Background
A number of trials17,18,42,43,117,118 focusing on young people (aged 12–21 years) have reported significant positive effects of brief interventions on a range of alcohol consumption measures. Our systematic review (reported in Work package 2: exploratory modelling of the interventions) suggested that eBIs can significantly reduce alcohol consumption compared with minimal or no intervention controls, and have the added advantage of being more acceptable and easier to implement than more traditional face-to-face interventions. Our study of the prevalence of risky drinking among an adolescent population (aged 10–17 years) reported in Work package 1: screening prevalence study of alcohol consumption and alcohol use disorders in adolescents aged 10–17 years attending emergency departments found that about one in four young people presenting to EDs was consuming three or more drinks on one or more occasion over the preceding month, and that this level of consumption was associated with increased physical, social and educational adverse consequences. We also observed a steep transition in drinking prevalence between 13 and 17 years of age.
Several school-based interventions121 that target non-drinking adolescents have been found to delay the onset of drinking behaviours, and a recent study of adolescents122 found lower rates of substance misuse initiation among those exposed to a web-based intervention. Web-based alcohol interventions for adolescents also demonstrated significantly greater reductions in consumption and harm among ‘high-risk’ drinkers.123 However, changes in risk status at follow-up for non-drinkers or low-risk drinkers have not been assessed in controlled trials of brief intervention.
Recruitment of both ‘high-risk’ and ‘low-risk’ drinkers has the additional benefit of addressing a major concern among both young people and parents, namely that participation in a trial of this nature may identify the young person as drinking at a level that warrants concern and intervention. Young people interviewed as part of our patient and public involvement work in work package 2 indicated that they would prefer to take part in a trial if there was no implication that they had an ‘alcohol problem’ and were assured that information about their drinking would not be disclosed to parents or health-care staff. Recruitment of both high- and low-risk-drinking young people was more acceptable to both young people and their parents, as was emphasising participant confidentiality.
Thus, we conducted two linked RCTs that included both high- and low-risk drinkers and abstainers, informing them that the study sought to prevent alcohol-related harm in young people. In addition, embedded within the proposed study was an internal feasibility study conducted prior to proceeding to the main trial.
The trials protocol has been published in Deluca et al.124 and parts of this section have been reproduced from Deluca et al.124 This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
Objectives
Primary objective
The primary objective was to conduct two linked RCTs to evaluate the clinical effectiveness and cost-effectiveness of brief intervention strategies compared with screening alone. One trial focused on high-risk adolescent drinkers attending EDs and the other focused on those identified as low risk or abstinent from alcohol. In both trials our primary outcome measure was quantity of alcohol consumed at 12 months after randomisation.
Secondary objectives
The secondary objectives of each study were:
to identify key predictors of recruitment to the trials
to explore the process of intervention through key psychological constructs that may lead to further refinement of the proposed interventions
to identify prognostic factors related to better outcomes
to explore interactions between participant factors, setting factors, treatment allocation and outcomes.
Our primary (null) hypothesis was similar for both trials: PFBA and personalised feedback plus eBIs is no more effective than screening alone in reducing alcohol consumed at 12 months after randomisation as measured with the AUDIT-C. Our secondary (null) hypothesis relating to health economics states that PFBA and eBIs are no more cost-effective than screening alone.
Methods
The linked trials were granted ethics approval by the National Research Ethics Service London – Fulham (reference 14/LO/0721). The trials comply with the Declaration of Helsinki125 and Good Clinical Practice126 and have been registered as ISRCTN45300218.
Study setting and participants
The trials were carried out in 10 EDs across three regions of England: North East, Yorkshire and The Humber, and London. Data collection was carried out from 10 a.m. to 10 p.m., 7 days per week, over an 8-month period (October 2014–May 2015). During these screening hours, consecutive ED attenders who were between their 14th and 18th birthdays and who met the inclusion criteria but none of the exclusion criteria were approached by a researcher and invited to participate in the study once cleared by ED staff to do so.
Eligibility criteria
Inclusion and exclusion criteria were chosen to maintain a balance between ensuring the sample was representative of the ED population while also able to engage with both the relevant interventions and follow-up.
Inclusion criteria
The inclusion criteria were being aged between 14 and 17 years inclusive; being alert and orientated; being able to speak English sufficiently well to complete the research assessment; living within 20 miles of the ED; being able and willing to provide informed consent to screening, intervention and follow-up; if under aged < 16 years, being ‘Gillick competent’ or having a parent or guardian who was able and willing to provide informed consent; and owning a smartphone or having access to the internet at home.
Exclusion criteria
The exclusion criteria were having a severe injury; suffering from a serious mental health problem; being grossly intoxicated; specialist services being involved because of social or psychological needs; receiving treatment for an AUD or substance use disorder within the past 6 months; or currently participating in other alcohol-related research.
The inclusion and exclusion criteria were discussed with hospital nurses/doctors before a potential participant was approached and after clinical staff assessed the participant. We relied on their knowledge and professional judgement.
Those who were grossly intoxicated on attendance were not the population of interest. The study addressed those who consumed alcohol at levels at risk to their health, rather than alcohol-related attendances. Although it is possible that these two groups overlapped, we were mindful of the issue of informed consent for those who presented as grossly intoxicated; however, if their intoxicated state reduced to an acceptable level while they were in the ED, they were approached.
Those who met the inclusion criteria and none of the exclusion criteria and scored ≥ 3 on the screening questionnaire, AUDIT-C, were eligible for the high-risk study; those who scored < 3 on AUDIT-C were eligible for the low-risk study.
Consent procedure
The study was introduced to patients, and to their parent or guardian if they were aged < 16 years, as a study about alcohol, lifestyle and health, with the focus on preventing alcohol-related harm in all young people attending ED irrespective of their alcohol consumption. Patients aged < 16 years attending the ED without their parent or guardian were also approached to take part if ED staff confirmed that they were ‘Gillick competent’. We extended Gillick competency to consenting for participation in research on the grounds of minimal/no risk in taking part in this study, the potential direct benefit that they would gain from the advice received and the potential benefit to the wider society in the roll-out of the findings.64
The study was first introduced by ED staff and then explained in more detail by research staff, both verbally and using the patient information sheet. If the patient was under the age of 16 years and accompanied by a parent or guardian, the parent or guardian would also receive the patient information sheet. Patients, and parents or guardians if applicable, had up to 4 hours to ask any questions about the study and to decide whether or not to take part. To obtain the most valid self-report data, patients were told as part of the informed consent procedure that their answers, including those on alcohol consumption, would not be disclosed to their parent or guardian or the ED staff without their consent ().
Decision tree for consent. PIS, patient information sheet.
If patients agreed to participate, their informed consent was recorded using an electronic device (iPad), overseen by a research assistant who also introduced and delivered the allocated intervention to each patient in a private area of the ED. Consent to participate included permission to give the patient’s data and contact details to the research staff, to provide the research team with access to the patient’s ED records, and to participate in follow-up at 6 and 12 months after recruitment.
Screening and baseline assessment
After consent was given by the patient or their parent or guardian, as appropriate, the participant completed a screening and baseline assessment ( shows the sequence of tools administration). All participants scoring ≥ 3 on the AUDIT-C (high-risk drinkers) were randomised between three groups [two intervention groups (PFBA and eBIs) and the control group receiving screening alone]. Of those scoring < 3 on the AUDIT-C (low-risk drinkers or abstainers), one in three was randomly selected to continue with the study and then randomised between three analogous groups. Participants who scored < 3 but were not selected for the trial were thanked for their participation, given a £5 voucher and returned to the care of the ED staff.
Baseline sequence. APP, SIPS Jr City app; BA, brief advice; EQ-5D-5L, EuroQol-5 Dimensions, five-level version; ESP19, European School Survey Project on Alcohol and Other Drugs Q19; ESP19-C, European School Survey Project on Alcohol and Other Drugs Q19c; (more...)
The screening and baseline assessment includes demographic information and contact details; health and lifestyle questions; the AUDIT-C;54 questions 19, 21 and 22 from ESPAD;68 the Strengths and Difficulties Questionnaire,127 the EuroQol-5 Dimensions, five-level version (EQ-5D-5L);128 and a short service use questionnaire.129 This took approximately 10 minutes to complete.
To simplify and enhance data collection, we used a bespoke electronic interface (developed in work package 1), which automated question routing, showing participants only relevant questions. To maximise completion rates, we used an attractive graphical interface. Participants were able to skip questions or withdraw consent at any stage. All of the instruments have been designed and validated for those aged 14–17 years. The screening and baseline assessment was conducted by trained researchers with experience of working with adolescents, and all researchers had completed enhanced Disclosure and Barring Service checks prior to working in the ED. All information that participants gave was treated in confidence.
Participants were remotely randomised with equal probability, stratified by centre, between a screening only control group and one of the two interventions: a single session of face-to-face PFBA or personalised feedback plus a smartphone- or web-based brief intervention (eBI). All participants were eligible to receive treatment as usual in addition to any trial intervention.
Randomisation
Randomisation to trial participant or non-participant was conducted using a simple block randomisation, with a one in three probability of selection. For those selected as participants, randomisation to study group was conducted using strings of randomly selected block sizes, three or six, stratified by ED and gender. Each iPad within a centre had a separate pre-programmed allocation sequence derived by an independent party and made secure using encryption. Researchers engaged in the baseline assessment were not aware of allocated group until after outcomes had been completed. Participants were not blind to allocated group.
Interventions
Screening only group: treatment as usual
After completing the baseline assessment, participants in the screening arm were thanked for their participation, reminded that a member of the research team would contact them in 6 and 12 months to conduct a follow-up interview and returned to the care of the ED staff for usual care.
Personalised feedback and brief advice
The PFBA intervention is structured brief advice that takes approximately 5 minutes to deliver () in one session. It is based on an advice leaflet adapted for the target age group in this study from the SIPS brief advice about alcohol risk intervention.130,131 It is based on the FRAMES model:132
Feedback: Give feedback on the risks and negative consequences of alcohol use. Seek the patient’s reaction and listen.
Responsibility: Emphasise that the individual is responsible for making his or her own decision about his/her alcohol use.
Advice: Give straightforward advice on modifying alcohol use.
Menu of options: Give menus of options to choose from, fostering the patient’s involvement in decision-making.
Empathy: Be empathic, respectful and non-judgemental.
Self-efficacy: Express optimism that the individual can modify his or her alcohol use if they choose. Self-efficacy is one’s ability to produce a desired result or effect.
It is conveyed verbally to the participant by trained research assistants or nurses and tailored to their risk status (high or low). It was delivered in a quiet room in the ED.
The advice covers recommended levels of alcohol consumption for young people; gives feedback on the screening results and their meaning; provides normative comparison information on prevalence rates of high- and low-risk drinking in young people; summarises the risks of drinking and highlights the benefits of stopping or reducing alcohol consumption; outlines strategies that they might employ to help stop or reduce alcohol consumption; highlights goals they might wish to consider; and indicates where to obtain further help if they are unsuccessful or need more support.
Each participant received a copy of the leaflet, which included additional information about alcohol intoxication, alcohol poisoning, and alcohol and the law.
Personalised feedback plus a smartphone- or web-based brief intervention
The eBI smartphone intervention SIPS City is an offline-capable mobile web application that works on a variety of platforms but is optimised for recent iPhone (Apple Inc., Cupertino, CA, USA) and Android (Google Inc., Mountain View, CA, USA) phones (). It was developed for this research by the software developer Codeface Ltd (Hove, UK) in collaboration with the research team. It followed the recommendations from patient and public involvement, and it was developed using the concept of gamification so that users can navigate, explore, learn facts and figures about alcohol, receive personalised feedback and set goals in an engaging format. The content was adapted to provide the most pertinent information and advice for high- or low-risk drinkers and was similar in content to what was provided in the PFBA intervention arm described above in Personalised feedback and brief advice. Games components of the web application supported high-risk drinkers to reduce or stop their alcohol consumption and low-risk users to maintain abstinence or low-risk drinking.
SIPS Jr Street app with full view of East and West Streets.
The SIPS City app was formatted into a virtual reality of two streets, west and east, in which there were multiple buildings such as a general practice, a pub and a youth centre. To gain access to some buildings, participants had to collect a certain number of coins, which could be obtained from talking to characters on the street or by answering questions correctly. When interacting with people on the street, participants were directed to certain buildings depending on the problem that person was encountering, for example the doctor for alcohol poisoning. It was also possible to drive in the car of ‘Rod McDuff’s School of Motoring’, and facts regarding the risks of alcohol and drinking were portrayed while inside the car.
The first building was the participants’ home, where they could fill out a drinking diary and receive feedback from this. It was also possible to view information on units and a letter from the local A&E about the participant’s drinking. Interaction with a health worker at the general practice allowed a user to follow-up the A&E letter and set personal alcohol goals. There was a sexual health clinic building that provided information on the increase of sexual health risks with increased alcohol intake. After two coins had been obtained, access to East Street was granted. The pharmacy was here, which provided information on how to reduce the effects of a hangover. The school provided information on the harmful effects of alcohol in relation to education, which provided relatable information to those in the age group in this study.
Whenever possible, the app was installed, with the help of a research assistant/nurse, on the participant’s smartphone while they were attending A&E and the participant was encouraged to use it. In the instances when they did not have access to their phone (e.g. flat battery, left at home, no data plan), patients were introduced to a demonstration version of the app on a study device (iPad) and allowed to play with it while in A&E. An e-mail and short message service (SMS) were also sent to the patient within 24 hours with instructions on how to download and install the app on their smartphone once they were at home.
Two further remainders (e-mail and SMS) were sent in the following 2 weeks to those who had failed to install the app on their smartphone.
For participants without access to a smartphone but with access to the internet through other computerised devices, access to a web-based version of the application was provided along with appropriate instructions for its use.
After receiving their allocated intervention (including the screening only group), all participants were thanked for their participation, reminded that a member of the research team would contact them in 6 and 12 months to conduct a follow-up interview, given a £5 voucher to thank them for their time and returned to the care of the ED staff.
Intervention fidelity
Research assistants were responsible for recruiting participants and delivering the interventions. The research assistants were trained during a 2-hour training session, which covered the rationale and procedures of the trial, the importance of reducing alcohol consumption and the correct delivery of the interventions. Filmed examples of delivery were presented and discussed, and role-play sessions were undertaken.
During the trial, we assessed fidelity of the delivery of the PFBA interventions by audio-recording a random sample of 20% of intervention sessions for each researcher. Each recording was assessed by a senior clinician member of the team on whether or not key aspects of the intervention were delivered as intended against a predefined checklist. When necessary, feedback was provided to researchers to improve fidelity. These recordings were prespecified in the protocol analysis plan.
Follow-up assessments
All participants were followed up with a brief set of questions at 6 months after randomisation (), and then at 12 months for a full assessment (). Follow-up interviews were conducted over the telephone, face to face or electronically via self-completion web survey, as preferred by the participant. The telephone and face-to-face follow-ups were conducted by research assistants trained in the administration of the assessment tools and blinded to the group allocation of the participants. Letters of thanks were sent to participants after each follow-up stage. On completion of each follow-up interview, participants were sent a gift token for £5 by post in recognition of their participation. On completion of the 12-month follow-up, participants were additionally entered in to a prize draw to win an iPad Air (Apple Inc., Cupertino, CA, USA), iPad mini (Apple Inc., Cupertino, CA, USA) or iPod (Apple Inc., Cupertino, CA, USA).
List of tools and order of presentation at 6-month follow-up. Q, question; SU, service utilisation.
List of tools and order of presentation at 12-month follow-up. B, items A (lifetime) and B (last 12 months) in question 19; CSRI, Client Service Receipt Inventory; EQ-5D-5L, EuroQol-5 Dimensions, five-level version; ESP19, European School Survey Project (more...)
Outcome measures
Primary outcome measure
The primary outcome was the total amount of alcohol consumed in standard UK units (1 unit = 8 g of ethanol) over the previous 3 months, measured at the 12-month follow-up using the AUDIT-C, which was either self-completed by web survey or administered by researchers blinded to treatment allocation.
In the published protocol124 we intended to use the Timeline Followback interview (28-day version). However, this was subsequently changed to the AUDIT-C to facilitate completion rate at follow-up. The AUDIT-C is a much shorter tool (three items) and can be self-administered.
Calculation of weekly units from the AUDIT-C was conducted as follows. The extended AUDIT-C asked two questions regarding frequency and quantity of alcohol consumed. Question 1 asks about frequency, and these values are converted to weekly frequency using the following algorithm: never (0), monthly (0.25), two to four times per month (0.75), two or three times per week (2.5), four or five times per week (4.5) and six or more times per week (6.6). Question 2 asks about quantity on each drinking occasion and is converted to standard units using the following algorithm: none (0), one or two (1.5), three or four (3.5), five or six (5.5), seven to nine (8), ten to twelve (11), 13 to 15 (14) and 15 or more (15). Weekly units are calculated by multiplying converted values for frequency and quantity.
This value allocates participants to 1 of 35 categories of consumption. An ordinal is one in which values are ranked, A is greater than B, but the relative magnitude of A relative to B is unknown. The weekly consumption calculation not only ranks participants but also allows a derivation of the relative difference between participant drinking levels. The large number of data points and the ability to assess relative magnitude means that the weekly consumption can be taken as a continuous measurement variable. This implicit assumption was tested as part of the overall analysis.
Moreover, any ordinal scale with > 11 data points can be treated as continuous.133
Secondary outcome measures
Participants were also asked questions about the consequences of alcohol consumption using questions 19, 21 and 22 from ESPAD.68 Hazardous alcohol use was assessed using the extended AUDIT-C questionnaire54 at baseline and after 6 and 12 months. General health and functioning was measured using the Strengths and Difficulties Questionnaire127 at baseline and 12 months.
Economic outcome measures
The primary outcome measure for the economic evaluation in the trial was a preference-based measure calculated from the EQ-5D-5L. The EQ-5D-5L quality-of-life instrument is preferred by NICE for the economic evaluation of NHS interventions. The tool focuses on five dimensions of health: mobility, self-care, usual activities, pain/discomfort and anxiety/depression.128 The original EuroQol-5 Dimensions had three response categories (EuroQol-5 Dimensions, three-level version) for each dimension. A newly released validated version with five response categories (EQ-5D-5L) for each dimension, providing enhanced discriminatory power, was used in the study.134 EQ-5D-5L requires no more than a few minutes to complete and thus imposes minimal burden on participants.
The EQ-5D-5L scores were converted to health utilities (1 = perfect health, 0 = equivalent to dead) using a tariff provided by the EuroQol group derived from UK social preference surveys. Resulting utilities were combined with survival data (unlikely to be affected by the service) and expressed in quality-adjusted life-years (QALYs). The estimated incremental cost per QALY from the service can be compared with the willingness-to-pay (WTP) threshold of £20,000–30,000 per extra QALY currently used by NICE to determine whether or not an intervention is ‘cost-effective’ and hence recommended for use in the NHS.135
Process outcome measures
Expectancy was measured using the ESPAD question 2168 at baseline and 12 months after randomisation. Adherence to the eBI was assessed by monitoring remotely either when the smartphone device was connected to the internet or when the web application was accessed.
Analysis
Sample size calculation
For both studies, the sample size addresses the effect of interventions on the primary outcome measure (alcohol consumption at 12 months after randomisation). We aimed to detect a meaningful effect size difference of ≥ 0.3, based on literature relating to adults and similar to differences observed for adolescents; this would equate with a difference in weekly consumption between intervention and control of 0.1 units in the low-risk trial and 2 units in the high-risk trial.136 To detect this with a significance level of 5% and statistical power of 80% when using a two-sided continuity-corrected test requires 175 in each of the three groups, yielding a target of 525 analysable participants in each of the two trials.
As there was little prior research in this specific area, our sample size calculation was based on similar UK RCTs137,145 addressing alcohol use in primary care populations. These RCTs reported effect size differences between brief interventions and minimal intervention of 0.36 and 0.27.138,145 Similar effects have been reported from studies in the USA, and an effect size of 0.3 is considered clinically important for alcohol brief intervention studies.139 As there is no literature on what might be a clinically important difference for the low-risk trial, we hypothesised that a small effect size difference, of a similar magnitude to or greater than that for the high-risk trial, could be interpreted as an important effect.
Predicting that follow-up at 12 months would be 70%, we needed to randomise 750 high-risk drinkers and 750 low-risk drinkers. Based on the estimated prevalence of 24.2% for high-risk drinking (namely AUDIT-C ≥ 3) from our earlier survey, and a consent rate of 60% (see Work package 1: screening prevalence study of alcohol consumption and alcohol use disorders in adolescents aged 10–17 years attending emergency departments), we estimated a number needed to approach of 5165 potential participants over the recruitment period. Of these participants, our survey predicts that 2350 will be low-risk drinkers consenting to the study.
Statistical analysis
The outcomes for both trials were analysed in a similar manner. Analysis was conducted using an intention-to-treat principle, whereby participants were analysed as members of their allocated group irrespective of treatment received. All analysis was conducted using SAS® software 9.4 (SAS Institute Inc., Cary, NC, USA) and conducted blind to allocated group.
The analytical approach employed a mixed-effects model, with a fixed effect for allocated group and a random effect for ED. The covariates age, gender and baseline alcohol consumption were included as baseline covariates, as these are known to influence outcome. The distribution of the primary outcome was assessed prior to analysis and, if necessary, appropriate transformations were undertaken. A sensitivity analysis was undertaken using a non-parametric approach and assessed change in consumption. Wilcoxon rank-sum indices were generated and analysed using a similar mixed-effects approach. The influence of missing data was assessed using a series of multiple imputation models, and these were synthesised to assess the sensitivity of the observed results to missing data. Secondary outcomes were assessed using a similar mixed-model approach and adjusting for respective baseline values. To explore the value of the findings, we performed a post hoc analysis and calculated the Bayes’ factor of the primary outcome, comparing eBI and PFBA with control.
Two exploratory analyses were undertaken. The first was to investigate the relationship between potential prognostic pre-randomisation factors and alcohol consumption at 12 months. The factors included were alcohol expectancy, alcohol-related problems and demographics, and any interaction between these factors and intervention group. An initial analysis explored the relationship between alcohol consumption and each factor individually, with factors or interaction terms with a p-value of < 0.2 combined to create a full model. Backward elimination was used, retaining factors with a p-value of < 0.2 in the final model. If an interaction term had a p-value of < 0.2 but the p-value for the main effect was > 0.2, both terms were retained in the model. A second exploratory analysis explored the relationship between eBI usage and alcohol consumption at month 12 for those allocated to eBI using a linear regression approach, controlling for baseline alcohol consumption and gender.
We estimated in a sample size calculation that we would assess 70% of those allocated at baseline and we achieved this end. In our analysis, we explored the nature of missing data at 12 months post randomisation using multiple imputation and assessed the impact of these imputation models on the observed outcome using sensitivity analyses. The derived models, which assume potential bias in loss to follow-up, had no effect on the outcomes observed, so these data without imputation were employed for the primary analysis.
Cost-effectiveness analysis
Individual-level data were used to estimate mean differential costs between interventions. As data were not normally distributed, 95% CIs were calculated using a non-parametric bootstrapped method.140 This was also done for effects, the EQ-5D-5L score and QALYs at 6- and 12-month follow-up. Difference in QALYs was estimated using the area under the curve method.
Sensitivity analysis
Cost-effectiveness results [mean total costs and effects, hence the incremental cost-effectiveness ratio (ICER)] are subject to uncertainty or sampling error. A joint uncertainty in costs and effects was investigated via a stochastic sensitivity analysis. Using a large number of non-parametric bootstrapped replications (n = 10,000) of costs and effects (jointly), this uncertainty was quantified through a 95% CI of the ICER.141,142 Based on the above bootstrapped replications, a two-dimensional cost-effectiveness plane was created, plotting the joint uncertainty in costs and effects between two groups. Furthermore, a cost-effectiveness acceptability curve was undertaken to show the probability that an intervention was cost-effective at a range of WTP values (£20,000 and £30,000 per QALY gain in the UK).
Valuation of resource use
All NHS resource use was reported in appropriate physical units and valued using relevant unit costs.143 All figures were based on 2014 costs. As costs were incurred only over 12 months, discounting was not necessary. The cost of screening and delivering the two interventions were ascertained by prospectively monitoring resource inputs to each arm of the trial at 6- and 12-month follow-up, including training, and valued using standard methods.141
Training costs
All resources involved in training were costed, including:
trainer time in preparing for training sessions, in travelling to training sessions and in delivering the training sessions (and anything else); this was costed by using the number of trainers and their salary or university/NHS grade/band
trainee time in travelling to training session and in attending training session; costed accordingly as in (a)
expenses incurred by trainers or trainees (e.g. train/bus fares, taxis, parking); for car travel, the travel time reported above was be converted into motoring costs
cost of any materials used (either described or in pounds sterling spent).
NHS and non-NHS costs
Effects on NHS and non-NHS costs was based on information gathered on patient contact with primary care, secondary care, specialist health services, social service and criminal justice, and other resources. These were collected prospectively using the appropriately modified version of the Client Service Receipt Inventory (CSRI). The CSRI captures any resource implication for the last 6 months. Service utilisation in CSRI was valued using local costs and, when possible, supplemented by national resources and information from previous alcohol studies.130,144,145 Appropriate unit costs were used to derive a cost of any NHS resource [e.g. hospitalisation, general practitioner (GP) visit] or non-NHS resource (e.g. cost of criminal offence) use.143
Missing data
Multiple imputation was used to handle missing values related to individual EQ-5D-5L input variables, with EQ-5D-5L utility values calculated from the imputed variables. Ten imputations were calculated. For missing costs, it was first determined whether costs were truly missing or truly zero, and for the truly missing costs the average costs for each intervention were imputed.
Results
Low-risk drinkers trial
Participant flow
Participant progress throughout the trial is presented in the CONSORT (Consolidated Standards of Reporting Trials) flow diagram (). Of the 7854 attendees, 5016 were approached (63.9%). All reasons for exclusion are reported in . Approximately 1% (n = 83) were intoxicated at the time of presentation and not approached for participation in the study. Twenty-five patients were excluded, because they did not own a smartphone or have internet access to receive the intervention. Of the patients approached, a total of 3326 met all of the inclusion criteria and consented to be screened (66%). Of these patients, 2571 (77.3%) scored < 3 on the AUDIT-C and were eligible for the low-risk study. One-third of these potential participants (n = 884) were selected at random and randomly allocated into one of the three groups.
The CONSORT flow diagram of the linked trials.
Sample characteristics
Demographic and outcome variables were similar across all three groups at baseline (). Overall, the mean age of those participating in the study was 15.1 years, 51% were female and 62.5% of the sample classified their ethnicity as white. Participants’ mean age at the time of first drink was 13.8 years and mean weekly alcohol consumption was low at 0.14 units of alcohol.
Demographic and baseline outcomes by allocated group in the low-risk trial
Main outcomes in the low-risk trial
The primary outcome, weekly alcohol units consumed at month 12, was derived from the AUDIT-C. As consumption was positively skewed, we explored transformations using the Box–Cox transformation approach and identified a cube-root transformation as appropriate to fit the data.
Outcomes at 6 and 12 months were back-transformed and are presented in . Mean differences and associated 95% CIs are presented in . No differences were observed between the groups for the primary outcome at 6 or 12 months. A sensitivity analysis employing the Wilcoxon rank-sum of the change score demonstrated similar results, as did an assessment of multiple imputation of missing values. A similar pattern was observed for secondary outcomes.
Adjusted least mean squares and 95% CI for outcomes at 6 and 12 months by allocated group in the low-risk trial
Adjusted least mean squares difference vs. control and 95% CI for outcomes at 6 and 12 months by allocated group in the low-risk trial
A post hoc analysis was also performed for the Bayes’ factor comparing eBI and PFBA with control: 0.05 [standard error (SE) 0.13] and 0.05 (SE 0.18), respectively. These results indicate that the null result is a true null finding of no effect of either intervention.
An analysis exploring potential interactions between quantity of alcohol consumption at baseline and allocated group found no significant interactions for the low-risk study (F = 1.78; p = 0.17).
Our exploratory analysis of prognostic factors that may impact on alcohol consumption at month 12 identified a number of significant positive predictors: higher baseline consumption, lower age of first drink, older age, being female, greater positive alcohol expectancy and greater alcohol-related problems (see Table 15, Appendix 1).
For those allocated to eBI, 103 (35.0%) participants actually engaged with the intervention after leaving the ED. No relationship was identified between engagement with the intervention and alcohol consumption at month 12.
Cost-effectiveness analysis in the low-risk trial
Cost-effectiveness analysis compared both the eBI and PFBA intervention groups with the control group for all societal costs () and for NHS/Personal and Social Services (PSS) costs only (). The analyses show that, for both the societal cost perspective and the narrower NHS/PSS perspective, the eBI is dominated by the control, whereas the PFBA intervention generates ICERs of £130,822 (societal) and £120,693 (NHS/PSS) per QALY gained, respectively.
Results of the cost-effectiveness analysis, societal perspective, in the low-risk trial
Results of the cost-effectiveness analysis, NHS/PSS perspective, in the low-risk trial
From the societal cost perspective, probabilistic sensitivity analysis (PSA) indicated that approximately 9% of the simulations for eBI compared with control were cost-effective at both the £20,000 and the £30,000 WTP thresholds, whereas approximately 26% and 30% of the simulations for PFBA compared with control were cost-effective at the £20,000 and £30,000 WTP thresholds, respectively ().
Results of the PSA, societal perspective, in the low-risk trial
From the NHS/PSS cost perspective, PSA again indicated that approximately 9% of the simulations for eBI compared with control were cost-effective at both the £20,000 and the £30,000 WTP thresholds, whereas approximately 31% and 33% of the simulations for PFBA compared with control were cost-effective at the £20,000 and £30,000 WTP thresholds, respectively ().
Results of the PSA, NHS/PSS perspective, in the low-risk trial
The deterministic analyses and PSA show that it is highly unlikely that either intervention is cost-effective at either the £20,000 or the £30,000 WTP threshold when compared with the control intervention in low-risk patients.
High-risk drinkers trial
Participant flow: high-risk trial
Participant progress throughout the trial is presented in the flow diagram (see ). Of the 7854 attendees, 5016 (63.9%) were approached. A total of 3326 participants consented to be screened (66.0%) and, of these, 756 (22.7%) participants scored ≥ 3 on the AUDIT-C and were eligible for the high-risk study.
Sample characteristics: high-risk trial
Demographic and outcome variables were similar across all three groups at baseline (). Overall, the mean age of those participating into the high-risk study was 16.1 years, 50.2% were female and 84.9% of the sample classified their ethnicity as white. Mean age at first drink was 13.5 years and mean weekly alcohol consumption was higher than in the low-risk trial, at 4.7 units of alcohol.
Demographic and baseline outcomes by allocated group in the high-risk trial
Main outcomes in the high-risk trial
The primary outcome, weekly alcohol units consumed at month 12, was derived from the AUDIT-C. As consumption was positively skewed, we explored transformations using the Box–Cox transformation approach and identified a cube-root transformation as appropriate to fit the data.
Outcomes at 6 and 12 months were back-transformed and are presented in . Mean differences and associated 95% CIs are presented in . No differences were observed between the groups for the primary outcome at 6 or 12 months. A sensitivity analysis employing the Wilcoxon rank-sum of the change score demonstrated similar results, as did an assessment of multiple imputation of missing values. A similar pattern was observed for secondary outcomes.
Adjusted least mean squares and 95% CI for outcomes at 6 and 12 months by allocated group in the high-risk trial
Adjusted differences from control and 95% CIs for outcomes by allocated group
We computed the Bayes’ factor comparing eBI and PFBA with control: 0.08 (SE 0.16) and 0.08 (SE 0.36), respectively. These results indicate that the null result is a true null finding of no effect of either intervention.
An analysis exploring potential interactions between quantity of alcohol consumption at baseline and allocated group found no significant interactions for the high-risk study (F = 0.27; p = 0.76).
Our exploratory analysis of prognostic factors that may impact on alcohol consumption at month 12 identified a number of significant positive predictors: higher baseline consumption, lower age of first drink, older age, being female, greater positive alcohol expectancy and greater alcohol-related problems (see Table 15, Appendix 1).
For those allocated to eBI, 84 (33.3%) actually engaged with the intervention after leaving the ED. No relationship was identified between engagement with the intervention and alcohol consumption at month 12.
Cost-effectiveness analysis
Cost-effectiveness analysis compared both the eBI and PFBA intervention groups with the control group for all societal costs () and for NHS/PSS costs only (). The analyses show that, for both the societal cost perspective and the narrower NHS/PSS perspective, the eBI is dominated by the control, whereas the PFBA intervention generates ICERs of £7580 (societal) and £6213 (NHS/PSS) per QALY gained, respectively.
Results of bootstrapped cost-effectiveness analysis from societal perspective
Results of bootstrapped cost-effectiveness analysis from NHS/PSS perspective
From the societal cost perspective, PSA indicated that approximately 28% of the simulations for eBI compared with control were cost-effective at the £20,000 WTP threshold and 27% at the £30,000 WTP threshold; whereas approximately 54% and 55% of the simulations for PFBA compared with control were cost-effective at the £20,000 and £30,000 WTP thresholds, respectively (). Although PFBA has a chance of being cost-effective when compared with control, the distribution of the bootstrapped ICERs show that there is a wide distribution ().
Results of the PSA, societal perspective, in the high-risk trial
Cost-effectiveness plane: PFBA vs. control, societal perspective, in the high-risk trial.
From the NHS/PSS cost perspective, PSA again indicated that approximately 30% of the simulations for eBI compared with control were cost-effective at the £20,000 WTP threshold and 29% at the £30,000 threshold; whereas approximately 54% and 56% of the simulations for PFBA compared with control were cost-effective at the £20,000 and £30,000 WTP thresholds, respectively (). Again, although PFBA has a chance of being cost-effective when compared with control, the distribution of the bootstrapped ICERs show that there is a wide distribution ().
Results of the PSA, NHS/PSS perspective, in the high-risk trial
Cost-effectiveness plane: PFBA vs. control, NHS/PSS perspective, in the high-risk trial.
The deterministic analyses and PSA show that it is highly unlikely that the eBI is cost-effective at either the £20,000 or the £30,000 WTP threshold when compared with the control intervention in high-risk patients, although there is an approximately 55% chance that the PFBA intervention is cost-effective compared with the control.
