The study was a parallel-arm, single-blind, individually randomised controlled trial exploring the clinical effectiveness and cost-effectiveness of brief intervention for excessive alcohol consumption among adults aged ≥ 19 years who attend sexual health clinics. The trial was an integrated clinical and economic evaluation and compared the effects of brief intervention with the effects of control treatment on excessive alcohol consumption, sexual behaviour, health-related quality of life and costs in the 6 months after randomisation.
Ethical approval was obtained from West London Research Ethics Committee 3 (10/H0706/29) and the study protocol was registered with Controlled Clinical Trials (ISRCTN 99963322) prior to the start of data collection.
Changes to original protocol
Prior to the start of the study, but following trial registration, one of the clinics where we were due to recruit participants withdrew from the study as it was unable to provide input from an alcohol health worker (AHW). Recruitment was therefore restricted to three sites. Another clinic started to provide additional support for young people aged ≤ 18 years aimed at promoting sexual health. This intervention included discussion of alcohol use. We therefore changed our eligibility criteria to include only those aged ≥ 19 years.
We made one additional change to the study after commencement. The original sample size for the study was set at 320, which was judged to be large enough to detect clinically important differences in levels of alcohol consumption among those offered active and control treatment. In the first few months of the trial it became clear that the rate of recruitment was higher than we had originally anticipated. With the support of the funder and the independent Trial Steering Committee and following approval of the Research Ethics Committee we increased the sample size to 760 in order to have sufficient power to examine clinically important differences in our main secondary outcome: the proportion of participants who reported any unprotected sex at follow-up.
The public and patients were involved at several stages of the study (see Appendix 1 for further details).
Study setting and sample
Study participants were recruited from three sexual health clinics in central and west London. These clinics serve a diverse population of over 500,000 people with high levels of alcohol misuse and poor sexual health.22,23
To participate in the study people had to be aged ≥ 19 years, be drinking excessively according to the Modified-Single Alcohol Screening Question (M-SASQ)24 and be willing to provide written informed consent to take part in the study. We excluded any person who was unable to communicate in English sufficiently well to complete baseline questionnaires, anyone who did not have an address or contact telephone number and anyone who believed they may not have been contactable in 6 months’ time.
Study interventions
The SHEAR had two treatment conditions: brief intervention for excessive use of alcohol and control treatment.
Brief intervention
Brief intervention was based on that used in a previous trial conducted in an emergency department16 and was found previously to be acceptable to clinicians in a sexual health clinic.9 The intervention is designed to be used by busy front-line clinicians such that it can be delivered within 2 or 3 minutes. The intervention is designed to deliver the treating clinician after they have dealt with the person’s presenting complaint. The intervention consists of four components:
confirming the current level of alcohol use and brief feedback that alcohol use at that level has the potential to harm health
making a link between alcohol and clinic attendance
written information on alcohol and health in the form of a leaflet recommended by the Department of Health: ‘How much is too much?’
25the offer of an appointment with an AHW.
This form of brief advice is shorter than some other forms of intervention, which can take 5–10 minutes to deliver. It is focused on these four simple tasks. Verbatim text that can be used to deliver each of these is available at: www.alcohollearningcentre.org.uk/_library/PAT_2011_Paddington_Alcohol_Test.pdf.
On days when participants were recruited from the clinics, an AHW was available to see those who were willing to receive further help. The appointment with the AHW lasted up to 30 minutes and used the FRAMES approach.17,18 In the case of any participant who was drinking at a harmful or dependent level, the AHW had the option of arranging a follow-up appointment or referring the participant to local alcohol services for individual alcohol counselling and other services. In the event that the participant was unable to attend an appointment that day, he or she was offered an appointment at a later date or telephone-based support and advice.
Control treatment
Those randomised to control treatment were offered a copy of the leaflet ‘Five Choices to Help You Stay Healthy’.26 This provides general information on health and prevention of ill health including information on alcohol use, diet, exercise and cigarette smoking and details of how to obtain further information about health and lifestyle.
Training and support for the delivery of brief intervention
A short training session was delivered at each of the hospital sites before participants were recruited. This session was incorporated into existing staff meetings. In the session, we provided background to the study, an overview of study logistics and details of each of the four components of the brief advice that clinicians were asked to provide those in the active arm of the trial. Clinicians were asked to use recommended text for delivering each of the four components of the intervention and encouraged to use web-based information at www.alcohollearningcentre.org.uk.27
In addition to this, the lead researcher (RS) spoke to front-line clinicians on the days when recruitment was taking place. She provided support and advice to clinicians, gave feedback on their performance and checked that brief advice was being delivered in accordance with the trial protocol.
All AHWs who took part in the study were experienced practitioners who had undertaken specific training in counselling people who misuse alcohol. Three were employed by the NHS and one was employed by a statutory organisation (Turning Point). All AHWs received regular clinical supervision. AHWs were encouraged to discuss work with trial participants along with other patients they saw during these sessions.
Treatment fidelity
In order to assess treatment fidelity, clinicians delivering brief advice and AHWs were asked to complete a treatment proforma for each person they saw. These proforma can be found in Appendices 2 and 3 of this report. The proforma completed by clinicians was based on one we used in a previous trial.16 Front-line clinicians were asked to indicate whether or not they had delivered each of the four components of brief advice and AHWs were asked to complete a longer proforma which recorded the number and length of session(s), interventions delivered during the session(s) and further information of referrals that were subsequently made. A member of the research team was on hand to check completion of these proforma and to support and advise clinicians on delivering brief advice if required. Proforma were examined at the end of the study to identify the proportion of those in the active arm of the trial who received brief advice and brief intervention.
Outcome measures
Primary and secondary outcomes
The primary outcome was mean weekly units of alcohol consumed during the previous 90 days. The main sexual health outcome of interest was having had any unprotected vaginal or anal sex in the past 3 months; this was referred to as our ‘main secondary outcome’. Both of these variables were measured at follow-up. The other secondary outcomes are detailed below in the report.
Baseline
Basic demographic and clinical data on age (years), gender, ethnicity and reason for presentation were extracted from clinic records at baseline and checked with the participants.
Alcohol consumption was assessed using the M-SASQ. The M-SASQ is a brief validated measure of excessive alcohol use that is acceptable to patients in general medical settings.24 It consists of a single question – for men: ‘How often do you drink more than 8 units of alcohol on one occasion?’ and for women: ‘How often do you drink more than 6 units of alcohol on one occasion?’ To help people answer this question they are shown a card which describes what 1 unit of alcohol is. Those drinking this amount once a month or more were considered eligible to participate in the trial. The question on alcohol was embedded in a series of four other questions asking about diet, exercise and smoking. In addition, eligible participants were asked about:
Sexual behaviour during the last 3 months using key variables that have been validated in other studies.
28 The variables comprised: number of sexual partners; number of people with whom they had unprotected sex with (vaginal or anal sex without a condom); any incidence of regretted sex; and how long they had known their last sexual partner before they had sex with them.
Health-related quality of life using the European Quality of Life-5 Dimensions (EQ-5D).
29 This is a generic preference-based measure for describing and valuing health-related quality of life assessed in five domains (mobility, self-care, usual activities, pain/discomfort, anxiety/depression). Utility scores are then derived from the EQ-5D, with higher scores indicating a better quality of life.
Six-month follow-up
Follow-up data were obtained by a telephone interview carried out by a researcher who was masked to the participant’s allocation status. The following outcomes were examined:
Alcohol consumption in the last 90 days using the Form 90. The Form 90 is a validated alcohol consumption assessment tool which provides a detailed day-by-day account of alcohol use in the 90 days prior to the interview.
30 Data from this questionnaire were used to calculate the primary outcome – mean weekly units of alcohol consumed during the previous 90 days. Secondary alcohol-related outcomes were mean units consumed per drinking day, percentage of days abstinent and whether the participant was drinking excessively according to the M-SASQ criteria.
Sexual behaviour in the last 90 days was assessed by a set of questions including total number of sexual partners; number of partners with whom the participant had had unprotected sex; any incidence of regretted sex; any incidence of unprotected sex after drinking alcohol and while feeling drunk; how long participants knew their last sexual partner before they first had sex with them; unplanned pregnancy; and any new diagnosis of a STI.
Service use data for the economic evaluation were collected using the Adult Service Use Schedule (AD-SUS), an interviewer-assessed instrument designed by one of the authors and based on previous economic evaluations in similar adult mental health and addiction populations.
31 The AD-SUS records the number and duration of contacts with a range of health and social service professionals, all hospital contacts and medications taken. Data on uptake of the brief intervention were collected from records to avoid participants revealing their treatment group to the research assessors. Data on indirect time, including preparation and supervision, were collected directly from the treating clinician.
Study procedures
Recruitment
At each clinic where recruitment took place we displayed posters in the waiting room providing information about the study. On days when recruitment took place, clinic staff gave all those attending the service a postcard with information about the study and asked people whether or not they would be willing to meet a researcher. If they agreed, the researcher explained the rationale for the study and gave the person a copy of the patient information leaflet. The researcher encouraged potential participants to spend as much time as they wanted asking questions about the study and considering whether or not they wish to take part. Average waiting times in these clinics between presenting to reception and seeing a clinic doctor are over 2 hours. This ensured that potential participants had sufficient time to hear about the study, consider whether or not they wanted to participate and complete the baseline assessment. Prior to completing the baseline assessment, participants signed and dated the informed consent form. For those willing to provide consent, eligibility to participate in the study was assessed and baseline clinical and demographic data were collected. Baseline assessments were completed using a computer-assisted self-completion interview.32,33
Contact details were then sought to enable the researchers to contact the participant at follow-up. Researchers were assisted in recruitment by clinical studies officers of the UK Mental Health Research Network.
The researcher provided all those who were ineligible with written information about health and lifestyle if they wanted this.
Randomisation
Participants were randomised via an independent remote automated telephone-based service operated by the Clinical Trials Unit at the University of Aberdeen, UK. Permuted blocks stratified by the clinic were used, with allocation ratio between arms of 1 : 1 and block sizes randomly assigned to four or six.
The researcher notified the treating clinician as to which arm of the trial the participant was in. All other members of the trial team, including researchers involved in the collection of follow-up data, were masked to treatment allocation.
Follow-up
Three months after randomisation, study participants received a telephone call, text or e-mail, thanking them for taking part in the study and reminding them that they would be contacted in 3 months’ time to complete the follow-up interview. They were also asked whether or not their contact details were likely to change during this period. If, at 6 months, our attempts to contact a participant were unsuccessful, and consent had been given, the researchers checked the participant’s contact details against those given during any subsequent visits to the clinic and contacted a nominated family member or friend. Follow-up interviews were carried out by telephone.
Masking of raters
Data were held securely and were password protected. Details of allocation status were held separately and were not accessible to the researchers involved in collecting follow-up data. Researchers involved in recruiting study participants played no part in follow-up interviews. Information on receipt of brief advice and brief interventions was gathered separately from proforma completed by clinicians and AHWs.
Participant honoraria
Participants who completed the follow-up interview were offered a gift voucher for £15 in recognition of their help with the study and to compensate them for any inconvenience they experienced.
Sample size
In the absence of information on weekly alcohol consumption among people attending sexual health clinics we based our sample size calculation on data from a previous trial set in an emergency department.16 We calculated that 97 evaluable participants would be needed per treatment arm to have 80% power to detect a difference in mean weekly alcohol consumption of 23.4 units with a standard deviation (SD) of 58.0 units (or 0.40 standardised difference), using a 5% level of statistical significance. However, a clustering effect may occur in the intervention arm due to different clinicians delivering the intervention. Power calculation formulae for a partially clustered design have been reported by Roberts and Roberts.34 Based on an average cluster size of 7 and an intraclass correlation coefficient of 0.04 in the intervention arm, a total of 112 evaluable patients in each arm (16 clusters of 7 in the control group) would provide above 80% power to detect such a difference. This corresponds to an inflation factor for clustering (design effect) of 1.15. Expecting a 30% drop-out rate at 6 months, we therefore aimed to recruit 320 participants.
During the first few months of the trial it became clear that the rate of recruitment was higher than we anticipated; the sample size was therefore modified to provide additional power to test both the primary and main secondary hypotheses (a reduction in unprotected sexual intercourse).
The final sample size was based on a practical size of 380 per arm (760 in total). If 65% of participants had unprotected sex in the control group, compared with 50% in the intervention arm, the power to detect such an effect would be above 90%, assuming 25% drop out, and a clustering design effect of 1.15. The power would remain above 80% if the absolute difference was above 13%.
Statistical analysis
A detailed statistical analysis plan was developed and published online before analysis.35 We used the statistical package Stata (StataCorp LP, College Station, TX, USA; version 12) for all of the descriptive analysis, graphs and regression models. All analyses were carried out according to randomisation arm (intention to treat), and two-sided p-values were considered significant when < 0.05. Descriptive analyses, including tables and graphs of baseline demographic and clinical variables, were conducted.
The primary outcome was compared between arms using random-effects linear regression adjusted for age (years), sex, clinic and M-SASQ measured at baseline. A random effect was included in the intervention arm to take into account any possible clustering by the clinician delivering the intervention, and residuals were allowed to differ by arm. This corresponds to the analysis suggested for partially nested trial design by Walwyn and Roberts.36 Sensitivity analyses were then performed to confirm the validity of the result. Different hierarchical models were fitted and standard errors were calculated using approaches more robust to non-normally distributed residuals, such as robust standard errors or non-parametric bootstrapping. Ten thousand bootstrap resamples were obtained to achieve stable estimates. Results of direct mean comparison (t-test) and of adjustment for imbalanced baseline characteristics were also explored.
The main secondary outcome was analysed using random-effects logistic regression and adjusting for unprotected sex at baseline. As for the primary outcome, various sensitivity analyses were also performed. Other secondary outcomes were compared using appropriate regressions or tests and adjusted for age (years), sex, clinic and the corresponding outcome variable at baseline. As the addition of a clinician random effect was found to have little effect on results, it was ignored for the comparison of secondary outcomes.
Baseline data were missing for one participant, and mean imputation was used in this case for adjusted analyses.37 Baseline characteristics of participants who dropped out from the trial were compared with the completers. Multiple imputation by chained equations was performed to impute the primary and main secondary outcomes at the follow-up visit. The imputation model included the important predictors of missingness and outcomes. Predictive mean matching was used to impute the alcohol consumption. Imputation was performed stratified by randomisation arm, and clustering by clinician was ignored in the imputation and analysis model. In order to reach negligible Monte Carlo error, 500 imputations were performed. Further sensitivity models allowing for missing not-at-random mechanism were also considered.38 They were based on a pattern mixture approach, considering a large range of possible differences in outcomes between participants who completed the follow-up and those who did not. Mean difference and confidence intervals (CIs) were estimated using the ‘rctmiss’39 user-written command in Stata.
To assess for possible heterogeneity of the intervention effect, primary and main secondary outcomes are also reported by the following subgroups: gender, age (< 25 years, 25–35 years, > 35 years), number of sexual partners in the 6 months preceding baseline (1 vs. > 1), and sexual orientation (heterosexual vs. non-heterosexual). We then tested for the presence of an interaction term between the subgroups and treatment arm. Age (years) was additionally tested as a continuous variable and considered to have a linear effect in the regression. For participants in the intervention arm, the alcohol consumption was also described by categories of intervention received.
Economic analysis
Estimation of costs
The economic evaluation took a NHS/Personal Social Service perspective, as recommended by the National Institute for Health and Care Excellence (NICE),40 and included all hospital contacts (inpatient, outpatient, accident and emergency), community health and social services (primary health care, community health services and social services) and medication.
All unit costs were for the financial year 2010–11. A summary of unit costs applied is listed in . Costs for NHS hospital contacts were sourced from NHS reference costs 201141 and community health and social service costs were taken from the annual unit costs of health and social care publication from the University of Kent42 or from relevant websites (as outlined in ). The cost of medications were calculated based on averages for British National Formulary chapters and were taken from Prescription Cost Analysis.44
Unit costs and sources used in economic evaluation
The cost of the intervention was estimated using the micro-costing (bottom-up) approach set out by the Personal Social Services Research Unit at the University of Kent, UK.42 We assumed that the brief advice was delivered by registrars, so used the median salary for registrars as the starting point. To this, employer national insurance and pension contributions were added as well as direct and indirect overhead costs to reflect hospital costs, administrative and managerial support costs, and capital costs. Total salary and overhead costs were then divided by the number of working hours per year, taken from Curtis,42 to calculate the cost per hour. Adjustments to this cost were made to reflect time taken by physicians in direct contact with patients and time spent on other activities. If the study participant saw an AHW in addition to the treating clinician then the cost of the AHW was added separately and reported as part of the cost of the intervention. The costs of the training were not included in the cost of the intervention.
Calculation of quality-adjusted life-years
Quality-adjusted life-years (QALYs) were calculated on the basis of the EQ-5D health state classification instrument, where health states are assigned a utility score using responses from a representative sample of adults in the UK.45 QALYs were calculated as the area under the curve defined by the utility values at baseline and 6-month follow-up and it was assumed that changes in utility score over time followed a linear path.46 An individual with perfect health would have an EQ-5D score of 1, which would translate to a QALY estimate of 0.5 QALYs over the 6-month follow-up.
Cost-effectiveness analysis
Differences in use of services between randomised groups at the 6-month follow-up are reported descriptively.
Differences in mean cost per participant were tested between groups using the Student’s t-test with ordinary least squares regression, and bootstrapping to confirm the validity of the results.47 Standard statistical tests were used because of the importance of the arithmetic mean in the analysis of cost data.48 The main analysis used cases for which complete data were available at follow-up and missing data imputation were not used. We used multiple imputation to test for the influence of missing cases in a sensitivity analysis.
Cost-effectiveness planes were produced to show the probability that: brief intervention is more effective and more costly than the control treatment; brief intervention is more effective and less costly than the control treatment; brief intervention is less effective and less costly than the control treatment; and brief intervention is less effective and more costly than the control treatment. The planes were constructed using regression models of total cost and outcome by treatment group, from which 10,000 bootstrapped resamples were run.
Knowledge of uncertainty around incremental cost-effectiveness is not sufficient for decision-making, which will depend on the how much society is willing to pay for improvements in outcomes. Cost-effectiveness acceptability curves (CEACs) were constructed, which show the likelihood that brief intervention is more cost-effective than the control treatment for different values a decision-maker is willing to pay for improvements in outcome.49
