Combining mirtazapine with SSRIs or SNRIs for treatment-resistant depression: the MIR RCT

Inclusion criteria (all must apply)

Eligible patients were those who met all of the following criteria.

• Aged ≥ 18 years and in primary care.

• Treated for depression for at least 6 weeks with any one of the following SSRI or SNRI antidepressants at recommended BNF doses – fluoxetine, sertraline, citalopram, escitalopram, fluvoxamine, paroxetine, duloxetine or venlafaxine (see Appendix 1, Table 25 for the adequate dose table).

• Adhered to their medication. Adherence to medication is difficult to measure. To operationalise our definition of treatment resistance, we used the Morisky four-item self-report measure of compliance,24 as adapted for CoBalT. 25 The Morisky measure has previously been validated against electronic monitoring bottles, with a score of zero (range 0–4) indicating at least 80% compliance. 18,26 Given the relatively long half-life of antidepressant medication, individuals who have forgotten to take one or two tablets were not excluded.

• Scored ≥ 14 points on the Beck Depression Inventory-II (BDI-II). 27

• An ICD-10 diagnosis of depression [assessed using the Clinical Interview Schedule – Revised (CIS-R)]. 28

Exclusion criteria (presence of any warrants exclusion)

General practitioners were asked to exclude patients who fulfilled any of the following exclusion criteria at the time of the record search:

• patients currently taking combined or augmented antidepressant treatment

• patients having their medication managed by a psychiatrist

• patients with dementia (formal diagnosis), bipolar disorder, psychosis or alcohol or substance abuse/dependence

• women who were pregnant, planning a pregnancy or breastfeeding

• patients who were unable to complete the study questionnaires

• patients who had had a previous adverse reaction to mirtazapine

• patients currently being treated with a monoamine oxidase inhibitor (MAOI), including moclobemide, or with other medical contraindications to mirtazapine.

Method 1: search of general practice computerised records to identify patients being treated for depression

The search of general practice computerised records identified all patients in the appropriate age range who had been prescribed SSRI or SNRI antidepressants for ≥ 6 weeks and who were currently being prescribed an antidepressant medication at an adequate dose for depression, as recommended in the BNF. 19

The GPs then screened this list of patients and excluded those patients who fulfilled any of the exclusion criteria listed above. A letter of invitation and a brief information leaflet about the study were sent by the general practice to the remaining potential participants. This letter sought permission for the research team to contact potential participants and to send a questionnaire asking about their depressive symptoms and adherence to antidepressant medication. Potential participants replied directly to the study team indicating whether or not they agreed to be contacted. One reminder was sent to those who did not respond to the initial letter of invitation.

On the reply slip, those who did not wish to participate were asked to indicate their age, sex and reason for non-participation. In addition, non-participants were asked to indicate their willingness to take part in a brief telephone interview to discuss their reasons for non-participation.

Anonymised data on the age and sex of those patients who were mailed an invitation to participate but who did not respond (or who refused to participate when invited during the consultation) were collected to assess the generalisability of the study findings.

Method 2: in-consultation recruitment

The GPs could also invite patients to take part in the study during a consultation. In such cases, they provided patients with an information leaflet about the study and obtained permission to pass their contact details to the research team. The research team then mailed a questionnaire to potential participants asking them about their depressive symptoms and adherence to antidepressant medication.

Postal screening: assessment of depressive symptoms and adherence to antidepressants

All those who agreed to be contacted by the research team (in response to either the postal invitation or a direct invitation from their GP during a consultation) were sent a postal questionnaire. This questionnaire collected data on the following:

• severity of depressive symptoms, using the BDI-II

• duration of antidepressant treatment, dose of medication and adherence to medication, using the Morisky measure of compliance

• sociodemographic variables (age, sex, marital status, ethnicity, educational qualifications, employment status, home ownership and financial difficulties).

One reminder was sent to those individuals who did not return a completed postal questionnaire within 2 weeks.

The GPs were then asked to review all of the patients who appeared to be eligible on the postal screening and to sign a form to confirm that these patients were suitable to be prescribed mirtazapine. Once GP approval was received, patients were contacted by a researcher by telephone and invited to attend a face-to-face appointment with a researcher to discuss participation in the trial and to assess their eligibility. The date, time and location of the baseline appointment were confirmed by letter. A detailed patient information sheet (PIS) and a detailed leaflet about mirtazapine were also enclosed; patients were asked to read both prior to attending the baseline appointment.

Patients who completed the screening process but were not eligible to participate received a letter informing them of this and thanking them for taking part. The letter explained that their GP had also been informed and would continue to care for them as usual. The GP received a letter which explained that the patient was ineligible as he or she had not met one or more of the eligibility criteria, but that the GP could refer the patient back to the trial if these factors changed. If the patient had given permission in the postal questionnaire, the GP also received a report that gave more detail about the inclusion criteria that were not satisfied, as well as the individual’s score on the BDI-II.

Baseline assessment

The baseline assessment was conducted in the patient’s home or at the GP’s surgery or on university premises. The researcher explained the study in detail and obtained written informed consent for the baseline assessment. If the potential participant agreed to the assessment, they completed the following questionnaires:

• BDI-II

• Morisky questionnaire (adherence to medication)

• Patient Health Questionnaire-9 items (PHQ-9),29 a brief measure of depression

• CIS-R, an in-depth psychiatric questionnaire that gives an ICD-10 diagnosis.

Patients were also asked for details of their prescribed medication, prior use of antidepressants and whether or not they were currently in receipt of psychological therapy. In addition, sociodemographic details were recorded (age, sex, ethnicity, marital status), together with information on a number of socioeconomic markers (employment status, housing situation).

Potential participants who scored ≥ 14 points on the BDI-II and who had an ICD-10 primary diagnosis of depression using the CIS-R, were told that they were potentially eligible to enter the trial [pending confirmation by the principal investigator (PI)] and were asked to provide further written consent for trial participation and indicate whether or not they were willing to be contacted about future research projects.

These potentially eligible participants were also asked to complete some further questionnaires including the:

• General Anxiety Disorder-7 (GAD-7) questionnaire30

• EuroQoL-5 Dimensions, five-level version (EQ-5D-5L),31 a brief measure of health-related QoL

• Short Form questionnaire-12 items (SF-12), a brief measure of mental and physical functioning32

• Antidepressant Side-Effect Checklist (ASEC), a measure of antidepressant side effects. 33

They were asked further questions about their history of depression, whether or not they had ever been referred to a psychiatrist and the strength of their preference for active treatment over placebo (as this could potentially affect medication adherence and outcomes). Additional information was collected on life events, financial stress, social support and use of alcohol. 34

Once the baseline assessment was complete, the local research clinician (PI) reviewed the baseline information and confirmed whether or not the patient was eligible for the MIR trial.

Informed consent

Prior to the start of the baseline assessment, patients were asked to provide written informed consent for the storage and processing of the data collected at the time of the assessment. This covered the data collected from both those who were found to be ineligible to participate in the trial as well as those who were eligible, thus, enabling the trial to be reported in line with the Consolidated Standards of Reporting Trials (CONSORT) guidelines. 35 Those patients who were identified as eligible to participate in the trial were asked to provide additional written informed consent for this purpose. The original signed and dated consent forms were held securely as part of the trial site file, with copies given to both the participants and their GPs for their records.

Randomisation, concealment of allocation and blinding

Following the baseline assessment, eligible and consenting participants were randomised using the automated randomisation service provided by the Bristol Randomised Trials Collaboration (BRTC). Randomisation was carried out by means of a computer-generated code to ensure concealment of allocation.

Randomisation was stratified by centre (n = 4) to ensure balance in terms of local differences. Minimisation was used to ensure balance in important prognostic indicators, that is, baseline BDI-II score (using approximate tertiles derived from CoBalT baseline scores: < 26 points; 26–34 points; ≥ 35 points), sex and whether or not the patient was currently receiving a psychological therapy (yes/no). We used minimisation with a probability weighing of 0.8 to reduce predictability. 36

Once this had been completed, the University Hospitals Bristol NHS Foundation Trust (UHB) pharmacy, which was the central trial pharmacy, was notified. A participant pack containing an initial 8-week supply of medication was sent by registered post from the UHB pharmacy either to participants’ GP surgery or, in exceptional circumstances, to their home.

Treatment group allocation

Participants were randomly assigned to one of two treatments: (1) one × 15-mg encapsulated mirtazapine tablet daily for 2 weeks followed by two × 15-mg encapsulated mirtazapine tablets for up to 50 weeks or (2) identical placebo tablets. Participants were free to withdraw from the medication at any time. Participants, clinicians, outcome assessors and the research team were blinded to allocation. All participants continued with their GP care and usual antidepressants as agreed by their GP. Clinicians were not restricted in their use of psychological services.

Unblinding was available through the trial pharmacy at all times in case of a medical emergency (‘emergency unblinding’). After the 12-week primary outcome had been completed, the code could also be broken at the request of a participant or his or her GP (‘elective unblinding’). Those who had not requested emergency or elective unblinding were unblinded at the end of the follow-up period or on withdrawal from the study. The trial team did not provide further supplies of the trial medication once participants had been unblinded.

Primary outcome

The primary outcome was change in the BDI-II score at 12 weeks post randomisation, measured as a continuous variable.

The BDI-II is a 21-item self-report instrument used to measure the severity of depressive symptoms occurring over the previous 2 weeks; it has been widely used in depression trials and extensively validated. 27 The 21 items are rated on a four-point severity scale (0–3) and are summed to give a total score (range 0–63 points). A higher score on the BDI-II denotes more severe depression.

Secondary outcomes

The BDI-II was also completed at 24 weeks and at 12 months to assess the longer-term effect of the intervention. Other outcome measures included at the 12- and 24-week and the 12-month follow-up assessments are listed below:

1. treatment response, measured as an improvement of ≥ 50% in BDI-II score at 12 weeks compared with baseline

2. the rate of remission of symptoms, defined as a BDI-II score of < 10 points at 12 weeks

3. depression, as measured using the PHQ-9

4. change in anxiety symptoms (measured using the GAD-7) at 12 weeks

5. all of the above outcomes at 24 weeks and 12 months

6. antidepressant use and adherence (using the Morisky questionnaire and additional questions)

7. QoL (measured using the EQ-5D-5L)

8. social and physical functioning (measured using the SF-12) at 12 and 24 weeks and 12 months

9. AEs, including any new symptoms or worsening of existing symptoms, consultations for a documented deterioration in illness and SAEs (self-reported or from a review of primary care notes) and side effects (measured using the ASEC) at 12 weeks and 12 months

10. cost-effectiveness from the perspectives of the NHS, patients and society (using self-report questionnaires at 12 and 24 weeks and at 12 months and using primary care practice data on consultations, services and prescriptions over the 12-month trial period).

See Appendix 1, Table 26 for a full schedule of the questionnaires used.

Handling missing items

For the BDI-II, PHQ-9 and GAD-7, the research team dealt with any missing data at an individual item level by adopting the following rule: if ≥ 10% of the items were incomplete, the data collected on that measure for that participant were disregarded. However, if < 10% of the items on a particular measure were missing, missing item(s) were imputed using the mean of the remaining items (rounded to an integer). Therefore, when an individual had completed 19 or 20 items for the primary outcome measure (BDI-II), the remaining one or two items were imputed. For all other measures (PHQ-9 and GAD-7), the 10% rule meant that only a single item would be imputed. Data were complete for the majority of the sample. The scoring manuals for the SF-12 and EQ-5D-5L, which require the application of complex scoring algorithms, indicated that, if any item was missing, the scale score should not be calculated.

A number of sensitivity analyses were conducted to assess the impact of missing primary outcome data on the main findings.

Packaging, labelling and dispensing

The labelling of medication packs was MHRA approved and conformed to Annexe 13 of good manufacturing practice standards and Article 13.3 of Directive 2001/20/EC. 38 Each medication pack had a medicine identification number, randomly generated to ensure that mirtazapine and placebo medicine packs were indistinguishable and, thus, maintain allocation concealment. The random numbers were generated by the BRTC and provided to the manufacturer, who used them to form the identifiers.

Sharp Clinical Services (UK) Ltd (Crickhowell, UK) provided qualified person services and distribution and project management. It shipped labelled and numbered packages to the central trial pharmacy (UHB Clinical Trials Pharmacy), where the trial medication was stored under controlled conditions. Storage was secure and there were delegation logs for access, for which the trial pharmacy took responsibility. The trial pharmacy dispensed individual participant packs and oversaw the packaging and posting of those packs. Participant packs contained no more than an 8-week supply of the trial medication and were posted by recorded delivery. All deliveries were logged to ensure drug accountability. The trial medication was shipped and stored in conditions in line with the manufacturer’s stability data.

Concomitant medication

Trial-stopping rules

The trial could be prematurely discontinued by the sponsor, chief investigator, regulatory authority or funder on the basis of new safety information or for other reasons given by the Data Monitoring and Ethics Committee (DMEC) or Trial Steering Committee (TSC).

The trial could be prematurely discontinued because of a lack of recruitment or on advice from the TSC. If the trial was to be prematurely discontinued, active participants would have to be informed and no further participant data would have been collected.

Original sample size calculation

The primary outcome was BDI-II score, reported as a continuous variable. It is difficult to estimate a clinically important difference in BDI-II score, although the NICE guideline panel for the first depression guideline39 suggests that this corresponds to about 3 points [standard deviation (SD) 0.35 points] on the Hamilton Depression Rating Scale (HDRS)40 for non-treatment-resistant patients and 2 points for those who are treatment resistant. The equivalent difference to 3 HDRS points for the BDI-II total score would be 3–4 points (SD 10–12 points in CoBalT), which is also a clinically important difference from the patient perspective. 41 With 200 participants in each group, we would have 91% power to detect a difference of 0.33 SDs at the 5% level. Allowing for a 15% loss to follow-up at 12 weeks, we wanted to recruit 472 patients.

For our secondary outcome, response rate, defined as a 50% reduction in symptoms using the BDI-II score, 200 patients in each group would yield 90% power to detect a difference of between 30% and 46% in response, or an odds ratio (OR) of 2, at a two-sided 5% significance level. We therefore aimed to recruit 120 patients from 24 general practices at each of the four recruiting centres.

Baseline data analyses

Baseline sociodemographic and clinical characteristics were described by treatment group to ascertain any marked imbalances and inform any additional adjustment of the primary and secondary analyses as appropriate. Continuous variables were summarised using the mean and SD [or median and interquartile range (IQR) if the distribution was skewed] and categorical data were summarised as frequencies and proportions. All antidepressant medication use at baseline was reported by class, name, dose and duration of treatment.

Stratification, minimisation and sociodemographic variables and measures of depression and treatment preference, assessed at baseline, were explored for associations with ‘missingness’ of the primary outcome measure at 12 and 24 weeks and 12 months. Comparisons were made using chi-squared tests for categorical variables and t-tests and Mann–Whitney U-tests for continuous variables.

Primary analysis

Analysis and reporting were in line with CONSORT guidelines,35 with the primary analyses being conducted on an intention-to-treat (ITT) basis. The primary analysis was the BDI-II score at 12 weeks post randomisation, measured as a continuous variable. Linear regression modelling was used to compare the randomised groups, adjusting for stratification and minimisation baseline variables and baseline BDI-II score.

Analyses of secondary outcomes were conducted by adjusting for the baseline measure of the outcome variable and stratification and minimisation variables. Linear regression was used for continuous outcomes and logistic regression models were used for binary outcomes.

Secondary analyses

Secondary analyses of the primary and all secondary outcomes included additional adjustment for variables demonstrating a marked imbalance at baseline (ascertained using descriptive statistics).

In all analyses, we have presented regression coefficients (or ORs for binary outcomes), with 95% CIs and p-values.

Subgroup analyses

We conducted a number of prespecified analyses to investigate any differential effects in subgroups defined by a number of factors. These were carried out by introducing appropriate interaction terms in the regression models. We carried out these analyses by baseline depression severity (BDI-II) and a multilevel measure of degree of treatment resistance based on duration of symptoms and prior treatment with antidepressants.

Sensitivity analyses

In all tables, missing data are indicated in the footnotes. For our analysis of the primary outcome, we investigated the influence of missing data using sensitivity analyses that made different assumptions: ‘best’- and ‘worst’-case scenarios and multiple imputation by chained equation (MICE) to impute missing data. 42,43 When using MICE, 25 data sets were generated and 10 switching procedures undertaken. The imputation model included all variables predictive of missingness, together with all of the variables used in the primary analysis. Results from the ITT analyses of complete cases are presented alongside these analyses.

Per-protocol analyses were conducted at 12 weeks and at 12 months for the primary outcome. As these are likely to be biased, we also adopted a complier-average causal effect (CACE)44 approach.

A further sensitivity analysis using CACE methods was conducted at 24 weeks and 12 months. If we define ‘compliers’ as those who had continued taking their trial medication up until 12 weeks, we could then estimate the effect of completing a 12-week course of mirtazapine on depression outcomes at the later follow-up points (24 weeks and 12 months).

Definitions

All adverse events

All AEs were reported by the chief investigator from the time that a signed and dated informed consent form was obtained until completion of the last trial-related procedure (collection of follow-up data 12 months after randomisation). Those occurrences meeting the definition of SAEs were reported using the SAE form (see Appendix 1), including SAEs spontaneously reported to the investigator within 30 days after the participant had completed the trial (including post-trial follow-up). UHB, on behalf of the sponsor, would evaluate any safety information that was spontaneously reported by a chief investigator beyond the time frame specified in the protocol. All AEs, regardless of seriousness, severity or presumed relationship to the trial drug, were recorded in the source document and the CRF, together with any measures taken. All PIs recorded in the CRF their opinion concerning the relationship of the AE to the trial therapy. UHB, on behalf of the sponsor, assumed responsibility for appropriate reporting of AEs to the regulatory authorities.

Serious adverse events

All SAEs were reported to UHB (who monitored SAEs on behalf of the sponsor) and the relevant PI by a delegated member of the research team within 24 hours of their knowledge of the event. The chief investigator and trial manager were also informed. All SAEs that were not resolved by the end of the trial (i.e. by the end of the primary care notes review follow-up period), or that were not resolved on discontinuation of the participant’s participation in the trial, were followed until (1) the event resolved/stabilised/returned to baseline, (2) the event could be attributed to other factors unrelated to the trial or (3) it became unlikely that additional information could be obtained.

The death of a participant was considered a SAE, as was any event requiring hospitalisation (or prolongation of hospitalisation) that occurred during the course of a participant’s participation. Exceptions to this were hospitalisations for:

• social reasons, in the absence of an AE

• in-clinic protocol measures

• surgery or procedure planned before entry into the trial (this was documented on the CRF).

Suspected unexpected serious adverse reaction

All relevant information about a SUSAR that occurred during the course of the trial was reported to the MHRA and the relevant ethics committee by UHB, on behalf of the sponsor, as soon as possible (fatal or life-threatening SUSARs were reported within 7 days; those that were not fatal or life-threatening were reported within 15 days).

The expectedness of an AE was determined by whether or not it was listed in the Summary of Medicinal Product Characteristics (SMPC), the BNF and/or the study protocol.

Expected AEs and adverse reactions are presented in Appendix 1, Box 2.

Direct access to source data/documents

The PIs and trial site teams allowed monitors (from UHB, on behalf of the sponsor), persons responsible for the audit, representatives of the REC and representatives of the regulatory authorities to have direct access to source data/documents. This was reflected in the PIS. Trial monitoring was undertaken on behalf of the sponsor by UHB using their monitoring SOP.

Before the trial

The PIs and trial sites allowed the monitors to visit the sites and facilities where the trial took place to ensure compliance with the protocol requirements. The University of Bristol’s Green Light procedure was implemented in each of the other collaborating centres in order to document preparedness to conduct recruitment locally. A monitoring plan was agreed prior to commencement of the trial.

During the trial

The PIs allowed the monitors and/or the sponsor to:

• inspect the sites, the facilities and the material used for the trial

• meet all members of their team involved in the trial

• consult all of the documents relevant to the trial

• check that the CRFs had been filled out correctly

• directly access source documents to compare data therein with the data in the CRFs

• verify that the trial was carried out in compliance with the protocol and local regulatory requirements

• carry out trial monitoring at regular intervals, depending on the recruitment rate and arranged between the chief investigator and monitor.

All information dealt with during these visits was treated as strictly confidential.

Quality assurance during the trial

We employed standard strategies to ensure the quality of the data; for example, a random sample of 20% of the CRFs was checked by the trial research team against entries in the database. Recruiting sites were asked to perform a self-audit on all entries and provide a report to the Bristol trial centre (who would report to the trial sponsor). A 10% sample audit was conducted by the UHB monitoring team.

The content of the database was validated at two stages, as follows:

1. At the data-entry stage, validation rules were set to run on submission of data in order to direct researchers to fields that required completion, should any essential fields be missed, and to flag up anomalous or incomplete entries so that researchers could correct data prior to final submission of the electronic CRF.

2. Management information regarding data quality and completeness at centre, site and patient level was generated from data within the trial database and used by the trial manager to inform the implementation and monitoring of the trial.

Standard operating procedures were developed to address each aspect of quality control and the quality assurance procedures.

Extension to the trial in 2015

We were obliged to request additional funding for the trial to meet the recruitment target set out in the trial protocol. At the end of September 2014, we had recruited and randomised 200 participants. The recruitment period as set out in the original application was to have ended on 28 February 2015. The final target was 470 participants and it was not likely that we would achieve this within the designated time frame. Recruitment rates during the study had been good and to target. However, there were delays in set-up and in obtaining the necessary permissions from regulatory bodies and finalising contracts with the supplier of the trial medication. An extension of 7 months was awarded and we recruited 480 participants in the revised timescale.

Elective unblinding after the primary outcome at 12 weeks

In the initial study design we proposed that participants would remain blind to allocation for 52 weeks. However, we were advised by the REC that this was not ethically acceptable and that we should offer elective unblinding after the 12-week primary outcome. Therefore, after the 12-week follow-up and data collection (the primary outcome), participants who wished to continue in the study had the following two options:

1. to continue with the trial medication, blind to allocation, with the understanding that they could be unblinded at any time if they wished

2. to be unblinded to their allocation, which would mean that they would no longer receive the trial medication but would remain in the study.

The randomised blinded design was, therefore, preserved intact up until the primary outcome at 12 weeks. However, after that point, participants could discover whether they had been taking placebo or the active treatment. Those who were unblinded then had the option to discontinue their treatment or to consult their GP and request treatment with mirtazapine. Therefore, in addition to the two blinded groups who continued to receive the trial medication, there were now four unblinded groups, as follows, all of whom were included in the remaining follow-ups:

1. those who had been taking placebo up to 12 weeks and who decided to request mirtazapine from their GP

2. those who had been taking placebo up to 12 weeks and who did not request mirtazapine

3. those who had been taking mirtazapine up to 12 weeks and who decided to continue with the drug, now supplied by their GP

4. those who had been taking mirtazapine up to 12 weeks and who decided not to continue with this drug.

Search of general practice computerised records to identify patients being treated for depression

There were 26,898 patients aged ≥ 18 years who were currently being prescribed antidepressants for ≥ 6 weeks at an adequate dose identified through searches of the general practice computerised records. A further 166 individuals were referred to the study directly by their GP. GPs excluded 7932 individuals who were ineligible. Fifty patients were invited as a result of an administrative error and responded, of whom 15 opted in to participate in the trial.

In total, 19,016 patients were mailed a letter of invitation to participate in the study, asking for their permission for the research team to contact them (although 50 of these were invited in error having been excluded by their GPs). GPs invited 166 patients to take part during consultation. Of these, 2180 patients (11%) responded to the letter of invitation and opted in, 4702 (25%) responded and opted out and 12,134 (64%) did not respond to the invitation (see Figure 2).

When available, information on age and sex were recorded from GP records and are summarised in Appendix 2, Tables 28 and 29.

Assessment of depressive symptoms and adherence to antidepressant medication

A screening questionnaire was mailed to 2299 potential participants, of whom 296 individuals did not complete the assessment screen and 60 declined. Of those who returned their questionnaire, 827 were not eligible; mostly this was because their BDI-II score was below the threshold of 14 points, they were not adhering to their antidepressant medication or they did not meet the criteria for an adequate dose/duration of treatment. Other reasons for ineligibility at this stage were pregnancy, breastfeeding, treatment with mirtazapine or amitriptyline and being part of another Clinical Trial of an Investigational Medicinal Product (CTIMP). In total, 1116 patients were eligible at the postal screen stage. However, 105 decided not to participate, 71 were excluded at the pre-baseline check and 52 were excluded by their GP as not suitable. A total of 137 were excluded because their checks were not completed before the closure of recruitment.

When available, information on age and sex were recorded from GP records and are summarised in Appendix 2, Table 30.

Baseline assessment of eligibility to participate in the randomised controlled trial

A total of 751 potential participants attended a face-to-face baseline appointment with a researcher to establish their eligibility to participate in the MIR trial and obtain written informed consent. In total, 270 individuals were ineligible to participate. The majority (n = 268) were ineligible because they did not meet the ICD-10 criteria for depression or failed to meet the BDI-II threshold. In total, 480 individuals were eligible to participate, gave written informed consent and were randomised.

The available information on age, sex and socioeconomic status for those who did and did not attend a baseline assessment is summarised in Table 3.

Summary of recruitment by centre

There were 480 participants recruited to the study from the four study centres. Bristol recruited the most participants (n = 177), Exeter recruited 122 participants, Hull/York recruited 99 participants and Manchester/Keele recruited 82 participants. A summary of recruitment by centre is presented in Appendix 2, Table 31.

A summary of the overall predicted and actual recruitment is provided in Figure 3. Owing to delays in start-up, a 7-month extension was needed to meet the final target.

FIGURE 3.

Predicted and actual recruitment.

Losses to follow-up

Follow-up rates were slightly higher in the group allocated to usual treatment and placebo and the difference increased after 12 weeks (see Figure 4). At 12 weeks, 88.8% of participants allocated to usual care and mirtazapine and 90.8% of participants allocated to usual care and placebo were contacted for follow-up. By 24 weeks, 81.3% of participants allocated to usual care and mirtazapine were contacted compared with 86.6% of participants allocated to usual care and placebo. At 52 weeks, these proportions were 79.3% and 83.3%, respectively.

Missing data

The pattern of missing data was explored by identifying variables recorded at baseline that were associated with ‘missingness’ of the primary outcome at 12 weeks’ follow-up. Table 5 presents the frequency and proportion of missing data by baseline characteristic.

At each time point, when participants were successfully contacted, few did not complete sufficient components of the BDI-II questionnaire to calculate a score (see Figure 4).

The pattern of missing data was further explored by identifying variables recorded at baseline that were associated with ‘missingness’ of the BDI-II score at 24 weeks and 12 months, as presented in Appendix 2, Tables 37 and 38.

Depressive symptoms at 12 weeks measured using the continuous Beck Depression Inventory-II score

The mean BDI-II score at 12 weeks was 18.0 points (SD 12.3 points) in participants randomised to the usual care and mirtazapine group compared with 19.7 points (SD 12.4 points) in those randomised to usual care and placebo (Table 6). Higher scores indicate more severe depression, with a maximum possible score of 63 points.

In the primary ITT analysis, participants in the intervention group had a slightly lower BDI-II score than those in the placebo group (see Table 6). The difference – corrected for baseline BDI-II score, centre, baseline BDI-II score terciles, sex and whether or not the patient was receiving psychological therapy at baseline – was less than the minimum clinically relevant difference that we sought to detect (0.33 SDs, which in this case was 4.1 points on the BDI-II) and the confidence interval (CI) overlapped zero, which includes the possibility that there is no difference between the two treatment groups. Further adjustment for variables that suggested some imbalance between the treatment groups at baseline (history of depression, duration of current course of antidepressants and suicidal ideation) did not meaningfully change the effect estimate (difference –2.12 points, 95% CI –4.25 to 0.02 points; p = 0.052).

Beck Depression Inventory-II as a continuous score at 24 weeks’ and 12 months’ follow-up

Table 8 summarises the means and differences in mean BDI-II sores at 24 weeks’ and 12 months’ follow-up. In both groups the mean BDI-II scores declined over time. At 24 weeks’ follow-up, those in the intervention group had a BDI-II score that was 0.85 points lower than those in the placebo group after adjustment for baseline BDI-II score and stratification and minimisation variables, with a 95% CI of –3.12 to 1.43 points. At 12 months’ follow-up, the difference between the two groups was 0.17 points (95% CI –2.13 to 2.46 points). In both cases the CIs overlapped zero, suggesting no evidence of a difference between the groups. Additional adjustment for the variables that were imbalanced at baseline increased the size of the differences somewhat but the 95% CIs continued to overlap zero (24 weeks: difference –1.26 points, 95% CI –3.57 to 1.05 points, p = 0.285; 12 months: difference –0.29 points, 95% CI –2.61 to 2.03 points, p = 0.807).

Our prespecified statistical analysis plan included a repeated-measures analysis incorporating outcomes at 12 and 24 weeks and 12 months post randomisation to examine whether any treatment effects were sustained, diminished or emerged later. This analysis was not conducted, given that differences between the treatment groups at these time points showed no evidence of an important difference.

‘Response’ to treatment at 12 and 24 weeks’ and 12 months’ follow-up

Table 9 summarises the percentages and ORs of ‘response’ to treatment (defined as a reduction in depressive symptoms on the BDI-II of ≥ 50% relative to baseline) at 12 and 24 weeks’ and 12 months’ follow-up. At each of the time points, the intervention group had an increased odds of response compared with the usual care group [OR 1.39, 95% CI 0.94 to 2.07; number needed to treat (NNT) = 12]. However, the CIs at each time point overlapped 1. Additional adjustment for the variables that were imbalanced at baseline did not make any meaningful difference to the observed difference at 12 weeks’, 24 weeks’ or 12 months’ follow-up (12 weeks: OR 1.45, 95% CI 0.97 to 2.16, p = 0.074; 24 weeks: OR 1.04, 95% CI 0.69 to 1.57, p = 0.842; 12 months: OR 1.05, 95% CI 0.69 to 1.58, p = 0.834).

‘Remission’ of depression symptoms at 12 and 24 weeks’ and 12 months’ follow-up

Table 10 summarises the percentages and ORs of ‘remission’ (defined as having a BDI-II score of < 10) at 12 and 24 weeks’ and 12 months’ follow-up. Those in the intervention group had an increased odds of ‘remission’ at 12 weeks compared with those in the placebo group and the CI overlapped 1 (OR 1.29, 95% CI 0.82 to 2.02; NNT = 18). Similar results were seen at 24 weeks and by 12 months the OR approached 1. The ORs tended to be slightly increased after adjustment for those variables that were imbalanced between treatment groups at baseline, but the CIs continued to overlap 1 (12 weeks: OR 1.39, 95% CI 0.88 to 2.21, p = 0.157; 24 weeks: OR 1.31, 95% CI 0.83 to 2.09, p = 0.249; 12 months: OR 1.02, 95% CI 0.65 to 1.60, p = 0.940).

Anxiety symptoms at 12 and 24 weeks’ and 12 months’ follow-up

Anxiety symptoms were assessed at baseline and 12 and 24 weeks’ and 12 months’ follow-up using the GAD-7 questionnaire and were analysed as a continuous score. Higher scores indicate more severe symptoms of anxiety, with a maximum possible score of 21. Table 11 summarises the mean and differences in mean GAD-7 scores at 12 and 24 weeks’ and 12 months’ follow-up. At 12 weeks, after correction for baseline GAD-7 score and minimisation and stratification variables, those in the intervention group had a mean GAD-7 score that was 0.98 points lower (less anxious) than those in the placebo group, with a 95% CI ranging from a 1.93-point decrease to a 0.03-point decrease. Differences between the two treatment groups were smaller at subsequent time points and CIs overlapped zero. Additional adjustment for the variables that were imbalanced at baseline did not alter the results greatly (12 weeks: difference –1.00 points, 95% CI –1.97 to –0.02 points, p = 0.045; 24 weeks: difference –0.71 points, 95% CI –1.73 to 0.31 points, p = 0.172; 12 months: difference –0.22 points, 95% CI –1.31 to 0.86 points, p = 0.685).

Adherence to antidepressants at 12 weeks’ follow-up

Adherence to the study medication at 12 weeks’ follow-up was assessed using information on participant-reported breaks in treatment and the Morisky questionnaire. Table 12 summarises the percentages and ORs of adherence to the study medication at 12 weeks. Although adherence was > 70% in both groups, those in the intervention group had a 0.55-fold odds of adherence to the study medication at 12 weeks compared with those in the placebo group. Adjustment for variables that were imbalanced at baseline did not alter this result (OR 0.56, 95% CI 0.34 to 0.92; p = 0.021).

Although adherence at 24 weeks and 12 months formed part of our planned secondary outcome analyses, as patients were offered the chance of unblinding at 12 weeks while continuing to be followed up, the results of these analyses became impossible to interpret and, therefore, are not presented here.

Quality of life at 12 and 24 weeks’ and 12 months’ follow-up

Quality of life was measured using the EQ-5D-5L questionnaire as well as the SF-12 questionnaire, from which aggregate scores for social and physical functioning were calculated.

Adverse events related to the trial medication at 12 weeks’ and 12 months’ follow-up

Depression symptoms assessed using the Patient Health Questionnaire-9 items at 12 weeks’ follow-up

The primary outcome and many of the secondary outcomes were derived from the BDI-II questionnaire and, to explore the consistency of these results, responses to the PHQ-9 at 12 weeks’ follow-up were reviewed in a post hoc analysis. The PHQ-9 scores nine symptoms of depression yielding a maximum score of 27 points; higher scores indicate more severe symptoms of depression.

Table 17 summarises the mean and difference in mean PHQ-9 sores at 12 weeks. Participants in the intervention group had a mean PHQ-9 score at 12 weeks that was 1.05 points lower than that in the placebo group after adjustment for baseline PHQ-9 score and stratification and minimisation variables, with a 95% CI of –2.14 to 0.04 points. The direction of association was the same as that observed for the BDI-II score outcomes. Additional adjustment for the variables that were imbalanced at baseline made the difference slightly larger (difference –1.24 points, 95% CI –2.34 to –0.13 points; p = 0.029).

Aim

The aim of the economic evaluation was to assess the cost-effectiveness of mirtazapine plus usual care compared with placebo plus usual care in patients with TRD. The economic evaluation was carried out using data collected as part of the MIR RCT.

The intervention

Mirtazapine is a well-established inexpensive treatment for depression, at £0.04 per tablet (either 15-mg or 30-mg strength),48 with a typical dose being one tablet per day. If it were to show a meaningful improvement in QoL [here measured in quality-adjusted life-years (QALYs)], there is, ex ante, a high probability that it will be cost-effective, all other things being equal.

Form of analysis

The economic evaluation was a within-trial analysis, based on the data collected during the 12-month trial period. All costs that were incurred in the 12 months following randomisation were included. The perspective for the primary analysis was that of the NHS and personal social services (PSS). Personal costs to patients and productivity costs for patients and carers who missed work were included in a secondary analysis from the societal perspective.

The primary analysis was a cost–utility analysis of the cost per QALY when comparing the two groups. Results are also presented for a cost-effectiveness analysis for the primary trial outcome of BDI-II score at 12 weeks post randomisation.

Resource-use categories

The analysis was based on the costs associated with caring for all patients from randomisation until 12 months. The following resource use data were included in the analysis.

Data collection

Unit costs

Missing data

When data were missing, they were assumed to be missing at random and multiple imputation (MI) was used. MI was conducted in Stata 14. The Stata MI commands were used to generate 50 imputed data sets using chained equations, which were combined for analysis using Rubin’s rule as specified as part of the -mi estimate- command. The imputation model for the analysis included the stratification and minimisation factors used for randomisation (site, baseline BDI-II score, whether or not a participant was undergoing talking therapy), age, sex, initial treatment allocation, continuous baseline EQ-5D-5L and BDI-II scores and a range of cost and quality of life variables.

Missing data were imputed for:

• EQ-5D-5L index scores at baseline and 12, 24 and 52 weeks

• mean patient-reported resource use costs at 12, 24 and 52 weeks

• mean costs of GP consultations and prescriptions.

Uncertainty arising from patient variation

Uncertainty arising from individual patient variability is reported as standard errors (SEs) for the point estimates of the mean costs and QALYs. Uncertainty in the incremental cost-effectiveness ratio reported for complete-case analyses is reported as 95% CIs, based on 1000 bootstrap replications. We also report incremental net monetary benefit (NMB) with CIs for a range of willingness-to-pay (WTP) values and cost-effectiveness acceptability curves (CEACs).

Discounting

Costs and outcomes were not discounted as the study and analysis was limited to a 12-month period. All costs were estimated in 2016 prices and have been inflated as necessary using the PSSRU Pay and Prices inflation index. 55

Resource use and costs

Cost–utility

Cost-effectiveness analyses are reported, relating the costs of each strategy to change in:

• QALYs at 12 weeks and 12 months post randomisation

• BDI-II scores at 12 weeks and 12 months post randomisation.

The primary outcome of interest for the economic evaluation was the incremental cost per QALY at 12 months’ follow-up. In keeping with the primary trial outcome, the primary analysis for the economic evaluation was based on complete cases only. A secondary analysis based on the imputed data is also reported.

Primary analyses

12-week cost–utility

No clinically important differences in QALYs or costs were observed between the two groups at 12 weeks’ follow-up (Table 21). At the primary outcome end point of 12 weeks, participants in the active treatment arm reported QALYs of 0.163 and participants in the control arm reported QALYs of 0.162, a difference of 0.002 QALYs (95% CI –0.002 to 0.005 QALYs). This difference is not meaningful. The active treatment arm had higher costs (£65) than the control arm (£63). The difference in costs was small (£2, 95% CI –£27 to £31) and is also not meaningful.

TABLE 21 - Incremental cost-effectiveness ratios, NHS and PSS perspective, complete-case analysis

ITT analysis adjusted for baseline EQ-5D-5L index score, BDI-II score and the stratification and other minimisation variables, bootstrapped SEs.

ITT analysis adjusted for baseline BDI-II score and the stratification and other minimisation variables, based on complete cases, bootstrapped SEs.

Primary analyses	Group
	Intervention		Control		Difference
	Costs (SE), £	QALYs (SE)	Costs (SE), £	QALYs (SE)	Costs (SE), £	QALYs (SE)
QALYsa
12 weeks	67 (11)	0.163 (0.001)	63 (10)	0.162 (0.001)	4 (15)	0.002 (0.002)
52 weeks	261 (52)	0.734 (0.009)	192 (49)	0.724 (0.009)	69 (71)	0.009 (0.013)
BDI-II scoreb
	Costs (SE), £	BDI-II score (points) (SE)	Costs (SE), £	BDI-II score (points) (SE)	Costs (SE), £	BDI-II score (points) (SE)
12 weeks	67 (11)	17.804 (0.748)	63 (10)	19.822 (0.743)	4 (15)	–2.128 (1.055)
52 weeks	261 (52)	16.7 0.824	192 (49)	16.8 0.805	69 (71)	–0.166 (1.149)

Assuming a WTP threshold of £20,000 per QALY, the expected NMB of the active treatment over placebo is £144, but with 95% CIs ranging from –£1413 to £1701 (Table 22).

TABLE 22 - Incremental NMB of mirtazapine compared with placeboa

iNMB, incremental net monetary benefit.

ITT analysis adjusted for baseline EQ-5D-5L score, baseline BDI-II score and stratification and other minimisation variables.

Analysis	WTP threshold per QALY (95% CI),a £
	£10,000	£20,000	£30,000	£50,000
Complete-case analysis
NHS and PSS perspective
iNMB at 52 weeks	25 (–267 to 137)	118 (–419 to 655)	211 (–582 to 1004)	398 (–914 to 1709)
iNMB at 12 weeks	72 (–706 to 851)	144 (–1413 to 1701)	216 (–2119 to 2552)	360 (–3532 to 4253)
Societal perspective
iNMB at 52 weeks	51,926 (–160,178 to 264,030)	103,852 (–320,356 to 528,060)	155,778 (–480,534 to 792,090)	259,630 (–800,890 to 1,320,149)
iNMB at 12 weeks	2419 (–5475 to 10,313)	4839 (–10,950 to 20,627)	7258 (–16,425 to 30,940)	12,096 (–27,374 to 51,567)
Imputed data analysis
NHS and PSS perspective
iNMB at 52 weeks	15 (–32 to 62)	34 (–48 to 116)	54 (–66 to 174)	92 (–106 to 290)
iNMB at 12 weeks	25 (–270 to 321)	118 (–417 to 654)	212 (–575 to 999)	398 (–898 to 1695)
Societal perspective
iNMB at 52 weeks	–123 (–397 to 152)	–103 (–386 to 180)	–84 (–381 to 213)	–45 (–382 to 292)
iNMB at 12 weeks	–676 (–1443 to 90)	–583 (–1478 to 312)	–490 (–1560 to 581)	–303 (–1798 to 1192)

The CEAC (see Figure 5) illustrates the proportion of times that each intervention is cost-effective based on the bootstrapped samples across a range of WTP estimates. Figure 5 illustrates that, based on the bootstrapped simulations, at 12 weeks mirtazapine was more likely to have the highest NMB. At a WTP threshold of £20,000 per QALY, mirtazapine was the more cost-effective option in 77% of the simulations, rising to 79% at a WTP threshold of £30,000 per QALY.

FIGURE 5.

Cost-effectiveness acceptability curve: probability of highest NMB at 12 weeks across a range of WTP estimates per QALY.

A CEAC based on the WTP for an additional 1-point improvement in BDI-II score at 12 weeks is presented in Figure 6. From this, we can see that mirtazapine is likely to have a higher expected NMB depending on the WTP for a 1-point change in BDI-II score. The shape of this curve is driven by the fact that mirtazapine provides a small improvement in BDI-II score relative to placebo. In most of the bootstrap simulations, mirtazapine is both more costly and more effective.

FIGURE 6.

Cost-effectiveness acceptability curve: probability of highest NMB at 12 weeks across a range of WTP estimates for a unit change in BDI-II score.

Fifty-two-week cost–utility

At 52 weeks, participants in the treatment arm had an incremental gain of 0.009 QALYs (95% CI –0.016 to 0.035 QALYs) and an incremental cost of £69 (95% CI –£74 to £206) compared with participants in the placebo arm (see Table 21). The NMB at 52 weeks is positive across the reported levels of WTP for an additional QALY – at a WTP of £20,000 per QALY, the NMB is £118 (95% CI –£419 to £655) (see Table 22). Again, consideration of the 95% CI shows a substantial degree of uncertainty around the results.

The CEAC at 52 weeks (Figure 7) shows similar results to that at 12 weeks. At a WTP of £20,000 per QALY, active treatment was the more cost-effective option in 69% of the simulations. At a WTP of £30,000 per QALY, it was more cost-effective in 71% of simulations.

FIGURE 7.

Cost-effectiveness acceptability curve: probability of highest NMB at 52 weeks across a range of WTP estimates per QALY.

Figure 8 shows the cost-effectiveness planes for the incremental cost per QALY at 12 and 52 weeks. The overall uncertainty in the results is clear.

FIGURE 8.

Cost-effectiveness planes from a NHS and PSS perspective: incremental cost per QALY. (a) 12 weeks; and (b) 52 weeks.

Secondary analyses

Ethics approval

Ethics approval was given by South East Wales Research Ethics Committee Panel C (reference: 12/WA/0352).

Design

We conducted semistructured interviews with people who were invited to participate in the trial but who declined, trial participants (who completed the trial and who withdrew) and GPs, to generate in-depth data on reasons for declining to participate, perspectives of those participating in the trial, reasons for completion of the trial or stopping the study medication and views and experiences of GPs about managing people with depression, with a focus on prescribing.

Sampling and recruitment to the qualitative study

General practitioners

General practitioners were recruited from practices participating in the MIR trial. Sampling of GP practices was purposive and selection was based on the geographical location of the practice and the sex and experience of the GPs, to give maximum variation in the available population. Sampling was also contingent on whether or not the practice had patients who had been randomised to the MIR trial.

General practitioners and practice managers were initially contacted by e-mail, with the information sheet provided as an attachment, inviting them to discuss their experiences of managing people with depression and their views of using a combination of antidepressants. If there was no response to the e-mail, telephone calls were made to the practices. If a GP expressed interest in participating in an interview, the research assistant arranged a telephone interview at a time that was convenient to the GP.

Consent

Informed consent to participate in the qualitative study was obtained from all participants in the study. People invited to take part in interviews about their reasons for declining to participate in the MIR trial were provided with a short questionnaire during the trial recruitment phase on which to indicate consent to be contacted about their reasons for non-participation. When these people were contacted by telephone, they were asked to confirm that they had consented to take part in a telephone interview and for their verbal consent to record the telephone interview. Participants were reminded about their rights as a participant to confidentiality and to withdraw their consent, at the beginning and the end of the conversation, respectively. This procedure for gaining consent to conduct and record telephone interviews was also followed when contacting GPs for telephone interviews for the study.

Patients participating in the trial were sent an information sheet in the post after their 12-week outcome measure had been taken by a researcher, providing details about participation in the qualitative study. Those participants who agreed to an interview were asked to sign a written consent form to indicate their consent to participate and the researcher confirmed this consent verbally before commencing the interview. The majority of interviews with patient participants were held face-to-face, in their own homes.

Data generation

Data were generated during semistructured interviews conducted either by telephone (with patients who declined participation in the trial and GPs) or face to face (with trial participants who completed or withdrew from the trial medication).

Topic guides for the groups of interview participants (decliners, trial participants and GPs) were produced prior to the commencement of the first interviews (see Appendix 4). These topic guides were amended and adapted in response to the initial analysis of interviews and after discussion within the research team. This process was iterative throughout the period of data generation to better explore the concepts and themes that emerged as being important to the interviewees and required more in-depth examination during interviews with subsequent participants.

Data generated from each of these data sets were collected during each interview by the qualitative research assistant, using a digital audio recorder. These audio files were downloaded to, and stored securely on, a secure network at the site where the qualitative research team was based (Keele University) and handled in accordance with the research institute’s SOPs. Audio files were transcribed verbatim by an authorised transcription company and transcripts were stored securely by the research assistant in accordance with SOPs.

Each interview was checked against the audio recording, cleaned and anonymised prior to analysis.

Analysis

Analysis of the data was conducted both independently and subsequently collectively by three members of the qualitative research team: the qualitative research assistant, a research fellow (Heather Burroughs) and the co-investigator with responsibility for managing the qualitative study (Carolyn Chew-Graham). The researchers were from different professional backgrounds (psychology, health services research and anthropology, and academic primary care), which increases the trustworthiness of the analysis. 78 Transcripts were read and re-read, themes arising during analysis and data collection were discussed within the team and the topic guides were modified as data collection and analysis progressed. The qualitative research team met regularly to discuss and agree coding, revisiting the data from each data set in an iterative manner to verify coding and themes generated from these codes. This process was repeated until it was agreed that each data set had reached saturation. Analysis was initially conducted within each data set and then comparison was carried out across the data sets. NVivo 10 (QSR International, Warrington, UK) was used to store data and aid analysis.

Patient and public involvement and engagement

The qualitative research team worked with a patient and public involvement and engagement (PPIE) group to discuss transcripts from the data sets. The aim was for the lay partners to add value to the analysis by bringing their own perspectives and identifying further areas for the researchers to explore in the interviews. 79 Four meetings were held with the PPIE group over the course of data generation.

Details of participants

A total of 60 interviews were conducted: 23 with ‘decliners’ (mean duration 11 minutes and 27 seconds), 23 with ‘completers’ and ‘withdrawers’ from the study medication (mean duration 47 minutes and 35 seconds) and 14 with GPs (mean duration 19 minutes and 31 seconds).

The 23 ‘decliners’ who we interviewed were distributed fairly evenly between the four centres, with seven from Bristol, six each from Exeter and Keele and four from Hull. Fifteen of the interviewees were female and eight were male, with an age range of 27–76 years. All were still blinded at the point of interview.

Illustrative data are presented with the sex and age of the participants given as identifiers in the case of trial participants (‘completers’ and ‘withdrawers’ from the study medication) and ‘decliners’. Data from GP transcripts are identified with numbers that reflect the chronological order in which the interviews were conducted.

Decliners

Table 24 lists the reasons that people gave for not wanting to participate in the MIR trial.

The most common reason for declining participation was not wanting to take part in a trial (49%). Thirty-six per cent of people invited indicated that they did not want to take mirtazapine; it is not known if this was because of prior experience of being prescribed this antidepressant. Not wanting to take more than one antidepressant was a reason given by 20% of people for declining to participate in the MIR trial. Interestingly, 17% of people suggested that they were not taking an antidepressant, even though they were being prescribed a SSRI or SNRI according to their GP records; 39% of people said that they planned to stop taking their current antidepressant.

Analysis of the ‘decliner’ data suggested that key themes were ‘the hard work of managing depression’, uncertainties about antidepressants, including questions over the value of a second antidepressant, and attainment and maintenance of a hard-won equilibrium.

Completers and withdrawers from the study medication

Analysis of data generated from interviews with participants who completed the trial or who withdrew from medication suggested four main themes, which resonated with the analysis of the ‘decliner’ data: (1) help-seeking at a point of crisis, (2) hoping for another option, (3) experiences of the trial, emphasising side effects, which led to (4) a dilemma of whether or not to continue the combination of medication.

General practitioner perspectives

The narratives of GPs about their management of people with depression suggested that it is complex, individualised and involves a good deal of effort when medication is used. GPs described the ‘hard work’ of managing people with depression, resonating with the patients’ perspectives. Observations about people presenting at a crisis point linked with the patient data sets.

Summary of findings

This embedded qualitative study within the MIR trial explored reasons why people approached to participate in the MIR trial declined, the experiences of people participating in the MIR trial and the perspectives of GPs about the role of a combination of two antidepressants for people with TRD.

The narratives of people who declined to participate in the MIR trial, but who were willing to be interviewed about this decision, and the participants in the trial all described the hard work involved in managing depression and the importance of a ‘crisis point’ in precipitation of help-seeking. The people who declined to participate in the MIR trial described feeling that they were at an equilibrium and feared that participating in the trial would disturb that hard-won equilibrium. Patients may be ‘resistant to change’, rather than the depression simply being ‘treatment resistant’.

Experiencing a crisis point seemed to be a motivator to make a change and the offer of trying a second antidepressant within a trial could act as that motivator. Some ‘decliners’ expressed uncertainties about the role of medication and feared that a second tablet would not help, or could not see the logic of a second medication.

The people who participated in the trial (whether they completed the study or withdrew from medication) described the constant hope that something would help their mood, which was something that the trial offered. Withdrawal from the study medication was sometimes put down to reported side effects, particularly weight gain. Of note, at the time of the interview, those people who withdrew did not know whether they had been taking placebo or mirtazapine. People who withdrew described reaching a point at which they had to balance improvement in their mood with side effects from the second agent and, when the balance tipped against side effects, they made the decision to withdraw from the medication.

The narratives of GP respondents reflected the patient data, describing the ‘hard work’ needed to manage people with depression and how ‘treatment resistance’ is about more than a poor response to an antidepressant: it is about a complex patient presenting the GP with a challenge in management. GPs described how patients often present at a crisis point and how the MIR trial offered an option to manage such patients. Some GPs reported confidence in combining antidepressants, but others did not. All expressed concern over the dangers of leaving people on two antidepressants, reflecting that, in their experience, people did remain on a SSRI without review or monitoring and with no attempt to withdraw the drug.

Comparison with previous literature

The most common reason for people declining participation in the MIR trial was not wanting to take part in a research study, as in Barnes et al. 76 But, in contrast to the study by Barnes et al.,76 in which the ‘decliners’ expressed negative feelings about the treatment offered in the trial, in our study ‘decliners’ expressed the view that they were at an equilibrium, which would be disturbed if they chose to participate in the trial, although a second antidepressant was not thought to be logical by a number of the ‘decliners’ we interviewed.

Hughes-Morley et al. 80 described how individuals declined participation because they judged themselves ineligible or not in need of the trial therapy. Similarly, our analysis suggests that the ‘decliners’ we interviewed were not in need of the trial intervention, the second antidepressant.

Trial participants described experiencing a crisis as a motivator to participate in the trial, which resonates with Schofield et al. ,81 who reported that participants described their first course of antidepressants as typically occurring when they had ‘hit rock bottom’, having exhausted all other possibilities. Schofield et al. 81 reported that participants described a period of experimentation in which it was usual to stop and restart medication, often several times. Ultimately, these recurring cycles lead to participants becoming more expert about their condition and better able to make an informed decision about medication. This expertise resonates with our analysis.

Schofield et al. 81 also reported that, for older people, there was often an acceptance that their condition, and medication use, would be long term. This acceptance might resonate with the concept of ‘equilibrium’ that we describe in this study.

In a meta-ethnography, Malpass et al. 75 outline the role of the GP in supporting decision-making about antidepressant use and facilitating concordant relationships with patients regarding antidepressant use. Not having this discussion at entry into the trial may account for some people declining to participate.

Malpass et al. 75 recommend that GPs are aware of the competing demands that patients experience at a decision-making juncture. We identified the importance of a ‘crisis’ in influencing a decision to try something new and a sense of equilibrium in resisting change.

Strengths and limitations

Exploring the reasons for non-participation in trials is still unusual; this study describes how interviews to determine such reasons can be incorporated into a trial recruitment procedure and illuminates the reasons people declined. It should be noted that people invited to participate in the MIR trial did not necessarily have TRD – they were people being prescribed one antidepressant in primary care. That 39% of respondents indicated that they were planning on stopping their antidepressant suggests that a good proportion of those invited to participate in the MIR trial may not have been depressed. None of the ‘decliners’ interviewed expressed any experience of or concerns about mirtazapine as a particular antidepressant, even though sampling included those who had given ‘I do not want to take mirtazapine’ as a reason for declining to participate in the MIR trial.

A limitation of the study is that people who responded to the decliner questionnaire, and the invitation to participate in a telephone interview about their reasons for declining to participate in the MIR trial, were a self-selected group, which may limit the generalisability of the findings. Perhaps people with stronger opinions might be more likely to respond. In addition, the telephone interviews were relatively short (mean duration < 12 minutes).

A strength of this study was involvement of the PPIE group in the analysis,79 although some challenges arose, including difficulties in recruiting a diverse range of members of the public to carry out the role, and the research team sometimes experienced difficulties keeping the PPIE members on track.

Implications

Exploring reasons for declining to participate in the MIR trial suggests that some people who are already taking one antidepressant may be reluctant to take a second. Potential participants were wary of possible side effects, but were also unconvinced of the logic behind such a combination. They considered that, if one antidepressant is not working, it was not plausible that a second would be effective. In addition, people described being in a state of equilibrium and reluctant to make a change, reflecting that this equilibrium is ‘hard won’ and they are unwilling to risk disturbing this. Understanding a patient’s view on medication is important for GPs when discussing antidepressants. The key finding of the importance of achieving an equilibrium is vital for researchers to understand and incorporate into recruitment strategies in future trials of treatments for depression.

A crisis seems to be the motivator to try something new, including taking a combination of antidepressants, and this is the point at which people often consult their GP for help. If a patient receives an invitation to participate in a trial when they are at such a point, they may be more willing to participate. Whether or not this affected the characteristics of patients recruited to the MIR trial is not known, but we might postulate that people who felt that their mood was particularly low and who were searching for something to help were the people recruited into the MIR trial. This does not necessarily mean that people recruited were more depressed, but their perceptions of their mood might have been that things were particularly difficult at the time that referral into the MIR trial was offered by their GP or the letter of invitation to participate was received.

Varied levels of confidence were reported by GPs in initiating prescribing of two antidepressants; some GPs reported that this had ‘crept into’ their practice following advice from a psychiatrist with previous patients. Other GPs reported that combining antidepressants was not something that they were comfortable with. All GPs, however, expressed concern over the risks of continued prescribing of two antidepressants, leaving people on a combination long term, and felt that guidance would be needed to ensure safe prescribing and withdrawal of one agent.

This qualitative study highlights the concern expressed by all GP respondents that people could then be left on two antidepressants, given that many already remain on one antidepressant long term and there is a reluctance from both patient and GP to make changes. This has implications for the education and training of GPs about the management of TRD, the importance of regular review and how to facilitate continuity of care for people with depression.

Comparison with non-pharmacological interventions for treatment-resistant depression

Although this is an extensive field and it is not practical to address it in a comprehensive way in this context, it is important to mention CoBalT. 21 This is a study that compared the addition of CBT to usual care for TRD, including the continuation of an antidepressant, and much of the design and recruitment strategy of the MIR trial is based on that of CoBalT. As a result, the populations in these two studies are more similar than in many TRD studies because of the wide variability in definitions of TRD in the literature, and a comparison is, therefore, of interest. In CoBalT, CBT was found to be more clinically effective and cost-effective than usual care and the OR of response in the intervention group was 3.26 (95% CI 2.10 to 5.06) at 24 weeks. However, participants in the usual care group in CoBalT were not blind to treatment allocation and were not offered any additional treatment, including placebo, and improved less than either group in the MIR trial. Although it is difficult to make a direct comparison between the two studies, it is important to note the relatively poor outcomes for TRD patients who are not offered further interventions.