Methods

Trial objectives

The overall aim of the trial was to investigate the effects of OROS-MPH on reducing ADHD symptoms in young male prisoners aged 16–25 years who meet the DSM-5 diagnostic criteria for ADHD. The following study questions are addressed in three main objectives:

• Primary objective – what is the efficacy of OROS-MPH in reducing inattention and hyperactivity/impulsivity in young male prisoners who meet the diagnostic criteria for DSM-5 ADHD?
• Secondary objective: what is the efficacy of OROS-MPH in reducing secondary outcomes that are key indicators of behavioural and functional impairments used in the management of young male prisoners in the UK? These include emotional dysregulation, antisocial behaviour in the prison, violent attitudes (a measure linked to aggression) and reports of behaviour from prison staff.
• Tertiary objective: are improvements in secondary behavioural outcomes mediated by improvements in ADHD symptoms or emotional dysregulation?

Trial design

Concerta in Adult Offenders with ADHD (CIAO)-II is a Phase IV, 8-week, parallel-arm, double-blinded, randomised, placebo-controlled trial of an extended-release formulation of MPH (OROS-MPH). ADHD symptoms, emotional dysregulation, attitudes towards violence, and behavioural outcomes were compared between the two trial arms.

Participants in the trial were young male prisoners aged 16–25 years who met the DSM-5 criteria for ADHD. Participants were randomised to 8 weeks of treatment with either OROS-MPH or placebo, titrated over the first 5 weeks to balance ADHD symptom improvement with adverse effects. A total of 200 participants were randomised and allocated in a 1 : 1 ratio to either the drug or placebo. By the end of the trial, 101 participants had been randomised to receive OROS-MPH and 99 had been randomised to the placebo arm. Randomisation was conducted by the King’s Clinical Trial Unit (KCTU), with blinding of investigators, including pharmacy staff, and participants. On completion of the 8-week trial, OROS-MPH or another medication for ADHD was offered to both the OROS-MPH and placebo groups as part of their ongoing clinical care.

Changes to protocol after trial commencement

A full list of all minor and substantial amendments to the protocol can be found (see Appendix 1, Table 31). Following the start of the trial, two minor amendments were made to the protocol, related to a clarification regarding the assessment of intelligence quotient (IQ), and to clarify that the Reactive–Proactive Aggression Questionnaire (RPQ) measure would be collected at baseline only. In addition, a substantial amendment was implemented on 18 December 2018. This amendment was approved by the Medicines and Healthcare products Regulatory Agency (MHRA), the Health Research Authority (HRA), Her Majesty’s Prison and Probation Service (HMPPS) (formerly known as the National Offender Management Service) and the Scottish Prison Service (SPS). Changes included an amendment to the timing of the database lock and clarifications to the wording describing the inclusion and exclusion criteria. The revised inclusion and exclusion criteria definitions had been applied from the start of the trial, but were not sufficiently well documented in the protocol. In addition, one of the secondary outcomes, the number of Incentive and Earned Privileges (IEPs) was dropped because these data are not recorded in Scottish prisons. No other outcomes were changed. A report on the final approved protocol is published in Trials.52 This report refers to the final approved protocol.

Eligibility criteria for participants

Eligibility for the study was checked and recorded once a consent form (consent II: consent for the clinical trial) had been signed and baseline assessments had been completed, prior to randomisation. Using an algorithm that applied the DSM-5 criteria to the DIVA 2.0 interview data, cases were checked to ensure that they met the diagnostic criteria for DSM-5 ADHD. A clinical review by a psychiatrist trained in the diagnostic assessment of ADHD was then completed to check all of the inclusion and exclusion criteria. Mental capacity required for consent was evaluated by the clinician, including ability to understand all information, rating scale and interview questions, and to provide detailed descriptions of symptoms and behaviour.

The initial exclusion criteria of an IQ score of < 60 was based on the Wechsler Abbreviated Scale of Intelligence^® Second Edition (WASI-II) estimates of IQ. This was later changed to include those with a WASI-II IQ estimate of < 60, for whom the score’s 95% CI spanned 60, and clinical assessment by the psychiatrist concluded that IQ was > 60 and that the participant had the mental capacity to provide informed consent. As there are no validated IQ tests for the visually impaired, including the WASI-II, this criterion was based on clinical judgement for participants with this impairment. This was also the procedure for anyone unable to complete the WASI-II assessment because of the severity of their ADHD symptoms or other mental health problems. This approach was applied to one participant with a low initial WASI-II test score.

Inclusion criteria

The inclusion criteria were as follows:

• male
• aged between 16 and 25 years at time of consent for screening
• ability to speak English, defined as sufficient to complete study assessments
• able to provide informed consent, defined as understanding the information sheet and being able to make an informed decision, taking into account pros and cons of trial participation
• meeting clinical diagnostic criteria for DSM-5 ADHD.

Following the DSM-5,1 the diagnostic criteria for ADHD were defined as five or more symptoms of ADHD in either the inattentive or hyperactive–impulsive symptom domains, and six or more symptoms of ADHD in either the inattentive or hyperactive–impulsive symptom domains before the age of 12 years. When it was not possible to gain enough clinical information to score childhood symptoms of ADHD, the operational criteria applied were adapted to include evidence of several ADHD symptoms with impairment starting before the age of 12 years, and five or more symptoms in either domain currently, with moderate to severe impairment. This meets the requirement under DSM-5 for several symptoms starting before the age of 12 years.

In addition, we required a persistent non-episodic course of ADHD symptoms, impairments from the symptoms in two or more clinical or psychosocial domains, and that the symptoms were present in two or more settings.

Exclusion criteria

The exclusion criteria were as follows:

• lacks capacity to give informed consent
• moderate or severe learning disability, defined as IQ of < 60
• serious risk of violence to the researcher
• current major depression, psychosis, mania or hypomania
• past history of bipolar disorder or schizophrenia, defined as those with episodic mania/hypomania or psychosis unrelated to acute drug intoxication, but not chronic emotional dysregulation (irritability, frustration, anger or emotional–mood instability)
• medical contraindications to the use of stimulants, including glaucoma, hypertension, cardiovascular disease or structural heart problems
• taking contraindicated medications during the 4 weeks prior to randomisation
• showing drug-seeking behaviour or craving, defined as drug-seeking behaviour that was unusually severe and likely to affect the titration protocol because of unusual and excessive demands for drugs, or if there was a current withdrawal syndrome from an addiction disorder with drug dependency
• receiving any ADHD medication between consent for screening and randomisation.

Trial setting

Participants were recruited from two prisons. HMP YOI Isis in London, England, opened in 2010 for sentenced young adults aged 18–25 years, but was later extended to include older prisoners. This is a category C prison, defined as suitable for prisoners who cannot be trusted in open conditions but who are unlikely to try to escape. There are two house blocks with four wings each, with mixed single and double cells, plus a segregation unit located between the house blocks. Operational capacity was 621 in January 2016. The prison focuses on rehabilitation and resettlement, and all prisoners are engaged with full-time education, occupation and rehabilitation programmes. There is a broad-based educational and vocational curriculum for young sentenced prisoners and available activities are, among others, mathematics, English, business studies, painting and decorating, waste management, mechanics and barbering.

Her Majesty’s Young Offender Institution (HM YOI) Polmont in Falkirk, Scotland, is a holding facility for young offenders in Scotland aged 16–21 years, with sentences ranging from 6 months to life. All participants were sentenced prisoners when screened for entry to the trial. It has the capacity to hold 760 prisoners and, since 2016, female prisoners have been housed in a separate wing. Individuals may stay in Polmont until age 23, depending on circumstances. In March 2019, Polmont had 370 male young offenders and 90 female young offenders. There are three house blocks and a segregation unit. Various activities are offered at Polmont, such as education and vocational training, and rehabilitation programmes such as Pathways, Moving Forward and Making Changes.

Recruitment procedure

Following consent to be screened for ADHD (consent I), screening questionnaire data were collected by the research teams using the Barkley DSM-IV ADHD symptom rating scale.30 Prisoners who screened positive were invited to complete the DIVA 2.0.31 This was followed by a clinical assessment by a psychiatrist trained in the diagnostic assessment of ADHD, including collateral information obtained from an informant whenever feasible. Participant’s expected release date could change at any time and additional convictions and charges could be added. Hence, it was not always possible to predict if a participant would be transferred, released or deported. For this reason, it is possible that some participants initially excluded because of a high risk of release could have taken part in the trial.

Following clinical review, prisoners who met the diagnostic criteria for ADHD and who met the other eligibility criteria for the trial were invited to take part in the clinical trial. Eligibility for the trial was further checked and recorded once the consent form for the clinical trial (consent II) had been signed and baseline assessments had been completed, prior to randomisation. Using an algorithm that applies the DSM-5 criteria to the DIVA 2.0 data, cases were checked to ensure that they met diagnostic criteria for DSM-5 ADHD. A clinical assessment and review by a psychiatrist trained in the diagnostic assessment of ADHD checked all inclusion and exclusion criteria prior to randomisation.

Intervention

Trial medication

The trial medication consisted of 18-mg capsules of OROS-MPH or placebo capsules.

Capsules were overencapsulated and packaged in bottles of 46. Each bottle was assigned a unique randomisation number and the randomisation system allocated the right bottle to each participant. Overencapsulation has been successfully adopted in previous studies to generate matched placebo to OROS-MPH, including the previous Swedish prison trial.33 OROS-MPH was provided as commercial stock by Janssen-Cilag Ltd. Piramal Healthcare UK Ltd (Morpeth, UK) (arranged by Modepharma Ltd, Beckenham, UK) provided the trial kits by overencapsulating the OROS-MPH capsules and manufacturing the placebo to match, and by providing approved packaging and labelling, Qualified Person Certification and distribution for 220 patients. Janssen-Cilag Ltd provided the Summary of Product Characteristics (SmPC), which was updated throughout the trial.

The overencapsulated active capsules were repacked in high-density polyethylene (HDPE) plastic bottles without the need for a stability programme, as 18-mg OROS-MPH has marketing authorisation for both HDPE and blister packaging. Placebo capsules were manufactured once. Trial medication overencapsulation and packaging was undertaken in two campaigns to accommodate a trial duration of up to 3.5 years. OROS-MPH 18-mg capsules typically have a maximum shelf life of 3 years from the date of manufacture; however, by the time the product was repacked for the clinical trial, the remaining shelf life was likely to be < 2.5 years. Overencapsulation used DBcaps^® (Capsugel, Inc., Morristown, NJ, USA) capsules, which are designed specifically for the blinding of clinical trial medication.

The reason for overencapsulating the OROS-MPH and placebo capsules, rather than manufacturing a matching placebo tablet, was because OROS-MPH capsules have printing on them and are of a distinct shape that would be difficult to manufacture and might have infringed copyright. We sought advice on this from previous investigators who used OROS-MPH, and from companies that provide drug and placebo supplies for studies. Studies on the use of DBcaps have shown that encapsulation of capsules results in a lag time of 2–3 minutes in disintegration, compared with the unencapsulated capsules.53–55 The pharmacokinetic properties of 18-mg OROS-MPH prolonged-release capsules indicate release over several hours: following oral administration of OROS-MPH, the drug overcoat dissolves, providing an initial maximum drug concentration at about 1–2 hours. The MPH contained in the two internal drug layers is gradually released over the next several hours. Peak plasma concentrations are achieved at about 6–8 hours, after which plasma levels of MPH gradually decrease (section 5.2 of the SmPC56).

Prescribing procedure

All prescriptions were completed by the trial psychiatrists at each site. Each patient was allocated a kit consisting of a labelled carton containing four labelled bottles, each containing 46 of the 18-mg OROS-MPH capsules or 46 of the placebo capsules. Each kit and its bottles were labelled according to annex 13 of the European good manufacturing practice (GMP)57 and had their own unique treatment pack number. The centralised randomisation system allocated the correct treatment pack number to each participant during the trial. Trial medication was delivered as prescribed daily, with participants observed by prison health-care nursing staff to ensure that they swallowed the capsules. There is a daily record of how many capsules were given out and taken by the participants, recorded by the nursing staff.

Titration procedure

All trial medication was titrated in the same way. Treatment started at an initial dose of one capsule of trial medication for 1 week. The number of capsules was then increased weekly over the following 4 weeks, in increments of one capsule, to a maximum of four capsules. This reflected a dose range for the active medication of 18, 36, 54 and 72 mg. Titration upwards was stopped if all 18 ADHD symptoms were scored as negligible or absent (score of 0 or 1 on the CAARS-O) or if unacceptable adverse effects were reported. The trial medication could also be reduced by one capsule if there was a limiting adverse event (AE), in which case there were no further increases in dose for the duration of the trial or, potentially, the cessation of trial medication in some cases. Individual titration protocols are routinely used in clinical practice as they optimise the control of ADHD symptoms while minimising potential adverse effects, and have been used successfully in previous efficacy trials of OROS-MPH.58,59

Medication was provided either once or twice daily, in the morning and/or at noon, up to the maximum daily dose. This provided flexibility to tailor the dose to the individual effects reported by participants and to maximise adherence to trial medication within the constraints of the prison system for delivery of controlled medications. A maximum dose of four capsules, equivalent to 72 mg of OROS-MPH, was included for this trial because previous trials and clinical consensus indicate that there is considerable individual variation in dosing to obtain the optimal balance of good control of ADHD symptoms while minimising common adverse effects. Although current licensing for OROS-MPH is up to 54 mg, this is based on dose levels for children and adolescents, rather than young adults, and NICE recommend a daily dose of MPH in adults to a maximum of 100 mg per day.13,14 For OROS-MPH, the British National Formulary60 recommended doses up to a maximum of 108 mg in adults.

Unblinding

An emergency unblinding service was provided, but was not required at any stage of the trial. It was set up so that any health-care staff would be able to call an emergency number that was added to the labels for both the bottle and the box with the trial kit. In addition, code-breaking envelopes were printed and held at the company providing the unblinding service in case of an emergency, but were not used. Unblinding did not take place for any participants during the trial.

Adherence to trial medication

Based on experience from the pilot study, we envisaged that adherence to medication would present a challenge for around 20% of participants. This could be for several reasons. Some offenders may not feel motivated to take the trial medication if they experience adverse effects or do not feel that they are improving. They may also take medication intermittently because of strict prison regimes that allow only a brief time for leaving their cells to obtain medication or difficulty in getting to the location where medicines are dispensed. These cases were not expected to contribute to missing data, but could affect the overall outcome of the intention-to-treat (ITT) analyses.

In the pilot study at HMP YOI Isis, we accrued considerable experience in managing the expectations of offenders and providing the support needed to help participants adhere to the trial protocol. The following steps were therefore adopted to maximise adherence to medication:

• In the pilot study, minor adverse effects were the most common reason for non-adherence to medication, affecting 13% of participants. This was linked to the observation that this population may be more sensitive to minor adverse effects, particularly changes in appetite, and sleep problems. To minimise this effect, we took care to identify the early signs of minor adverse effects such as appetite loss and to adjust the medication dose accordingly by systematic evaluation of ADHD symptoms and minor adverse effects during the titration phase of the trial.
• Seven per cent of the pilot sample did not wish to take medication in the mornings and the protocol for the current trial was adjusted to allow for medication to be taken later in the day, until noon. This allowed for participants who got up later in the day, who worked mainly in the afternoons or who had a strong preference for mid-day dosing. This flexibility in dosing time reflects dosing decisions in non-trial patients and can provide a better match to participants’ daily routines.
• During the pilot study, prison staff did not always let patients out of their cells to receive medication or remind participants to get up on time. To resolve this problem, we initiated the use of research staff, whenever possible, to assist in the delivery of medication by checking that prisoners were always out of their cells in time to receive trial medication.
• In the pilot study, treatment was disrupted because of religious customs for several participants. We took care to check that participants were not randomised during periods when religious customs might interfere with adherence to the trial protocol.
• In the pilot study, daily adherence to the trial medication reduced when participants were not reviewed weekly. One of the findings in the pilot study was the importance that prisoners gave to the weekly follow-up meetings during the titration process, during which they could discuss their symptoms and response to the treatment process in addition to completing trial assessments. We also provided brief support on an ad hoc basis or in the form of advice, if approached by prisoners when visiting the prison wings.

Concomitant treatments

All concomitant medications were recorded. Concomitant treatments were allowed with medications that are not contraindicated with MPH. Use of the following medications in the 4 weeks prior to the start of treatment with OROS-MPH was not permitted based on potential adverse drug interactions: clonidine, coumarins, monoamine oxidase inhibitors, moclobemide and rasagiline. There were no exclusions for this reason.

Measures

A schedule of trial assessments is provided in Table 1.

Baseline-only measures were collected for all participants prior to randomisation, and included descriptors of the trial population and putative moderators of the treatment response. The primary outcome was the investigator-rated CAARS-O score at 8 weeks after the initial trial prescription. There were 13 secondary outcome measures (see Table 2) also assessed at 8 weeks. Baseline (pre-randomisation) values of the primary and secondary outcome measures were collected so that we could assess change. In addition, week-5 values were collected for variables that featured as putative mediators (see Table 2).

Once a participant had signed consent II for the RCT, baseline measures were obtained through face-to-face administration of the questionnaires. Help in reading or understanding questions was provided by research staff in most cases. Because of the large number of baseline measures, these were frequently collected across two sessions. All baseline measures were obtained prior to randomisation.

Data such as days incarcerated, release date, prison regime, adjudication, attendance at offender behaviour and educational programmes, demographic details and medical history were obtained by accessing prison and medical health records. The assessments were completed at baseline and at week 8 for the subset of outcome measures taken from prison records. Reports on the behaviour of participants were obtained from prison and education staff, reflecting behaviour at baseline and at week 8. Not all of the participants attended educational or rehabilitation programmes; hence, it was not possible to obtain measures from education staff for all participants. In addition, it was not always possible to obtain ratings from education staff because they did not always know the pupils or their names, and because of frequent staff changes and difficulties getting hold of teachers. The researchers tried to overcome these issues and found that the best option was to obtain the measure over the telephone.

Investigator-rated measures

These measures were completed by a trained psychiatrist or research assistant following interview assessments with participants:

• The DIVA 2.0 – this is a semistructured diagnostic interview designed to capture the diagnostic symptoms and other criteria for DSM-IV ADHD.31 It was used in this study as the main diagnostic instrument for the diagnosis of ADHD. Diagnoses were mainly based on descriptions of ADHD symptoms in daily life from the person being assessed, as well as other observations during the clinical assessment. In a few cases when it was not possible to obtain sufficiently clear descriptions of symptoms, particularly if there was no account of ADHD symptoms from before the age of 12 years, informants were contacted to further inform the DIVA 2.0 ratings. Since the start of the study, DIVA-5 has been published, which applies a different algorithm for DSM-5 criteria, although the individual items have not changed. The diagnostic algorithm applied to DIVA 2.0 was modified for the DSM-5 criteria, as defined in the inclusion criteria for this trial.
• The Mini International Neuropsychiatric Interview (MINI) 7.0.1 – the MINI 7.0.1 is a semistructured interview assessment used to capture DSM-IV diagnostic criteria for common mental health disorders.62 The sections completed included major depressive episode, suicidality, manic episode, hypomanic episode, panic disorder, agoraphobia, social anxiety disorder, obsessive–compulsive disorder, post-traumatic stress disorder (PTSD), psychotic disorder and mood disorder with psychotic features, generalised anxiety disorder, and ASPD.
• The MINI checklist – in addition to diagnostic categories evaluated at baseline only, the MINI 7.0.1. includes a cross-disorder symptom checklist, with ratings from 0–10. We used this at baseline and also as an 8-week outcome assessment.
• The Zanarini Rating Scale for Borderline Personality Disorder (ZAN-BPD) – the ZAN-BPD is a validated measure for the assessment of symptoms of borderline personality disorder (BPD).63 We used this measure as a baseline moderator.
• The CAARS-O64 – the 8-week CAARS-O assessment was the primary outcome measure for this study. The CAARS-O was also used as a secondary outcome at week 5, and to assist the psychiatrist in titrating participants to the optimal trial medication dose at weeks 1–4. The CAARS-O consists of the 18 DSM-IV ADHD symptoms, rated on a four-point Likert scale (0: not at all, never; 1: just a little, once in a while; 2: pretty much often; and 3: very much, frequently). This scale and other very similar scales have been extensively validated as outcome measures in previous clinical trials of adult ADHD.
• The Wender–Reimherr Adult Attention Deficit Disorder Scale (WRAADDS) – we used the emotional dysregulation items from the WRAADDS.65 We applied the emotional dysregulation items from an interview assessment of the WRAADS items, following previous publications on the treatment response of emotional symptoms in ADHD.17,66
• The Adverse Events Scale (AES)67 – this captures common known adverse effects associated with the use of MPH and other stimulant medications.
• The Clinical Global Impression (CGI) scale68 – this was used by the research psychiatrist to give an overall rating of clinical severity, and a clinical impression of the clinical response and adverse effects of the trial medication. This was the only outcome obtained by the clinician (rather than the researcher) who was directly engaged in the titration of trial medication for each participant.
• Physical health – body mass index (BMI), heart rate and blood pressure were measured to monitor the safety of the trial medication, as recommended.14

Participant self-rating scales

Self-rating scales were given to the participants. The scale questions were usually read out to participants, who gave their response accordingly:

• The RPQ69 – this scale is included as a baseline moderator capturing proactive and reactive forms of aggression.
• The Weiss Conduct Disorder scale (Weiss-CD) – this scale was included to capture conduct disorder symptoms as a descriptor and baseline moderator.
• The Alcohol Use Disorders Identification Test for Consumption (AUDIT-C) – this is validated as a quick screen for alcohol use problems.70 This was used as a quick screen to capture alcohol use in the year prior to the current prison sentence.
• The National Institute on Drug Abuse (NIDA) Quick Screen – this is a quick screen for drug use problems, adapted from the single-question screen for drug use in primary care by Saitz et al.71
• The Childhood Trauma Questionnaire (CTQ)72 – this was used to capture reports of childhood trauma that might moderate the clinical response to MPH.
• The Barkley ADHD Self-Rating Scale for DSM-IV ADHD symptoms30 was used as an initial screening instrument. It is not reported as a baseline or outcome variable. Participants were considered to screen positive for ADHD if they scored two (often) or three (very often) for four or more symptoms in either the inattentive or hyperactive/impulsive symptom domains.
• The Affective Reactivity Index-Self (ARI-S)73 – this is a self-rating scale for irritability.
• The Mind Excessively Wandering Scale (MEWS)74 – this is a self-rating scale that captures excessive spontaneous mind-wandering, an aspect of psychopathology that is closely associated with ADHD and a good predictor of ADHD-associated impairment in daily life.
• The Brief Symptom Inventory (BSI)75 – is a self-rating scale that captures comorbid symptoms reflecting common aspects of psychopathology. Subscales include nine symptom dimensions: somatisation, obsession–compulsion, interpersonal sensitivity, depression, anxiety, hostility, phobic anxiety, paranoid ideation and psychoticism.
• The Maudsley Violence Questionnaire (MVQ)76 – this was designed to capture beliefs associated with violence. The machismo subscale relates to embarrassment over backing down and justification of violence in response to threat and attack. The acceptance subscale includes the overt enjoyment and acceptance of violence in everyday life. In previous research, the machismo subscale showed a greater relationship to actual violence76 and a greater reduction in the pilot study for this trial.29 We report on the total scale in this trial.
• The Clinical Outcomes in Routine Evaluation – Outcome Measure (CORE-OM)77 – this is a scale that captures subjective well-being, problems and symptoms, life-functioning, and risk and harm. It was designed to measure psychological distress before and after treatment.

Data from prison records and prison and education staff

Data were collected from prison records and prison and educational staff, relating to behaviour in the previous 56 days. For cases of individuals new to custody presenting with significant behavioural problems, the retrospective baseline reporting period was 56 days to allow for initial behavioural problems that may arise when people first enter prison. For 16 participants (nine in Isis and seven in Polmont), the time reported was shorter because they had not been in the prison for a sufficient length of time; this was reported as a protocol deviation. The following data were collected:

• Critical incidents – critical incidents were taken from prison records for the period of 8 weeks prior to randomisation (baseline) and for the 8 weeks of the trial (8-week outcome). This item consisted of the number of adjudications for antisocial behaviour and reports of rule breaking at HMP YOI Isis and HM YOI Polmont.
• Modified Overt Aggression Scale-Prison (MOAS-P) – ratings of aggressive behaviour by prison staff for the previous week were obtained using the Modified Overt Aggression Scale (MOAS).78
• Behaviour Report Card-Prison (BRC-P) – ratings of behaviour by prison staff for the previous week were obtained using behaviour report cards79 by prison officers.
• Modified Overt Aggression Scale-Education (MOAS-E) – ratings of aggressive behaviour by education staff for the previous week were obtained using the MOAS. This item was optional, depending on whether or not prisoners attended education sessions.
• Behaviour Report Card-Education (BRC-E) – classroom behaviour report cards scored by education staff were obtained for the previous week. This item is optional, depending on whether or not prisoners attended education sessions.
• Incentives and earned privileges – the number of positive IEPs for positive engagement in education, occupational and rehabilitation programmes was obtained from prison records for the 2 months prior to randomisation, and after 8 weeks of the trial. IEPs were available at HMP YOI Isis only; for this reason, they were not included as an outcome in the statistical analysis plan (SAP).

Primary and secondary outcome measures

Details of the primary and secondary outcome measures at week 8 are listed in Table 2.

Primary outcome

The primary end point was the level of ADHD symptoms measured on the investigator-rated CAARS-O at 8 weeks post treatment initiation. This addresses the question of the efficacy of OROS-MPH on ADHD symptoms in young male adult offenders meeting the DSM-5 diagnostic criteria for ADHD. The investigator-rated CAARS-O score is a common outcome measure used in previous treatment trials of ADHD in the community, and measures the same list of 18 symptoms used as the primary outcome in nearly all other studies of adult ADHD.32

Secondary outcomes

The secondary outcomes addressed important questions about the effects on comorbid symptoms and behavioural impairments that are commonly seen in offenders with ADHD. Ratings of behaviour from prison records and staff included the following: critical incidents recorded in the prison records for the 8-week period from initiation of the trial medication to the week 8 assessment, ratings of aggressive behaviour over the previous week by prison staff using the MOAS-P at 8 weeks, behaviour over the previous week reported using BRC-P from prison staff at 8 weeks, and engagement with educational activities (including the number of scheduled educational sessions, the proportion of scheduled educational sessions attended and reports of disruptive behaviour in education sessions reported at 8 weeks using the BRC-E and over the course of the trial using the MOAS-E). The reports from education staff were possible only for those participants who attended education or rehabilitation sessions. Ratings of symptoms and behaviour reported by participants included attitudes towards violence captured using the MVQ at 8 weeks; general distress and psychopathology captured using the CORE-OM and BSI at 8 weeks; excessive spontaneous mind-wandering captured using the MEWS at 5 and 8 weeks; symptoms of emotional dysregulation and irritability captured using the WRAADDS and ARI-S at 5 and 8 weeks; and the CGI scale at 5 and 8 weeks as a general measure of overall clinical improvement.

Mediator measures

To address the tertiary objectives, mediation hypotheses were tested using the following putative mediators and outcome variables: the CAARS-O hyperactive/impulsivity and inattention subscores, and the WRAADDS for emotional dysregulation at week 5 were considered as putative mediators. These measures were hypothesised to mediate treatment effects on secondary behavioural outcomes (critical incidents and BRC-P at week 8).

Participant timelines

A flow diagram of participant timelines is illustrated in Figure 1.

FIGURE 1 Flow diagram of participant timelines. a, Prior to consent to be screened for ADHD, potential participants were provided with information about the trial. Initial consent (screening and diagnostic step) allowed for the use of screening questionnaires (Barkley ADHD self-rating scale for DSM-IV ADHD symptom); b, if ADHD was suspected, a diagnostic interview for ADHD was carried out, which included the DIVA interview for adult ADHD. Diagnosis of ADHD and suitability for the trial were then confirmed by a medical assessment prior to consent for the trial. (Figure not reproduced here; see the original article.)

Consent

There were two stages of consent. Initial consent, consent I (screening and diagnostic step) allowed the use of the self-rated Barkley ADHD scale30 as a screening questionnaire for ADHD, followed by a diagnostic assessment using the DIVA 2.0 for adults.31 This was followed by a review of the diagnosis by a trained psychiatrist. During these pre-trial steps, prisoners who failed to meet any of the eligibility criteria at any stage were not invited to continue in the trial. At HMP YOI Isis, routine screening was already in place using this scale and a pre cohort was identified before the start of the trial. These participants were asked to sign consent I at the time of the diagnostic interview. Routine screening did not take place at HM YOI Polmont before the start of this trial.

Following the pre-trial screening stage, individuals who met the diagnostic and eligibility criteria for the trial were invited to take part in the clinical trial, at which stage they were invited to take part in the RCT. Full information about the trial, and subsequent consent to participate (consent II), was completed with the trial psychiatrist after the diagnosis had been verified by clinical review. Participants completed baseline measures after signing consent II, to take part in the trial. Once the baseline measures had been completed, eligibility criteria were further checked, documented and entered into the trial database.

Research visits

Informed consent was divided into consent I, for diagnostic screening and identification of potential participants for the RCT, and consent II, required prior to obtaining baseline data and randomisation to OROS-MPH or placebo.

Following consent for screening (consent I) and confirmation of the diagnosis of ADHD and eligibility by a psychiatrist trained in the assessment of ADHD, information sheets and consent forms for the controlled trial (consent II) were provided to and discussed with potential participants (visit 1). Information sheets were reviewed and informed consent obtained for the clinical trial (visit 2). No limit was set for the time taken between visits 1 and 2 within the time frame of the project. Potential participants were encouraged to take as much time as they needed to reach a fully informed decision about participation in the trial.

Baseline data were then collected from participants (visit 3), prison and health-care records, and prison and educational staff. Once the patient baseline data had been collected and eligibility confirmed following medical review by one of the trial psychiatrists, participants were randomised to treatment with placebo or OROS-MPH (visit 4). Trial prescriptions were completed and given to the pharmacy. Medication was started within 1 week of randomisation, but usually within 2 or 3 days. At the end of week 1, participants were reviewed, and trial medication titrated according to their clinical response and adverse effect profile (visit 5, week 1 titration). Symptoms of ADHD were measured using the CAARS-O, AEs were checked using the AES, and heart rate and blood pressure were checked and recorded. This titration procedure was repeated at the end of weeks 2, 3 and 4 (visits 6–8) and was expected to remain unchanged. Five weeks after the start of the trial medication (visit 9, week-5 assessment), the maintenance prescription was checked and could be changed for a final time. The final dose was then maintained for the last 3 weeks of the trial, unless participants withdrew from ongoing treatment. At the week-5 assessment, outcome measures were completed by a research investigator (not the clinician who titrated the medication): the CAARS-O, WRAADDS and MEWS; heart rate and blood pressure; weight; and the AES. The CGI scale was completed by the psychiatrist who titrated the medication dose, as this measures a clinician’s view on the global clinical response to treatment; however, for 42 participants, the CGI scale was obtained by the researcher instead of the psychiatrist. The final visit, visit 10, was completed 3 weeks later, 8 weeks after the initiation of the trial medication. At this visit, all outcome measures were completed. As far as possible, the information on clinical response derived during the titration visits (weeks 1–4) was not shared with other members of the research team. Thus, potential unblinding based on the observed clinical response and AEs was minimised. Researchers who were engaged in generating outcome data also conducted data entry and undertook internal source data verification (SDV) checks, as well as checking data queries.

Sample size

The sample size of 200 young male offenders was determined as follows. The primary outcome was ADHD symptoms, measured using the CAARS-O, a total 18-item ADHD symptom score. The results of a single-arm open-label pilot study of young male prisoners with ADHD who were given MPH showed a mean decrease of 25.0 points with a standard deviation (SD) of 9.1.29 This suggested a large standardised effect size of SMD 2.75. It could reasonably be assumed that at least 20% of this effect might be attributed to the effects of MPH. On this basis, this trial is powered to detect a standardised effect size of d = 0.55. Assuming a SD of 9.1, this would translate into a treatment difference of 5.0 points. This effect size is consistent with the results of a 2018 meta-regression analysis,32 which estimated the effect of treatment to be a SMD of 0.49 (95% CI 0.08 to 0.64). The sample size calculation used G*Power version 3 (Heinrich Heine University Düsseldorf, Germany)80 and was based on the use of a t-test to compare the means of the treatment groups. To have 90% power at the 5% significance level to detect a standardised effect of SMD 0.55, this trial would need to collect outcome data on 142 participants. Inflating for the expectation that loss to follow-up may be as high as 25%, a minimum of 190 participants are required, with the target for the trial set at 200. A 25% loss to follow-up was expected to be easily achieved, as, in the pilot study, 10% of participants left the prison as a result of unexpected transfers from the prison, whereas problems with adherence to trial medication were rarely followed by problems obtaining follow-up assessment data.

Interim analyses and stopping guidelines

No interim analyses were planned or performed. Stopping guidelines were based on recruitment rate and a review of feasibility for recruitment. All of the original recruitment targets were met. At milestone 3, the Data Monitoring Committee (DMC) reviewed progress with recruitment and safety, and recommended continuation of the trial.

The key milestones established before the start of the trial were as follows:

• milestone 1 (3 months) – five participants recruited and trial medication initiated
• milestone 2 (14 months) – 65 participants randomised and trial medication initiated
• milestone 3 (24 months) – 120 participants randomised and trial medication initiated.

Milestone 1 was reached by the end of September 2017, with a total of 75 participants randomised (Isis, 47; Polmont, 28). By milestone 3, after the first 24 months of the trial, a total of 152 participants were randomised (Isis, 91; Polmont, 61). During this period, it was established that no other recruitment site was required.

Randomisation and blinding

Randomisation type and sequence generation

Randomisation to OROS-MPH or placebo was in a 1 : 1 ratio. Randomisation was at the participant level and was conducted using an online system set up by the KCTU, ensuring reliability and credibility in the randomisation process. Randomisation was stratified by prison, with variable block sizes to ensure that equal numbers of patients were allocated to the two arms in each prison stratum.

Randomisation allocation concealment mechanism and implementation

The randomisation process was conducted by trial psychiatrists, who entered the participant information (research identification number and participant name initials) into the KCTU randomisation system. The trial psychiatrists then completed initial trial prescriptions. Each participant was allocated a kit (labelled carton) containing four labelled bottles, each containing 46 OROS-MPH (18 mg) or placebo capsules. Each kit and its bottles were labelled according to annex 13 of the European GMP for investigational medicinal products (IMPs)57 and had their own pack number, which was matched to the randomisation number. The centralised randomisation system allocated the correct treatment kit number to each participant during the trial.

Blinding

Blinding was maintained for all trial investigators, including the on-site researchers, pharmacy staff and trial manager. The statistical team remained blinded to trial arm allocation until after the analyses laid out in the analysis plan were completed. Whenever possible, outcome measures were obtained by an assessor who was not involved in the titration procedures to minimise the potential for bias when rating outcome variables. This was possible for all assessments apart from 16 assessments when it was not possible because staff had to cover for each other, for example during sick leave. Researchers engaged in generating outcome data also entered the data into the trial database and conducted internal SDV checks, as well as resolving any data queries.

The main analysis data set did not include any trial medication dosage data to ensure that the trial statistician remained blinded. We did, however, conduct a sensitivity analysis to assess efficacy for those showing good compliance with the trial medication, as well as generating tables to describe the doses prescribed and taken over time by trial arm. The sensitivity analysis excluded those participants who took trial medication on < 75% of the days on which it was prescribed, which was thought to be a reasonable criterion for poor adherence to trial medication. These analyses were done last to minimise the risk of unblinding.

Withdrawal of subjects

Participants had the right to withdraw from the trial at any time for any reason. Health-care staff had the right to withdraw participants from the trial if they thought that the trial was having an adverse effect on the participants; this was not applied for any participant in the trial.

However, when participants discontinued trial medication, we invited them to remain in the trial to complete trial assessments, thereby minimising loss of data. When participants decided to withdraw from the trial, all efforts were made to report the reason for withdrawal as thoroughly as possible. Withdrawal from the trial was defined as withdrawal from taking the trial medication and from providing further follow-up assessment data. There was a significant group who continued to provide follow-up assessments, but who had withdrawn from taking trial medication.

Owing to potential concerns about the interaction of trial medication with unknown psychoactive substances, if a participant disclosed to any member of the research team that they had used ‘spice’ (synthetic cannabis or other unknown psychoactive substance) or there were suspicions of use reported by health-care or prison staff, while participating in the trial, a clinical evaluation was made. If this was current use (defined as within the previous 2 days), the trial medication was stopped. If this occurred earlier in the trial and was considered an isolated incident, the trial medication could continue. If the trial medication was stopped, the participant remained in the trial and was asked to complete trial assessments. A clinical assessment was made on a case-by-case basis to consider the safety of restarting trial medication 48 hours after stopping the trial medication.

Statistical analyses

A detailed SAP was developed by the trial statisticians in collaboration with the chief investigator, was reviewed by the DMC, and was reviewed and approved by the Trial Steering Committee (TSC) before the trial database was locked. Here we provide a summary of the statistical analysis approaches employed according to this plan. However, note that the decision of some analyses could be finalised only after having sight of the data, namely the necessity to employ multiple imputation (MI) to handle the missing data generating process and the distributional assumptions for some of the non-questionnaire secondary outcomes.

An ITT approach was used for all primary and secondary week-8 outcomes, that is participants were analysed in the groups to which they were randomised, irrespective of adherence to trial medication. The primary outcome measure, the CAARS-O, and the secondary outcome measures, MEWS, WRAADDS, ARI-S, CORE-OM, BSI, MVQ and CGI scale, were continuous variables. Their modelling relied on normal assumptions for error terms and the treatment effects were quantified by trial arm differences (and standardised differences). The MOAS-E, MOAS-P, BRC-P and BRC-E had also been expected to follow normal distributions and critical incidents had been expected to follow a Poisson distribution. However, on review, the residuals or all of the MOAS-P, MOAS-E, BRC-P, BRC-E and critical incidents were noticeably positive skewed, and possibly overdispersed or zero inflated, and so could not be modelled by a normal distribution (see Table 2).

The number of education sessions scheduled at 8 weeks was defined as being scheduled to attend any of offender training, vocational training or education sessions. We had intended to use logistic regression to analyse education as a binary variable of whether or not any education sessions were scheduled, but, at baseline, almost all (187/200) of the participants had some form of education session scheduled. At week 8, this increased to 191 out of 200 participants. Owing to this lack of variability in the binary education outcome, we instead analysed the underlying count variable ‘number of any form of education scheduled between baseline and week 8’ using a negative binomial distribution to allow for positive skewness and overdispersion of this variable. The proportion of education sessions attended out of those scheduled was also described by arm.

Education outcomes were considered for only the subpopulation of prisoners enrolled in education at baseline. A total of 187 trial participants were enrolled in some form of education at baseline. However, within this subpopulation, only 83 MOAS-E baseline forms were completed. Similarly, only 67 BRC-E forms were completed. We considered an actual sample size of less than half of the intended size at baseline to be too small to attempt meaningful formal inference. Thus, the education outcomes MOAS-E and BRC-E were summarised descriptively only.

The BRC-P was approximated by a negative binomial distribution, which is appropriate as, although not directly defined as a count of incidents, the questionnaire is a weighted proxy count of incidents of inappropriate behaviour. Thus, BRC-P was modelled in the same way as the number of education session scheduled.

The MOAS-P exhibited large zero inflation and too few remaining data points to allow modelling of the distribution (143/200 participants who were included in the trial were rated ‘0’ on the MOAS-P at 8 weeks). We therefore dichotomised this outcome to give a binary variable (1 = participant had any aggressive event, 0 = no aggression) and analysed it using logistic regression.

For critical incidents, which is a count of incidents, the data were positively skewed and zero inflated. However, the residuals were no longer zero inflated after baseline count of incidents was included in the analysis model. A negative binomial model, rather than a Poisson distribution, was employed to allow for overdispersion. In addition, the time in prison was included in the model as an offset to model the fact that the number of critical incidents reported at week 8 are proportional to the time spent in prison between randomisation and withdrawal from or the end of the 8-week trial.

Need for multiple imputation

Formal trial arm comparisons were carried out by MI, more specifically by using the flexible multivariate imputation by chained equations approach.81 This was necessary because withdrawal from treatment was found to be predictive of missing primary outcomes (i.e. missing CAARS-O scores at 8 weeks). Withdrawal from treatment was defined as withdrawing completely from the trial through choice; through release, transfer or deportation; or withdrawing from treatment only. The association between treatment withdrawal during the trial and missing data in the primary outcome at 8 weeks was tested using Fisher’s exact test and found to be predictive (p < 0.001). Thus, a MI approach was pursued to allow for a missing data generating mechanism that was missing at random, with the observed variables allowed to drive missingness, including withdrawal from treatment.

We empirically assessed whether or not baseline variables were predictive of missing data using a forward logistic regression modelling approach. The following baseline variables were considered: all comorbidities as recorded using the MINI checklist, each individual MINI diagnosis, subscores of the WRAADDS (emotional over-reactivity, affective lability, temper), the MVQ, the CGI scale, the AES, IQ score, the CTQ, the number of days in prison prior to randomisation, the RPQ and the proactive and reactive subscales, the ZAN-BPD score, Weiss-CD, the MOAS as recorded by education staff (MOAS-E) and prison staff (MOAS-P), alcohol use, illicit drug use, ethnicity, their highest level of education, previous treatment for ADHD and employment. All baseline variables were considered unless one of the categories of the variable occurred too infrequently to expect to observe both missing and observed data (defined as < 21 observations). Variables were considered to be potentially important, and were later considered for inclusion in the imputation model, if they were statistically significant at a liberal 10% test level in a logistic regression of missingness at 8 weeks on the variable in question, trial arm and prison (the randomisation stratifier). A single potential predictor variable found that only employment prior to incarceration was predictive of providing CAARS-O data at 8 weeks, with those who were in employment, self-employed or in education being more likely to provide final outcomes (p = 0.075).

Multivariate imputation by chained equations was used with regression models for imputation of missing values in continuous variables, with logistic regression models for imputation of missing values in binary variables and with negative binomial models for imputation of missing values in counts.

Analysis model

The analysis models used to estimate treatment effects included the respective outcome variable as the dependent variable; and trial arm, baseline values of the outcome (if available) and randomisation stratifiers (prison sites) as explanatory variables. Depending on the distributional assumptions made for the various secondary outcomes, OROS-MPH effects were quantified by trial arm differences, odds ratios or incidence rate ratios (see Table 2).

Imputation model

For each outcome variable, the imputation model included (1) all of the variables of the analysis model, (2) measures of the outcome variable at baseline and at 4 and/or 5 weeks, if available, and (3) known predictors of missingness (withdrawal from treatment during the trial and employment prior to incarceration). List item 1 is stipulated by MI theory; item 2 was undertaken to improve the precision of the inferences and also to allow outcome measures at earlier time points to drive drop out at later time points; and item 3 accommodates identified predictors of missingness and allowed us to make a more realistic missing-at-random assumption.

Sensitivity analyses

Four sensitivity analyses were conducted. Sensitivity analyses were carried out for the primary outcome only (CAARS-O score at 8 weeks). The first sensitivity analysis assessed the assumption that the treatment effect is constant only within defined time windows. That is, only data recorded at the week-8 visit date ± 2 days were included in this sensitivity analysis. The second sensitivity analysis looked at the effect of delayed treatment initiation. Again, those who failed to start treatment within 7 days of randomisation were removed from the analysis of the primary outcome. The third sensitivity analysis evaluated the impact of IQ on the primary outcome. Those who did not record an IQ of > 60 on the WASI-II were excluded from this analysis. The fourth sensitivity analysis targeted the efficacy of OROS-MPH by considering the subgroup of trial participants who were considered to have complied with treatment. This was defined as taking any trial medication on at least 75% of the days on which it was prescribed. Those who withdrew or were released, transferred or deported were excluded.

Mechanisms analyses

Treatment effect modification by prespecified putative moderators CTQ, ZAN-BPD and reactive and proactive subscores of the RPQ was investigated for the primary outcome (i.e. CAARS-O score at 8 weeks). Moderators were investigated one at a time. For each putative moderator, the modelling for CAARS-O at week 8 was extended to include this variable and an interaction between the moderator and the trial arm. To accommodate these variables in the imputation step, the ‘just another variable’ approach to MI with interactions was used, that is both the fully observed moderator and the product term were simply added to the list of predictors variables available for imputation of missing values. The statistical significance of the interaction term then informs regarding treatment effect modification.

Treatment effect mediation of behavioural outcomes by 5-week improvements in ADHD symptoms or emotional dysregulation (measured using the WRAADDS) was assessed by fitting the following six single mediator models:

1. exposure – trial arm: mediator – CAARS-O hyperactivity/impulsivity subscore; outcome – BRC-P
2. exposure – trial arm: mediator – CAARS-O inattention subscore; outcome – BRC-P
3. exposure – trial arm: mediator – CAARS-O WRAADDS emotional dysregulation score; outcome – BRC-P
4. exposure – trial arm: mediator – CAARS-O hyperactivity/impulsivity subscore; outcome – critical incidents
5. exposure – trial arm: mediator – CAARS-O inattention subscore; outcome – critical incidents
6. exposure – trial arm: mediator – CAARS-O WRAADDS emotional dysregulation score; outcome – critical incidents

Based on complete cases only, causal mediation analyses were performed using the Stata^® (StataCorp LP, College Station, TX, USA) command ‘paramed’.81 To base this parametric mediation modelling on linear regression models, the respective outcome variables were log-transformed (BRC-P and critical incidents were count outcomes). Baseline values of the mediator and the outcome variable, prison, age, educational level and IQ were included as covariates in the mediator and outcome models because they were either hypothesised confounders of the mediator–outcome relationship or a randomisation stratifier (prison). The causal mediation analyses allowed for an interaction between trial arm and the respective mediator. To quantify mediation, we reported the estimated natural indirect effect and the natural direct effect. All analyses were carried out in Stata version 15.1.