).tw. or Sertraline/ (1930)

11. 4 or 5 or 6 or 7 or 8 or 9 or 10 (16,639)

12. (refract$ or resistan$ or nonrespon$ or unrespon$ or fail$ or (incomplet$ adj respon$) or (no$ adj2 respon$)).tw. (167,518)

13. (inadequat$ respon$ or (sub$ adj2 respon$) or (poor$ adj respon$)).tw. (8907)

14. 12 or 13 (174,891)

15. treatment outcome clinical trial.md. (19,559)

16. treatment effectiveness evaluation/ (12,499)

17. (random$ or rct or placebo$).tw. (117,674)

18. ((singl$ or doubl$ or tripl$ or treb$) adj (blind$3 or mask$3 or sham$ or dummy)).tw. (16,252)

19. 15 or 16 or 17 or 18 (136,659)

20. (comment or editorial or letter).dt. (31,151)

21. 20 not treatment outcome clinical trial.md. (30,861)

22. 19 not 21 (135,679)

23. antidepressant drugs/ (13,273)

24. (antidepress$ or anti-depress$).tw. (26,031)

25. 23 or 24 (28,058)

26. (augment$ or adjunct$ or combin$ or add$ or potentiation).tw. (527,381)

27. polypharmacy/ (536)

28. Drug Augmentation/ (740)

29. 26 or 27 or 28 (527,592)

30. (Lithium or Camcolit or Liskonum or Priadel or Li-Liquid or Litarex or Lithonate or Phasal or Lithny or Licio or Lithii or Litio or Litium or Lityum or dilithium or litu or Cibalith-S or Eskalith or Lithane or Lithobid or Lithonate or Lithotabs).tw. (8478)

31. ((atypical adj antipsychotic?) or (atypical adj anti-psychotic?) or Amisulprid$ or Solian or Aripiprazol$ or Abilify or Clozapin$ or Clozaril or Klotsapiini or Klozapin$ or FazaClo or Fazalco or Denzapine or Zaponex or Olanzapin$ or Olantsapiini or Zyprexa or Zypadhera or Paliperidon$ or Invega or Xeplion or Quetiapin$ or Seroquel or Risperidon$ or Riszperidon or Rysperydon or Risperdal or Ziprasidon$ or Geodon).tw. (14,792)

32. exp Lithium Carbonate/ or Lithium/ (5297)

33. neuroleptic drugs/ or aripiprazole/ or clozapine/ or olanzapine/ or quetiapine/ or risperidone/ (20,778)

34. 30 or 31 or 32 or 33 (30,510)

35. 25 and 29 and 34 (1550)

36. 11 or 35 (17,771)

37. 3 and 14 and 22 and 36 (720)

Search strategy

1. *depression/ or *major depression/ (107,356)

2. depress$.tw. (302,295)

3. 1 or 2 (321,132)

4. serotonin antagonist/ or serotonin uptake inhibitor/ or fluoxetine plus olanzapine/ or fluvoxamine maleate/ or ssri$.tw. or (serotonin adj2 inhibitor$).tw. (46,279)

5. (citalopram or citalopramum or celexa or cipramil).tw. or citalopram/ (14,277)

6. (escitalopram or S-citalopram or lexapro or cipralex).tw. or escitalopram/ (4407)

7. (fluoxetin$ or fluokset$ or felicium or prozac or prozep or prozit or sarafem or selfemra or symbyax).tw. or fluoxetine/ or fluoxetine plus olanzapine/ (32,341)

8. (fluvoxamin$ or fluvoksami$ or faverin or luvox).tw. or fluvoxamine/ or fluvoxamine maleate/ (10,747)

9. (paroxetin$ or parokset$ or seroxat or paxil or pexeva).tw. or paroxetine/ (19,872)

10. (sertralin$ or lustral or zoloft).tw. or sertraline/ (16,373)

11. 4 or 5 or 6 or 7 or 8 or 9 or 10 (79,964)

12. (refract$ or resistan$ or nonrespon$ or unrespon$ or fail$ or (incomplet$ adj respon$) or (no$ adj2 respon$)).tw. (1,520,915)

13. drug treatment failure/ (10,532)

14. (inadequat$ respon$ or (sub$ adj2 respon$) or (poor$ adj respon$)).tw. (40,713)

15. 12 or 13 or 14 (1,555,120)

16. randomized controlled trial/ (284,436)

17. (random$ or rct or placebo$).tw. (707,279)

18. ((singl$ or doubl$ or tripl$ or treb$) adj (blind$3 or mask$3 or sham$ or dummy)).tw. (129,146)

19. 16 or 17 or 18 (787,496)

20. (editorial or letter).pt. (1,110,740)

21. 20 not (16 or clinical trial/) (1,070,423)

22. 19 not 21 (785,365)

23. exp animals/ not exp humans/ (1,237,993)

24. 22 not 23 (761,970)

25. (anti-depress$ or antidepress$).tw. (52,230)

26. antidepressant agent/ (59,124)

27. 25 or 26 (85,174)

28. (augment$ or adjunct$ or combin$ or add$ or potentiation).tw. (3,077,561)

29. drug combination/ (53,343)

30. 28 or 29 (3,112,487)

31. exp lithium/ or exp lithium derivative/ or lithium carbonate/ or lithium chloride/ or lithium citrate/ (46,132)

32. atypical antipsychotic agent/ or ziprasidone/ or risperidone/ or quetiapine/ or paliperidone/ or olanzapine/ or clozapine/ or clozapine derivative/ or clozapine n oxide/ or aripiprazole/ or amisulpride/ (46,689)

33. (Lithium or Camcolit or Liskonum or Priadel or Li-Liquid or Litarex or Lithonate or Phasal or Lithny or Licio or Lithii or Litio or Litium or Lityum or dilithium or litu or Cibalith-S or Eskalith or Lithane or Lithobid or Lithonate or Lithotabs).tw. (29,256)

34. ((atypical adj antipsychotic?) or (atypical adj anti-psychotic?) or Amisulprid$ or Solian or Aripiprazol$ or Abilify or Clozapin$ or Clozaril or Klotsapiini or Klozapin$ or FazaClo or Fazalco or Denzapine or Zaponex or Olanzapin$ or Olantsapiini or Zyprexa or Zypadhera or Paliperidon$ or Invega or Xeplion or Quetiapin$ or Seroquel or Risperidon$ or Riszperidon or Rysperydon or Risperdal or Ziprasidon$ or Geodon).tw. (27,771)

35. or/31-34 (96,393)

36. (27 and 30 and 35) or 11 (81,812)

37. 3 and 15 and 24 and 36 (1303)

Search strategy

PART ONE: REVIEW, REVIEWER AND STUDY INFORMATION

Study ID:

Reviewer name:

Date of completion of this form:

Title of paper/abstract:

Source (journal, year, volume, pages):

Authors:

Language of publication:

Type of paper (e.g. full paper/abstract/poster):

PART TWO: VERIFICATION OF STUDY ELIGIBILITY

PART THREE: INFORMATION ABOUT THE STUDY

Characteristics of the trial

Country(ies) where the clinical trial was conducted:

Sponsors of the clinical trial:

Any conflicts of interest reported for any of the researchers?

Date the clinical trial was conducted:

Type of clinical trial design (e.g. parallel, crossover, or cluster trial):

If the trial was of crossover design, are there pre-crossover results reported?

Was the trial multicentre? If so, how many centres were there?

Characteristics of the patients

Inclusion criteria

• How and where were patients enrolled, were any patient risk factors used?

• What details of the antidepressant(s) patients had failed to respond to are provided?

• How was treatment resistance defined?

Exclusion criteria

• Were specific groups of people excluded?

Total number of people randomised:

Information on the age of the patients:

Information on the sex of the patients (m/f):

Information on the ethnicity of the patients:

Information on patients' medical history (i.e. previous depression):

Type of intervention:

Intervention 1 SSRI  +  XX (where XX  =  atypical antipsychotic or lithium or no treatment)

SSRI name and brand:

SSRI dose and regimen used [e.g. 80  mg once daily (OD)]:

Delivery of SSRI [e.g. per os (p.o.) tablet/dissolvable/enteric coated]:

Number of doses of SSRI given per day (with SD/SE if given):

Duration of SSRI treatment in days (with SD/SE if given):

What was XX (name and brand)?

XX dose and regimen used (e.g. 80  mg OD):

Delivery of XX (e.g. p.o. tablet/dissolvable/enteric coated):

Number of doses of XX given per day (with SD/SE if given):

Duration of XX treatment in days (with SD/SE if given):

Number of patients randomised:

Intervention 2 SSRI  +  YY (where YY  =  lithium or atypical antipsychotic or placebo or no treatment)

SSRI name and brand:

SSRI dose and regimen used (e.g. 80  mg OD):

Delivery of SSRI (e.g. p.o. tablet/dissolvable/enteric coated):

Number of doses of SSRI given per day (with SD/SE if given):

Duration of SSRI treatment in days (with SD/SE if given):

What was YY (name and brand)?

YY dose and regimen used (e.g. 80  mg OD):

Delivery of YY (e.g. p.o. tablet/dissolvable/enteric coated):

Number of doses of YY given per day (with SD/SE if given):

Duration of YY treatment in days (with SD/SE if given):

Number of patients randomised:

Was the formulation and appearance of YY (e.g. lithium) matched to that of XX (e.g. atypical antipsychotic)?

Were any additional interventions given to either or both groups?

Did the RCT carry out any subgroup analyses of interest? (i.e. Different durations of depression, different classes of previous antidepressants, different genders, age, different severities of depression or different number of prior antidepressants.)

If yes, please give details here:

Did the RCT provide any details of maintenance therapy +/− outcomes?

PART FOUR: CLINICAL TRIAL QUALITY

Please describe the method of randomisation and allocation concealment used in the clinical trial:

Please describe the method of blinding and who was blinded in the clinical trial:

How would you describe the trial's design to minimise bias for (please tick):

How would you rate the trial's overall risk of bias?

 Low risk Unclear High risk

Do you have any additional comments you would like to make about this clinical trial?

Ideally, would you like further information about the clinical trial from the authors (if so, please give details)?