Adjunctive Medication Management and Contingency Management to enhance adherence to acamprosate for alcohol dependence: the ADAM trial RCT

Acamprosate for relapse prevention in alcohol dependence

Acamprosate and naltrexone have been recommended by the National Institute for Health and Care Excellence (NICE), in combination with a psychological intervention, as first-line treatments to support relapse prevention in alcohol dependence. 21 Disulfiram is regarded by NICE as a second-line treatment due to a more limited evidence base and potential adverse effects. In the past 10 years, there has been a 58% increase in prescriptions for these medications issued, however, in the past 3 years, there has been a slight decrease. 2 Naltrexone has been licensed for use in the UK to support relapse prevention in alcohol dependence since May 2013 and can also be used to treat opioid dependence. As the reason for prescription is not recorded in the NHS prescription service, the prescribing data for this in relation to alcohol dependence cannot be directly compared against acamprosate. 2 However, Prescription Cost Analysis data produced by the NHS Business Service Authority for 2012 shows that there were 117,417 acamprosate prescription items compared to just 17,790 prescription items for naltrexone. 13,22 Since naltrexone has been granted marketing authorisation for alcohol dependence, this difference seems not to have changed. 23 As these data include prescriptions for naltrexone to treat opiate dependence, the number of prescription items for alcohol dependence is likely to be fewer than indicated. Due to this difference in prescribing practices and differences in dosing regimens and side effects between medications, which may affect adherence, acamprosate was the focus for the current research.

Acamprosate modulates the glutamatergic system and attenuates the imbalance between inhibitory (GABA) and excitatory (glutamate) neurotransmitters in the brain during alcohol withdrawal, reducing the conditioned effect of alcohol and the negative reinforcement of the addictive behaviour. 24–27 A meta-analysis found acamprosate to have a moderate effect on maintenance of abstinence in people with alcohol dependence. 26 These results have been supported by a more recent systematic review and meta-analysis that found acamprosate to be the only alcohol relapse prevention intervention with enough evidence to conclude that it is better than placebo in supporting detoxification and for alcohol-dependent patients to maintain abstinence for up to 12 months in primary care settings. There was additional evidence that acamprosate might be effective longer term but the evidence was weak. 28

Adherence to acamprosate

Poor adherence to medication is a common problem, particularly in chronic conditions, a greater risk of poor adherence has also been associated with substance use disorder. 29,30 Although the available evidence supports the efficacy of acamprosate in clinical trials, poor adherence to the medication may pose a problem for effectiveness in clinical practice. We conducted a systematic review of the rates of adherence to acamprosate reported in clinical trials. We found that the mean adherence rate reported ranged from 54.2%31 to 95%. 32 This variation in adherence may be partially explained by differences in the definition of medication adherence (e.g. percentage of prescribed medication taken or percentage of those taking 80% or more of prescribed medication) and measurement of adherence. Several different methods are used in clinical trials to monitor medication adherence, with variation in the confidence in their accuracy. For example, counting returned medication and self-report of adherence may be considered ‘low’ confidence measures, electronic monitoring of pill bottle opening ‘medium confidence’ and supervised dosing ‘high confidence’. 33 Medication non-adherence in clinical practice is likely to be significantly greater than that seen in clinical trials that often offer payment for participation, adherence support and frequent monitoring appointments.

The median duration of acamprosate pharmacotherapy has been found to be only 2.1 months with just 27.7% of those prescribed acamprosate persisting for 6 months as recommended by NICE. 34 Therefore, patients may not be gaining the maximum benefits from the medication. 21 Medication effectiveness can be further limited by underdosing, overdosing or taking medication at incorrect intervals. 35 The Testing Combined Pharmacotherapies and Behavioral Interventions in Alcohol Dependence trial, a large US-based trial, found an association between poor adherence to both acamprosate and naltrexone and lower percentage days abstinent and higher percentage days heavy drinking. 36,37 Furthermore, poorer alcohol outcomes were also identified in those who were non-adherent early in the Testing Combined Pharmacotherapies and Behavioral Interventions in Alcohol Dependence trial compared to those who were non-adherent later in the trial. 36,37

Reasons for non-adherence

Poor adherence to a medication may be due to multiple and complex reasons. 29 The complexity of the dose regimen, with greater dose frequency and complexity of instructions, has been associated with poorer adherence. 38–40 Side effects were commonly reported as impacting on adherence in the Testing Combined Pharmacotherapies and Behavioral Interventions in Alcohol Dependence trial. 36,37 Poorer adherence was also found for the trials combined therapy group compared to the single active therapy group, which may be explained by an increase in side effects experienced by participants taking both acamprosate and naltrexone. 30

Horne has proposed that patients weigh up the potential costs and benefits of medication when making adherence decisions. 41,42 Greater adherence to medications in chronic health conditions such as asthma, diabetes, cardiac disease and cancer has been associated with a greater perception of the benefits of the medication. Conversely, greater concern about potential side effects of medication has been associated with poorer adherence. 42,43 The development of effective interventions to address the uncertainty that some patients feel towards the benefits of medication and their concerns about the potential adverse effects is a priority to improve adherence in the treatment of chronic health conditions. 35,43

Medication Management

The NICE alcohol treatment guidelines21 recommend monthly supervision for the first 6 months while taking acamprosate. There is a wide variation in the type and frequency of support received in clinical practice with support usually delivered by a combination of primary care and specialist addiction services.

Psychosocial interventions to maximise adherence to medications for alcohol relapse prevention, including acamprosate, have been used successfully in clinical trials. A six-stage, manualised intervention called BRENDA (Biopsychosocial evaluation, Report, Empathy, Needs assessment/goals, Direct advice, and Assessment)44 has been found to be beneficial for improving adherence to medication for alcohol relapse prevention. 45,46

The manual-based psychosocial intervention, Medical Management, was developed from BRENDA and other Medication Management (MM) interventions47 for the Testing Combined Pharmacotherapies and Behavioral Interventions in Alcohol Dependence trial48 and it has since been used in clinical trials of other medications. 49,50 Medical Management provides education, support and practical advice to service users about their alcohol drinking behaviour and medication to support adherence. An initial session, lasting up to 60 minutes, identifies the rationale for taking the prescribed medication, provides an overview of the dosing regimen, highlights the importance of taking the medication as it has been prescribed and an individualised plan to support adherence is developed. In addition, the service users’ diagnosis, treatment goals and participation in support groups are discussed. The initial session acts as a foundation for the preceding shorter Medical Management sessions that last up to 30 minutes. Despite the successful inclusion of the MM interventions BRENDA and Medical Management in clinical trials, research to support their use in clinical practice is lacking.

The role of the pharmacist

The community pharmacist’s role has extended beyond medication supply to improve public health,51–55 HIV prevention in opioid dependence56–58 and supporting adherence to medication. 59–61 A joint statement by the Royal College of General Practitioners and the Royal Pharmaceutical Society in 2011 identified a role for suitably qualified community pharmacists to contribute to care planning and treatment interventions in substance use disorder (RCGP and RPS, 2011). This has been followed by a recent report of the commission on the future models of NHS care delivered through pharmacy. 62

Healthy Living Pharmacies (HLPs) aim to provide a range of health promotion interventions, with alcohol dependence as one of the conditions targeted. 55,63,64 Community pharmacists and support staff have expressed positive attitudes towards providing extended services in alcohol and substance use disorder when adequate training is provided. 65–67

Contingency Management

Engagement and retention in treatment for alcohol dependence are often suboptimal and are related to poor treatment outcomes. 68,69 Contingency Management (CM) has been found to improve substance use disorder treatment retention and engagement70–74 as well as increase adherence to prescribed pharmacotherapies such as naltrexone for opiate relapse prevention75 and improve prevention, diagnosis and treatment outcomes for HIV, tuberculosis and hepatitis control in substance use disorders. 76 There is also evidence of an increased rate of abstinence of cannabis use over and above evidenced-based treatment (individual Motivational Enhancement Therapy/Cognitive Behavioural Therapy),77 reduced tobacco and alcohol use among adults not in treatment for substance use disorders,78 and it is effective for cocaine, tobacco, opiates and cocaine, and polysubstance use. 79,80

Research on the use of CM in the treatment of alcohol dependence is limited. 21,80 A systematic review81 concluded that CM continues to be a highly effective intervention for a range of substance use disorder treatment and follow-up outcomes, showing sustained growth and high treatment efficacy, and recommends further dissemination and implementation of CM to increase its impact.

Petry et al. 82 examined the use of CM and the common concerns of its implementation and recommend that more research needs to be done to promote a better understanding of CM and its benefits. It is this lack of understanding that leads to CM being underutilised despite its clearly effective results in substance use disorder treatment outcomes. Subsequently, a systematic review of the dissemination and implementation of CM has been conducted. 83 The findings report the importance of organisational input and ongoing supervision and consultation to optimise the effects of CM as well as including barriers to implementing CM, the main one being cost despite evidence of CM’s cost-effectiveness. 84

Studies can use a prize-based protocol with incentives of variable magnitude based on abstinence and/or treatment participation. 85,86 Alternatively, fixed monetary incentives or monetary incentives on an escalating scale may also be used, for example, to improve substance use disorder treatment retention and drug use and post-traumatic stress disorder outcomes,87 improve retention and abstinence with stimulant users in outpatient psychosocial treatment programs,88 improve hepatitis B vaccination adherence and completion in injecting drug users and to reduce heavy alcohol consumption. 89

Why this research is needed now

The effectiveness of acamprosate to support alcohol relapse prevention has been well documented. 21,26,28,46,48 The full benefit of acamprosate in clinical practice has not been maximised due to poor adherence and insufficient duration of its use. Supporting patients in taking acamprosate as prescribed through the application of MM has the potential to help improve adherence and increase the clinical effectiveness of acamprosate. However, there is currently insufficient evidence to determine the most effective form of intervention to support adherence.

The delivery of MM within the pharmacy setting is aligned with the development and expansion of the role of the pharmacist. 55,62–66,90,91 However, the effectiveness of interventions to increase medication adherence for alcohol dependence delivered by pharmacists is currently not known.

Research into the application of CM in alcohol dependence treatment has been recommended by NICE21 to build on the existing small, but growing, body of evidence. 70,78,92 The financial cost of delivering CM is low (< £10 per session) and extensive training to deliver the intervention is not required. If shown to be effective, CM has considerable potential to be adopted within NHS services and community pharmacy to enhance alcohol dependence treatment.

Objectives

• To conduct a definitive three-arm, randomised controlled trial (RCT) of the effectiveness of MM with and without CM compared to Standard Support (SS) alone in enhancing adherence to acamprosate in alcohol dependence relapse prevention.

• To estimate the cost-effectiveness of MM with and without CM compared to SS alone in enhancing adherence to acamprosate in alcohol dependence relapse prevention.

• To assess the impact of adherence to acamprosate for alcohol dependence relapse prevention on abstinence and reduced alcohol consumption.

Primary hypothesis

The primary null hypothesis was:

• MM and MM + CM will be no more effective than SS alone in terms of per cent adherence to acamprosate, 6 months post randomisation.

Secondary hypotheses

The secondary null hypotheses were:

• MM and MM + CM will be no more cost-effective than SS alone at 6 months post randomisation.

• MM and MM + CM will be no more effective than SS in terms of the percentage of possible doses of acamprosate taken, at 12 months post randomisation.

• MM and MM + CM will be no more cost-effective than SS alone in terms of quality-adjusted life-years (QALYs) at 12 months post randomisation.

• No greater adherence to acamprosate will be associated with improved alcohol outcomes, namely a higher percentage of days abstinent, fewer units of alcohol per drinking day, reduced relapse to any drinking and reduced relapse to heavy drinking at 6 and 12 months post randomisation.

• Service user beliefs about medication, and therapeutic relationship with care providers, will not be moderated by medication adherence at 6 and 12 months post randomisation.

Methods

The full methods and results have been published elsewhere. 95

The study received local NHS approvals, and NHS research ethics approval from the West of Scotland Research Ethics Committee (Ref: 15/WS/0048).

Results

Medication Management

The focus groups with service users and pharmacists highlighted the importance of clear guidance but with room for the conversation to be led by the needs of the service user. The importance of excellent communication skills and knowledge of alcohol dependence was also noted. This was incorporated into the guidance documents produced and the pharmacist training (see Project Webpage Document). Treatment goals were identified during the welcome call to help tailor the intervention to the participants’ needs. A printed summary letter was sent to participants that they were encouraged to edit and goals were revised during the subsequent calls. We developed a partnership with Celesio/Lloyds Pharmacies who provided essential input into how the MM calls were delivered. They suggested using a call centre with a small pool of specially trained pharmacists to deliver the intervention. This allowed continuity of the pharmacist delivering the intervention for participants, which had been highlighted as important during both the focus groups and meeting with SURG for building rapport. Text message reminders were also incorporated as suggested by SURG. The length and number of MM sessions were guided by the original48,93 MM manual as well as conversations with SURG and Celesio/Lloyds pharmacies, with a longer initial ‘welcome call’ (~30 minutes) followed by shorter (10–15 minutes) follow-up calls weekly for 6 weeks, fortnightly for 6 weeks and monthly for 3 months (12 calls in total). More frequent calls were completed at the start of treatment when the greatest support may be required with calls becoming less frequent over the 6-month period.

Contingency Management

To maximise the effectiveness of CM, three key principles have been identified in the research literature: clear and objective verification of the treatment goal, immediacy of the reinforcement and significant magnitude of the reinforcement. 102,103 These principles were taken into consideration in conjunction with the results of the focus group and patient and public involvement work when developing the CM schedule (see Chapter 3, Table 3).

Treatment arm 1: Standard Support

All participants in the trial were prescribed acamprosate (two 333 mg tablets morning, afternoon and evening if the service user’s body weight was 60 kg or above, or two 333 mg tablets in the morning and one 333 mg tablet in the afternoon and evening if the service users body weight was below 60 kg, according to the manufacturer’s Summary of Product Characteristics) as soon as possible following alcohol abstinence in addition to the psychosocial care normally provided. The decision to initiate acamprosate was determined by the treating clinician in the specialist alcohol service in conjunction with the service user.

Based on the current service provision of the five study centres, SS comprised monthly dispensing of prescribed acamprosate from the Lloyds community pharmacy, monthly monitoring of the service user for 3 months by the specialist alcohol service, and then returned to the care of their GP for monthly monitoring in accordance with the NICE guidelines104,105 and current NHS clinical practice.

Treatment arm 2: Standard Support plus Medication Management (SS + MM)

Participants followed the same care pathway as those in the SS arm of the trial with the addition of MM, which was delivered by a central telephone support service by trained pharmacists. The MM intervention was adapted from the Medical Management intervention developed by Pettinati et al.,106 for the Testing Combined Pharmacotherapies and Behavioral Interventions in Alcohol Dependence study, a randomised controlled clinical trial of naltrexone and acamprosate for alcohol dependence. A freely available comprehensive manual was published by the Testing Combined Pharmacotherapies and Behavioral Interventions in Alcohol Dependence research group. This was used as a basis for the MM intervention for the proposed research. Adaptation was made in consultation with service users and pharmacists to ensure that it is suitable and acceptable in the context of a UK central pharmacy telephone support service, delivered by trained pharmacists (see Chapter 2).

MM was delivered once a week for the first 6 weeks, reducing to once a fortnight for the following 6 weeks, and then monthly for 3 months, following the same schedule as the Testing Combined Pharmacotherapies and Behavioral Interventions in Alcohol Dependence study. MM was delivered by telephone by a trained pharmacist based in a central telephone support service in the UK provided by Celesio/Lloyds Pharmacy. The initial MM session lasted approximately 30–45 minutes and acted as a foundation for the subsequent sessions, which lasted approximately 10–15 minutes each. Each participant was assigned a specific pharmacist based in the central telephone support service to deliver each of the MM sessions for that participant where possible. The pharmacist sent a text message reminder the day before the appointment and called the participant to deliver the MM session at an agreed time.

MM provided support in developing strategies to help participants to manage their medication including the rationale for taking acamprosate, adhering to the dose regimen and managing side effects, education about their medication and alcohol dependence, and supporting participants’ efforts to change their drinking behaviours. Treatment goals were identified to tailor the intervention to the participant and an individual plan for maintaining adherence was developed with the participant in the initial session to guide the successive MM sessions. Over the 6-month period, the pharmacist sent four summary letters to the participant highlighting the participant’s individual aims, goals and key information regarding their MM plan.

Treatment arm 3: Standard Support plus Medication Management with Contingency Management (SS + MM + CM)

Evidence shows that there can be barriers to participation in MM sessions. 107 To optimise attendance, participants in this arm followed the same care pathway as those in the SS + MM arm of the trial but with the addition of CM. Incentives in the form of Love to Shop vouchers (not redeemable for alcohol) were provided to reinforce attendance at MM sessions by telephone with the pharmacist. The CM procedure and value of the incentives have been informed by the available literature on CM in substance use disorder80 and alcohol dependence104 and focus groups with service users with experience with treatment services for alcohol dependence (see Chapter 2).

Participants received £5 in the form of a voucher for each MM session completed. In addition, they received a £10 bonus voucher for completing four calls in succession, a £20 bonus for completing eight calls and a £30 bonus for completing all 12 calls, with a total value of up to £120 for completing all support sessions (see Table 3). After each MM session, a SMS text message was sent to the participant to inform them that they had been awarded a voucher, the magnitude of the voucher and the total voucher value received to date. Vouchers were given on completion of the 6-month follow-up visit by the research team unless the participant withdrew from the trial or requested part payment earlier.

Inclusion criteria

• Adult, aged 18 years or older.

• An International Statistical Classification of Diseases and Related Health Problems, Tenth Revision, diagnosis of alcohol dependence.

• Abstinent from alcohol at baseline assessment.

• In receipt of a prescription for acamprosate.

• Willing and able to provide informed consent.

Exclusion criteria

• Diagnosis of a severe physical/mental illness likely to preclude active participation in treatment or follow-up.

• Unable to understand verbal English at a level necessary to engage in the intervention and follow-up assessments.

• Concurrent dependence on an illicit substance (other than cannabis).

Secondary outcome measures

Secondary, process and economic outcome measures are detailed in Table 4. At baseline assessment, participant demographics were collected as well as a history of use of acamprosate, other relapse prevention medication use, and previous medically assisted detoxification using a medical history checklist devised specifically for the trial. The substance use section of the Alcohol, Smoking and Substance Involvement Screening Test-Lite (ASSIST-Lite)104 was administered to obtain a history of any substance use.

Severity of dependence was measured at the baseline assessment and at 6- and 12-month follow-ups using the SADQ. 108 The SADQ is a 20-item self-complete questionnaire containing items representing five domains of the alcohol dependence syndrome: (1) physical withdrawal signs, (2) psychological withdrawal signs, (3) withdrawal relief drinking, (4) tolerance and (5) reinstatement following a period of abstinence.

Alcohol consumption was measured using the TLFB Form 90,109 administered at initial baseline assessment and at 6- and 12-month follow-ups after initiation of acamprosate. Percentage days abstinent, units of alcohol per drinking day (1 UK unit = 8 g alcohol), relapse to any drinking and relapse to heavy drinking (8 +/6 + units for males/females on a single occasion) were derived.

Participants’ beliefs about medications were assessed using the BMQ. 110 The BMQ assesses an individual’s beliefs about medication specific to them and their health, as well as their general beliefs about medication. This questionnaire was administered at baseline assessment and at 6-month follow-up. The measure was also used to evaluate the impact of MM on beliefs and concerns about medication and the association with adherence to acamprosate. The MMAS-8 assesses non-adherence and was administered at months 2, 4, 6 and 12. 111–113

Participants in the SS + MM and SS + MM + CM groups were asked to rate their therapeutic relationship with the telephone pharmacist at 6 months using the STAR rating scale. 114 Therapeutic relationship (or alliance) has been found to predict clinical outcome across a range of mental disorders115 including alcohol dependence. 116 This was used as an additional process measure to explore the impact of therapeutic relationship on medication adherence and clinical outcome.

Alcohol-related problems were assessed at baseline assessment using the APQ. 8 The APQ is a 46-item questionnaire assessing potential problems with psychological, physical, social, legal, employment, relationships and parenting that may be experienced due to alcohol. The AUQ117 assesses current urge for alcohol using eight items which cover three factors: desire for a drink (four items); expectation of positive effect from drinking (two items) and inability to avoid drinking if alcohol was available (two items). This questionnaire was administered at baseline assessment and at 6- and 12-month follow-ups.

Measures for the economic evaluation, collected at baseline, 6- and 12-month follow-ups, included the EQ-5D-5L measure of health-related quality of life118 suitable for the calculation of QALYs and the AD-SUS, based on a version designed for use in alcohol and drug populations119 and adapted for the purpose of the ADAM trial. These measures are described in more detail in Chapter 5.

Participants were asked at each bi-monthly research visit whether they have experienced any side effects from the medication. In addition, reasons for non-adherence were recorded.

Analysis by treatment allocated

Analyses all available data for participants who were randomised, regardless of whether they complied with allocation. This data set includes participants who were withdrawn from the trial post randomisation. These analyses are a lower bound estimate of treatment effects as they represent the effect of offering an intervention, rather than the effect of receiving that intervention.

Complier average causal effects

We assessed treatment effects in the presence of non-compliance, with compliance measured at the individual level and including all those allocated as part of the trial. Our approach for assessing treatment effects under non-compliance was via the instrumental variables (IV) framework. 121 The benefit of using an IV approach is that randomisation is maintained in the analysis, which is crucial for estimating unbiased treatment effects. 122 CACE weights the ATA treatment effect by the proportion of compliers (see Equation 1):

If the proportion compliant is 1.0 (i.e. perfect compliance), then the CACE estimate is the same as the ATA estimate, but otherwise the impact of this approach is to increase the magnitude of the treatment effect.

Complier average causal effects use a two-stage least squares (2SLS). The first stage model uses treatment received (T) as the outcome, with random allocation (Z) as the independent variable (see Equation 2):

Based on the stage 1 model, we then calculate predicted values of treatment received (T^) for use in stage 2. The second stage model predicts the substantive outcome (Y, e.g. days abstinent) using the predicted values of treatment received (T^) based on the stage 1 model (see Equation 3):

The CACE analysis was conducted using 2SLS estimation with the ivregress command in Stata^® (StataCorp LP, College Station, TX, USA). A linear regression approach was used to assess per cent adherence with medication and per cent days abstinent from alcohol at 6 months post randomisation.

Missing data

The proportion of missing data and patterns of missingness were examined for the primary outcome, per cent adherence to medication at month 6. Levels of missing data are reported along with any systematic occurrences of missing data observed in the data sets. Where outcomes are derived scores, individual item scores were checked for systematic missingness by comparing the proportion of missing values by SADQ category (≤ 30 and > 30), current prescription of relapse prevention medication, community alcohol treatment service and allocated group.

To avoid loss of efficiency, we imputed missing primary outcome values using multiple imputation using the MI commands of Stata; we employed 50 iterations of a multiple imputation model and combined these using MIcombine. We then tested the extent to which the imputed model deviates from the model generated using the observed data. This approach makes an underlying assumption that the missing data are missing at random (MAR). To test a potential assumption that data were not missing at random (MNAR), we conducted a sensitivity analysis using a pattern mixture approach and multiple imputation to explore the sensitivity of the primary outcome results to departures from the MAR assumption, and this was implemented using the rctmiss command in Stata. 123,124

Statistical analysis methods

Analysis of the study is presented in accordance with CONSORT guidelines. 125 The primary analysis is an ATA and is based on all available data for participants who were randomised, irrespective of whether they complied with their allocation or not.

Diagnostic tests and plots were conducted to assess the assumptions of normality for per cent adherence to acamprosate at month 6. There were significant departures from normality (Project Webpage Document) and the distribution overall and by group was a bimodal distribution with a zero-inflation. As the primary outcome is a percentage, it can be recalculated as a proportion between 0 and 1 by dividing the per cent adherence by 100. The fractional nature of the outcome allows for a fractional logistic model126 to be fitted, assuming the variance in response is proportional to a binomial distribution, and employing a logit link function to maintain bonds between 0 and 1. Fixed effects were included for allocation (SS/SS + MM/SS + MM + CM) and stratification covariates (SADQ; ≤ 30 or > 30, other relapse prevention medication; yes or no, site). Results are initially presented as odds ratios, but marginal effects can be derived to present the mean difference in per cent adherence between the groups and the associated 95% CI. Secondary analysis of the primary outcome encompasses multiple imputation to assess missing data as MAR, sensitivity analysis to assess missing data being MNAR and inclusion of data on adherence to allocated intervention using two pre-specified scenarios incorporated into a CACE analysis. Analysis was undertaken to explore differences between SS + MM + CM versus SS and then to explore differences between SS + MM versus MM. As significant effects were observed on the primary outcome measure, an analysis exploring SS + MM + CM versus SS + MM was then undertaken.

Secondary outcomes are analysed in a similar manner by assessing distributional assumptions and conducting an appropriate regression with the inclusion of the same covariates as the primary analysis. Where baseline values for the secondary outcome are available, these are also included as covariates.

Stepwise regression was conducted to model the between pre-randomisation factors and per cent adherence to Acamprosate and per cent days abstinent at month 6. Interaction terms with treatment allocation were included in the model and a significance level of 0.1 was used as a threshold to determine which variables were maintained in the final model reported. Baseline variables included initially in the model include age, gender, marital status, ethnicity, employment status, number of children, age of first drink, weekly and daily drinking, frequency and quantity of alcohol consumption, severity of alcohol dependence, alcohol urges and alcohol-related problems. This analysis was augmented with an additional analysis for the SS + MM and SS + MM + CM groups where the same dependent variable was assessed with the same independent variables with the addition of therapeutic alliance.

Deviations from the statistical analysis plan

The APQ was not collected at 6 and 12 months, which is a deviation from the trial protocol. In addition, two key changes were implemented that deviated from the statistical analysis plan. These changes were agreed with the trial steering committee.

1. Measuring the primary outcome

Our primary outcome was per cent adherence to acamprosate at month 6. At the design stage, we explored the current literature on measuring adherence and settled on a hierarchy of measure. Our first measure was to be the MEMS, where medication is placed in a pill bottle and a cap measures every time the cap was opened. As acamprosate is taken in three daily doses (666 mg three times per day or 666 mg once followed by 333 mg twice a day), we planned to calculate the denominator as the number of days in the period between baseline and month 6 multiplied by three and the numerator the number of MEMS registered openings; this would allow an objective measure of adherence. At month 6, MEMS data were available for 514 participants, 70%, yet only one participant was 100% adherent and the majority, 383 (75%) had zero adherence with a mean adherence of 23% (95% CI 21% to 26%). Exploration of qualitative notes taken at the time of follow-up suggested large numbers of participants complied with their medication but found the MEMS device inconvenient and cumbersome, rather than taking medication out of the device many either stopped using it at all or took their daily medication out on a weekly basis. An alternative objective measure was pill counts conducted by the pharmacist when participants came to collect their prescribed medication, but it was clear early in the trial that this method was fraught with complications. Many participants did not actually attend the pharmacy to collect medications, those that did often did not bring their remaining pills for counting and when pill counts were conducted, they often failed to include details of what medication had been previously dispensed. Pharmacy pill counts were not considered a reliable source of adherence data. Our third source of adherence data was self-report, where we asked participants at month 6 to estimate the per cent of days they had been adherent to medication in the past 28 days. At month 6, self-report data were available for 514 participants, 70%, of whom 257 (50%) stated they had been completely non-adherent, 70 (15%) had been 100% adherent and the mean adherence was 42% (95% CI 38% to 46%). There was a mean difference between self-report and MEMS of 12.3% (95% CI 8.7% to 15.9%). Considering the known issues with the MEMS data, we decided to use the self-report adherence as the primary outcome. Our reasoning was based on evidence that in the trial, the MEMS device underestimated adherence, evidence that self-report is a valid and reliable method for estimating adherence127 and that self-report estimates over shorter period are reliable proxies for adherence over longer periods for those with chronic conditions. 128

1. Exploring the impact of participants being followed up late at the month 6 primary end point

In our trial protocol, we stated that the primary end point would be 6 months after randomisation. As we are dealing with a relatively hard-to-reach group, we allowed a window of + 60 days around the 6-month follow-up point. It became apparent prior to analysis that some participants had 6-month follow-up assessments conducted later than 6 months plus 60 days. We identified 45 participants that fell into this group, 9% of those followed up at month 6. Exploring baseline variables for those followed up late versus those not, identified no differences in terms of demographics, allocation, or outcomes (Project Webpage Document). We did note that the majority of those followed up late indicated zero adherence to acamprosate at month 6, 34 out of 45 (76%). To place maximum confidence on our findings, we proposed to include an additional sensitivity analysis of the primary outcome where we include and exclude these participants to explore any impact of late follow-up on the outcomes observed.

Late follow-up sensitivity analysis

Our first sensitivity analysis incorporates those who followed up on time and those who followed up late at month 6 (please see Appendix 1, Table 24). Table 12 presents the results of this sensitivity analysis. There are no significant deviations from the primary analysis conducted so all participants are included in the analysis.

Missing primary outcome data

To explore whether missing primary outcome data are MAR we conducted a logistic regression analysis to explore what variables might predict missing outcome data. Variable selection was undertaken by exploring associations at the 0.10 p-level. The analysis identified age, site, gender and per cent days abstinent at baseline as being predictive of missing primary outcome data. Using these predictive variables, we imputed 50 iterations of a multiple imputation model and pooled these using the micombine command in Stata.

We used a similar fractional regression approach as the primary analysis to generate odds ratios comparing SS + M + CM versus SS, SS + MM + versus SS and SS + MM + CM versus SS + MM. As the mi command set does not allow for the generation of marginal effect, we plotted odds ratios and 95% CIs for each of the comparisons in order to visually explore whether there are significant deviations between the observed and imputed values of the primary outcome. Appendix 1, Figures 21–23 presents the plots and it is confirmed that the influence of missing data, based on a MAR assumption, does not influence our interpretation of the results.

To explore the influence of missing data being MNAR, we employed a sensitivity analysis to explore departures from the MAR assumption. Varying deltas of 0.1 ranging from −0.5 to 0 were employed in intervention group alone, control group alone and both intervention and control group. Analysis was undertaken using the rctmiss command in Stata. Visual inspection of the data indicates even at large departures from the MAR assumption the impact on the treatment effect is small; these are presented in Appendix 1, Figures 24–26.

We can be confident that the treatment effects are robust under different interpretations of the nature of missing data.

Complier average causal effects analysis

We defined our compliance criteria as two separate scenarios: scenario 1 where participants in the intervention groups adhered to at least 6 (50%) MM calls and scenario 2 where they adhered to 12 (100%). Of those allocated to one of the intervention groups, 105 (58%) adhered to 50% of the MM calls in the SS + MM group and 128 (69%) in the SS + MM + CM group. Compliance with 100% of MM calls also differed by allocated group with 20 (11%) complying in the SS + MM group and 73 (39%) complying in the SS + MM + CM group.

We conducted a CACE analysis using the ivregress 2sls command in Stata. The dependent variable was per cent adherence to acamprosate at month 6 adjusted for stratification covariates. Table 13 presents a comparison of ATA effects and CACE effects for each of our scenarios.

Under scenario 1, 50% adherence, the mean difference in per cent adherence increases from 10.6% (95% CI 19.6% to 1.62%) to 12.45% (95% CI 17.8%) in the SS + MM + CM versus SS comparison. This indicated that at 50% compliance, those allocated to SS + MM + CM have a mean of 12.4% more adherence to acamprosate at month 6 than those allocated to SS alone. When comparing the SS + MM versus the SS group, a previous non-significant difference of 3.14% (95% CI 12.8% to −6.5%) becomes a significant difference of 13.2% (25.4% to 1.15%). Under scenario 2, the differences are larger in magnitude, the SS + MM + CM group having 22.2% (95% CI 29.7% to 14.7%) more adherent days than the SS group alone. While the magnitude of difference is larger for the SS + MM versus SS comparison 31.8% (95–60.5% vs. 3.10%), this comparison is based on a small number of participants (n = 20), and should be interpreted with caution.

Method of economic evaluation

The primary economic analysis was a cost-utility analysis where outcomes were expressed as QALYs, as recommended by NICE. 129 A secondary analysis explored cost-effectiveness in terms of the primary clinical outcome which was adherence to relapse prevention medication.

Perspective

The primary economic perspective was the NHS and personal social services (NHS/PSS) perspective preferred by NICE. This covers all hospital, community health and social services. A sensitivity analysis based on a broader perspective including impact on the criminal justice sector was planned, however, this became redundant because criminal activity was not reported by any patient at any time point.

Time horizon

The primary time horizon of the economic analyses using both QALYs and relapse medication adherence was the 6-month follow-up, consistent with the primary clinical analysis. A secondary analysis was carried out at the 12-month follow-up using QALYs.

Outcomes

The primary economic outcome was QALYs calculated using the EQ-5D-5L measure of health-related quality-of-life scores at baseline, 6- and 12-month follow-ups. The EuroQol-5 Dimensions (EQ-5D) is a non-disease-specific measure for describing and valuing health-related quality of life. 118 The measure covers five health domains (mobility, self-care, usual activities, pain/discomfort, anxiety/depression) which are rated on five levels (e.g. no, slight, moderate, severe, extreme problems).

The EQ-5D-5L (five levels) is a revised version of the EuroQol-5 Dimensions, three-level version (EQ-5D-3L). It has been established that the EQ-5D-3L can be used with confidence in economic evaluations for alcohol dependence. 114 The health states from the EQ-5D-5L were given a utility score using responses from a representative sample of adults in the UK. 125 These utility scores, however, have not been accepted by NICE which instead recommends using EQ-5D-3L values based on a cross-walk algorithm. 126 Accordingly, a cost-utility analysis based on the EQ-5D-3L values is presented as a sensitivity analysis.

Economic model

The costs and benefits of interventions for alcohol problems, such as alcohol-related complications and mortality, extend well beyond the usual time horizons of clinical trials. Economic modelling can be utilised to project costs and QALYs beyond the trial end points. A de novo patient level, multistate life table model was built using published risk equations130 to estimate the effects of changing alcohol consumption on costs and QALYs over a 20-year time horizon.

Measurement of resources

Valuation of resources

A unit cost was applied to each resource used to calculate the total cost of resources used by each study participant (summarised in Table 16). All unit costs are for the financial year 2018–19, uprated where necessary using the GDP deflator. 137

Intervention cost

Data analysis

Data completeness

Full service use data at the 6-month follow-up point were available for 76% (141/185) in the SS + MM + CM group, 67% (122/182) in the SS + MM group and 69% (255/372) in the SS alone group, which was 70% of the total number randomised. Full service use data for the 6 months prior to the 12-month follow-up were available for 56% (104/185) in the SS + MM + CM group, 53% (96/182) in the SS + MM group and 52% (193/372) in the SS alone group, which was 53% of the total number randomised.

Resource use

Resource use reported over the periods from baseline to 6-month follow-up and from 6-month to 12-month follow-up is summarised in Table 17.

Interventions

Patients in the SS + MM + CM group attended more MM sessions than the SS + MM group (mean 7.7 sessions per participant vs. 6.1). Patients in the SS + MM + CM group received an average of around nine vouchers out of a possible 12; 98% received at least one voucher.

Health and social care services

Apart from less reported use of supported accommodation in the SS + MM + CM group (mean 3.1 weeks per participant) compared to the SS + MM group (mean 15.1 weeks) and the SS group (mean 10.4 weeks) over the 6-month follow-up, the use of other health and social care services was broadly similar across the three randomised groups over this period.

Over the period from the 6-month to the 12-month follow-up, the use of supported accommodation was highest in the SS + MM group (mean 14.5 weeks per participant), compared to the SS + MM + CM group (mean 9.1 weeks) and the SS group (mean 8.5 weeks). The use of all other health and social care services was broadly comparable across all three groups.

Alcohol relapse prevention medication

The SS + MM + CM group were prescribed acamprosate for a longer period on average (mean 120 days) compared to the SS + MM group (mean 101 days) and the SS group (mean 100 days) over the 6-month follow-up period. The pattern remained similar when including use of other relapse medication, with the highest use in the SS + MM + CM (mean 124 days) compared to the SS + MM group (mean 110 days) and the SS group (mean 102 days). Over the 6- to 12-month follow-up period, the SS group were prescribed acamprosate for a longer period on average (mean 47 days) compared to the SS + MM + CM group (mean 44 days) and the SS + MM group (mean 37 days). However, these differences reduced when all relapse medications were combined (mean 49 days SS; 47 SS + MM + CM; 44 SS + MM).

Crime

Participants did not report committing any criminal offences over the 6-month or 12-month follow-up periods. This is consistent with data suggesting that most alcohol-related crimes, such as assault and domestic abuse, are due to alcohol abuse rather than alcohol dependence. 142 At 6 months, 28 (5%) of participants followed up (SS + MM + CM n = 5 (4%), SS + MM n = 8 (7%), SS n = 15 (6%)) reported being victims of crime including fraud (n = 2), personal theft (n = 7), house burglary (n = 4), criminal damage (n = 1), assault (n = 9), harassment (n = 4) and domestic violence (n = 1). At 12 months, 33 (8%) of participants followed up [SS + MM + CM n = 9 (9%), SS + MM n = 10 (10%) SS n = 14 (7%)] reported being victims of crime including fraud (n = 2), personal theft (n = 6), house burglary (n = 6), criminal damage (n = 2), assault (n = 9), harassment (n = 5) and indecent assault (n = 3).

Costs

Primary cost-utility analysis

EQ-5D-5L scores

EQ-5D-5L health states are summarised in Figure 3 based on the complete data. The graph shows that the average health state values at baseline were similar for all groups (range 0.786–0.809). At the 6-month follow-up point, the average health state values were also similar for all groups (range 0.778–0.805), although with small declines seen in the SS + MM + CM and SS groups. At the 12-month follow-up, the SS + MM + CM and SS + MM treatment groups recorded small gains compared to baseline of + 0.044 and + 0.026, respectively, while the SS group recorded a slight decline (−0.014).

FIGURE 3.

Mean EQ-5D-5L health state scores over time by group.

Cost-utility analysis at 6-month follow-up

In the primary economic evaluation, a cost-utility analysis at the 6-month follow-up point, the SS + MM + CM group achieved small gains in QALYs compared to the SS group at a lower cost per participant. Thus, the SS + MM + CM group dominated the SS group (better outcomes, lower costs). The SS + MM group also achieved small gains in QALYs but at an additional cost per participant compared to the SS group. The ICER (additional cost per additional QALY) for the SS + MM group was £95,000 per QALY, which is well above the NICE £20,000–30,000 per QALY threshold.

The scatterplots in the cost-effectiveness planes in Figures 4 and 5 illustrate uncertainty in the joint distribution of costs and outcomes generated using non-parametric bootstrap methods for SS + MM + CM versus SS and SS + MM versus SS, respectively, at the 6-month follow-up point. For SS + MM + CM (see Figure 4), the scatterplot illustrates that for the majority of replications, total costs were lower and total QALYs were higher for SS + MM + CM than SS. For SS + MM (see Figure 5), the scatterplot illustrates that for the majority of replications, total costs and QALYs were higher for SS + MM when compared to SS.

FIGURE 4.

Scatterplot showing the bootstrapped mean differences in imputed costs and imputed QALYs for SS + MM + CM compared to SS at 6 months.

FIGURE 5.

Scatterplot showing the bootstrapped mean differences in imputed costs and imputed QALYs for SS + MM compared to SS at 6 months.

The CEAC in Figure 6 shows that there was a 78% to 80% probability that SS + MM + CM was cost-effective compared to SS, using the NICE willingness-to-pay thresholds of £20,000–30,000 per QALY at the 6-month follow-up point. The CEAC in Figure 7 shows that there was a 22% to 27% probability that SS + MM was cost-effective compared to SS, using the NICE willingness-to-pay thresholds of £20,000–30,000 per QALY at the 6-month follow-up point.

FIGURE 6.

Cost-effectiveness acceptability curve showing the probability that SS + MM + CM is cost-effective compared to SS alone in terms of QALYs at 6 months.

FIGURE 7.

Cost-effectiveness acceptability curve showing the probability that SS + MM is cost-effective compared to SS alone in terms of QALYs at 6 months.

Secondary cost-utility analysis at 12 months

At the secondary 12-month follow-up point, the SS + MM + CM group made gains in QALYS but at a higher cost per participant compared to the SS group. The ICER (additional cost per additional QALY) was £9517 per QALY, which is below the NICE £20,000–30,000 per QALY threshold. The SS + MM group also achieved gains in QALYs with an additional cost per participant compared to the SS group. The ICER (additional cost per additional QALY) for the SS + MM group compared to the SS group was £24,133 per QALY, also below the NICE threshold.

For SS + MM + CM, the 12-month scatterplot (see Figure 8) illustrates that in the majority of replications, total costs were greater and QALYs also greater compared to SS. For SS + MM, the 12-month scatterplot (see Figure 9) is similar to the SS + MM + CM versus SS comparison, with the majority of replications also showing higher costs and QALYs for SS + MM compared to SS.

FIGURE 8.

Scatterplot showing the bootstrapped mean differences in imputed costs and imputed QALYs for SS + MM + CM compared to SS at 12 months.

FIGURE 9.

Scatterplot showing the bootstrapped mean differences in imputed costs and imputed QALYs of SS + MM compared to SS at 12 months.

The CEAC in Figure 10 shows that the probability that SS + MM + CM was cost-effective compared to SS ranged between 61% and 68% using the NICE willingness-to-pay threshold of £20,000–30,000 per QALY. The CEAC in Figure 11 shows that the probability that SS + MM was cost-effective compared to SS ranged between 48% and 581%, using the same thresholds.

FIGURE 10.

Cost-effectiveness acceptability curve showing the probability that SS + MM + CM is cost-effective compared to SS alone in terms of QALYs at 12 months.

FIGURE 11.

Cost-effectiveness acceptability curve showing the probability that SS + MM is cost-effective compared to SS alone in terms of QALYs at 12 months.

Secondary cost-effectiveness analysis at 6 months

The primary clinical outcome was adherence to acamprosate medication expressed as percentage of days adherent at the 6-month follow-up. Adherence was highest in the SS + MM + CC group (47%, n = 141,) followed by SS + MM group (41%, n = 122) and lowest in the SS group (38%, n = 255).

The results of tests of differences in imputed and adjusted percentage of adherent days at the 6-month primary clinical follow-up point are presented in Table 21. Note that multiple imputation in the economic analysis included baseline costs and baseline utilities as covariates in the chained equations leading to slightly different imputed estimates to the clinical analysis. Adherence was better in both treatment groups compared to control, reaching statistical significance in the SS + MM + CM group.

In terms of the primary clinical outcome at 6-month follow-up (acamprosate medication adherence), costs were lower and outcomes significantly better in the SS + MM + CM group compared to the SS group, so SS + MM + CM again dominated SS alone. For the SS + MM group, outcomes were better but costs were higher compared to the SS group, meaning that assessment of cost-effectiveness depends upon the maximum threshold value decision-makers would be prepared to pay for a unit change in outcome.

The scatterplots in the cost-effectiveness planes in Figures 12 and 13 illustrate uncertainty in the joint distribution of costs and outcomes generated using non-parametric bootstrap methods, for SS + MM + CM versus SS and SS + MM versus SS, respectively. For SS + MM + CM (see Figure 12), the scatterplot clearly illustrates that for the majority of replications, total costs are lower for SS + MM + CM versus SS and effectiveness is higher. For the SS + MM group (see Figure 13), most replications fall in the north-east quadrant where both costs and effects are higher compared to the SS group.

FIGURE 12.

Scatterplot showing the bootstrapped mean differences in imputed costs and imputed medication adherence for SS + MM + CM compared to SS at 6 months.

FIGURE 13.

Scatterplot showing the bootstrapped mean differences in imputed costs and imputed medication adherence for SS + MM compared to SS at 6 months.

The CEAC in Figure 14 shows that the probability that SS + M + CM is cost-effective compared to SS is between 89% and 95% when decision-makers are willing to pay between £0 and £50 per percentage point improvement in medication adherence over 6 months. However, the probability that SS + MM is cost-effective compared to SS was only 18% to 22% across the same range of willingness-to-pay values (see Figure 15).

FIGURE 14.

Cost-effectiveness acceptability curve showing probability that SS + MM + CM is cost-effective compared to SS in terms of acamprosate adherence at 6 months.

FIGURE 15.

Cost-effectiveness acceptability curve showing the probability that SS + MM is cost-effective compared to SS in terms of acamprosate adherence at 6 months.

Sensitivity analysis

Results of the sensitivity analyses are reported in Table 22. Both the complete case analysis (rather than imputation of missing data) and the analysis using the EQ-5D-3L value set (rather than the EQ-5D-5L) produced similar results to the primary analysis at 6-month follow-up, indicating that the results of the primary analysis were robust to variation of assumptions.

Cost-effectiveness outcomes modelled over 20 years

Running the patient-level, multistate life table model over 20 years after the end of the trial produced results in favour of both SS + MM + CM and SS + MM compared to SS alone. The SS + MM + CM group dominated SS achieving 0.03 more discounted QALYs at a lower discounted cost of −£401 per participant. The SS + MM group achieved QALY gains of 0.15 discounted QALYS but at a higher discounted cost of £64 per participant compared to the SS group. The ICER (additional cost per additional QALY) was £427 per QALY, well below the NICE £20,000–30,000 threshold.

Probabilistic sensitivity analysis was used to vary model key parameters to estimate model uncertainty. For SS + MM + CM, the 20-year scatterplot (see Figure 16) shows that 999 out of 1000 simulations fell in the south-east quadrant, indicting more QALY gains at lower cost compared to SS. For SS + MM, the 20-year scatterplot (see Figure 17) shows relatively greater gains in QALYs compared to SS but with over half of these at a higher cost.

FIGURE 16.

Scatterplot showing the bootstrapped mean differences in imputed costs and imputed QALYs for SS + MM + CM compared to SS modelled over 20 years.

FIGURE 17.

Scatterplot showing the bootstrapped mean differences in imputed costs and imputed QALYs for SS + MM compared to SS modelled over 20 years.

The CEAC in Figure 18 shows that the probability that SS + MM + CM was cost-effective compared to SS was 100% across all values of the NICE willingness-to-pay per QALY threshold plotted (£0–30,000). The CEAC in Figure 19 shows that the probability that SS + MM was cost-effective compared to SS was 100% across the range of the NICE willingness-to-pay threshold of £20,000–30,000 per QALY.

FIGURE 18.

Cost-effectiveness acceptability curve showing the probability that SS + MM + CM is cost-effective compared to SS in terms of QALYs modelled over 20 years.

FIGURE 19.

Cost-effectiveness acceptability curve showing the probability that SS + MM is cost-effective compared to SS in terms of QALYs modelled over 20 years.

SS + MM + CM versus SS

For SS + MM + CM versus SS, the primary economic analysis at 6-month follow-up using QALYs, the secondary economic analysis at 6-month follow-up using medication adherence and the economic modelling over a 20-year time horizon using QALYs all found SS + MM + CM to dominate SS (better outcomes at lower cost). At 12-month follow-up, although SS + MM + CM was not dominant, it generated more QALYs at an additional cost that was below the NICE cost per QALY threshold. There was, therefore, a higher probability of SS + MM + CM being cost-effective compared to SS alone in all analyses and at all time points.

SS + MM versus SS

For SS + MM versus SS, at 6- and 12-month follow-up and when modelled over 20 years, SS + MM achieved better outcomes at higher cost compared to SS. In terms of cost-effectiveness, SS + MM was not found to be cost-effective at 6-month follow-up but had a higher probability of being cost-effective compared to SS at both the 12-month follow-up (using the higher £30,000 per QALY NICE threshold) and when modelled over 20 years (over the full £20,000–30,0000 cost per QALY threshold).

Implications

The results of the primary economic analysis at the 6-month follow-up point suggest that MM was only cost-effective when supported by voucher incentives to encourage support session uptake. This finding was heavily influenced by lower total costs in the SS + MM + CM group as a result of lower use of residential rehabilitation facilities compared to both SS + MM and SS alone, which may be related to the significantly higher medication adherence seen in the CM group. Over the medium (12 months) and longer term (20 years), SS + MM + CM remained cost-effective compared to SS and there was a higher probability of SS + MM being cost-effective compared to SS. These results support the addition of MM to SS for alcohol dependence, with or without CM. However, the economic benefit was stronger when CM was included.

Recruitment and randomisation

This study aimed to compare the effects of pharmacist-delivered MM, with and without incentivisation provided by CM, compared to SS. The primary hypothesis was that MM plus CM plus SS (SS + MM + CM), and MM plus SS without CM (SS + MM), would result in greater medication adherence than SS alone. We also hypothesised that SS + MM + CM would result in greater adherence to MM, as well as greater adherence to medication than SS + MM.

We were able to recruit 739 participants, of whom 372 (50%) were randomly allocated to the SS group, 182 (25%) to SS + MM group and 185 (25%) to SS + MM + CM group, achieving our planned recruitment target, although due to changes in service provider contracts, this took longer to achieve than planned. We also managed to successfully follow up 518 (70%) participants at the primary end point, achieving the planned statistical power. Our follow-up rate at 12 months was lower than planned but this had no influence on the primary outcome analysis.

Comparing the characteristics of participants across study arms at baseline showed no differences between allocated groups, suggesting that the randomisation method minimised bias.

Primary outcome analysis and sensitivity analyses

As discussed in Chapter 3, we were unable to use the primary outcome measure as intended due to various limitations of the MEMS and pill count data. We therefore used self-reported adherence as our primary outcome measure. The potential limitations of this are discussed below.

Our primary ATA showed that the mean difference in per cent adherence to acamprosate was 10.6% lower in the SS group compared to the SS + MM + CM group which was statistically significant. This was also close to the pre-specified clinically important difference. Although the mean adherence was 3.1% lower in the SS group compared to the SS + MM group, this was not statistically significant, and a similar non-significant finding was seen comparing SS + MM and SS + MM + CM groups, with SS + MM 7.9% lower than SS + MM + CM.

We found that 45 (6.1%) participants had been followed up outside of the 2-month follow-up window at the primary end point. A sensitivity analysis, however, showed that this had no impact on the primary outcome analysis. Similarly, both multiple imputation and a sensitivity analysis to explore the impact of missing data indicated that we can be confident that the observed treatment effects are robust under different interpretations of the nature of missing data.

A CACE analysis was conducted under two scenarios: scenario 1 where participants adhered to at least 6 (50%) of the MM calls, and scenario 2 where they adhered to 12 (100%). In scenario 1, 58% in the SS + MM group adhered to 50% of the MM calls, compared to 69% in the SS + MM + CM group. In scenario 2, 11% in the SS + MM group adhered to 100% of MM calls compared to 39% in the SS + MM + CM group. The CACE analysis showed that compared to our intention-to-treat analysis, under scenario 1, the difference between SS + MM + CM and SS increased from 10.6% to 12.4% greater adherence to acamprosate at month 6. Comparing SS + MM to SS, the previously non-significant difference (3.1%) became statistically significant within the CACE analysis at a difference of 13.2%.

Under scenario 2, the differences became larger in magnitude at 22.2% comparing SS + MM + CM with SS, and 31.8% comparing SS + MM with SS (although this latter analysis needs to be interpreted cautiously given the small number of participants in the analysis). So, in both scenarios 1 and 2, the benefits of SS + MM + CM were significant and clinically important. This was the case for SS + MM only when compliance was incorporated into the analysis.

Secondary outcome analyses

The ATA analysis showed lower per cent days abstinent in SS compared to both SS + MM + CM and SS + MM, but these were not significant. However, when assessed under the scenario 2 CACE analysis, per cent days abstinent were significantly lower in the SS group compared with both SS + MM + CM and SS + MM groups. ATA analysis of other secondary outcome measures at both 6 and 12 months showed no significant differences between groups.

The analysis of compliance with MM calls showed that those who complied were significantly more likely to have improved alcohol use outcomes: 12.5% more per cent days abstinent in the SS + MM + CM group and 15.1% more in the SS + MM group compared to SS alone (although the differences in the SS + MM group should be taken with caution considering the small numbers involved). There was also a significant correlation between increased adherence to acamprosate and better drinking outcomes (r = 0.39; p < 0.001).

Health economic analysis

The primary economic analysis was a cost-utility analysis where outcomes were expressed as QALYs, as recommended by NICE. 129 A secondary analysis explored cost-effectiveness in terms of the primary clinical outcome, which was adherence to relapse prevention medication. The primary economic perspective was the NHS/PSS perspective preferred by NICE. This covers all hospital, community health and social services. A sensitivity analysis based on a broader societal perspective, including impact on the criminal justice sector was planned. However, this became redundant because criminal activity was not reported by any patient at any time point.

Full service use data at the primary follow-up point of 6 months were available for 70% of the sample randomised, and 53% at the 12-month follow-up point. We found that SS + MM + CM participants attended more MM calls than the SS + MM group (mean of 7.7 vs. 6.1), and participants in the SS + MM + CM group received an average of 9 vouchers out of a possible 12, with 98% receiving at least one voucher. The SS + MM + CM group were prescribed acamprosate for a longer period compared to the SS + MM and SS groups at month 6, although this reduced and was broadly similar between groups by month 12.

The main difference in health and social care costs at month 6 was the SS + MM + CM group reported much less use of supported accommodation (primarily residential rehabilitation) than either the SS + MM or SS groups. The use of all other health and social care services was broadly similar across groups at month 6. At month 12, the SS + MM group reported greater use of supported accommodation than SS + MM + CM and SS, but the use of all health and social care services was similar across groups.

Differences in quality of life between groups and across time points were small, but favoured the SS + MM + CM and SS + MM groups over the SS group. This translated into small differences in QALYs which were significantly higher only for the SS + MM group over SS at 12 months.

The primary economic analysis found that SS + MM + CM was cost-effective compared to SS alone, but that SS + MM was not cost-effective compared to SS alone. SS + MM + CM dominated SS alone, achieving more QALYs at lower cost, with an 89% probability that SS + MM + CM was more cost-effective compared to SS over the NICE £20,000–30,000 cost per QALY threshold range. The cost per QALY benefit of SS + MM versus SS was at an additional cost of £151,250 per QALY which is well over the NICE threshold. The results were robust in the planned sensitivity analyses.

Secondary analysis based on the main clinical outcome of medication adherence showed that SS + MM + CM dominated SS, and SS + MM also achieved better medication adherence but at an additional cost of £329 per one additional per cent of medication adherence. At 12 months, both SS + MM + CM and SS + MM had a higher probability of being cost-effective than SS alone at the NICE cost-per-QALY thresholds of £20,000–30,000 while SS + MM had a higher probability of being cost-effective than SS alone at the £30,000 cost-per-QALY threshold. An analysis of cost-effectiveness modelled over 20 years amplified these findings and showed that the probability both SS + MM + CM and SS + MM were cost-effective compared to SS was 100% across all values of the NICE willingness-to-pay threshold of £20,000–30,000 per QALY.

Generalisability

The study had minimal exclusion criteria compared to some other trials of relapse prevention medications and a large proportion of those approached to participate were both eligible and gave consent to take part. Given the large and diverse range of community addiction services in four regions across England from which participants were recruited, including both NHS and third sector treatment providers in urban, metropolitan and suburban/semi-rural areas, we are confident that our participants were representative of patients attending typical services in England at the time of the study. Participants had to be recently detoxified from alcohol in order to be eligible, so the sample will not be representative of all patients attending addiction services. However, they will be generally representative of people for whom initiation of acamprosate prescribing is recommended by NICE.

Implications for health care

Implications for research