Minocycline for negative symptoms of schizophrenia and possible mechanistic actions: the BeneMin RCT

Feasibility study

Large numbers of patients are needed to allow detection of subtle biological effects of psychiatric illness and drug action on brain structure and function. Imaging studies conducted across multiple centres offer major opportunities to bring patient recruitment into a manageable time frame. However, there are significant operational and statistical challenges, notably the addition of between-centre variance. To address these issues, the PsyGrid and NeuroPsyGrid25 consortia undertook a longitudinal calibration study in which 12 male volunteers were scanned at five centres under the same study protocol as we propose to use.

A voxel-based method of calculating statistical power for multicentre imaging studies was derived from these data and has been used as the basis for power calculations herein. 24 In addition, functional and structural MRI data were modelled at each brain voxel to estimate the partitioning of variance between the main effects of centre, subject, occasion and within-occasion order, as well as interactions of centre-by-occasion, subject-by-occasion and centre-by-subject.

Between-centre variance was limited to around 10% of the total. The main effect of subject was the largest variance partition for structural MRI (70–80%), and error (unexplained) variance was the largest for functional MRI (> 80%). Moreover, subject-by-centre interactions were generally 1–2% of the total variance. Therefore, there are no insurmountable obstacles to using MRI as an outcome variable in multicentre trials, and including a factor for centre in analysis falls within the guidelines of the Steering Committee of the International Conference on Harmonisation of Technical Requirements for Pharmaceuticals for Human Use, which cover the statistical analysis of magnetic resonance images.

Structural imaging

The loss of grey matter was assessed by computational segmentation of high-resolution T1-weighted images of the brain. As an adjunct to the measurement of cytokines in peripheral blood, a T2-weighted image (along with a proton-density-weighted image as part of a dual-echo sequence) was acquired to observe possible neuroinflammation. Furthermore, in combination with the T1 image, we also used multichannel texture analysis, pioneered in the assessment of multiple sclerosis lesions, to identify areas of abnormal MRI contrast associated with inflammation and to observe potential longitudinal changes.

Functional imaging of cognition

Echo-planar images depicting blood oxygen level dependent (BOLD) contrasts during performance of the N-back working memory task26,27 show engagement of an executive function network comprising the dorsolateral prefrontal cortex, the anterior cingulate cortex and the parietal cortex. Impaired functioning in this system has been strongly implicated in the pathogenesis of schizophrenia for several decades. 28 Evidence suggests that underlying deficits in intrinsic NMDA glutamate–gamma-aminobutyric acid (GABA) neurotransmission29 and impaired connectivity between elements of the network30 contribute to the impairment. Performance and fMRI measures in the N-back task should therefore be especially sensitive to any NMDA cognitive-enhancing effects of minocycline, particularly in schizophrenia. BOLD-sensitive data were also acquired while participants were resting (i.e. task absent) in order to investigate the endogenous dynamics and functional connectivity in frontotemporal circuits widely implicated in schizophrenia. 30

Research questions

If minocycline is started early in schizophrenia, does it improve negative symptoms or lessen their development over the next 12 months more than it does in established illnesses, and does this improve quality of life? Does minocycline work by its neuroprotective or anti-inflammatory actions or by its effects on glutamate?

Primary and subsidiary effectiveness predictions

• Prediction 1: minocycline minimises later negative symptoms when administered during the acute phase of psychosis, compared with standard care alone.

• Prediction 2: minocycline reduces weight gain and adverse metabolic changes associated with standard antipsychotic treatments.

• Prediction 3: improvements in negative symptoms will translate into improved functioning and quality of life.

Mechanistic hypotheses

Hypothesis 1: minocycline works by lessening a degenerative process that is most active in the acute phase of psychosis and is responsible for the development of negative symptoms. The hypothesis predicts that the loss of grey matter, known to occur during the early years following the onset of psychosis, will be lessened by minocycline treatment and that this will correlate with, and explain, improved negative symptoms.

Hypothesis 2: minocycline works by lessening an inflammatory process in the brain that gives rise to negative symptoms, possibly but not necessarily mediated by subtle neurodegeneration (see hypothesis 1). The hypothesis predicts that circulating proinflammatory cytokines will be lessened by minocycline treatment.

Hypothesis 3: minocycline works by ameliorating defective NMDA glutamate receptor function, which mediates negative symptoms. The hypothesis predicts that minocycline will improve cortical function as measured by fMRI activation during a working memory task and resting state connectivity, and reduce baseline concentrations of glutamate/glutamine, as measured by magnetic resonance spectroscopy. It also predicts that the benefits to negative symptoms wane when the drug is stopped. However, it is possible that glutamate actions could also be neuroprotective (see hypothesis 1) whether or not minocycline enhances glutamate function in the short term.

Some of the results of the BeneMin study have been published. 31

Inclusion criteria

• Male or female aged 16–35 years.

• Meeting the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV),32 criteria for schizophrenia or schizophreniform or schizo-affective psychosis as assessed by the research team.

• In an episode as defined by the presence of positive symptoms (scoring > 2 on items P1, 2, 3 or 6).

• In contact with early-intervention, community or inpatient services.

• Within 5 years of onset of symptoms.

• Intelligence quotient (IQ) of > 70.

• Male and female patients, and their partners, who were willing to use effective birth control, as defined in the patient information leaflet (PIL), throughout the study and for 7 days after stopping trial medication. Women had to have a negative pregnancy test.

• Able to understand and willing to give written informed consent.

• Fluent in English.

Exclusion criteria

• Current substance misuse diagnosis that in the opinion of the investigator may interfere with the study.

• Patients who, in the investigator’s judgement, pose a current serious suicidal or violence risk.

• Use of antibiotics of the tetracycline type within 2 months of baseline visit, or a history of sensitivity or intolerance.

• History of systemic lupus erythematosus or a history of it in a first-degree relative.

• Use of any investigational drug within 30 days of baseline visit.

• Relevant current or past haematological, hepatic, renal, neurological or other medical disorder that in the opinion of the PI/responsible medical officer (RMO) may interfere with the study.

• Taking medical treatments that could seriously interact with minocycline as described in the summary of product characteristics (SmPC) and judged by the PI/RMO.

• Clinically significant deviation from the reference range in clinical laboratory test results as judged by the investigator.

• Previous randomisation in the present study.

• Pregnancy or breastfeeding.

• Meeting the MRI scanning exclusion criteria (Manchester pro forma or equivalent).

Study withdrawal criteria

• The patient withdraws consent for any or no reason.

• Any adverse event (AE) considered to be related to active trial medication that is a threat to health or well-being, as determined by the local PI, the RMO or the patient.

• At the wish of the RMO.

• Safety reasons as judged by the PI, particularly if the patient becomes pregnant.

• Worsening of psychosis considered by the local PI, the RMO or the patient to be related to active trial medication.

• The patient is unable to comply with the restrictions on the use of concomitant medications listed in the SmPC.

• The patient is unable to tolerate the study medication.

Research assistant interview training and procedural standardisation

Detailed standardised operating procedures and checklists were drawn up for all ratings and procedures. The research team held 2-day meetings three times per year for harmonisation and problem-solving. Team teleconferences were held every 1–2 weeks. The main focus was conducting standardised SCI-PANSS interviews. Raters were trained and tested on the use of SCI-PANSS and how to make PANSS ratings using video recordings and group discussions. RAs scored PANSS negative symptoms on up to 11 different video recordings throughout the study to maintain consistency and inter-rater agreement. Agreement among the seven principal RAs of negative scores in up to 11 reference videos of SCI-PANSS interviews produced an intraclass correlation of 0.7.

Recruitment

Recruitment followed the procedures of PsyGrid. The RMO or another member of the clinical care team who knew the patient made the first approach. The local CRN CSOs had clinical research contracts with the CRN trusts, and they assisted with ascertaining suitable patients. Records were not accessed or screened by researchers until consent had been obtained. The RMO assessed diagnostic and other eligibility criteria using a checklist (Diagnostic and Eligibility Checklist) and invited potentially eligible patients to take part, at which stage patients received the PIL. The PIL included the information that travel expenses would be paid, and that their time and effort in attending each scanning session would be compensated with a £30 payment. The CSO/RA arranged a screening visit if patients wanted to participate. The CSO/RA logged the contact into a form on OpenCDMS, which allocated a patient identification number.

Screening visit

The RA obtained informed consent, having verified that the patient understood the PIL and what was involved in the study. Consent covered case-note review and the procedures in the PIL. Separate consent was obtained for gifting DNA for future genotyping studies in fully anonymised form to the University of Manchester. The CSO/RA applied a diagnostic checklist to the case notes on the basis of a Mini-International Nueropsychiatric Interview (MINI),33 reached a consensus diagnosis with the RMO and confirmed the presence of psychotic symptoms (scoring > 2 on items for hallucinations, delusions or suspiciousness). Blood was taken for renal and liver function tests. The participant undertook the IQ tests, and the CSO/RA tested the participants’ urine for drugs and for pregnancy in female participants.

The RA arranged the randomisation/scanning visit, having reviewed the screening investigations. The RAs ensured that the case report form checklist for the visit was complete and then logged the data into OpenCDMS.

Randomisation visit

This covered baseline ratings and up to 45 minutes of scanning. Participants could stop scanning at any time without withdrawing from the study. These activities could be spread over more than 1 day. The RAs checked consent to continue and completed the items listed. Saliva was collected using the Oragene^®-DNA OG-500 kit (DNA Genotek Inc., Ottawa, Canada), which was posted to the neuroscience and psychiatry unit. The patient attended their local scanner unit for a 45-minute scan. The RA, together with a radiographer, controlled the computer projection of the N-back task and recording of performance.

The RA informed OpenCDMS that the patient should be randomised. OpenCDMS then (1) allocated the patient to a treatment arm, (2) e-mailed the local pharmacy about which treatment ‘kit’ to use to dispense 3 months’ supply and (3) recorded when a kit was dispensed. Medication was collected by the RA.

Two-, 6- and 9-month visits

Table 1 gives a list of assessments. The RA received e-mail prompts for these visits from OpenCDMS. A second urine dipstick test for drugs of abuse was carried out at 6 months and reported to the local PI if positive.

Twelve-month final trial visit

During this visit, the screening visit safety measures and subsequent effectiveness measures were recapped. At all visits the patients’ cumulative clinical drug treatment was updated from the case notes. The MINI was repeated to ascertain diagnostic status. The scanning session was repeated. Trial medication was stopped.

Fifteen-month follow-up visit

Safety and efficacy measures were repeated.

Study medication and blinding

Minocycline (modified release) or matching placebo was taken as two 100-mg capsules per day for 2 weeks (the standard clinical dose), and increased to three 100-mg capsules per day for the remainder of the 12-month study period, added to standard APD therapy and routine care. A dose of 300 mg per day was selected to maximise exposure without increasing side effects. 34 Because it is a modified-release preparation, the capsules can be taken once, twice or three times per day to a maximum of three capsules in 24 hours, as preferred. Catalent Pharma Solutions (Somerset, NJ, USA) manufactured matching placebo and minocycline capsules, carried out quality control and assays as required by the Medicines and Healthcare products Regulatory Agency and carried out labelling and distribution of supplies to trust pharmacies.

Primary clinical outcome variable

1. Negative symptom severity as defined by negative symptom subscale score on the PANSS. This is the gold standard for comprehensively rating symptoms of schizophrenia. The negative symptom subscale is composed of seven items, each rated 1–7. 35

Secondary clinical outcome variables

1. Full-scale PANSS and positive syndrome subscale score.

2. Calgary Depression Scale for Schizophrenia (CDSS) (self-rating).

3. Functional outcome:

   • Global Assessment of Functioning (GAF) from DSM-IV32

   • Social Functioning Scale (SFS) self-rating in seven domains. 36

Primary biomarker outcome variables

1. Medial prefrontal grey matter volume (GMV) (H1).

2. Circulating cytokine IL-6 concentration (H2).

3. Dorsolateral prefrontal cortex BOLD response, percentage correct and connectivity during the N-back task (H3).

4. Cognitive outcomes:

   • Blyler Wechsler Adult Intelligence Scale (WAIS)III short form – current IQ37

   • IQ decline from premorbid IQ [Wechsler Test of Adult Reading (WTAR)], which predicts later negative symptoms37,38

   • Digit–symbol test – processing speed39

   • Verbal fluency – cognitive correlate of negative symptoms40

   • Auditory Verbal Learning Task – verbal learning. 41

5. Body weight and body mass index (BMI).

6. APD treatment in chlorpromazine equivalents.

Secondary biomarker outcome variables

1. Change in total and other regional GMV (H1).

2. Change in cytokine screen concentrations (H2).

3. Change in resting connectivity (H3).

Side effects and comorbidity

1. Substance misuse: urine drug screens.

2. Extrapyramidal symptoms (EPS):

   • the Simpson and Angus scale for pseudo-Parkinsonian symptoms and signs42

   • the Barnes Akathisia scale43

   • the Abnormal Involuntary Movement Scale (AIMS) for tardive dyskinesia. 44

3. APD subjective side effects: Antipsychotic Non-Neurological Side Effects Rating Scale

   (ANNSERS), developed in CUtLASS (Cost Utility of the Latest Antipsychotic drugs in Schizophrenia Study) and used in PsyGrid. 45

4. Seven-point treatment adherence scale. 46

5. Plasma minocycline concentration at 6 and 12 months.

Cytokine function

At the initial screen, a blood sample was collected into ethylenediaminetetraacetic acid (EDTA), and subaliquots of plasma were prepared for measurement of inflammatory markers. All markers were analysed at the King’s College London centre. Cytokines were measured using Meso Scale Discovery (MSD) (Rockville, MD, USA) V-PLEX sandwich immunoassays in accordance with the manufacturer’s instructions. MSD Proinflammatory Panel 1 (human) kits were used for the measurement of IL-1RA, IL-1β, IL-2, IL-4, IL-6, IL-8, IL-10, IL-12, IL-13, tumour necrosis factor alpha and interferon alpha. Patient blood samples were added to individual wells, together with a solution containing detection antibodies conjugated with electrochemiluminescent labels (‘sulfo-tag’, a MSD trademark). Eight standards (or ‘calibrators’) were created. All samples were measured in duplicate. With a few exceptions, all samples from the same patient were analysed together in the same plate. Each pair of plates (proinflammatory panel/customised cytokine panel) was analysed together so that the full range of analytes were measured in each patient sample in one session. High and low controls were used to assess variance between plates. The interassay coefficient of variations was < 10%. High-sensitivity C-reactive protein (hsCRP) and IL-1RA were measured at the Clinical Biochemistry Laboratory at King’s College Hospital. Further blood samples were collected at 6 and 12 months to evaluate treatment effects on inflammation and then again at 15 months to determine whether or not the impact had been sustained. Exploratory analyses of the relationship between changes in cytokine concentration and the effects of minocycline on clinical outcomes and on structural and functional imaging biomarkers were carried out. Plans to measure gene expression for cytokines in blood were not implemented.

Oragene kits were stored in the neuroscience and psychiatry unit; however, too few patients consented to donate DNA to make genotyping worthwhile.

Minocycline assay for compliance

After the end of the study, and with a minor amendment, spare plasma samples from the 6- and 12-month cytokine samples were assayed for minocycline. Only samples from the group allocated to minocycline were assayed, together with a few from the placebo group. Assays were carried out in the laboratory of Professor Nicholas Barnes at the University of Birmingham. On the day of analysis, patient serum was thawed and mixed with perchloric acid (Sigma-Aldrich, St Louis, MO, USA) to a final concentration of 0.2 M, to precipitate protein, which was removed by centrifugation (17,200 revolutions per minute for 5 minutes). Sample supernatant was transferred into autosampler vials (Thermo Scientific™ Chromacol™, Thermo Fisher Scientific, Waltham, MA, USA) for assay by high-performance liquid chromatography (HPLC)-ultraviolet/visible detection.

The HPLC system was a Dionex UltiMate™ 3000 RSLCnano UHPLC system with an automatic Rheodyne™ injector (Dionex Corporation, Sunnyvale, CA, USA) (130-µl fixed loop; 50-µl injection volume). Samples were separated using a Luna^® 5 µm C18 LC column (Phenomenex, Torrance, CA) using a mobile phase consisting of 25 mM KH₂PO₄ (pH 2.85) with 10% acetonitrile delivered at a flow rate of 1.0 ml per minute. After generation of a minocycline calibration curve, samples were run in batches of six before a positive control was run (20 µg/ml minocycline standard). Minocycline levels were quantified using a MWP-3000 RS diode array detector (UltiMate™ DAD-3000; ThermoFisher Scientific, Waltham, MA, USA) at a wavelength of 350 nm. The data were analysed by Thermo Fisher Scientific (Waltham, MA, USA) Dionex Chromeleon software to integrate peaks and calculate minocycline levels by referring to the calibration curve.

Screening and follow-up safety assessments

The clinical team and RAs confirmed the absence of symptoms and signs using a checklist and the recorded sitting blood pressure and heart rate. Blood samples were taken for haematology and differential white-cell count, and for clinical chemistry, to confirm normal renal and liver function. These assessments were repeated after 12 months in the trial or at the time of withdrawal from the study. The RAs tested urine for pregnancy in females (about 40% of the sample) with a dipstick test. This was repeated at each follow-up until the end of the treatment phase. In the event of a positive test, the patient was withdrawn from the investigational medicinal product and the general practitioner was informed. The participant was followed up for safety reporting to the conclusion of the pregnancy. AE reporting and procedures were followed (see protocol on the project web page: URL: www.journalslibrary.nihr.ac.uk/programmes/eme/0910023/#/; accessed 14 June 2019).

Sample size

The study was completed in 12 English and Scottish trusts, associated with six academic centres, each with a PI, a RA and an imaging centre. The chief investigator and research manager at the University of Manchester managed the trial. The academic centres had previously collaborated in the PsyGrid and NeuroPsyGrid consortia,25 which included four of the six imaging centres that had produced harmonised procedures through their involvement with the NeuroPsyGrid consortium. The sample size was derived from recruitment rates and PANSS scores from the PsyGrid study.

The study was designed to produce clinical and biomarker data in 170 patients completing 1 year of either placebo or minocycline add-on treatment (85 patients per group). This sample size has 90% power to detect a standardised effect size of 0.5 in the primary clinical outcome (e.g. a group difference in negative symptom scores of 3 units, assuming that the within-group standard deviation is equal to 6 units – as estimated from the Manchester-led MRI trial and PsyGrid clinical data) using a two-tailed t-test at a p-value of < 0.05. A difference of 3 units is the smallest effect that we would consider to have any clinical significance. A simple t-test produces a conservative estimate. Power was greater in practice using a repeated measures design and conditioning on relevant baseline covariates. For statistical reasons we chose not to base our sample size calculations on mediator variables or on their hypothesised relationship with the primary outcome. However, based on the NeuroPsyGrid data collected from five research centres, the minimal detectable difference in grey matter is 2% at 80% probability with the sample size calculated above. This is much smaller than the published MRI changes over 1 year. 14,15

Calculations about recruitment were based on previous experience with the PsyGrid CRN consortium and our previous minocycline study. In the Chaudhry et al. study2 in Pakistan and Brazil, 25% of those assessed were randomised and 29% of patients dropped out during the trial to an equal extent in both arms. We assumed a 25% dropout rate both from screening and from randomisation onwards. These figures were intended to be pessimistic. It was anticipated that dropout rates might be lower in the UK than in Pakistan and Brazil because of the more developed clinical care system and the involvement of patients and patient organisations in the design and monitoring of the study. The assumed dropout rates gave figures of 282 participants at screening and 226 participants at randomisation to produce completion in 170 participants. Each recruitment centre therefore needed to screen 2.1 patients per month and randomise 1.7 patients per month. These figures were achieved by several centres in the PsyGrid study. The RA workload was realistic at a rate of one combined MRI scanning and clinical rating session per week and one or two clinical follow-up ratings per week.

Derivation of primary and secondary mechanistic variables

Almost all procedures and tasks for MRI assessment were those implemented in the NeuroPsyGrid collaboration. MRI assessments were undertaken by patients at baseline and 12 months, and were analysed to test the mechanistic hypotheses. Differences between treatment arms and associations with outcome variables were tested both in a hypothesis-driven approach by focusing on a priori regions of interest, defined by an anatomical atlas, and in an exploratory manner at all intracerebral locations (i.e. whole brain).

T1-weighted high-resolution MRI images were acquired, from which estimates of the distribution of GMV were made using standard voxel-based morphometry. All images were registered in a standard anatomical space of the Montreal Neurological Institute template to facilitate between-subject analyses. Additionally, T2-weighted and proton density images were acquired with a dual-echo sequence, which, together with the T1 data, were used to identify areas of abnormal MRI contrast associated with inflammation in a whole-brain examination. Mean GMV was extracted from the medial prefrontal cortex for each patient (H1), as well as across the whole brain.

Echo-planar images depicting BOLD functional contrast were acquired during performance of the N-back working memory task, which engages an executive function network. BOLD-sensitive data were also acquired while participants were resting (i.e. without externally applied stimuli) to investigate endogenous dynamics and functional connectivity.

BOLD-sensitive data sets were initially pre-processed to correct for patient head movement. Root-mean-square head deviations were estimated for each participant across the acquired volumes to estimate residual head motion. At each intracerebral voxel, a general linear model was regressed onto the BOLD time series to estimate orthogonal contrasts of stimuli of the working memory task, namely 1-back and 2-back versus 0-back, and 1-back versus 2-back. The patterns of brain activation associated with the contrasts were estimated from the baseline data of all patients. Mean BOLD responses were extracted from regions of interest defined by the anatomical atlas and corresponding to significantly activated regions, namely the dorsolateral prefrontal cortex (H3). Mean reaction time and accuracy of response to task stimuli were also recorded.

Statistical testing of regions of interest imaging data (functional or structure) used similar statistical models to those used for clinical outcome measures, comprising a main effect of treatment, a main effect of time and a treatment-by-time interaction, along with covariates for centre, age and sex. Effects of treatment were identified by significant treatment-by-time interactions, which were further investigated for the direction of the effect by appropriate post hoc testing. These models were also adopted for testing behavioural measures during the working memory task as well as at each intracerebral voxel in whole-brain testing of imaging measures. To control for type I errors due to multiple comparisons, and to account for non-independent values in neighbouring voxels, robust, well-validated methods of non-parametric inference on spatially extended regions were used.

Overview

The statistical analysis was overseen by study statistician, PI Graham Dunn. A draft statistical analysis plan was submitted to the TSC for approval as part of its 9-month assessment (see Report Supplementary Material 1). There were no interim analyses, and all analyses were carried out after the collection of the final outcome measures. All statistical analyses of the clinical, cognitive and biomarker outcomes were analysed using Stata^® version 14 (StataCorp LP, College Station, TX, USA).

Summary statistics

Tables of summary statistics were generated for all primary and secondary outcomes and for selected baseline variables. These were inspected for preliminary evidence of treatment effects. Baseline variables were used to check the baseline balance of the randomised arm.

Treatment effect estimation

Treatment effects were reported using 95% confidence intervals (CIs), supplemented by their associated p-values. An effect was regarded as statistically significant if the p-value was ≤ 0.05, which was reduced to 0.01 when treatment effects at each of the four follow-up times were evaluated separately.

Note that the treatment effect is essentially the average outcome for the treated arm minus the average outcome for the placebo arm. So for a severity score (high is a poor outcome), a beneficial effect of minocycline would be indicated by a treatment effect estimate that is negative (i.e. below zero). For a function score (high is a good outcome), a beneficial effect would be indicated by a treatment effect estimate that is positive (i.e. above zero).

Primary clinical outcome

Treatment effects on severity of negative symptoms were estimated through the use of a random-effects regression model (using Stata’s xtreg command) after allowing for time of follow-up (2, 6, 9 or 12 months: treated as a categorical variable), centre and baseline (pre-randomisation) severity of negative symptoms. The effect of time of follow-up on treatment efficacy is evaluated by the treatment-by-time interactions and treatment effects at the four follow-up times, estimated accordingly. When there was no significant variation in the treatment effect over time (i.e. no treatment-by-time interaction), the interaction was dropped from the model and treatment efficacy common to all four follow-up times was estimated. All models contained centre-by-time and baseline severity-by-time interactions. Sensitivity of efficacy estimates for the effects of poor compliance with medication and other covariates on loss to follow-up was assessed through descriptive summary statistics and the use of inverse probability weighting as described below.

Secondary clinical outcomes

Similar methods to those used for the primary clinical outcomes were also used here, using mixed-effects models involving data from 2-, 6-, 9- and 12-month follow-up, or, where appropriate, using data collected at 6 and 12 months. For variables that were collected only at 12-month follow-up, a simple analysis of covariance (ANCOVA) model was employed, specifying treatment centre and corresponding baseline value of the outcome as covariates.

Primary biomarker outcome variables

Grey matter volume was measured at baseline and 12 months only, so a simple ANCOVA model was employed, specifying treatment centre and corresponding baseline value as covariates. IL-6 and hs-CRP, being measured at two follow-up times, were analysed in the same way as were negative symptoms.

Allowing for missing outcome data: construction of inverse probability weights

Treatment effect estimates may be sensitive to assumptions concerning mechanisms of attrition. An obvious candidate as a predictor of attrition is the level of compliance with allocated medication. Unfortunately, only 87% (180 out of 207) of the trial participants provided data on their levels of compliance (see Results) at one or more of the follow-up assessments, and it was therefore decided that this component of the trial analysis should be relatively informal; it is reported in Appendix 2. Rather than look at the full data set, the analyses were limited to looking at the effects of minocycline on negative symptom scores, the primary outcome variable only, at the 12-month follow-up assessment.

To investigate patterns of attrition, a logistic regression model was used to explore which baseline characteristics, together with treatment allocation and patterns of adherence to allocated treatment, predicted which participants provided negative symptom outcome data at 12 months after randomisation (separately for each randomised arm). If deemed necessary, the final models were used to generate an expected probability of providing outcome data (for each arm separately), and the reciprocal of this estimated probabilit was used as an inverse probability weight for use in the random-effects models for the primary outcome.

Route to recruitment

Patients were recruited from community mental health teams in a variety of ways from site to site and trust to trust. RAs kept in contact with their local community mental health teams. They reviewed the current caseloads with care co-ordinators or medical staff, and they attended outpatient clinics. The CSOs from the local NIHR Clinical Research Network teams were briefed about the study at site initiation visits and throughout the study. Generally, community mental health teams and CSO staff made initial contact with potential patients to request permission for RAs to approach them about the study and to talk through the PIL. The number of patients considered potentially eligible varied widely because in some centres computerised lists of patients were available. Altogether, 2227 patients were considered for inclusion and 572 received PIL visits. After 24 hours, 267 consented to be screened for inclusion and were allocated an identification number by OpenCDMS. A total of 229 patients attended a screening assessment, of whom 10 failed to meet the inclusion criteria, five did not consent to continue and four withdrew. A total of 207 patients were successfully randomised and allocated to a treatment code and pharmacy pack: 104 were assigned to receive minocycline and 103 were assigned to receive placebo.

Flow through the study

The Consolidated Standards of Reporting Trials (CONSORT) flow diagram (Figure 1) shows the numbers of participants remaining in the study, including those who missed an assessment but attended a subsequent one, up to and including the 12-month visit: the end of the treatment phase. The number of participants seen at each appointment for whom ratings were recorded is also shown. Overall, 78 out of 207 (38%) of the randomised participants dropped out during the 12-month treatment phase. The great majority (63/78, 81%) of the reasons were either participant’s request or loss to follow-up. In terms of the primary outcome measure at 6 months, scores were available for 66% of those randomised and, at 12 months, 61%. The numbers of participants dropping out were identical (39 on placebo and 39 on minocycline) and 33 and 3, respectively, withdrew and were lost to follow-up.

FIGURE 1.

The CONSORT flow diagram. f/u, follow-up; ID, identification; LTFU, lost to follow-up; w/d, withdrawn.

Table 2 shows the assignment to study treatment across the 12 allocation centres, each of which had a collaborating trust pharmacy, linked to Open CDMS, and the six universities of the PIs and RAs. It can be seen that each centre allocated a similar number of participants to placebo and minocycline. Each pharmacy and trust required a site initiation visit, so there were varying delays before centres were open for recruitment.

The course of recruitment in each centre is shown Figure 2. The first pharmacy was ready to dispense in London in December 2012, but the first patient was not randomised until 16 April 2013. The 207th patient was randomised in May 2015 and the last 12-month visit was on 9 September 2016. A steady randomisation rate of 11 per month was achieved throughout 2014 (see Figure 2). In October 2013, recruitment was lagging behind the target rate of 225 participants, with 172 completing the 12 months of the study. With the approval of the Efficacy and Mechanism Evaluation programme, the target was reduced to 207 on the grounds that retention was better than expected, and a new collaborating centre with a new PI, Rachel Upthegrove, was initiated, together with a new recruiting centre, the Cheshire Wirral Partnership. These new centres recruited at a rapid rate, and, together with improved recruitment in other centres, 207 participants had been recruited by May 2015 (Figure 3).

FIGURE 2.

Target, revised target and actual numbers recruited (blue line).

FIGURE 3.

Course of recruitment by centre. Vertical axis shows the cumulative number of participants recruited. CWP, Cheshire Wirral Partnership; KCL, King’s College London; UCL, University College London.

Overview of minocycline effects

Minocycline had no discernible influence on any outcome variable in terms of direction, magnitude or statistical significance. The treatment effects common to all follow-ups are summarised in Table 4, and none achieved statistical significance. Treatment effects are the difference in group means (minocycline minus placebo) attributable to treatment after controlling for baseline and centre, as described in the analysis plan. Treatment effects with negative signs indicate lower scores in the minocycline group than in the placebo group. Thus, negative treatment effects for severity variables indicate a beneficial effect of minocycline, whereas for functional scores, positive treatment effects indicate a beneficial effect of minocycline. There is no consistent tendency for beneficial effects with minocycline across the outcome measures. Because there are no statistically significant effects, statistical analyses are presented only for primary outcomes; other analyses are presented in Appendix 1.

Negative symptom subscale score

The minimum possible negative score is 7 and the maximum is 49. Scores decreased by slightly less than 3 points over the trial (Table 5). There are no significant treatment effects (Tables 6–8). Negative scores calculated from items N1 + N2 + N3 + N4 + N6 + G7 + G16 according to Marder et al. 48 produced very similar means, correlating > 0.9 at each time point with the standard negative score. There is no indication of a rebound worsening after cessation of treatment at 12 months (see Table 5).

Analysis for scores and other fixed models are systematically presented in the same way for each of the outcomes: (1) estimates of relevant model parameters (see Table 6), (2) estimates of treatment effects at the available follow-up times (see Table 7) and (3) estimate of treatment effects, assuming that there is no variation in the treatment effects across the different follow-up times (see Table 8).

Positive symptom subscale and total scale scores

Total scores at baseline of about 70 correspond to a mild to moderate overall severity of illness,47 and this is concordant with the GAF scores in Table 9. Mean scores improved in both groups to a similar extent (see Table 9).

Calgary Depression Scale for Schizophrenia (self-rating)

The mean CDSS scores of > 5 predict depression diagnosis with a sensitivity of 88% and a specificity of 69%. 49 This indicates a high prevalence of depressive symptoms. Although scores improve in both groups, this may reflect dropout of those with greater depressive symptoms (Table 10).

Global Assessment of Functioning

This measure rates whichever is the more severe of impaired social function and symptoms severity, with lower values indicating the greater impairment. The mean values lie in the 50–60 range, which represents a moderate level of symptoms or social impairment. Small improvements in mean GAF score occurred in both groups, although these were slightly greater in the placebo group than in the minocycline group (Table 11).

Social Functioning Scale

Greater scores indicate better social function, but there were no group differences, and the scores did not change over the course of the study (Table 12).

Cognitive performance

Cognitive scores tended to improve equally in both groups (Table 13).

Body weight and body mass index

Increases in BMI and body weight occurred with both placebo and minocycline, but there were no statistically significant differences between the treatment groups (Table 14).

Antipsychotic drug treatment

Drug treatments at baseline and at 12 months were grouped into main classes and each class was expressed as a percentage of the total number of drugs prescribed. There were no clear group differences in the pattern of treatment current at baseline or at 12 months. Almost all patients were taking second-generation (‘atypical’) antipsychotics rather than first-generation medications. Many patients in both groups were taking antidepressants at both times (Figure 4).

FIGURE 4.

Drug classes by treatment.

The most common oral antipsychotics used did not differ at baseline, and nor did the average daily dose, expressed in olanzapine equivalents (Table 15).

The groups were well matched with regard to oral antipsychotics, with olanzapine being the most frequent. The mean daily doses in olanzapine equivalents (mg/day) were closely similar (see Table 15).

Few participants were on depot/long-acting injection antipsychotic medication; the most common were paliperidone and its parent compound, risperidone (Table 16). The groups were well matched with regard to long-acting injections. ‘Clozapine’ was recorded as the depot name in three cases. This may have been a confusion with clopixol.

Grey matter volume

There were no systematic trends in GMV over time and no treatment effects (Tables 17 and 18).

Circulating interleukin 6 and high-sensitive C-reactive protein concentrations

There were no systematic trends in cytokine concentrations over time and no treatment effects (Tables 19–22).

N-back performance and functional magnetic resonance imaging blood oxygen level-dependent response in dorsolateral prefrontal cortex

There were no statistically significant treatment effects on performance or on fMRI BOLD responses during the N-back task (see analysis in Deakin et al. 31 and Report Supplementary Material 1).

Total and other regional grey matter volumes

Substance misuse

There were more cannabis-positive urine tests in the placebo group at baseline and 6 months than in the minocycline group (Table 23). Confining an exploratory analysis to those with cannabis-free urine did not reveal a treatment-responsive group.

Extrapyramidal symptoms side effect summaries

The prevalence of EPS side effects was low and there were no group differences (Tables 24–26).

Non-neurological side-effect scores were minimal and did not differ between the groups (Table 27).

Seven-point treatment adherence scale

Adherence scores were bimodally distributed (Table 28); most participants rated their adherence maximally, but the percentage of scores of < 3 (%) increased to 23–25% by 12 months.

There were no group differences in the distribution of adherence scores in Table 28. Analysis of the effect of treatment on subscales in participants with maximal self-rated adherence (score of 7) did not reveal a statistically significant benefit or a trend towards a benefit of minocycline on the primary or secondary symptom ratings (results not shown).

Low treatment adherence scores at 2 months predicted approximately 80% non-completion (i.e. dropout) at 12 months, and this was equally true in both treatment arms (Table 29). The data formed part of the sensitivity analysis reported in Appendix 2.

Plasma minocycline concentrations at 6 and 12 months

Minocycline was assayed in 56 blood samples from the minocycline group. It was detectable in 32 (57%) at 6 months and 35% at 12 months (Table 30). Analysis of the effect of treatment on subscales in the 32 participants with detectable minocycline levels compared with the placebo group did not reveal a statistically significant or trend towards benefit of minocycline (results not shown).

Adverse events and reactions

Sixty-seven AEs were reported for participants in the placebo group and 60 were reported in the minocycline group in the per-protocol reporting period between randomisation and 15-month follow-up (Table 31). They occurred in 14 MedDRA system organ class categories (www.meddra.org/), with few marked differences in frequency. The most common MedDRA categories were gastrointestinal (category 7; n = 31), psychiatric (category 19; n = 24), neurological (mostly headache; category 7; n = 24) and dermatological (category 23; n = 18) (see Table 29). Worsening of psychiatric illness was twice as common in the placebo group as in the minocycline group (16 compared with 8 of the 207 total randomised). Eleven adverse reactions (i.e. AEs that are judged possibly related to trial medication) occurred in the placebo group and five occurred in the minocycline, consisting of skin rash, headache and gastrointestinal upset.

Serious adverse events

Hospitalisations were more common in the minocycline group (n = 18) than in the placebo group (n = 11) (Table 32). There were 15 admissions in 10 minocycline-treated patients, and 10 admissions in six placebo-treated patients. The admissions were due to worsening of psychiatric illness, with a combination of intensification of psychosis, dysphoria and suicidal ideation and intent. None was rated by the local PIs as a reaction to trial medication or as a suspected unexpected serious adverse reaction; other factors were noted, such as poor adherence to medication and alcohol and substance misuse.

No deaths occurred during the trial. Two deaths occurred after completion of the minocycline group. One patient died of an overdose 9 months after their 12-month study visit. Another developed a deep-vein thrombosis and died of a pulmonary embolus 2 months after the follow-up visit (5 months after the end of trial treatment). Although this event occurred outside the reporting period, a thorough search of the literature and web resources was made, and no cases of deep-vein thrombosis or pulmonary embolus associated with minocycline use have been described. The patient was prescribed clozapine and was a smoker, both of which are known to be predisposing factors.

Social Functioning Scale 1: social engagement/withdrawal

Social Functioning Scale 2: interpersonal behaviour/relations

Social Functioning Scale 3: independence-performance

Social Functioning Scale 4: recreation

Social Functioning Scale 5: pro-social activities

Social Functioning Scale 6: independence-competence

Social Functioning Scale 7: employment/occupation

Estimated treatment effects: processing speed scores and current intelligence quotient

Body weight (kg) at 12 months

High sensitive C-reactive protein