Randomised controlled trial of Antiglucocorticoid augmentation (metyrapone) of antiDepressants in Depression (ADD Study)

Background

This task assesses the ability of a participant to identify each of the six basic facial emotions, and previous research has shown an impaired ability to identify emotions in depression, an abnormality that resolves with clinical improvement. 97 Early improvement in recognition of happy faces has been reported to predict response to antidepressant treatment. 98

Each emotion (happiness, sadness, anger, fear, surprise, disgust) was presented for 1.0 second at three intensities (30%, 50% and 70%) by four actors, together with 12 presentations of neutral expressions. The inter-trial interval was 4.5 seconds, participants pressed a labelled key to indicate their choice of emotion.

The data were analysed as accuracy (hit rate measured as the proportion of correct identifications of each emotion to the total number of responses to that emotion) and misattributions (false alarm rate measured as the proportion of misattributions of a specific emotion to the total number of responses to other emotions). RTs are not reported as there was no instruction to react rapidly, and the need to choose between seven response keys makes interpretation of the results unclear.

Patients and control subjects were compared using baseline data by an ANCOVA with Huynh–Feldt correction: emotion type was a within-subject factor, group was a between-subject factor, and age and sex were covariates, given evidence that they can influence face emotion recognition. 97 Illustrative post hoc analysis was by ANCOVA for each emotion separately. Uncorrected degrees of freedom are presented for clarity.

The effect of metyrapone treatment in patients was assessed using multiple ANCOVA for each emotion with the post-treatment value as the dependent variable and pretreatment value as the covariate.

Newcastle

We investigated levels of blood-oxygen-level-dependent (BOLD) signal changes in response to negative emotional stimuli in a subgroup of patients in the Newcastle centre using fMRI. Two emotional task paradigms were utilised, which are known to recruit both limbic and prefrontal brain regions. The facial emotion processing (FEP) task was based largely on work by Hariri et al. ,100 and included two emotional conditions [‘emotional matching’ (EM) and ‘emotional labelling’ (EL)] and one control condition [‘shape matching’ (SM)]. The emotional enhancement of memory (EEM) task was based on work by Kensinger and Corkin101 and included encoding conditions for negative arousing, negative non-arousing and emotionally neutral words.

The FEP task presented triplets of stimuli to participants, with two stimuli at the bottom left and right of the screen, and one stimulus in the centre top. In all conditions participants had to indicate via button press which of the two stimuli at the bottom matched the stimulus at the top. In the EM condition, all three stimuli were greyscale photographs of faces (from the Ekman set),102 showing either anger or fear, but from two different actors, with one of the bottom faces being repeated at the top. Thus theoretically no explicit processing of the emotion in the faces was required. In the EL condition, the bottom images showed one face displaying anger and one face displaying fear; the centre-top stimulus was an emotional label (‘anger’ or ‘fear’). Thus participants had to explicitly identify which face showed the indicated emotion. The SM condition showed simple elliptical and circular shapes, filled with random pixels of a similar distribution of greyscale values as the faces. Similar to the EM condition, the top shape was exactly identical to one of the bottom two shapes. Each triplet was presented for 3.5 seconds in blocks of five trials. Within each condition, the order of trials was randomised. Blocks of trials were separated by periods of baseline that were between 15.25 and 15.75 seconds long. Three blocks per condition were presented in one of the two predefined orders of blocks.

In the EEM task, participants were presented with individual words displayed in the centre of the screen, and had to decide whether or not the word was ‘concrete’ or ‘abstract’. Unbeknown to participants, all words fell into one of three categories according to their emotional valence and arousal, based on published word norms: negative arousing words, negative non-arousing words and neutral words. Individual words were presented for 2400 milliseconds, followed by a fixation cross of 600 milliseconds. They were presented in short blocks of three or four words from the same category. Two blocks of trials were always presented directly after each other. They were then followed by a short block of three trials of a distractor task. In the distractor task, participants were presented with days of the week and had to make a decision whether or not the day was in the first or second half of the week. The day directly in the middle of the week, here defined as Thursday, was never presented. A baseline period of 9 seconds followed each distractor block. In total, 28 words of each category were presented. There were two fixed orders of trials that were counterbalanced across participants. Owing to the repeated nature of the task for the patient group, all participants were informed that they would be asked to remember the presented words after the scan. Data from this recognition test will not be presented here.

Scanning took place on an Achieva^® 3.0T MR scanner (Philips Healthcare, Best, the Netherlands). The functional tasks used standard echo planar imaging (EPI) sequences [FEP task: voxel size 2.5 × 2.5 × 4 mm³, 35 slices, repetition time (TR) = 2500 milliseconds, echo time (TE) = 30 milliseconds; EEM task: 3 × 3 × 4 mm³, 33 slices, TR = 2250 milliseconds, TE = 30 milliseconds]. To reduce motion-related distortions and artefacts, slices were orientated sagittally to keep the most common type of motion (head tilts) within slices. REST (REgional Saturation Technique) slabs were positioned above and below the head. In addition to these functional sequences, a standard T1-weighted, 1-mm isotropic anatomical scan was performed. The anatomical scans were utilised in the preprocessing of the functional data.

Functional image processing was undertaken using Statistical Parametric Mapping (SPM8) (Wellcome Trust Centre for Neuroimaging, University College London, London, UK; www.fil.ion.ucl.ac.uk/spm) and followed standard procedures, including slice timing correction, realignment and co-registration to the anatomical scan. First-level models were computed in native space and included regressors for the different stimulus types, the estimated realignment parameters and their first derivative. Contrast images from these models were normalised to standard space using parameters estimated from the anatomical scan and spatially smoothed with a 8-mm full width at half maximum Gaussian kernel. Second-level models were created using the results from all included participants. Whole-brain analyses were performed to identify if the different task conditions resulted in differential BOLD signal across both groups of participants in the amygdala. If so, BOLD signals were extracted from the significant clusters. In addition, BOLD signals were extracted from anatomically defined regions of interest of the amygdala using probabilistic cytoarchitectonic maps. 103 Extracted BOLD signals were analysed using repeated measures ANOVA, with condition as within-subject factor, group as between-subject factor, and age, sex and premorbid IQ as covariates.

Manchester

Subjects were scanned using a parallel, between-subjects design on a 1.5-tesla (1.5-T) scanner. The scanning session, which took place between 12 noon and 3 p.m., consisted of a brief localisation magnetic resonance imaging (MRI) scan, followed by two fMRI protocols in which participants undertook the n-back task followed by the encoding emotional pictures memory task. After a 6-minute structural MRI scan, subjects completed the retrieval emotional pictures memory task.

All computerised tasks were run on a PC in E-Prime 2.0 (Psychology Software Tools Inc., Sharpsburg, PA, USA) and back-projected onto a screen visible to the participant via two mirrors attached to the head coil. The task responses were acquired from the participant using a fibre-optic button box held in the right hand. All participants received standardised training on how to perform the tasks prior to the scan.

The task used by Whalley et al. 104 was adapted to create an in-scanner recognition memory task. Additional images from the IAPS battery were used to lengthen the task.

For the encoding task, a total of 60 images were shown in five blocks of six positive and six neutral images (each block lasting 48 seconds). These were followed by a block of rest, consisting of a fixation cross, which also lasted 24 seconds each. There were five blocks of each condition shown, making the task length 360 seconds in total. Participants were asked to indicate, using the button box, whether or not they felt that the picture was ‘emotional’, and instructed to remember the images.

During the recognition segment of the task, a total of 96 images were shown in six blocks of eight positive images and six blocks of eight neutral images (each block lasting 48 seconds). These were followed by a block of rest, consisting of a fixation cross, which also lasted 24 seconds each. There were six blocks of each condition shown, making the task length 432 seconds in total. All images shown in the encoding section were re-shown. Participants were asked to indicate, using the button box, whether or not they recognised the image from the encoding task.

A blocked version of the n-back task was adapted from the task of Koychev et al. 105 It consisted of four blocks of zero-back and four blocks of two-back, each lasting 32 seconds. These required three correct responses from 13 stimuli per block using the button box. These blocks were interspersed with four blocks of rest consisting of a fixation cross, lasting 20 seconds.

Scanning was carried out on an Intera^® 1.5-T MRI scanner (Philips Healthcare, Best, the Netherlands) with prospective motion correction. Data were acquired with T2*-weighted, gradient echo EPI. Full brain coverage was used with TR = 2 seconds, TE = 40 milliseconds, 3.5 mm in-plane resolution, 4.5-mm slice thickness with 0.5-mm slice gap and 29 slices.

Data analysis

Behavioural and demographic data were analysed using Statistical Package for Social Science version 20.0 (SPSS Inc., Chicago, IL, USA). Behavioural data from both tasks were analysed using repeated measure ANOVA with one within-subject factor condition (difficulty: e.g. emotional valency – positive, neutral or zero-back, two-back) and one between-subject factor (participant group), as well as independent t-tests. The outcome measures used for the tasks were speed of response, number omitted, number of false positives and number correct. Results are presented as mean and SD, except when adjusted means from ANCOVA analyses are reported with standard error of the mean (SEM). Error bars in figures are also plotted as SEM for clarity, unless otherwise stated.

Data from fMRI were processed and analysed using SPM8. Scans were spatially preprocessed using standard protocols. They were realigned, using the first scan as a reference, normalised into the Talairach and Tournoux106 stereotactic space using Montreal Neurological Institute (MNI) templates then spatially smoothed using a 8-mm Gaussian kernel. A mask was created, which was a sum of the regions of interest (ROIs) for each task and used as single comparison in the analysis.

For both of the emotional picture tasks (encoding and retrieval), positive and neutral conditions were contrasted with rest. Second-level processing was performed using one-sample t-tests to explore the main effect of task on positive pictures minus rest, neutral pictures minus rest, all pictures minus rest and vice versa, and a two-sample t-test to explore the effect of depression on each level of the task (HVs minus patients and vice versa). For the encoding task, our ROIs were the amygdala, hippocampus and parahippocampal gyrus, dorsolateral prefrontal cortex (dlPFC) and inferior temporal area [Brodmann area 9 (BA9), BA20 and BA45]. For the retrieval task, the a priori regions of interest were amygdala, hippocampus, parahippocampal gyrus, anterior cingulate cortex (BA32), insula, dlPFC and frontopolar areas (BA9 and BA10).

For the n-back task, both levels of n-back were contrasted with rest. Second-level processing was performed using one-sample t-tests to explore (1) the main effect of task at the highest working memory load (two-back minus zero-back) and (2) the effect of depression on each level of the task (HVs minus patients and vice versa). Our ROIs were lateral premotor cortex (BA6), dorsal anterior cingulate cortex (BA32), dlPFC (BA9, BA46) ventrolateral prefrontal cortex (BA44, BA45, BA47), medial posterior parietal and inferior parietal lobule (BA7, BA40) and medial cerebellum, as per Symonds et al. 107

Within these areas we report as significant areas family-wise error (FWE) corrected peak level p(FWE) < 0.05 for the region of interest. Other areas surviving p(FWE) < 0.05 at a whole-brain level are also reported for interest.

Acute effects of hydrocortisone in treatment-resistant depression using pharmacological functional magnetic resonance imaging

We aimed to examine the effect of an acute bolus of hydrocortisone on patients with TRD in comparison with age- and sex-matched HVs. Using pharmacological functional magnetic resonance imaging (phMRI),107 we investigated whether or not BOLD signal is altered after acute administration of 100 mg of intravenous hydrocortisone in a between-subject, placebo-controlled parallel design study of 30 patients and 30 control subjects.

Electroencephalography

To examine the predictive ability of these identified EEG variables in response to metyrapone augmentation, patients consenting to EEG recordings had recordings made prior to commencement of treatment with metyrapone. Of particular interest was whether or not a high slope in the LDAEP analysis, predictive of response to SSRIs, would predict response to metyrapone, suggestive of an enhancement of serotonergic function. To examine both the effect of emotion on episodic memory retrieval and the putative ERP marker of LTP, EEG was recorded prior to treatment with metyrapone in patients and a cohort of healthy control subjects, with patients tested again at week +5.

Electroencephalography was recorded from 29 active silver/silver chloride electrodes attached to the scalp using an EEG cap (EasyCaps, Germany) with electrodes placed according to the international 10–20 system. Two further electrodes were placed on each mastoid process (right mastoid as a reference; left mastoid as an active channel), two on the outer canthi of the eyes to record horizontal eye movements [horizontal electrooculogram (HEOG) channel] and one on the nasion with linked electrodes under the midpoint of each eye to record vertical eye movements [vertical electrooculogram (VEOG) channel]. All electrode impedances were kept at < 5 kiloohms (kΩ). Recordings were made using a Neuroscan^® SynAmps2^® amplifier (Compumedics Germany GmbH, Singen, Germany) with an analogue–digital conversion rate of 500 Hz, and high- and low-pass filters of 0.5 Hz and 30 Hz, respectively.

Emotional Source Memory Task recordings

The Emotional Source Memory Task is divided into two elements: a study phase and a test phase. In the study phase, participants were presented with an emotionally valenced background picture of an event or scene drawn from the IAPS and were asked to rate its valence (7-point Likert scale, ranging from strongly negative to strongly positive). Following this, the background picture was superimposed with a neutral object. Participants were asked to make an association between the object and the event/scene in the background picture. Following completion of the study phase and a 5-minute delay, the test phase entailed a surprise memory test throughout which EEG was recorded. During the test phase, old and new neutral objects were presented, and participants had to make a recognition judgement responding either ‘old’ or ‘new’ via a keypad. If the participant recognised the object from the study phase, and responded as such, a follow-up recollection response was required. For this, the participant was asked to recall the valence of the associated background presented at study with the neutral object and respond via a keypad (positive, neutral, negative, don’t know). Analysis focused on behavioural data and ERPs of correctly identified old objects that were also associated with a correct identification of background valences (Hit–Hits), with relation to the emotional valence rating of the associated background given during the study phase.

Long-term potentiation recordings

During each testing session visual evoked potentials (VEPs) were recorded. The visual stimuli and the experimental procedure were based on those used by Teyler et al. 69 The stimulus (a series of chequerboards) was presented on a computer monitor (circular chequerboard stimuli subtending 8 degrees of visual angle, check size 0.3 degrees of visual angle). Participants were instructed to fixate on a red square in the centre of the screen. During baseline testing, the stimuli were presented at a frequency of 0.8 Hz (duration 33 milliseconds), and there were two baseline blocks of 100 flickers each. Baseline testing was followed by a block of photic tetanus when the identical chequerboard stimulus is presented at 8.6 Hz for 1000 flickers. This was followed by 2 minutes’ rest with eyes closed, followed by two post-tetanus test blocks (with stimulus presentation as per baseline testing) at 4-minute intervals. EEG was recorded during the two baseline and two post-tetanus test blocks.

Electroencephalogram analysis

All EEG analysis was conducted using Neuroscan 4.5 software. EEG recordings were arithmetically adjusted to a linked mastoid reference. Blink correction was performed on the basis of VEOG recordings using the Neuroscan’s ocular artefact reduction function. Residual artefacts were removed, whereby any epoch, for which any channel (including HEOG but not VEOG) had a voltage deflection of > +75 µV or < –75 µV, was rejected.

It did not prove worthwhile analysing the EEG predictors of response, nor the ESMT ERPs for reasons described below. However, the LTP EEG data were analysed. This is described in Chapter 3.

Continuous EEG recorded during the VEP test phases was epoched from –100 milliseconds to + 500 milliseconds around the presentation of the visual chequerboard stimulus. Epochs were artefact corrected and rejected as described above. Retained epochs were averaged to generate VEPs. To determine time windows within which N1, P1 and N1b amplitudes would be determined, peak detection using Neuroscan algorithms was run on the grand-averaged VEP data (all time points, all subjects) for each individual channel. The focus of analysis was on the Oz electrode, as this has been shown to demonstrate the largest and most robust VEP. 69 The average amplitude of N1, P1 and N1b was calculated as an area under the curve for a 20-milliseconds window centred at 160 milliseconds for N1, 248 milliseconds for P2 and 204 milliseconds for N1b, being halfway between N1 and P2. The magnitude of VEP potentiation was calculated as the difference between averaged N1, P2 and N1b components pre-tetanic stimulation and the first post-tetanic VEP measure.

Eligibility criteria

• Diagnostic and Statistical Manual of Mental Disorders-Fourth Edition (DSM-IV)108-defined major depressive episode, assessed using the Structured Clinical Interview for DSM (SCID) research version. 109

• HDRS17110 score of ≥ 18 at weeks –2 and 0 (see below).

• Massachusetts General Hospital Treatment Resistant Depression (MGH-TRD) staging score of 2–10 as a measure of treatment refractoriness. 111 This cut-off is based on MGH-TRD scores seen in primary and secondary care, but short of tertiary care, in the NHS. 112

• Single-agent or combination antidepressant treatment that includes a serotonergic drug (SSRI, tertiary amine tricyclic, venlafaxine, duloxetine or mirtazapine). At the point of randomisation, patients must have been on their current antidepressant medication, at the current dose, for a minimum of 4 weeks.

• Aged between 18 and 65 years. An upper age limit was included to reduce rates of physical health comorbidities that could complicate decisions around the safety of prescription of a drug (metyrapone) that may lower cortisol levels.

Exclusion criteria

• Other DSM-IV axis 1 diagnosis, other than an anxiety disorder considered to be secondary to a primary diagnosis of depression, confirmed using SCID.

• Physical comorbidity that would make metyrapone inappropriate, including untreated hypothyroidism, disorders of steroid production, cardiac failure, angina, myocardial infarction in the last 3 years and renal failure.

• Pregnancy or breastfeeding.

• Use of medication that would interfere with metyrapone.

• Dependence on alcohol or other drug(s) in the past 12 months, and/or current harmful use of such substances (defined as meeting SCID criteria for harmful use or dependence).

• Current or recent participation in a research study that could interfere with results.

Eligibility criteria

• No evidence of axis 1 disorder on the DSM-IV,108 assessed using the SCID research version. 109

• Aged between 18 and 65 years.

Exclusion criteria

• Physical comorbidity as per above (see Patients).

• Pregnancy or breastfeeding.

• Use of medication as per above (see Patients).

• Dependence on alcohol or other drug(s) in the past 12 months, and/or current harmful use of such substances (defined as meeting SCID criteria for harmful use or dependence).

• Current or recent participation in a research study that could interfere with results.

Statistical methods used to compare groups for primary and secondary outcomes

The primary outcome was the MADRS score at 5 weeks post randomisation adjusted for the MADRS score at baseline. The mean difference between treatment groups was estimated using a mixed model, assuming a normal error structure. The dependent variable was the MADRS score at 5 weeks; the baseline score was included as a covariate. Difference between centres was included as a random effect; origin of care (primary or secondary) was included as a fixed effect. Unadjusted estimates of the difference between groups are also presented (variation between centres and differences between origin of care were removed from the model). The analysis was undertaken on an intention-to-treat basis (i.e. groups were defined by randomised allocation rather than the actual treatment received).

Further secondary analysis of the primary outcome was prespecified. The changes in MADRS scores in the two randomised groups 5 weeks post randomisation were compared using an independent sample t-test. Two further binary outcomes were defined. Response was defined as a reduction in MADRS score that was less than or equal to half of the score at baseline; remission was defined as a MADRS score that was ≤ 10. These variables were analysed using the mixed-model approach described above, but with the assumption of a binomial error structure and logit link function.

Secondary outcomes (including the CAS, BDI and STAI) and QoL (EQ-5D health tariff and EuroQol health scale) were analysed using the mixed-model approach described above.

A secondary clinical objective was to consider the persistence of the effect of metyrapone. The prespecified analysis to assess persistence was to extend the mixed-model approach described above to include repeated measures on individuals. Additional explanatory variables included in the model were change over time (differences between scores observed at different visits) and an interaction between these changes and treatment group.

A further clinical objective was to assess how well metyrapone was tolerated. Side effects were assessed using the TSES and YMRS. The YMRS was analysed using the mixed-model approach described above. For the TSES, the relative risk (RR) of individual symptoms in the two groups was calculated.

The number of non-serious adverse events (AEs) reported by patients in each of the randomised groups was compared using a negative binomial regression model.

Methods for additional analyses, such as subgroup analyses and adjusted analyses

In the SAP, further per-protocol analyses were specified, in which groups were defined on the basis of compliance with medication, based on measurements of 11-deoxycortisol. (However, see Chapter 4, Discussion, Limitations.) Additional survival analysis of time to response and time to remission, based on a Cox proportional hazards model, was also specified.

Important changes to statistical methods as proposed

The primary analysis in the original protocol was an independent t-test of change in MADRS scores (baseline to week +5). The main reason for specifying this as the primary analysis was that it corresponds to the stated sample size calculation. The main disadvantage for specifying it as the primary analysis are the following recommendations:

1. ANCOVA is usually to be preferred to an analysis of change scores (see, for example, Van Breukelen,113 Senn114).

2. The primary analysis should take into account the stratification factors used in the randomisation of patients (Parzen et al. ,115 Kahan and Morris116).

Consequently, when formulating the SAP, the primary analysis was defined as ANCOVA, and the analysis of change scores was defined as a secondary analysis of the primary outcome. The final version of the SAP was approved prior to the breaking of the blind and is given in Appendix 1.

In retrospect, the proposed secondary analyses of persistence and times to response and remission should have been specified as being conditional on metyrapone being proven to be clinically efficacious at the primary end point. With no observed reduction in depression 5 weeks after randomisation, there was no rationale for testing hypotheses based on differences between groups at further time points. For outcomes for which there was no evidence of a difference between groups 5 weeks post randomisation, summary statistics are presented for the subsequent visits.

The TSES was not analysed as a single score (as had originally been specified in the SAP), after a review of the literature describing its development and use indicated that this was not appropriate. 88

Dates defining recruitment and follow-up

The first patient was randomised at the beginning of March 2011 and the last patient in mid-December 2012. The last 24-week follow-up was in late June 2013.

Recruitment was initially slow largely due to the complexity of the protocol which some clinicians and patients found too daunting. This required the development of a recovery plan, agreed by the funder at month 10, which included some changes to the protocol to reduce of the number of exclusions related to medical comorbidities (criteria shown above are final ones). The presence of any axis 1 disorder in addition to a major depressive episode was initially an exclusion criterion. This was relaxed to allow inclusion of patients with an axis 1 diagnosis of an anxiety disorder. There was much greater emphasis by the MHRN and others on promoting the study within clinical teams. The CLRN and Northumberland, Tyne and Wear Trust supported the study by funding junior clinicians through the Research Capability Funding process and these were an invaluable help, in particular with patients’ screenings. Several media events were also very helpful in raising awareness of the study and enhancing recruitment. Adverts were also placed in some local publications in Newcastle and Leeds, and this was a major reason for the higher proportion of primary care patients in Newcastle than, for example, Manchester.

Why did the trial end or stop?

The trial was stopped at the end of the study, as agreed with the Efficacy and Mechanism Evaluation (EME) Board after the last patient’s last visit. Owing to enhanced recruitment (n = 165 randomised), we were able to exceed the revised target of 140, giving us 84% power to detect the effects hypothesised in the sample size determinations. The trial period included a 7-month extension.

Baseline data

Baseline data of demographic and clinical characteristics are shown in Table 5. The placebo and the metyrapone groups were well balanced on key demographic variables and clinical characteristics.

In general, for data arising from interview-administered questionnaires, there were very few missing data items (Table 6). There was a greater number of missing items in the data arising from self-completion questionnaires, but, overall, the level of missing data was very low and only a small number of missing values were imputed.

In general, for data arising from interview-administered questionnaires, there were very few missing data items. There was a greater number of missing items in the data arising from self-completion questionnaires, but, overall, the level of missing data was very low.

Numbers analysed

Analysis was on the basis of intention-to-treat by original assigned groups (placebo n = 82, metyrapone n = 83). Details of the numbers analysed for each outcome measure, at each time point, for both groups, can be found in Appendix 2.

Outcomes and estimation

Results for the primary and secondary clinical outcomes and secondary analysis of it, as outlined above and described in detail in the prespecified SAP (see Appendix 1), are described below. There is one exception to this: the prespecified analysis excluding patients who discontinued treatment during the 3-week treatment period and another utilising the 11-deoxycortisol measure of apparent concordance with treatment are not included, as these analyses have not been completed owing to lack of availability of the data (see Analyses not yet complete, below).

Primary outcome measure: Montgomery–Åsberg Depression Research Scale

The primary outcome measure was depression as measured by the MADRS instrument. The key objective of the trial was to ascertain whether or not the strategy of treating patients with metyrapone resulted in a clinically significant reduction in depression at 2 weeks post treatment (5 weeks post randomisation) when comparing treatment with placebo.

Baseline

The mean MADRS score at baseline was 26.0 (SD 5.8) for patients recruited from primary care and significantly higher at 29.4 (SD 6.0) for patients recruited from secondary care; the mean difference at this time point was 3.4 [95% confidence interval (CI) 1.5 to 5.3]. Mean scores were non-significantly higher for patients recruited in Manchester than with the other two sites but this is primarily due to the higher proportion of patients recruited from secondary care in that site. The distribution of baseline scores by site and origin is shown in the box plots in Figure 2.

FIGURE 2.

Box plots of MADRS scores by site and origin of patient.

The horizontal line in the centre of each box corresponds with the median score; the lower and upper edges of the box correspond to the lower and upper quartiles of the distribution, respectively. The ‘whiskers’ show the distribution of the remaining values.

The distribution of scores in the two trial arms is shown in Figure 3. Details of the box and whisker plot are as described for Figure 2.

FIGURE 3.

Box plots of MADRS scores by group to which randomised.

Effect of treatment group on Montgomery–Åsberg Depression Research Scale score

Summary statistics for the MADRS at all visits are given in Table 7, and the distributions of scores are illustrated in Figure 4.

FIGURE 4.

Montgomery–Åsberg Depression Research Scale scores over time for patients treated with either metyrapone or placebo. Data plotted as mean, with error bars representing the SEM.

Examination of the summary statistics and Figure 4 suggests that there is a reduction in the mean scores at all of the follow-up visits compared with baseline, but that there is little difference at any time point in the groups randomised to metyrapone or placebo. The issue of whether or not this result may be effected by differential dropout between the groups is discussed below (see Ancillary analyses) and later in this chapter (see Figures 11 and 12).

Primary outcome analysis The prespecified primary analysis was an ANCOVA of week +5 MADRS scores, with baseline scores included as a covariate, adjusted for differences between randomisation strata (study site and primary or secondary care origin of the patients). The difference between groups was estimated on an intention-to-treat basis, with patients being included in the group to which they were randomised.

Fitting a simple regression model, including only baseline MADRS score and a binary indicator of group to which randomised, yielded the following regression coefficients:

• baseline MADRS: 0.88 (95% CI 0.63 to 1.12)

• difference between groups (metyrapone–placebo): –0.43 (95% CI –3.48 to 2.63).

The MADRS score at 5 weeks was highly correlated with the score at baseline; the difference between groups at 5 weeks did not differ significantly from zero.

Variation between sites was then included as a random effect, and the difference between patients recruited from primary care and those recruited from secondary care was added as a fixed effect. There was little change to the estimated effect of metyrapone; the estimated difference between groups, based on the adjusted model, was –0.51 (95% CI –3.48 to 2.46).

Further analysis of Montgomery–Åsberg Depression Research Scale data

Change in MADRS scores between baseline and week +5: a prespecified secondary analysis was to calculate the change in MADRS scores from baseline to week +5 for each patient and to compare group means using an independent sample t-test. Summary statistics for baseline, week +5 and change scores are given in Table 8.

There was a significant reduction in MADRS scores of approximately 6 points between baseline and week +5. The difference between the changes in the metyrapone group and the change in the placebo group was = –0.39 (95% CI –3.37 to 2.67) and did not differ significantly from zero (t₁₄₁ = 0.252; p = 0.80). The estimated effect of metyrapone is very similar to that obtained from the ANCOVA.

Analysis using atypical Montgomery–Åsberg Depression Research Scale scores

These were calculated using the highest score from items 4a and 4b and items 5a and 5b on the MADRS. The mean atypical MADRS score at baseline in the metyrapone group was 28.5 (SD 6.7) and 28.9 (SD 5.4) in the placebo group. For both groups there was a significant reduction in atypical MADRS scores of approximately ‘7’ between baseline and week +5. The difference between the changes in atypical MADRS scores in the metyrapone group and the change in the placebo group was –0.42 (95% CI –3.36 to 2.63) and did not differ significantly from zero (t₁₄₁ = 0.275; p = 0.79).

Persistence of change in Montgomery–Åsberg Depression Research Scale scores

For each individual, the MADRS score was recorded on up to five occasions following randomisation. These repeated measures were analysed using a mixed model in which we assume that for each patient the MADRS scores varied randomly about an individual mean score (with SD σ_e), but that this mean varied randomly across patients (with SD σ_u). A normal distribution was assumed for each of the random effects. Fitting an initial model with only a constant term (corresponding to the assumption that the mean MADRS score is the same at all five follow-up visits) in addition to the random effects, the estimated mean MADRS score across the five follow-up visits was 21.6 (95% CI 20.1 to 23.0). There was significant variation of scores both within patients [σ_e = 6.57 (95% CI 6.17 to 6.99)] and between patients [σ_u = 8.58 (95% CI 7.54 to 9.78)].

As we are interested primarily in changes in depression, the second step was to include the baseline MADRS score as a covariate. This model indicated a significant reduction in depression from baseline at each of the five follow-up visits. Baseline depression explained some of the variation between individual patients (the estimate of σ_u fell to 6.69) but there were still significant differences between patients.

Adding in differences between randomisation strata (sites and origin of patient – either primary or secondary care) did not explain very much additional variability. With the baseline score as a covariate there were no significant differences between these strata. The estimated difference in depression across all five follow-up visits between patients on metyrapone and patients on placebo was 0.75 (95% CI –1.59 to 3.10).

There was no evidence of any trend in depression over the follow-up visits. Adding a linear trend over visits to the model, the estimated mean change in MADRS score between consecutive visits was –0.27 (95% CI –0.65 to 0.10) and did not constitute a clinically important change in depression.

One of the objectives of the ADD Study was to ascertain whether or not the effect of metyrapone persists over time. Given that there is no obvious impact of the drug at 2 weeks post treatment, this objective is moot. However, fitting an interaction between the effect of metyrapone and time demonstrates that the interaction is not significant; there is no evidence that changes in depression over time differ between treatment groups.

Response rates

Response was defined as a reduction of ≥ 50% in MADRS score. The proportion of responders at each time point is shown in Table 9.

At each visit the proportion of patients who reported a MADRS that was equal to or less than half their baseline score was between 19% and 26%. At the primary end point of 5 weeks post randomisation, 21% of patients were deemed to be responders. Responder rates for each trial arm at 5 weeks are given in Table 10.

Response rates were almost identical in the two groups at this time point. The odds ratio (based on a logistic regression model adjusted for site and origin of patient: metyrapone/placebo) = 0.95 (95% CI 0.41 to 2.20). The difference between groups was not clinically or statistically significant but the wide CI (the odds could be less than half or more than double) suggests that the study was not adequately powered to detect differences in this binary measure of outcome.

Remission rates

A patient was defined as being in remission if their MADRS score was ≤ 10. Remission rates broken down by visit are shown in Table 11.

The proportions of patients in remission appear to be similar at each visit. The proportion of patients in remission 5 weeks post randomisation by study group are shown in Table 12.

Remission rates at this time point were similar in the two groups. The odds ratio (based on a logistic regression model adjusted for site and origin of patient: metyrapone/placebo = 0.97 (95% CI 0.40 to 2.55). Again, the width of the CI is large (the odds of ‘success’ in one group could be less than half or more than double that in the other).

Suicidal thoughts

During the trial, the AE of increased suicidality was reported (as was to be expected in such a population). The DMEC looked at this AE with the patients divided into two – still blinded – groups. DMEC recommended no change to the conduct of the trial and that item 10 on the MADRS (‘Suicidal thoughts’) was examined in the two groups during analysis, and this was incorporated into the SAP.

The prespecified analysis was an ANCOVA of week +5 suicidality score, with baseline score included as a covariate, adjusted for differences between randomisation strata. The difference between groups was estimated on an intention-to-treat basis, with patients being included in the group to which they were randomised.

Fitting a simple regression model including only baseline MADRS score and a binary indicator of group to which randomised yielded the following regression coefficients:

• baseline MADRS item 10 score: 0.73 (95% CI 0.59 to 0.87)

• difference between groups (metyrapone – placebo): –0.13 (95% CI –0.52 to 0.26).

Suicidality score at 5 weeks was highly correlated with the score at baseline; the effect of metyrapone did not differ significantly from zero.

Variation between sites and the difference between patients recruited from primary care and patients recruited from secondary care were then added as fixed effects. There was little change to the estimated effect of metyrapone; the estimated difference between groups, based on the adjusted model, was –0.15 (95% CI –0.53 to 0.24).

The estimated impact of metyrapone on suicidality score (as measured by item 10 on the MADRS) was a change of –0.15 (95% CI –0.53 and 0.24).

Beck Depression Inventory

Effect of metyrapone on Beck Depression Inventory scores 5 weeks post randomisation

As defined in the SAP, scores obtained 5 weeks post randomisation were analysed using ANCOVA, with baseline scores included as a covariate. This suggested a difference of –2.65 (95% CI –6.53 to 1.23) between the effects of metyrapone and placebo. Adjusting for random variation between centres and a difference between patients originating from primary and secondary care gave an adjusted estimate of –2.65 (95% CI –6.41 to 1.10), which was not significantly different from zero. The adjusted estimate differs very little from the unadjusted one. Although not significant, the direction of effect is consistent with that hypothesised. However, the estimated magnitude of the effect (–2.65) was small in comparison with the average change between baseline and 5 weeks post randomisation of 6.14 (95% CI 4.19 to 8.09) in the two groups combined.

Persistence of effect

Given that the estimated impact of metyrapone at 5 weeks was not statistically significant, it was not felt appropriate to undertake an analysis to assess the persistence of the effect over time. BDI scores over time are plotted in Figure 5.

FIGURE 5.

Beck Depression Inventory scores over time for patients treated with either metyrapone or placebo. Data plotted as mean, with error bars representing the SEM.

Clinical Anxiety Scale

Effect of metyrapone on Clinical Anxiety Scale scores 5 weeks post randomisation

Based on ANCOVA of CAS scores recorded 5 weeks post randomisation (with baseline scores included as a covariate) the unadjusted estimate of the effect of metyrapone was 0.55 (95% CI –1.21 to 2.30). Adjusting for differences between centres and patient origin of care the estimated effect of metyrapone on CAS scores was 0.46 (95% CI –1.20 to 2.12). These estimates correspond to differences between groups that are neither statistically nor clinically significant.

Persistence of effect

Lower anxiety levels were recorded at all follow-up visits compared with those observed at baseline. However, there was very little difference between the groups at any time point. Given the lack of a difference between metyrapone and placebo at week +5, no further analysis was conducted. CAS scores over time are plotted in Figure 6.

FIGURE 6.

Clinical Anxiety Scale scores over time for patients treated with either metyrapone or placebo. Data plotted as mean with error bars representing the SEM.

State–Trait Anxiety Inventory

The STAI comprises two scales: state anxiety and trait anxiety. Because the study investigated the effect of treatment with metyrapone over the short term, only state anxiety is described here. This scale evaluates the current state of anxiety, asking how respondents feel ‘right now,’ using items that measure subjective feelings of apprehension, tension, nervousness, worry, and activation/arousal of the autonomic nervous system.

Effect of metyrapone on state anxiety scores 5 weeks post randomisation

Based on an ANCOVA, the estimated effect of metyrapone relative to placebo at week +5 was an increase in state anxiety of 1.2 (95% CI –0.7 to 3.1). This is consistent with the larger fall in anxiety between baseline and week +5 in the group randomised to metyrapone than in the other group that can be seen in Figure 7. However, the difference between groups is not statistically significant (p = 0.21). Adjusting for origin of patient care and differences between centres resulted in very little change in the estimated impact; adjusted estimate was an increase in anxiety of 1.2 (95% CI –0.6 to 3.0).

FIGURE 7.

State anxiety scores over time for patients treated with either metyrapone or placebo. Data plotted as mean with error bars representing the SEM.

Persistence of effect of metyrapone on State–Trait Anxiety Inventory scores

Given that there is no obvious effect on state anxiety 5 weeks post randomisation, persistence was not formally assessed. However, it is clear from Figure 7 that there was very little change in anxiety levels in either group of the period of follow-up.

EuroQol health tariff

Effect of metyrapone on EQ-5D health tariffs 5 weeks post randomisation

Quality-of-life scores appear to improve in both groups between baseline and the visit 5 weeks post randomisation (Figure 8).

FIGURE 8.

European Quality of Life-5 Dimensions tariffs over time for patients treated with either metyrapone or placebo. Data plotted as mean with error bars representing the SEM.

Based on a simple ANCOVA model with baseline tariff value included as a covariate, the estimated effect of metyrapone was a change in tariff value of 0.014 (95% CI –0.073 to 0.101). Adjusting for variation between sites and origin of patient care, the estimated effect of metyrapone was 0.015 (95% CI –0.069 to 0.099). There was no evidence of a beneficial effect of treatment with metyrapone on EQ-5D QoL tariff value.

EuroQol visual analogue scale

Effect of metyrapone on EuroQol visual analogue scale at 5 weeks post randomisation

The EQ-VAS scores appear to improve in both groups between baseline and the visit 3 weeks post randomisation (Figure 9).

FIGURE 9.

EuroQol visual analogue scale scores over time for patients treated with either metyrapone or placebo. Data plotted as mean with error bars representing the SEM.

The estimated effect of metyrapone at 5 weeks post randomisation (ANCOVA with baseline scores included as a covariate) was 5.7 (95% CI –0.8 to 12.1). Adjusting for origin of patient and variation between centres, the adjusted estimate is 5.6 (95% CI –0.7 to 12.0).

Persistence of effect

It is interesting to observe that the largest difference between groups was observed at week +5. It is likely that this is a chance result, with the difference between groups being very much smaller at subsequent visits.

Young Mania Rating Scale

Effect of metyrapone on Young Mania Rating Scale scores

Mean and SD YMRS scores for each randomised group are reported for all visits (Figure 10).

FIGURE 10.

Young Mania Rating Scale scores over time for patients treated with either metyrapone or placebo. Data plotted as mean with error bars representing the SEM.

The estimated difference between patients randomised to metyrapone and placebo at 5 weeks post randomisation (ANCOVA of week +5 scores with baseline scores included as a covariate) was –0.04 (95% CI –0.54 to 0.46). Adjusting for origin of patient and variation between centres resulted in almost no change to the estimate; the adjusted estimate is –0.04 (95% CI –0.52 to 0.45). There is no evidence of different levels of manic symptoms between patients randomised to metyrapone and placebo.

Ancillary analyses

Missing data

There may be some suggestion that patients randomised to metyrapone were less inclined to attend follow-up visits than patients randomised to placebo. This is shown in Figure 11.

FIGURE 11.

Cumulative survival (or retention) against visit number (visits at which MADRS was recorded are numbered sequentially).

Fitting a Cox proportional hazards model, the estimated hazard ratio was 0.57 (95% CI 0.35 to 0.93); patients randomised to metyrapone were less likely to be retained in the study than other patients. It is not clear whether this is a chance finding (we were unlucky with the randomisation) or a genuine effect of taking the active drug.

In order to investigate whether or not retention was related to baseline depression, the baseline MADRS score was included in the Cox regression model. The hazard ratio corresponding to a unit increase in MADRS score was 1.01 (95% CI 0.97 to 1.04). This suggest that time to drop out was not related to baseline depression.

To further explore the reasons for non-attendance at follow-up visits, the relationship between change in symptoms between baseline and week +3 (the period of treatment with metyrapone or placebo) and attendance at the primary outcome time point at week +5 was examined. Figure 12 shows the change in MADRS scores between baseline and 3 weeks for patients who did or did not attend at week +5.

FIGURE 12.

Change in depression between baseline and 3 weeks post randomisation by whether or not the patient attended the primary end point 5 weeks post randomisation. Lines correspond with individual patients.

The premise behind joint modelling is that whether or not a patient responds is influenced by outcome; specifically patients with poorer outcomes are less likely to respond than other patients. The expectation would be that the lines in the upper half of Figure 12 (corresponding to non-responders at week +5) would have a greater downwards gradient than those in the lower plot (responders at week +5). There is very little evidence of such a pattern either from inspection of the plot or examination of the mean changes in depression from baseline to the visit at week +3, which are shown in Table 13.

TABLE 13 - Reduction in depression from baseline to week +3

Status at week 5	Reduction in depression baseline to week +3
	Mean	SD	95% confidence limits of the difference
			Lower	Upper
Non-attenders	4.80	9.88	–2.26	11.86
Attenders	6.54	8.86	5.05	8.03

Those who attended at week +5 had a slightly greater reduction in depression (mean change = 6.5) than those who failed to attend (mean change = 4.8). The difference in these changes was 1.7 (95% CI –4.0 to 7.5). Although not statistically significant, the direction of effect is as hypothesised – there is a tendency for patients who drop out to be those with worse outcomes. In practice, dropout rates were very similar in the two groups and if anything were higher in patients randomised to metyrapone. Thus any adjustment to the estimated impact of metyrapone based on joint modelling would be to further reduce the size of the estimated reduction in depression. There is no evidence from the joint modelling to suggest a clinically beneficial effect of metyrapone on depression.

Hypothalamic–pituitary–adrenal axis function

Hypothalamic–pituitary–adrenal axis function was determined in order to gain an understanding of the mechanism of action of the drug. Salivary cortisol was measured at 11 p.m. and then on five occasions, 15 minutes apart from wakening, on the following day, as per the standard protocol for the CAR. Patients collected such samples at baseline, immediately after cessation of treatment and 2 weeks later. Healthy control subjects collected samples at baseline. Areas under the curve [area under the curve ground (AUCg) and area under the curve increase (AUCi)] were determined using trapezoidal method for the morning cortisols. 117 The control subjects and patients did not differ in their cortisol parameters (Figure 13 and Table 14).

FIGURE 13.

Salivary cortisol (µg/dl) at 11 p.m. and on awakening the following morning, and at 15-minute intervals for 60 minutes thereafter in the two groups.

Baseline AUCg did not impact on clinical response to treatment. This was analysed using repeated measures ANCOVA with ‘visit’ as the within-subject factor (with MADRS score at week +3 and MADRS score at week +5 as the two levels), ‘treatment’ (metyrapone and placebo) as the between-subject factor, and baseline MADRS score and baseline AUCg as the covariates (F_{1, 93} = 1.275 df; p = 0.26).

Metyrapone did not significantly impact cortisol in the short or longer term. Change in HPA axis function (measured using δAUCg, i.e. the difference between AUCg at baseline and AUCg at week +3) did not impact on clinical response to treatment. This was analysed using repeated measures ANCOVA with ‘visit’ as the within-subject factor (with MADRS score at week +3 and MADRS score at week +5 as the two levels), ‘treatment’ (metyrapone and placebo) as the between-subject factor and baseline MADRS score and δAUCg as the covariates (F_{1, 81} < 1; not significant).

Childhood trauma

The CTQ revealed that patients had experienced more childhood adversity than control subjects (Table 15).

Emotional abuse, emotional neglect, physical abuse, physical neglect and sexual abuse did not predict response to medication (analysed using repeated measures ANOVA with these subscores on the CTQ entered as covariates: data not shown).

Spatial working memory

Treatment (patients)

Valid data were available for 43 patients (20 metyrapone, 23 placebo) tested at baseline and at 5 weeks. This analysis was conducted on totals for both the BSE and WSE. Results of the ANCOVA are presented in Table 17. There were no significant main effects of treatment for either outcome measure.

Attentional Network Test

Treatment (patients)

Valid data were available for 38 patients (17 metyrapone, 21 placebo) tested at baseline and at 5 weeks. This analysis was conducted on the alerting, orientating and conflict indices. Results of the ANCOVA are presented in Table 19. There were no significant main effects of treatment on any ANT measure.

Object-location memory

Treatment (patients only)

Valid data were available for 38 patients (16 metyrapone, 22 placebo) tested at baseline and at 5 weeks. This analysis was conducted on the POM, OLB and COM. Results of the ANCOVA are presented in Table 21.

Digit span

Treatment (patients only)

Valid data were available for 45 patients (22 metyrapone, 23 placebo) tested at baseline and at 5 weeks. This analysis was conducted on all indices. Results of the ANCOVA are presented in Table 23. There were no significant main effects of treatment on any measure.

Face Emotion Recognition Task

Treatment (patients only)

Valid data were available for 22 patients (8 metyrapone, 14 placebo) tested at baseline and at 5 weeks. Results of the ANCOVA are presented in Tables 26 and 27.

Misattribution

Emotional Memory Task

Treatment (patients only)

Valid data were available for 41 patients (19 metyrapone, 22 placebo) tested at baseline and at 5 weeks.

Recall: No significant effect of treatment was found. Results of the ANCOVA are presented in Table 31.

Recognition: No significant effects of treatment were found. Results of the ANCOVAs are presented in Tables 32 and 33.

Affective GoNoGoTask

Treatment (patients only)

Valid data were available for 38 patients (17 metyrapone, 21 placebo) tested at baseline and at 5 weeks. No significant effect of treatment was found. Results of the post hoc ANCOVAs are presented in Table 36.

Functional magnetic resonance imaging

Facial emotion processing task

Whole-brain analysis showed a cluster of significantly higher (p < 0.05, FWE corrected) BOLD signal in the right amygdala in both the EM condition and the EL condition than the SM condition (Figure 14). Table 37 shows the estimated marginal means of the BOLD signal in the three conditions in this cluster. Controlling for age, sex and premorbid IQ [National Adult Reading Test (NART)], there was a significant difference between patients and control subjects in the EM condition, but not in the EL or SM conditions, with patients showing higher BOLD values than control subjects.

TABLE 37 - Average BOLD signals (and SD) in the FEP task from a cluster showing significant differences between the facial processing conditions (EM, EL) and the control condition in the right amygdala (MNI 18, –8, –16; k = 96)

n.s., not significant.

	EM	EL	SM
Patients	0.352 (0.371)	0.408 (0.446)	–0.120 (0.295)
Control subjects	0.067 (0.371)	0.334 (0.449)	–0.227 (0.298)
Group effect	F = 5.076; p = 0.033	F < 1 (n.s.)	F = 1.112; p = 0.301

FIGURE 14.

Significant cluster of BOLD signal increases in the EM condition (a) and EL condition (b) over the control condition SM across the entire sample (p < 0.05, FEW corrected, k = 10). Numbers in parentheses are MNI coordinates (in mm) of the local signal peaks.

Estimated marginal means for the BOLD signal of the three conditions in anatomically defined amygdala regions, separately for the left and right hemisphere, are shown in Table 38. There were no significant differences between the two groups in any of the conditions in either region, although there was a statistical trend for higher BOLD signal for patients in the left amygdala for the EM condition.

TABLE 38 - Average BOLD signals (and SD) in the FEP task for anatomically defined regions of interest of the left and right amygdala

n.s., not significant.

	EM	EL	SM
Left amygdala
Patients	0.280 (0.389)	0.450 (0.492)	0.244 (0.328)
Control subjects	0.018 (0.391)	0.373 (0.496)	0.108 (0.329)
Group effect	F = 3.838; p = 0.061	F < 1 (n.s.)	F = 1.455; p = 0.238
Right amygdala
Patients	0.317 (0.319)	0.307 (0.384)	0.114 (0.253)
Control subjects	0.141 (0.321)	0.357 (0.387)	0.072 (0.251)
Group effect	F = 2.605; p = 0.118	F < 1 (n.s.)	F < 1 (n.s.)

In the Manchester centre patient sample

Twenty-seven patients were eligible for this study, one of whom subsequently withdrew consent. Twenty-six patients participated in the final study (13 females). Twenty patients were right handed, six ambidextrous and 1 left handed. Given the literature regarding laterality118 and hand preference combined with the relative difficulty to recruit and retain patients with TRD, it was decided to take a pragmatic approach and recruit patients with a preference for left-handedness and perform a post hoc sensitivity analysis. One scan from the patient group was excluded from analysis owing to movement artefact, and another subject was withdrawn from the study owing to the discovery of a structural abnormality; therefore, data are presented for 24 patients.

Thirty-two right-handed HVs were recruited from advertisements by the University of Manchester. Thirty were eligible (15 females). All of the female subjects were on long-term hormonal contraception to control for the effects of menstrual cycle on endogenous cortisol levels to allow reliable within-subject comparison119 but subjects were otherwise medication free, drank < 21 units of alcohol/week and were caffeine free on the scan days.

One male HV was later excluded owing to a head-coil malfunction; therefore data are reported from 29 subjects.

The mean age of patients was 44.1 years (SD 7.7 years; range 26–57 years) compared with mean age 36.47 years (SD 5.94 years; range 31–55 years) for the HVs. The mean NART IQ score for the patient group was 109.7 (SD 10.0) and for the HVs 113.3 (SD 10.4). There was no significant difference between the patient and HV groups in either age or IQ (p = 0.22 and p = 0.15, respectively). The baseline GRID-HAMD score for the patient group was 23.74 (SD 3.7) compared with 0.28 (SD 0.5) for the HVs. There was a significant difference in handedness between the HV group and the patients, with one left-handed individual and five ambidextrous individuals in the TRD group; however, the results were not substantially altered by the removal of this individual in post hoc sensitivity analysis.

Behavioural results

Functional magnetic resonance imaging results

Emotional pictures encoding task

The main effect of task (all participants, all encoding images minus rest) showed increased BOLD signal in bilateral parahippocampal, bilateral inferior frontal gyrus (BA47, BA9), left temporal gyrus (BA20) and right hippocampus in keeping with findings of meta-analyses of encoding tasks literature. 120,121

When the HVs were compared with the patients, patients with TRD showed a significant decreased BOLD signal at whole-brain level in the posterior cingulate compared with the HVs during the encoding task [HVs minus patients, all encoding images minus rest contrast, BA31; x = 9.5, y = –35, z = 35, Z = 3.47 p(FWE) (whole brain corrected) = 0.043. There were no significant findings in the pre-hypothesised ROIs]. (see Table 40 and Figure 15). There were no significant results in the opposite subtraction.

TABLE 40 - Results of the three fMRI tasks

BA, Brodmann area; K, cluster size at p < 0.005; L, left; Pts, patients; R, right.

Whole-brain corrected p(FWE) < 0.05.

p(FWE) < 0.05 following a priori ROI analysis.

Area	BA	Side	Coordinates (MNI)			Cluster		p(FWE)
			x	y	z	K	Z
Emotional Pictures Encoding task
Encode–rest HV–Pts
Posterior cingulate	31	R	9.5	–35	35	77	3.47	0.043a
Emotional Pictures Retrieval task
Retrieve–rest HV–Pts
Insula	40	L	–46.5	–21	20	56	3.85	0.046b
Positive–neutral HV–Pts
Anterior cingulate	33	L	–4.5	17.5	25	18	3.95	0.033b
n-back task
2back–0back HV–Pts
dlPFC	9	L	–25.5	7	50	68	4.53	< 0.001a

FIGURE 15.

Significant BOLD decrease in the patients compared with the control subjects in the posterior cingulate (9.5, –35, 35), p(FWE) whole brain = 0.043 during the encoding task. Control subjects minus patients, all encoding images minus rest, error bars indicate SEM.

Emotional pictures retrieval task

The main effect of task (all participants, all retrieval tasks minus rest) showed increased BOLD signal in the parahippocampus bilaterally, the left inferior temporal lobe (BA36), the left lingual gyrus (BA18) and left BA9 and BA6. There was no significant change in BOLD signal in the hippocampus.

In the a priori ROI of the insula, the BOLD signal was increased in the patients compared with the control subjects while completing the retrieval task, compared with the rest condition (patients minus control subjects, retrieval images minus rest contrast) [posterior insula, BA40; x = –46.5, y = –21, z = 20, Z = 3.85 p(FWE)(ROI analysis) = 0.046] (see Table 40 and Figure 16). In the anterior cingulate, another a priori ROI, the BOLD signal was reduced in the patients compared with the control subjects while retrieving positive images, compared with neutral images (control subjects minus patients, positive images minus neutral images contrast) [anterior cingulate, BA33; x = –4.5, y = 17.5, z = 25, Z = 3.95 p(FWE)(ROI analysis) = 0.033] (see Table 40 and Figure 17).

FIGURE 16.

Significant BOLD increase in the patients compared with the control subjects in the posterior insula (–46.5, –21, 20), ROI p(FWE) = 0.046 during the retrieval task. Patients minus control subjects, all retrieval images minus rest, error bars indicate SEM.

FIGURE 17.

Significant BOLD increase in the patients compared with the control subjects in the anterior cingulate (–4.5, 17.5, 25) ROI p(FWE) = 0.033 during the retrieval task. Control subjects minus patients, positive minus neutral, error bars indicate SEM.

n-back task

The main effect of the task (using all participant scans), the two-back compared with zero-back subtraction showed increased BOLD signal bilaterally in the dlPFC, medial posterior parietal and lateral premotor cortices, in keeping with the areas reported in a previous meta-analysis. 122 In the control subjects minus patients two-back minus zero-back contrast, the patients showed a decrease in BOLD signal compared with the control subjects in the dlPFC [BA6; x = –25.5, y = 7, z = 25, Z = 50 p(FWE)(whole brain corrected) = < 0.001] (see Table 40 and Figure 18). However, there was no difference seen in our ROI areas.

FIGURE 18.

Significant BOLD increase in the control subjects compared with patients in the dlPFC (–25.5, 7, 50) whole-brain p(FWE) < 0.001 in the two-back minus zero-back contrast. Error bars indicate SEM.

Acute effects of hydrocortisone in treatment-resistant depression using pharmacological functional magnetic resonance imaging

The final sample consisted of 20 patients with TRD and 27 HVs. The a priori regions of interest were the hippocampus, hypothalamus, amygdala and dorsolateral and ventrolateral prefrontal cortex (all areas rich in corticosteroid receptors). We failed to demonstrate a difference between the TRD and HV groups, and also a main effect of hydrocortisone, in contrast with our previous study. 107 Reasons that may have accounted for this result include lack of power and artefact.

Electroencephalography

At study initiation it had been intended to recruit 50 patients to the EEG mechanistic studies, giving approximately 25 treated with metyrapone and 25 with placebo. Given the slow recruitment to the study initially, two decisions were made – first, there was an acceptance of a lower sample size than the originally planned 190, and, second, emphasis was given to recruitment to the main treatment element of the study, including patients who did not wish to consent to recruitment to the mechanistic studies. A total of 34 patients were recruited for the EEG studies. Twenty-five of these had MADRS data at the primary end point at week +5. Of these, 14 were treated with placebo and 11 with metyrapone, with five responders and six non-responders. These numbers did not provide sufficient power to warrant examination of the EEG variables suggested to predict response to active treatment, and therefore this was not pursued further.

With regard to the ESMT, the issue of the small patient sample size was compounded by a methodological issue. A large amount of horizontal eye movement was found in patients during the test phase. This led to a high epoch reject rate. EEG analysis of the ESMT task was therefore not conducted owing to an insufficient number of both patients and control subjects meeting EEG signal-to-noise ratio inclusion criteria. For a satisfactory signal–noise ratio, at least 10 epochs would need to be present in each of three task outcome measures (neutral, negative and positive correctly recalled items). Only three patients and one control met this criterion.

With regard to the LTP analysis, this was conducted on 32 patients and 20 control subjects. The mean VEP waveforms for patients and control subjects from the first pre-tetanus and first post-tetanus assessment are shown in Figure 19. Data for the individual components of the VEP are shown in Figures 20–22.

FIGURE 19.

Mean VEP waveforms of patients (green lines) and control subjects (black lines) before (solid lines) and after (dotted lines) visual tetanic stimulation.

FIGURE 20.

Shows N1 amplitude, together with the magnitude of enhancement from the mean of the two pre-tetanus amplitudes to the first post-tetanus amplitude (error bars are SEM).

FIGURE 21.

Shows P2 amplitude, respectively, together with the magnitude of enhancement from the mean of the two pre-tetanus amplitudes to the first post-tetanus amplitude (error bars are SEM).

FIGURE 22.

Shows N1b amplitude, respectively, together with the magnitude of enhancement from the mean of the two pre-tetanus amplitudes to the first post-tetanus amplitude (error bars are SEM).

There was no significant difference in magnitude of either the first or second pre-tetanus N1, P2 or N1b components between the patients and the healthy control subjects (all p-values > 0.2). As can be seen, there was an increase in amplitude of N1 and N1b post tetanus, especially in the healthy control subjects, although little change in amplitude of P2. In the healthy control subjects there was a significant difference in N1 and N1b amplitudes in the post-tetanus test 1 block compared with the mean of the pre-tetanus blocks (t = 2.17, p = 0.035; t = 3.51; p = 0.001 respectively), but this was not seen for the P2 component, or for any component in the patient group (p > 0.2 for all comparisons). Comparison of the magnitude of the increase in N1 amplitude showed a significantly greater increase for the control subjects than the patients (t = 2.01; p = 0.043). The difference in magnitude of increase in N1b between patients and control subjects, however, did not reach significance (t = 1.52; p = 0.146).

It was not possible to examine the effect of metyrapone on the LTP EEG measure owing to insufficient data being available.

Harms

Toronto Side Effects Scale

Across all 32 symptoms rated by this scale, there was no difference between the randomised groups in terms of incidence. There were two symptoms where the difference was significant at the 5% level (delayed ejaculation and weight loss both being greater in the placebo group), but this is only to be expected given that there are 32 symptoms rated. No differences were seen in the incidence of central nervous system side effects. The data for this are shown in Table 46. It is particularly noteworthy that the rates of dizziness and postural hypotension are very similar in both randomised groups.

In the full patient sample

1. To assess whether metyrapone changes patients’ HPA axis function:

   1. HPA function will be assessed by calculating the Area Under the Curve (AUC) for the CAR data, measuring the peak change in cortisol on waking by comparing the initial waking sample with the maximum, as well as examining the diurnal change in cortisol using the CAR and the 11 p.m. saliva sample. The effect of metyrapone will be compared with placebo in relation to HPA axis function at week 3 (end of metyrapone treatment) using ANOVA covarying for the baseline at week 0.

2. To assess whether changes in HPA axis function are seen 2 weeks after metyrapone treatment:

   1. HPA axis function assessed as above comparing data at week 5 covarying for week 0.

3. To assess whether the change in HPA axis function correlates with clinical response:

   1. The correlation between change in HPA axis function (CAR AUC and maximal peak, and diurnal cortisol, both at week 3 and week 5 compared with week 0) with change in MADRS, CAS, BDI, STAI, EQ-5D (week 5 compared with week 0) will be examined.

4. To assess whether baseline HPA axis function predicts clinical response:

   1. The correlation between HPA axis function (CAR AUC and maximal peak, and diurnal cortisol) with change in MADRS, CAS, BDI, STAI, EQ-5D (week 5 compared with week 0) will be examined.

5. To assess if type and severity of childhood trauma [as assessed by the Childhood Trauma Questionnaire (CTQ) scores] predicts clinical response:

   1. The correlation between CTQ with change in MADRS, CAS, BDI, STAI, EQ-5D (week 5 compared with week 0) will be examined.

6. To assess whether the change in, or baseline assessment of, HPA axis function has specific genetic underpinning and whether this relates to, or explains better, the clinical response:

   1. The relationship between clinical response (as described above) and HPA axis function compared with a range of candidate polymorphisms will be examined.

In subsamples of patients

1. To determine the nature and extent of neuropsychological abnormalities in patients with TRD compared with healthy control subjects:

   1. This will be studied using the various outcome measures from the neuropsychological tasks employed, comparing the patients with healthy control subjects. Analysis will include controlling for age, gender and IQ (as assessed using NART).

2. To determine whether neuropsychological performance improves with metyrapone treatment:

   1. This will be studied using the various outcome measures from the neuropsychological tasks employed, comparing performance at week 5 to that at week 0, in patients treated with metyrapone compared with those treated with placebo. This will be conducted both using an intention-to-treat population and using a ‘per-protocol’ population.

3. To determine whether clinical improvements with metyrapone augmentation correlate with improvements in neuropsychological function and particularly (a) changes in cortisol sensitive tasks, e.g. spatial working memory and/or (b) 5-HT sensitive tasks, e.g. emotional processing:

   1. The correlation between changes in neuropsychological function (as measured above) with changes in clinical symptoms (MADRS, CAS, BDI, STAI) at week 5 compared with week 0 will be examined.

4. To compare visual cortical long-term potentiation (LTP) in patients with TRD and healthy control subjects:

   1. The degree and persistence of LTP induced in visual cortex by the tetanic presentation of a flashing checkerboard in patients (at week 0) will be compared with that in healthy control subjects. Analysis will include controlling for age, gender and IQ (as assessed using NART).

5. To examine if visual cortical LTP is altered by treatment with metyrapone:

   1. The degree and persistence of LTP at week 5 compared with week 0 will be examined in patients treated with metyrapone and placebo. This will be conducted both using an intention-to-treat population and using a ‘per-protocol’ population.

6. To compare emotional source memory performance, and its electrophysiological underpinnings, in patients with TRD and healthy control subjects:

   1. Behaviour responses, and related ERPs, recorded from patients at week 0 will be compared with those from healthy control subjects. Analysis will include controlling for age, gender and IQ (as assessed using NART).

7. To examine if emotional source memory performance, and its electrophysiological underpinnings, is altered by treatment with metyrapone:

   1. The Behaviour responses, and related ERPs, at week 5 compared with week 0 will be examined in patients treated with metyrapone and placebo. This will be conducted both using an intention-to-treat population and using a ‘per-protocol’ population.

8. To determine whether EEG predictors of conventional antidepressant response, and in particular specific predictors of response to serotonergic antidepressants, predict response to metyrapone augmentation:

   1. The correlation between a variety of EEG variables (including LDAEP slope, alpha power, alpha hemispheric asymmetry, anterior cingulate located theta power, and combinations of these), recorded at week 0 will be correlated with the change in clinical symptoms (MADRS, CAS, BDI, STAI) at week 5 compared with week 0 will be examined.

9. To compare the degree of activation of the amygdala in response to emotional faces or emotional words in patients with TRD and healthy control subjects:

   1. The fMRI BOLD response to emotional stimuli, particularly in relation to activity in the amygdala, recorded from patients at week 0 will be compared with those from healthy control subjects. Analysis will include controlling for age, gender and IQ (as assessed using NART).

10. To determine whether abnormalities in the degree of activation of the amygdala pretreatment is modified by metyrapone treatment:

    1. The change in fMRI BOLD response to emotional stimuli, particularly in relation to activity in the amygdala, at week 5 compared with week 0, in patients treated with metyrapone compared with those treated with placebo. This will be conducted both using an intention-to-treat population and using a ‘per-protocol’ population.

11. To determine if changes in the degree of activation of the amygdala with metyrapone post treatment correlate with clinical outcome:

    1. The correlation between change in amygdala activation at week 5 compared with week 0 with change in MADRS, CAS, BDI, STAI, EQ-5D (week 5 compared with week 0) will be examined.

12. To compare neural activity during episodic and working memory tasks in patients with TRD and healthy control subjects:

    1. The fMRI BOLD response during episodic and working memory tasks recorded from patients at week 0 will be compared with those from healthy control subjects. Analysis will include controlling for IQ (as assessed using NART) and if required age and gender.

13. To determine whether abnormalities in neural activity during episodic and working memory tasks is modified by metyrapone treatment:

    1. The change in fMRI BOLD response during episodic and working memory tasks at week 5 compared with week 0, in patients treated with metyrapone compared with those treated with placebo.

14. To determine if changes in neural activity during episodic and working memory tasks correlate with clinical outcome:

    1. The correlation between change in BOLD response during episodic and working tasks at week 5 compared with week 0 with change in MADRS, CAS, BDI, STAI, EQ-5D (week 5 compared with week 0) will be examined.

15. To determine whether patients with TRD have altered hippocampal response to hydrocortisone compared with healthy control subjects.

    1. The phMRI BOLD response related to the administration of hydrocortisone in patients at week 0 will be compared with those from healthy control subjects.

16. To determine if the hippocampal response to hydrocortisone is modified in patients by metyrapone treatment:

    1. The change in phMRI BOLD response related to the administration of hydrocortisone at week 5 compared with week 0, in patients treated with metyrapone

    2. The change in phMRI BOLD response related to the administration of hydrocortisone at week 5 in patients treated with metyrapone compared with those treated with placebo.

17. To determine the effect of genetic variations on cognitive and emotional processing and on predictors of treatment outcome using candidate genes with functional effects on HPA axis and monoamine function.

18. To examine the degree to which differences in ratings of depression between the scales used to assess mood at baseline predict response to metyrapone, including a focus on the degree of disparity between observer and self-ratings of mood:

    1. The difference in ratings of mood using the HDRS, MADRS and BDI at baseline will be correlated with the change in mood (assessed using MADRS) at week 5 compared with week 0.