Pilot trial methods

Sections of this chapter have been reproduced from Holloway et al.,1 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) licence, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The text below includes minor additions and formatting changes to the original text.

APPRAISE was a two-arm, parallel-group, individually randomised pilot study of a self-efficacy-enhancing psychosocial alcohol intervention for men on remand in prison.1 It was intended to provide the evidence to support the design of a future definitive multicentre RCT, for which new funding would be sought.

The study aligned to the MRC framework,43 using mixed methods with two linked phases.

• Phase I involved a two-arm, parallel-group, individually randomised pilot study. The pilot evaluation provides data to inform the feasibility of a future definitive multicentre RCT.
• Phase II comprised an embedded process evaluation. The MRC framework for process evaluation guided the planning, design and proposed conduct of the process evaluation.44

This chapter presents the methods of the Phase I pilot trial with the pilot trial findings presented in Chapter 4, and relates to research Objectives 1 and 4 in our protocol. Objectives 2 and 3 are discussed in later chapters.

The primary research objectives of the Phase I pilot study were:

Objective 1: to pilot the study measures and evaluation methods to assess the feasibility of conducting a future definitive multicentre, pragmatic, parallel-group RCT.

1. Is it feasible to conduct a future multicentre RCT of a self-efficacy-enhancing psychosocial alcohol intervention for men on remand?
2. Can we obtain reasonable estimates of the parameters necessary to inform the design and sample size calculation for a future definitive multicentre RCT? This includes standard deviations of potential continuous primary outcomes and estimates of recruitment, retention and follow-up rates.
3. How well do participants complete the questionnaires necessary for a future definitive RCT?
4. Can we collect economic data needed for a future definitive RCT?
5. Can we access recidivism data from the PNC databases for trial participants?
6. Can we access health data from routine NHS data sources for trial participants?

Objective 4: to assess whether operational progression criteria for conducting a future definitive RCT are met across trial arms and study sites and, if so, develop a protocol for a future definitive trial. Operational progression criteria are based on previous research results.7

1. Do the two prisons invited to the study agree to take part?
2. Based on knowledge from previous data, do at least 90 eligible participants consent to take part and be randomised across the trial arms?
3. Do at least 70% of participants who consent to the trial receive the intervention?
4. Do at least 60% of those who received the intervention get followed up at 12 months across trial arms and study sites?

Setting and site selection

The study occurred in two study sites hosting men on remand within the prison estate of the Scottish Prison Service (SPS) and His Majesty’s Prison and Probation Service (HMPPS), one in Scotland and one in England respectively.

The complexity of the prison estate cannot be overestimated. Our learning from previous work had highlighted how essential it was to engage with our colleagues within the prison system with regard to the study. Prison governors and associated personnel at both prison sites were involved in discussions regarding the study while we were developing the initial funding application. We had built good operational knowledge as well as relationships with key personnel over the course of securing funding.

The prison governor at each site was the primary gatekeeper and contact for the study team. This proved vital in previous prison research work and again for this study. In the months leading up to data collection commencing, while waiting for ethics approval, we had several meetings with the governors at both study sites and also with appointed research study contacts for day-to-day operations. Visits were made to both prison sites to share details of the study with prison officers, health professionals and CGL over several months. The strong relationships formed were essential to our understanding of the complexities of the prison estates and the day-to-day work carried out there, providing insight into how we could undertake the research study with minimal disruption to the estate operations.

Eligibility criteria

Men detained on remand had to meet the following criteria:

• Inclusion criteria
  • detained in either the SPS Scottish study site or His Majesty’s Prison Service (HMPS) North East England study site
  • have been in the prison setting for 3 months or less
  • aged 18 years and over
  • informed consent given
  • ≥ 8 on the AUDIT screening tool51
• Exclusion criteria
  • previously recruited to the study
  • unable to give informed consent or deemed incompetent/unable to make an informed decision regarding consent
  • identified as a risk to self and/or others by prison staff
  • judged to be under the influence of an illicit substance by prison or research staff
  • currently taking Antabuse
  • on a segregative rule under the prison rules
  • not able to understand the documents, which are in the English language, or agree to the research assistant (RA) working with them to understand them.

Prison and/or research staff were responsible for making a subjective judgement as to whether or not a remand person was under the influence of an illicit substance and whether the level of intoxication was likely to have an influence on risk or capacity to understand/consent.

Participant selection and recruitment

The prison staff were key to facilitating the selection and recruitment of study participants at both sites. At the England study site, peer induction workers were also key to this process. The RAs spent time developing good working relationships with prison staff and peer induction workers on the wing. Information regarding males on remand entering the estate was made available to the RA with the selection and recruitment process across both study sites, as outlined in Table 4.

Consent

The RA at each site reviewed the Participant Information Sheet (PIS) with men on remand who met the eligibility criteria, provided a written and verbal description of the study, answered any questions or queries and then invited them to consider participating. The RA obtained informed consent from those willing to participate. Those who did not wish to participate at this point were thanked for their time and received no further interaction with regard to participating in the study.

We were aware that freedom of consent can easily be undermined for prisoners and those in the criminal justice system, which means these individuals may be more vulnerable to exploitation or abuse by researchers: learning disabilities, illiteracy and language barriers are prevalent within these populations. The prevalence of these characteristics alongside the power differential between researcher and potential participant meant that particular care was needed to ensure that valid, freely given and fully informed consent was achieved. For the purposes of this research, we considered valid consent as underpinned by adequate information being provided to the potential study participants and that they had the capacity to decide for themselves. A capable person would:

• understand the purpose and nature of the research
• understand what the research involves, its benefits (or lack of benefits), risks and burdens
• understand the alternatives to taking part
• be able to retain the information long enough to make an effective decision
• be able to make a free choice; and
• be capable of making this particular decision at the time it needs to be made.

Participants were asked to consent to their data being used for the present study as well as consenting to long-term follow-up (beyond this study) and for appropriate linkage to be conducted.

Stopping rule/discontinuation/withdrawal criteria

Participants were reminded during the trial that they were free to withdraw at any time without having to give a reason and without it affecting their care in the prison setting or on liberation. Where a participant wished to withdraw, we honoured their wish. We also made attempts to ascertain and document the reason for withdrawal, recording this within the case report form (CRF). Participants who requested withdrawal from the study intervention were asked if they would be willing to remain in the study for the purposes of follow-up data collection. Data provided at the point of withdrawal were retained and used in the study analysis, where the patient consented to this.

Potential reasons for participants being withdrawn from the study were: participant withdrawal of consent; RA discretion in the best interest of the participant to withdraw; the RA was informed by Prison, NHS or service staff that it was in the best interests of the participant to withdraw; an adverse event required discontinuation of participation in the trial; or termination of the trial by the sponsor.

Sample size

The aim was to recruit at least 180 participants in total comprising 90 participants per study arm across the two study sites. From prison data obtained from correspondence with SPS and the Ministry of Justice (MoJ), we had estimated that approximately 50% of participants would be liberated while the rest would remain incarcerated, leaving 45 participants per study arm (90 in total) across the two study sites.

The target sample size

1. would enable us to calculate two-sided 95% confidence intervals around proportions recruited, liberated, and dropping out in each study arm with half-widths of < 0.15
2. exceeds the 30 per group recommendation of Lancaster et al.,53 and the 35 participants per group recommendation of Teare et al.,54 for estimating key unknown design parameters, for example standard deviation with sufficient precision when the primary outcome is continuous; and
3. would ensure that within each study arm within each site we were satisfying the minimum 12 per group rule of thumb of Julious for pilot trials.55

Participant screening

The AUDIT tool51 is considered the ‘gold standard’ alcohol screening instrument and was developed by the World Health Organization (WHO). It is a 10-item screening tool designed to assess alcohol consumption, drinking behaviours and alcohol-related harms. The score is the total sum derived from the 10 questions. Scores range from 0 to 40. A score of 8–15 indicates hazardous/increasing-risk alcohol use, 16–19 indicates harmful/high-risk alcohol use and 20–40 indicates likely dependence. The AUDIT has been used previously in the criminal justice system,7 and was found to be feasible and acceptable with men on remand in the PRISM-A study.40

After obtaining informed consent, the RAs screened all study participants using the AUDIT. Those with a score of ≥ 8 were considered eligible for the study. Those scoring < 8 on AUDIT were thanked for their time and took no further part in the study. Those with an AUDIT score of ≥ 8 completed the baseline assessment and measures and were then randomised.

Randomisation and blinding

The randomisation process was supported by the Edinburgh Clinical Trials Unit (ECTU). Due to the restriction of the use of electronic equipment and no access to mobile phones by the research team and RAs while in the prison setting, we were unable to use a randomisation system, such as an interactive voice response telephone system or one accessed over the web. Allocation was conducted at the level of the participants, randomised to the active or control intervention using stratified block randomisation by site, via opaque sealed envelopes, based on a predetermined random number allocation carried out by the ECTU. Allocation was conducted by researchers opening the next sequential sealed opaque envelope after consent had been provided and the baseline assessment completed. Therefore, RAs were blind to group allocation at baseline.

The RAs were not involved in the delivery of the active and control interventions but were aware of study allocation of participants as the trial progressed. Allocation concealment was used whereby neither the person delivering the interventions nor the participant was aware of the study allocation until they were irrevocably entered into the trial. Both the trial statistician (RP) and health economists (AS, JB) were blind to group allocation prior to the final analysis and only had details of study participants by study number.

The trial statistician and health economists were only to be unblinded if requested to do so by the Data Monitoring and Ethics Committee Project Steering Group (DMEC), due to safety concerns. This was not necessary during the study and unblinding took place at the final analysis stage.

Self-report primary outcome measure

The proposed primary outcome measure for a future definitive study was total alcohol consumed (in units) in a 28-day period. This can be ascertained using the 28-day timeline followback questionnaire (TLFB-28).43 Three other variables can be derived from the data: per cent days abstinent, drinks per drinking day (secondary outcome measures) and total number of days where alcoholic drinks are consumed. The TLFB-28 was to be conducted by the RA at time points (TP) 1 and 2.

Self-report secondary outcome measures

Secondary outcomes measures were to be completed at TP0 (baseline), TP1 (6 months) and TP2 (12 months).

Alcohol use frequency, quantity (on a typical occasion) and binge drinking were assessed using the AUDIT.34 In order to assess the utility of using Average Drinks per Day derived from the AUDIT at 12 months as a primary outcome measure we planned to randomise the order of presentation of TLFB-28 and AUDIT and conduct a levels-of-agreement analysis, using TLFB-28 as a gold standard, to explore whether AUDIT is an acceptable proxy for TLFB-28. The order of presentation of TLFB and AUDIT was to be randomised in advance by a secure remote randomisation service. (See section ‘Follow-up plans post-COVID 19’ for changes to the use of AUDIT and TLFB-28 at TP1 and TP2.)

The following study measures were to be used across the three time points.

Warwick-Edinburgh Mental Well-being Scale

Warwick-Edinburgh Mental Well-being Scale56 comprises 14 items with five response categories, ranging from ‘None of the time’ to ‘All of the time’. The items are all worded positively and cover both feeling and functioning aspects of mental well-being. Items are scored on a range from 1 to 5, providing a total score between 14 and 70.

Drinking Refusal Self-efficacy Questionnaire – revised

This revised version of the DRSEQ57 assesses a person’s belief in his/her ability to resist alcohol. There are 19 items, each scored on a five-point Likert scale. Responses range from 1 (I am very sure I could NOT resist) to 6 (I am very sure I could resist), providing a total score between 19 and 114. A lower score indicates less self-efficacy.

Three subscales can also be calculated: Social Pressure (SP) (score range 5–30), Emotional Relief (ER) (score range 7–42) and Opportunistic (OP) (score range 7–42).

Negative Alcohol Expectancy Questionnaire

The NAEQ58 assesses the extent to which people expect negative consequences to occur if they drink. There are 60 items in the questionnaire. Responses are measured on a five-point Likert scale ranging from 1, which is ‘highly unlikely’, to 5, which is ‘highly likely’, providing a total score between 60 and 300. A lower score indicates more negative expectancies.

The items, themselves, are grouped onto three dimensions (same-day, next-day and continued-drinking expectations). The next-day and continued-drinking dimensions are combined to form the distal expectancy subscale (score range 39–195) and the same-day, the proximal subscale (score range 21–105).

EuroQol-5 Dimensions, five-level version

The EQ-5D-5L59 comprises two parts. The first is a visual analogue scale and the second is a descriptive system. There are five dimensions within the descriptive system: mobility, self-care, usual activities, pain/discomfort, anxiety/depression. Each has five levels: no problems, slight problems, moderate problems, severe problems, and extreme problems. Participants convey their health state by identifying the most appropriate statement in each of the five dimensions. Each dimension score is combined to describe health state. An X is used for the visual analogue score to identify the health of the participant on that particular day.

Readiness to Change Ruler

The Readiness to Change Ruler is used to assess a person’s readiness to change their alcohol use.10 There are six readiness to change rulers on the assessment form. Each ruler represents a linear scale from 1 to 10, and the participant marks on the ruler their current position in the change process.

The Readiness to Change Ruler was completed immediately prior to and immediately following the initial intervention session. This was carried out by the interventionist as part of the initial intervention delivery.

Economic Form 90

We adapted the terminology of the Economic Form 9060 (the Economic Form 90 was originally designed for American context) (see Chapter 7, Methods) to produce a service user questionnaire to determine social costs including in the domains of health and social care use, criminal justice involvement, unemployment, and welfare (see Appendix 9).

Baseline data collection and assessment of outcomes

Baseline data (TP0) collection was conducted prior to randomisation to prevent observer or respondent bias to questions asked or answered at baseline by knowledge of the randomisation group, with follow-up data collected at TP1 (6 months) and TP2 (12 months).

The baseline assessment data were collected via researcher-led completion of hard-copy questionnaires embedded within the Baseline CRF (TP0). Due to restrictions on the use of electronic equipment in the prison estate, we were unable to use electronic devices to record study data on site. The RA was able to answer any questions and offered clarification as needed. The baseline assessment took approximately 35 to 60 minutes. Participants also completed the Participant Locator Form (see Appendix 10). The Locator Form was used to collect information from participants about a range of methods, modes and formats of contacting them at follow-up if they were liberated. Immediately after, participants were randomised, informed of their allocation,61 and thanked for their time.

The baseline CRF recorded demographic data on age, known court dates, ethnicity, postcode [Scottish Index of Multiple Deprivation (SIMD)], relationship status, and educational achievement. In addition, baseline data on proposed secondary outcome measures were recorded via the Readiness to Change Ruler,10 the EQ-5D-5L,59 the Warwick-Edinburgh Mental Well-being Scale,56 the Drinking Refusal Self-efficacy Questionnaire – revised,57 the Negative Alcohol Expectancy Questionnaire,58 and a revised version of the Economic Form 90,60 which recorded data relating to hospital and or A/E visits, general practitioner (GP) registration and visits, engagement with Community Psychiatric Nurse, engagement with any addiction services, emergency ambulance use, engagement with social worker, current living arrangements, ever been homeless, employment status, number of arrests and number of court visits.

Control condition (care as usual)

The control condition (care as usual) consisted of similar components at both study sites.

England control condition

All men held on remand are screened by a reception nurse on reception. If a substance misuse need is identified in the screening, they are referred to the Drug and Alcohol Recovery Team (DART) and the team then sees them the following day to complete the triage and comprehensive assessment. Those men who decline the offer of the DART service are then revisited to be given harm-reduction information and a point of contact if they need to self-refer further into their sentence. Posters are displayed around all of the establishment and anyone or any service can refer an individual to DART.

They refer to the relevant community substance misuse treatment service for an individual’s release and ensure the service user understands when and where their appointment is. They also offer additional support through the gate with their Reconnect Navigators, who will support those individuals who are more complex and struggle to engage to their first appointment.

Scotland control condition

For those service users who expressed that they are looking for alcohol support, the support is person-centred and based on the person’s needs. Following assessment, individuals would agree a recovery plan with their worker, the recovery coordinator, and support Scotland Mental Well-Being Scalemay be alcohol awareness, harm reduction, motivational support and relapse prevention. The frequency of this would be dependent on the individuals’ needs and would not be time-limited. People with alcohol issues could be referred to the relevant region’s alcohol service for counselling support if they wanted this. If the individual was returning to the region where the prison was, they would be offered through-care support from a recovery support service. This would include practical support as well as substance misuse support to help them reintegrate back into the community.

Intervention condition

Following completion of baseline assessments and randomisation, those allocated to the intervention arm were invited to attend their first of three intervention sessions. This was delivered face to face by the identified interventionist at each study site. The APPRAISE intervention was delivered at both sites by CGL.

The in-prison and post-liberation elements of the APPRAISE intervention are described in full detail in Chapter 2.

Follow-up data collection

We worked with our PPI co-applicant and former community justice mentor, who mentored men in prison and on liberation, to advise on the strategies to maximise retention and follow-up. These strategies were based on her own experiences of following up her mentees on liberation and were also strategies used by CGL community justice workers. Each study participant provided their preferred method of follow-up by completing the Locator Form at baseline.

Follow-up assessments were to be attempted at 6 months (TP1) and 12 months (TP2) post-randomisation where the participant had been (a) not liberated, (b) liberated and in the community, or (c) liberated and then re-incarcerated. An 8-week window was in place for the follow-up to be completed for both of the time points. If a participant was not contacted at 6 months they were still approached again at 12 months. Follow-up was to be conducted by the RA at each study site. However, due to changes in personnel at both study sites during the study, all follow-ups for both study sites were not conducted by the RA who undertook recruitment and baseline measurements. Feedback from all RAs noted that participants responded to follow-up more positively when it was the RA they knew and remembered from recruitment and baseline. There was a sense of trust that had been built which was less evident when the new RAs were undertaking the follow-ups.

Follow-up plans pre-coronavirus disease discovered in 2019

Participants were initially followed up using one of the following methods, determined by the information they provided at baseline interview as part of the Locator Form completion:

• by phone call (at varying times and days of the week)
• via hard copies of the follow-up questionnaire sent by post with an accompanying letter to be completed and returned to the study team via a pre-paid envelope; two reminders were sent; or
• via hard copy in prison for completion with the RA.

Follow-up plans post-coronavirus disease discovered in 2019

As a result of the impact of COVID-19 on the planned follow-up strategy, modifications were made to the follow-up method, procedures, data collection format, measures, and CRFs. These were discussed by the project management group (PMG) and then at the Trial Steering Committee (TSC). Ethical approval for these amendments was sought and gained.

Additional follow-up approaches were added: text message, WhatsApp, Facebook and an electronic Qualtrics link to the survey sent by phone or e-mail. Follow-ups were to be conducted over the phone by the RA. Where study participants were identified as being incarcerated at follow-up, we would attempt to make appropriate arrangements with the relevant prison to establish if it was possible to visit and undertake the follow-up assessment. Due to the duration of the COVID-19 pandemic, access to the England study site was still not possible and for the Scotland site, although access had started to return, it was restricted and was only possible in a few instances, by which time the study had concluded.

Participants who had given permission to use social media were also sought on Facebook and one attempt to contact them was carried out (we were limited to one attempt as per ethics committee stipulation).

COVID-19 restrictions meant that we were unable to undertake face-to-face interviews. We therefore developed alternative formats of CRFs for participants to complete at 6- and 12-month follow-ups. Our amendment allowed us to send electronic copies via Qualtrics link to participants (if they had consented to us contacting them in this way) and would allow for self-completion using hard copies. We simplified the CRFs to accommodate this and removed the TLFB-28, which required in-person administration. We therefore replaced the AUDIT and TLFB-28 with the extended-item AUDIT-C questionnaire. Previous studies have shown AUDIT-C and the TLFB-28 demonstrating excellent levels of agreement on alcohol consumption.4 An overall total score of ≥ 5 is deemed as AUDIT-C positive, and indicates increasing or higher-risk drinking.62

The NAEQ includes three sub-sections. The final sub-section was removed as it was considered as having potential to cause distress when self-completed. This was discussed at length by the PMG and with particular input from our PPI co-applicant. The layout of the Health Econ Form was also simplified to support self-completion.

Statistical methods/analysis plan

The progress of participants through the APPRAISE pilot trial is presented in accordance with CONSORT  (Consolidated Standards of Reporting Trials) guidelines40 to allow descriptions of key parameters required for a future RCT: eligibility rates, consent, adherence, retention at follow-up and data completeness of outcome measures.

Overall statistical principles

Analyses were based on a modified intention-to-treat (ITT) population unless otherwise stated. The modified ITT population included all participants who were randomised into the APPRAISE study, if they had provided valid non-missing data for each outcome at the relevant time points. Participants were analysed according to the trial arm they were randomised to, irrespective of the treatment they received, continuing study eligibility, or compliance post randomisation. Participants who discontinued with the intervention, withdrew, or were non-compliant with the protocol were included in the ITT population, provided that valid outcome data were recorded for them.

It was assumed that any missing data were missing-completely-at-random (MCAR). Although a multiple imputation analysis was specified in the protocol paper, this approach is only valid when the proportion of missing data lies between 5% and 40%.63 Very high amounts of missing data at follow-up meant that no multiple imputation was performed and all missing data were left as missing.

Lancaster et al.53 recommend that the analysis of a pilot study should be mainly descriptive and focus on confidence intervals. Therefore, the analyses focused on calculating descriptive statistics and two-sided 95% confidence intervals rather than on hypothesis tests or p-values in general. Outliers were identified by viewing boxplots and/or histograms of the outcome variables of interest and were queried at the data-checking stage where an error had been suspected. All analyses included outliers as standard unless otherwise stated.

Primary outcome measure

The primary outcome was to be collected at TP0, TP1 and TP2. The primary outcome for the pilot study was alcohol consumed in the 28 days prior to TP2, assessed using the extended Alcohol Use Disorders Identification Test – Consumption (AUDIT-C). This was calculated using questions 1 and 2 of the AUDIT-C, which address frequency and quantity of alcohol consumed.64,65 Responses to question 1 were recoded; Never = 0, Monthly or less = 0.25, 2–4 times per month = 0.75, 2–3 times per week 2.5, 4–5 times per week = 4.5, 6 or more times per week = 7. Question 2 was recoded; 1–2 drinks = 1.5, 3–4 = 3.5, 5–6 = 5.5, 7–9 = 8, 10–12 = 11, 12–14 = 13 and more than 14 = 16. The weekly consumption was derived by multiplying the recoded values for question 1 and question 2 together and then multiplying the product by 4 to derive 28-day consumption.

The original AUDIT-C measure is known for right censoring quantity of consumption and as a consequence being less responsive to change in heavy drinkers. Extending the AUDIT-C response categories we aimed to have a trade-off between minimal completion burden and granular data. We initially tested our assumptions about categorisation of consumption in data sets that contained the extended AUDIT-C and TLFB and found minimal benefit of allocating a daily consumption value over 16 units per drinking day, something that has been further confirmed in other studies using the extended AUDIT-C to assess weekly consumption.66

Secondary outcome measures

The secondary outcome measures are the Readiness to Change Ruler,10 the EQ-5D-5L,59 the Warwick-Edinburgh Mental Well-being Scale,67 the Drinking Refusal Self-efficacy Questionnaire – revised,57 the Negative Alcohol Expectancy Questionnaire,58 and a revised version of the Economic Form 90.60 Measurements were recorded at baseline (TP0) using a combination of self-completion and researcher-led completion of a hard-copy questionnaire. These measures were to be collected again at 6 months (TP1) and 12 months (TP2) post randomisation where participants were (a) not liberated, (b) liberated and in the community or (c) liberated but back in prison.

Recruitment, adherence and follow-up retention analyses

Trial management data, the Intervention Delivery Logs, and follow-up questionnaires were used to assess recruitment, eligibility, adherence, and retention at follow-up. For recruitment and eligibility, the following trial management data (recorded from the initiation of the trial until full recruitment is reached) were considered by each study site where possible:

• number entering on remand
• number given information sheet
• number approached
• number consented
• number randomised and
• number followed up at each time point.

These numbers were used to calculate appropriate proportions, which are reported for the entire study and by site. Exact 95% confidence intervals were calculated for key overall proportions relating to recruitment, adherence and retention.

Descriptive summaries

Descriptive statistics were calculated for the following variables:

• baseline characteristics
• numbers receiving the in-prison and post-liberation intervention
• duration of the in-prison intervention sessions
• reasons why the in-prison intervention was not delivered
• numbers followed-up
• reasons for being unable to follow-up
• information on data completeness at baseline and follow-up time points
• the primary outcome (28-day alcohol consumption)
• secondary outcomes.

In general, categorical variables were presented as counts and percentages. Discrete and continuous variables were presented using the mean, median, standard deviation (SD), minimum, maximum, and number of responses, split by randomisation group for the entire study and by study site.

A side-by-side boxplot was produced to show the age distribution of participants randomised to each site.

Primary outcome analyses

Means and 95% confidence intervals were calculated for 28-day alcohol consumption in each randomised group and site.

For the primary analysis, the mean difference in alcohol consumption at 12 months post randomisation (TP2) between the two randomisation groups was calculated (with 95% confidence intervals), both overall and stratified by site. Linear regression was used, adjusting for baseline alcohol consumption (TP0).

The mean difference was calculated assuming that any missing outcome data were missing-at-random (MAR) and only patients with AUDIT-C data collected at TP0 and TP2 were included (i.e. we performed a complete-cases analysis).

Secondary outcome analyses

The psychometric measurements were treated as secondary outcomes, and descriptive statistics were calculated based on these, stratified by site and randomised group.

Table 5 contains information on the number and type of measurements for each psychometric tool.

Information on the derivation of the psychometric outcome measures is provided below, along with details of how missing individual item scores were accounted for in the calculation of total scores.

1. AUDIT-C: the primary outcome is alcohol consumed in the 28 days prior to TP2, assessed using the extended AUDIT-C. This is calculated using questions 1 and 2 of the AUDIT-C, which address frequency and quantity of alcohol consumed. Responses to question 1 were recoded; Never = 0, Monthly or less = 0.25, 2–4 times per month = 0.75, 2–3 times per week = 2.5, 4–5 times per week = 4.5, 6 or more times per week = 7. Question 2 was recoded; 1–2 drinks = 1.5, 3–4 = 3.5, 5–6 = 5.5, 7–9 = 8, 10–12 = 11, 12–14 = 13 and more than 14 = 16. The weekly consumption was derived by multiplying the recoded values for question 1 and question 2 together and then multiplying the product by 4 to derive 28-day consumption. The AUDIT-C was also used to calculate a total score, which was calculated as the sum of the first three questions of the AUDIT questionnaire. The first question was scored as 0 (never), 1 (monthly or less), 2 (2–4 times a month), 3 (2–3 times a week), 4 (4–5 times a week, or 6 or more times a week); the second question scored as 0 (1 to 2), 1 (3 to 4), 2 (5 to 6), 3 (7 to 9), 4 (10 to 12, 12 to 14, or more than 14); and the third question scored as 0 (never), 1 (less than monthly), 2 (monthly), 3 (weekly), 4 (daily or almost daily). If any participants answered ‘Never’ to AUDIT-C Q1 then their overall score was 0. Otherwise, if any of the first three questions were missing then the AUDIT-C was also missing.
2. The Readiness to Change Ruler is used to assess a person’s readiness to change their alcohol use. There are six readiness to change rulers on the assessment form. Each ruler represents a linear scale from 1 to 10, and the participant marks on the ruler their current position in the change process. The score for each question was regarded as the integer value closest to the mark made by the participant on the ruler. No missing-value imputation was performed.
3. EQ-5D-5L consists of five descriptive dimensions (mobility, self-care, usual activities, pain/discomfort, anxiety/depression) and a question on how the respondent would rate their health. Each descriptive dimension has a score of 1–5 with levels: no problems (1), slight problems (2), moderate problems (3), severe problems (4), and unable to or extreme problems (5). The visual analogue score consists of a scale numbered from 0 (‘best health you can imagine’) to 100 (‘worst health you can imagine’). No missing-value imputation was performed.
4. The WEMWBS consists of 14 items with five response categories. Each item was scored on a range from 1 to 5, and a total score was calculated with a possible range of 14 to 70. A higher WEMWBS score indicates a higher level of mental well-being. The total score was treated as missing if 50% or more of the individual items were missing. If fewer than 50% of the items were missing, then each of the missing individual item scores was imputed using the mean of the remaining non-missing questions before calculation of the total score.
5. The DRSEQ-R consists of 19 items with five response categories following a Likert scale. Each item is scored on a range from 1 to 6, providing a total score between 19 and 114. A lower score indicates less self-efficacy to resist alcohol. The total score was treated as missing if 50% or more of the questions were missing. If fewer than 50% of the items are missing, then each of the missing individual item scores was imputed using the mean of the remaining non-missing questions before calculation of the total score.
6. The NAEQ at baseline consists of 60 items with five response categories following a Likert scale. Each item is scored on a range from 1 to 5, with 1 indicating ‘highly unlikely’ and 5 indicating ‘highly likely’. The NAEQ was used to generate two expectancy scores: same day (sum of items 1–21) and next day (sum of items 22–39).68 Each expectancy score was treated as missing if 50% or more of the questions were missing. If fewer than 50% of the items were missing, then each of the missing individual item scores was imputed using the mean of the remaining non-missing questions before calculation of the expectancy score. A higher score indicates more negative expectancies of drinking alcohol.

Logic model analyses

An exploratory analysis was performed to explore key behavioural markers (mediators) of change and how these may influence the effect of the delivered intervention. The six proposed mechanisms of change were: self-efficacy, outcome expectations, goals, plans, social support and high-risk situations.

In exploratory analysis, Kendall’s tau correlation coefficients were calculated to assess the correlation between each of the mechanisms of change and the primary outcome (AUDIT-C) at 12 months.

Software

Statistical Product and Service Solutions (SPSS) version 27 (IBM Corp. Released 2020. IBM SPSS Statistics for Windows, Version 27.0. Armonk, NY: IBM Corp) was used to produce the side-by-side boxplot of age. RStudio version 1.4.1103 [RStudio Team (2021). RStudio: Integrated Development Environment for R. RStudio, PBC, Boston, MA. www.rstudio.com/] was used to analyse the screening and recruitment data and produce the recruitment graph. All other tables and analyses were produced using statistical analysis system (SAS) software version 9.4 (SAS Institute Inc., Cary, NC, USA).

Ethics approval

Ethics approval for the study was sought and secured from five different organisations:

1. Integrated Research Application System (IRAS) (East of Scotland Research Ethics Service) REC reference 19/ES/0068, IRAS project ID 261003
2. HMPPS Ministry of Justice, National Research Committee reference 2019240
3. SPS Research Access and Ethics Committee (RAEC) approval letter dated 13 June 2019
4. the University of Scotland School of Health in Social Science Nursing Studies Research Panel approval letter dated 26 September 2019
5. NHS R&D in Scotland (R&D No. 2019/0268).

The University of Edinburgh acted as trial sponsor (CAHSS 18/2/02). A model non-commercial agreement (mNCA) was in place for the England study site but not for the Scotland study site following discussions with Health Research Authority (HRA). The mNCA is strongly recommended as the agreement between sites for clinical trials/investigations; it is not mandatory for a non-commercial study. For the Scotland study site it was not considered the appropriate mechanism for agreement and therefore a site agreement between the Sponsor and the SPS site was drawn up and put in place. The agreement for this took many months of discussion with stakeholders and was a long protracted process causing delays to the start of recruitment at the Scotland study site.

Management of the study

Two groups were in place to support the research management of the study. The PMG was convened and comprised the chief investigator (CI), co-applicants, our PPI member who was also a co-applicant, the Project Manager and researchers from both study sites working on the project. Professor Jeremy Bray chaired the PMG. The frequency of the PMG meetings was determined by the needs of the project. The PMG’s responsibility was to ensure appropriate and timely study implementation. The constitution and composition of the PMG and TSC were in line with NIHR research governance guidelines.

The TSC was appointed to provide independent assessment of the study progress, adherence to the protocol, patient safety and well-being of study participants, consideration of new information of relevance to the progress of the study and support of decision-making on the merits of a future definitive trial. A subcommittee DMEC was also convened following approval from NIHR-PH as this was a pilot feasibility study. It was agreed that because the study was a feasibility study rather that a definitive trial, it was sufficient to have the group as a subcommittee rather than having it sealed off as a separate entity. The terms of reference (ToR) for the TSC with the ToR for the subcommittee added were agreed (see Appendix 11).

Research governance

The study was managed and conducted in accordance with the Medical Research Council’s Guidelines on Good Clinical Practice in Clinical Trials (www.mrc.ac.uk), which included compliance with national and international regulations on the ethical involvement of patients in clinical research (including the Declaration of Helsinki). Before the study commenced, all required approvals were obtained and all conditions of approvals were met. All staff assisting with the study were adequately informed about the protocol and their trial-related duties.

The study was conducted at each site in compliance with the study protocol and all protocol amendments. All forms, reports, and other records were designed to maintain participant confidentiality. All data were held in a secure environment with participants’ information identified by a unique participant identification code. All records were kept in a secure storage area in compliance with the Data Protection Act (2018) and the General Data Protection Regulation (GDPR) 2019 regarding collection, storage, processing and disclosure of personal information. Trial management support was provided by the Clinical Trials Unit we were working with.

Changes to the original study protocol

During the course of the study, a number of amendments were made to the study protocol, the majority as a result of the impact of the COVID-19 pandemic. Several of these changes have been discussed above, but this section provides a comprehensive catalogue of substantial changes. All proposed amendments to the study were discussed by the CI, PMG and TSC. Once agreed, the process involved discussion with the study sponsor, whose responsibility it was to determine if an amendment was substantial or not.

Substantial amendments were submitted to the East of Scotland Research Ethics Service Committee by the CI, on behalf of the sponsor. Once approval was obtained, changes to the study protocol were then implemented. (Table 6 provides details of the substantial protocol changes approved and made.)