Criteria for considering studies for this review
Types of studies
RCTs and cluster RCTs will be included.
Trials using a cross-over design will be included but only data from the first treatment phase will be used.
Any other study design, including quasi-randomised studies and non-randomised studies will be excluded from this review.
Types of participants
Participants must be aged 18 to 74 years.
If the study includes some participants that are aged under 75 years and some over the age of 74 years, we will exclude the study if the mean age of participants is over 74 years. Similarly, if the study includes some participants that are aged under 18 years and some aged 18 years or older, we will exclude the study if the mean age of participants is less than 18 years.
Definition of treatment-resistant depression
A primary diagnosis of unipolar depression that has not responded (or has only partially responded) to a minimum of four weeks of antidepressant treatment at a recommended dose (at least 150 mg/day imipramine or equivalent antidepressant (e.g. 20 mg/day citalopram)).
Studies that include participants who have not responded because of intolerance to antidepressant medication will be excluded.
While there have been initiatives to improve access to psychological therapies in England and elsewhere, access to psychological treatment is still limited and antidepressants are often the first-line treatment for adults with depression. Therefore, this review will not include studies of interventions for those who have not responded to psychological treatment.
Acceptable diagnoses include those based on criteria from DSM-IV-TR or earlier versions (APA 2000), International Classification of Diseases (ICD)-10 (WHO 1992), Feighner criteria (Feighner 1972) or Research Diagnostic Criteria (Spitzer 1978). Studies that have not used standardised diagnostic criteria will be excluded.
Studies of participants with comorbid schizophrenia or bipolar disorder will be excluded.
Studies including both unipolar and bipolar participants will be excluded unless data are available for the subgroup of unipolar participants.
Studies involving participants with comorbid physical conditions or other psychological disorders (e.g. anxiety) will be eligible for inclusion as long as the comorbidity is not the focus of the study.
Types of interventions
The experimental interventions are based on the 'next step' approach to the management of depression that has not responded to treatment with antidepressants:
increasing the dose of antidepressant monotherapy;
switching to a different antidepressant monotherapy;
augmenting treatment with another antidepressant;
augmenting treatment with a non-antidepressant.
Antidepressants can be grouped as TCAs, MAOIs, SSRIs, SNRIs and NaSSAs.
Non-antidepressant medications used as augmentors include antipsychotics (e.g. olanzapine), anxiolytics (e.g. buspirone), antimania drugs (e.g. lithium) and beta-blockers (e.g. pindolol).
Studies examining non-standard pharmacological approaches for treating TRD (e.g. sex hormones, vitamins, herbal medicines and food supplements) will be excluded.
Full details of comparisons to be made can be found in the Data extraction and management section.
Studies examining psychological interventions given in addition to antidepressant medication for individuals with TRD will be included in another review (Wiles 2013).
Types of outcome measures
1. Change in depressive symptoms as measured on rating scales for depression, either clinician rated (e.g. Hamilton Rating Scale for Depression (HAM-D; Hamilton 1960) or Montgomery-Asberg Depression Rating Scale (MADRS; Montgomery 1979)), or self report (e.g. Beck Depression Inventory (BDI; Beck 1961; Beck 1996) (or other validated measures)). Data on observer-rated and self report outcomes will be analysed separately.
2. Number of drop-outs within the trials, which indicates compliance rates with different treatments. Data on reasons for drop-out will be collected, where available, and will be summarised in narrative form.
3. Response or remission rates or both based on changes in depression measures - either clinician rated (e.g. HAM-D; Hamilton 1960)) or self report (e.g. BDI; Beck 1961; Beck 1996) or other validated measures). Response is frequently quantified as at least a 50% reduction in symptoms on the HAM-D or BDI but we will accept the study's original definition. Remission is based on the absolute score on the depression measure. Examples of definitions of remission include 7 or less on the HAM-D and 10 or less on BDI. Again, we will accept the study authors' original definition.
4. Improvement in social adjustment and social functioning including the Global Assessment of Function (Luborsky 1962) scores, where reported, will be summarised in narrative form.
5. Improvement in quality of life as measured on the Short Form (SF)-36 (Ware 1993), Health of the Nation Outcome Scales (HoNOS) (Wing 1994), or World Health Organization Quality of Life (WHOQOL) (WHOQOL 1998) or similar scale, where reported, will be summarised in narrative form.
6. Economic outcomes (e.g. days of work absence/ability to return to work, number of appointments with primary care physician, number of referrals to secondary services, use of additional treatments), where reported will be summarised in narrative form.
7. Adverse effects (e.g. completed/attempted suicides), where reported, will be summarised in narrative form.
Timing of outcome assessment
Outcomes at each reported follow-up point will be summarised. Where appropriate and if the data allow, outcomes will be categorised as short term (up to 12 weeks) and longer-term (12 weeks or longer).
Search methods for identification of studies
Cochrane Depression, Anxiety and Neurosis Group's Specialized Register (CCDANCTR)
The Cochrane Depression, Anxiety and Neurosis Group (CCDAN) maintain two clinical trials registers at their editorial base in Bristol, UK, a references register and a studies-based register. The CCDANCTR-References Register contains over 31,500 reports of trials in depression, anxiety and neurosis. Approximately 65% of these references have been tagged to individual, coded trials. The coded trials are held in the CCDANCTR-Studies Register and records are linked between the two registers through the use of unique Study ID tags. Coding of trials is based on the EU-Psi coding manual. Please contact the CCDAN Trials Search Co-ordinator for further details. Reports of trials for inclusion in the Group's registers are collated from routine (weekly), generic searches of MEDLINE (1950-), EMBASE (1974-) and PsycINFO (1967-); quarterly searches of the Cochrane Central Register of Controlled Trials (CENTRAL) and review-specific searches of additional databases. Reports of trials are also sourced from international trials registers via the World Health Organization's trials portal (ICTRP); drug companies; and handsearching of key journals, conference proceedings and other (non-Cochrane) systematic reviews and meta-analyses.
Details of CCDAN's generic search strategies can be found on the Group's website.
The CCDANCTR-Studies Register will be searched using the following terms:
Condition = ((depressi* or "affective disorder" or "mood disorder*") and ("treatment resistant" or recurrent))
The CCDANCTR-References will be searched using a more sensitive set of terms (keywords and subject headings) to identify additional untagged/uncoded references:
1. depressi* [Ti, Ab, KW]
2. (*refractory* or *resistan* or *recurren*) [Ti, Ab]
3. (augment* or potentiat*) [Ti, Ab]
4. (chronicity or "chronic depress*" or "chronically depress*" or "depressed chronic*" or "chronic major depressi*" or "chronic affective disorder*" or "chronic mood disorder*" or (chronic* and (relaps* or recurr*))) [Ti, Ab, KW]
5. ("persistent depress*" or "persistently depress*" or "depression persist*" or "persistent major depress*" or "persistence of depress*" or "persistence of major depress*") [Ti, Ab]
6. (nonrespon* or non-respon* or "non respon*" or "not respon*" or "no respon*" or "partial respon*" or "partially respon*" or "incomplete respon*" or "incompletely respon*" or unrespon*) [Ti, Ab]
7. ("failed to respond" or "failed to improve" or "failure to respon*" or "failure to improve" or "failed medication*" or "antidepressant fail*" or "treatment fail*") [Ti, Ab]
8. (inadequate* and respon*) [Ti, Ab]
9. "treatment resistant depression" [KW]
10. (recurrence or "recurrent depression" or "recurrent disease") [KW]
11. "drug resistance" [KW]
12. "treatment failure" [KW]
13. "drug potentiation" [KW]
14. augmentation [KW]
16. (1 and 15)
The WHO trials portal (ICTRP) and ClinicalTrials.gov will also be searched for 'treatment resistant' or 'treatment refractory' depression to identifying any additional ongoing or unpublished studies. Principal investigators will be contacted, where necessary, to obtain further details of ongoing/unpublished studies, or trials reported as conference abstracts only.
There will be no date or language restrictions applied to our search.
Searching other resources
Reference lists of all included studies and other relevant systematic reviews will be searched for papers that may meet inclusion criteria. Subject experts will also be contacted to ensure that all relevant published and unpublished studies have been considered for inclusion.
Data collection and analysis
Selection of studies
One review author (NW) will examine titles and abstracts to remove obviously irrelevant reports and then screen study abstracts against inclusion criteria using a standardised abstract screening form. In any case of uncertainty an over-inclusive approach will be taken and the full paper will be obtained along with those for the studies assessed as meeting the inclusion criteria. Two review authors will screen each paper for inclusion or exclusion from the review. If any disagreements arise these will be discussed with a third review author. If it is not possible to determine eligibility for a study, it will be added to the list of those awaiting assessment and the authors will be contacted for further information or clarification.
The study selection process will be documented using a PRISMA study selection flow diagram, which will be presented in the review.
Data extraction and management
Data regarding participants, interventions and their comparators, methodological details, treatment effects including dropouts and possible biases will be independently extracted by two review authors using a standardised data extraction form. If any disagreements arise these will be discussed with a third review author. The data extraction form will be piloted during the first phase of data extraction.
Information relating to study population, definition of TRD, sample size, interventions, comparators, potential biases in the conduct of the trial, outcomes, follow-up and methods of statistical analysis will be abstracted.
Main planned comparisons
Increasing the dose of antidepressant monotherapy compared with continuing on an antidepressant.
Switching to a different antidepressant monotherapy compared with continuing on an antidepressant.
Augmenting treatment with another antidepressant compared with continuing on an antidepressant.
Augmenting treatment with a non-antidepressant compared with continuing on an antidepressant.
Within each of these strategies, this review will summarise the evidence for each drug individually. For example, taking the fourth approach of augmenting with a non-antidepressant medication, the augmentator could be lithium, olanzapine, buspirone or pindolol. In this case, the evidence for each of these four drugs would be presented separately, rather than combined to summarise the evidence for a particular 'treatment approach'. This will maximise the clinical relevance of the findings.
Given the large number of possible combinations of medications that could be evaluated, it is not possible to provide an exhaustive list of all the potential comparisons. However, one example for each of the different approaches is given below.
Citalopram 40 mg/day compared with remaining on citalopram 20 mg/day.
Switching to mianserin compared with remaining on fluoxetine.
Addition of mianserin to fluoxetine compared with remaining on fluoxetine alone.
Nortriptyline plus lithium compared with remaining on nortriptyline alone.
Assessment of risk of bias in included studies
Two review authors will independently assess risk of bias for each included study using The Cochrane Collaboration's 'Risk of bias' tool (Higgins 2008a). If any disagreements arise these will be discussed with a third review author. The following criteria will be assessed:
sequence generation: was the allocation sequence adequately generated?
allocation concealment: was allocation adequately concealed?
blinding of participants, study personnel and outcome assessors for each outcome: was knowledge of the allocated treatment adequately prevented during the study?
incomplete outcome data for each main outcome or class of outcomes: were incomplete outcome data adequately addressed?
selective outcome reporting: are reports of the study free of suggestion of selective outcome reporting?
other sources of bias: was the study apparently free of other problems that could put it at a high risk of bias?
A description of what was reported to have happened in each study will be provided and a judgement on the risk of bias will be made for each domain within and across studies, based on the following three categories: low risk of bias; unclear risk of bias; high risk of bias.
Where studies provide little or no detail about randomisation, the authors will be contacted to seek clarification.
All risk of bias data will be presented graphically and described in the text.
Measures of treatment effect
Continuous outcomes will be analysed by calculating the mean difference (MD) between groups, if studies use the same outcome measure for comparison. If different outcome measures are used to assess the same outcome, the standardised mean difference (SMD) and 95% confidence intervals (CI) will be calculated. The SMD will be interpreted as follows: 0.2 represents a small effect, 0.5 a moderate effect, and 0.8 a large effect (Cohen 1988).
Dichotomous outcomes will be analysed by calculating the odds ratio (OR) and 95% CIs for each comparison, and then converted to risk ratios (RR) using the formula provided in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2008). When overall risks are significant, the number needed to treat for an additional beneficial outcome (NNTB) or for an additional harmful outcome (NNTH) to produce one outcome will be calculated by combining the overall RR with an estimate of prevalence of the event in the control group of the trials.
Unit of analysis issues
Cluster randomised trials
Results from cluster RCTs will be incorporated into the review using generic inverse variance methods (Higgins 2008). With cluster RCTs, it is important to ensure that the data have been analysed taking into account the clustered nature of the data. The intracluster correlation coefficient (ICC) will be extracted for each trial. Where no such data are reported, this information will be requested from study authors. If this is not available, in line with the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2008), we will use estimates from similar studies in order to 'correct' data for clustering, where this has not been done.
For cross-over trials, only results from the first randomised treatment period will be included in the analysis.
Studies with multiple treatment groups
Trials that have more than two arms (e.g. pharmacological intervention (A); pharmacological intervention (B); and control) can cause problems in pair-wise meta-analysis. If any studies are identified with two or more active treatment arms, then the following approach will be undertaken dependent on whether the outcome is continuous or dichotomous.
For a continuous outcome: means, standard deviations (SDs) and the number of participants for each active treatment group will be pooled across treatment arms as a function of the number of participants in each arm to be compared against the control group (Higgins 2008).
For a dichotomous outcome: active treatment groups will be combined into a single arm for comparison against the control group (in terms of the number of people with events and sample sizes), or the control group will be split equally (Higgins 2008).
Dealing with missing data
Where there are missing data, study authors will be contacted to obtain data. If an outcome is missing for more than 50% of participants this study will be excluded from the analysis. If SD data are unobtainable but are available for the majority of other included studies, these will be calculated by using an imputation method, which is based on the SDs of the other included studies (Furukawa 2006).
Missing dichotomous data will be managed using an intention-to-treat (ITT) analysis in which it will be assumed that participants who dropped out after randomisation had a negative outcome. In addition, best/worse-case scenarios will also be calculated, in which it will be assumed that dropouts in the intervention group had positive outcomes and those in the comparator group had negative outcomes (best-case scenario), and that dropouts in the intervention group had negative outcomes and those in the comparator group had positive outcomes (worst-case scenario), thus giving boundaries for the observed treatment effect.
Missing continuous data will be managed using a last observation carried forward (LOCF) analysis if LOCF data are reported, or on end point data alone.
Assessment of heterogeneity
Heterogeneity will be assessed using the Chi2 test, which provides evidence of variation in effect estimates beyond that of chance. The Chi2 test has low power to assess heterogeneity when there are few included studies or small numbers of participants, so the P value will be conservatively set at 0.1. Heterogeneity will also be quantified using the I2 statistic, which calculates the percentage of variability due to heterogeneity rather than chance. We expect, a priori, that there will be considerable clinical heterogeneity between studies, therefore I2 values between 50% and 90% will be considered to represent substantial statistical heterogeneity and will be explored further. However, the importance of the observed I2 value will depend on the magnitude and direction of treatment effects and the strength of evidence for heterogeneity (Higgins 2008). Forest plots generated using Review Manager 5 software now also provide an estimate of tau2, the between-study variance in a random-effects meta-analysis (RevMan 2011). Therefore, for the primary outcome, we will also use tau2 to give an indication of the spread of true intervention effects.
Assessment of reporting biases
Reporting bias will be managed by undertaking comprehensive searches for papers that will not be restricted to the English language. Outcome reporting bias will be determined for all included studies and trial protocols will be sought wherever possible. If outcome data are missing this will be requested from authors. Funnel plots will be used to help detect reporting biases if at least 10 studies are included (Higgins 2008). Formal assessment of asymmetry in the funnel plot will be examined using the Egger test (Egger 1997).
Given the potential for heterogeneity in the included interventions, we will use a random-effects model for all analyses. This approach will incorporate the assumption that the different studies are estimating different, yet related, intervention effects and it takes into account differences between studies even if there is no statistically significant heterogeneity. Heterogeneity will be tested formally using both the Chi2 test and I2 statistic (as outlined above). We will seek clinical advice in terms of combining treatment groups in order to ensure that findings are clinically meaningful.
If meta-analysis is not possible (e.g. due to insufficient data or substantial heterogeneity), a narrative assessment of the evidence will be given. This will summarise the evidence according to intervention type.
Subgroup analysis and investigation of heterogeneity
The degree of treatment resistance recorded at the point of entry to the trial is expected to have an impact on outcome. Therefore, the following subgroup analyses will be undertaken:
severity of depression (non-responders compared with partial responders): the severity of depression is expected to have an impact on outcome. Heterogeneity analyses are, therefore, planned to categorise severity according to whether participants were classified as being 'non-responders' or 'partial responders' at baseline;
Length of acute treatment phase: there were likely to be differences in the length of the duration of antidepressant treatment prior to trial entry that would be expected to affect outcome. Length of acute treatment phase was categorised as: four weeks or longer; 12 weeks or longer and six months or longer.
Such subgroup analyses will only be conducted when there are data from at least 10 studies to be included (Higgins 2008).
Sensitivity analyses were planned to explore how much of the variation between studies comparing pharmacological interventions for TRD was accounted for by between-study differences in:
study quality: allocation concealment is to be used as a marker of trial quality. Studies that have not used allocation concealment were to be excluded;
blinding: studies that were unblinded will be excluded;
attrition: studies with more than 20% drop-out will be excluded;
missing data: studies that have imputed missing data will be excluded;
funding source: studies funded by pharmaceutical companies will be excluded;
publication type: studies not published in full (conference abstract/proceedings, doctoral dissertation) will be excluded.
Summary of findings table
'Summary of findings' tables will be prepared for all relevant comparisons. Each table will include all outcomes.