This study employed a mixed design using both within-subject and between-subject independent variables (IVs). There was one within-subject IV with two levels (pre-intervention and post-intervention), and one between-subject IV with two levels (the intervention; either TSST or relaxation for 15 min). There were also several covariates, including alcohol-use measures and neurocognitive measures. The dependent variables were alcohol craving, quantified using as explicit craving (assessed via questionnaire), implicit craving (assessed via computer task) and number of drinks consumed on a voluntary drinking protocol.
Thirty-nine participants were recruited from staff and students at the University of Portsmouth (22 male and 17 female; mean age = 23.92 years [SD = 4.90]) using opportunity sampling, i.e. through internal advertising through e-mail and by word of mouth. The advertising informed participants that we were interested in investigating what leaves some more at risk of misusing alcohol. Participants were also informed that they would take part in a mild stress test; however, specific details of the procedure were withheld from participants. To confirm the suitability of the participants, they were initially sent a pre-screening questionnaire via e-mail. Exclusion criteria included age < 18 or > 55 years, previously or currently undergoing treatment for alcoholism or treatment for anxiety, depression or any other stress-related disorder. To be sure this was the case, participants also completed the Patient Health Questionnaire for Depression and Anxiety (PHQ-4; Kroenke et al. 2009) as a secondary screening for depression and anxiety and the Alcohol Use Disorders Test (AUDIT; Bush et al. 1998) to screen for undiagnosed alcohol dependence. Any participant who score > 5 on the PHQ-4 or > 20 on the AUDIT was subsequently excluded. Owing to their effects on salivary cortisol (sC) and alpha-amylase (sAA) levels, there were several other exclusion criteria. Female participants could not be pregnant, breastfeeding or currently taking oestrogen- and progesterone-based contraception. For all participants, participants could not take any of the following medications within the past 24 h: barbiturates, phenytoin, carbamazepine, meprobamate, glutethimide, alpha-methyldopa, corticosteroids, non-steroidal anti-inflammatory agents (e.g. aspirin, ibuprofen]), codeine, propranolol, beta-adrenergic agonists, cyproheptadine and psychotropic medications (including sedative hypnotics). The study was approved in its current form by the University of Portsmouth Science Faculty Ethics Board (ref. SFEC 2016–068).
Alcohol use and drinking behaviour
Alcohol use and drinking behaviour was evaluated using three measures: participants self-reported units of alcohol consumed per week, participants completed the AUDIT and participants completed the Binge Drinking Scale (BDS; Cranford et al. 2006).
Average alcohol use (units/week) was assessed through a single question ‘how many units do you typically consume in a week? Please note you cannot just count each drink as a unit of alcohol. The number of units depends on the different strength and size of each drink, so it can vary a lot. Here are some examples, single shot of spirits (25 ml, ABV 40%) = 1 unit, Alcopop (275 ml, 5.5%) = 1.5 units, small glass of wine (125 ml, ABV 12% = 1.5 units), large glass of wine (250 ml, ABV 12% = 3 units), bottle of beer/cider (330 ml, 5% ABV) = 1.7 units, can of beer/cider (440 ml, ABV 5.5%) = 2 units, pint of lower strength beer/cider (ABV 3.6%) = 2 units and pint of higher-strength beer/cider (5.2%) = 3 units’. The AUDIT, developed by the World Health Organisation (Babor et al. 2001) as a brief assessment of alcohol misuse for use in primary care and by researchers, was chosen to asses alcohol dependence. The AUDIT is scored on a scale of 0–40, where scores of > 20 would be considered dependent on alcohol, and > 30 severely dependent. The English version of the AUDIT has shown to have good psychometric properties for identifying alcohol dependence (Saunders et al. 1993; Stockwell et al. 1983). Additionally, the BDS was chosen to measure the level of binge drinking amongst the sample. Here participants are asked “What is the greatest number of drinks you have consumed in a 2-h period during the past 12 months?”. As the BDS originated in the US, Cranford et al. (2006) defined, binge drinking as consuming > 5 drinks for men or > 4 drinks for women on at least one occasion in the past 2 weeks. However, as the authors of this paper are based at a UK institution, we define binge drinking as consuming > 3 drinks in a single session, based on the NHS guidelines of binge drinking, i.e. consuming > 6 units in a single session (the mean unit/drink in the UK is 2).
Impulsivity, risk-taking, sensation seeking and decision making were assessed through the implementation of both questionnaire (explicit) and computer task (implicit) measures. Questionnaires included Barratt Impulsiveness Scale (BIS-11; Patton et al. 1995) and the Arnett Inventory of Sensation Seeking (AISS; Arnett 1994).The BIS-11 was chosen as it has been validated to have good psychometric properties in both research and clinical environments when quantifying the construct of impulsivity. Likewise, we employed the AISS to measure sensation seeking—an individual tendency closely related to both trait impulsivity and risk-taking (Magid et al. 2007).
Implicit impulsivity was assessed using a Stop-Signal computer task (SST; Logan et al. 1997). Within the SST, participants must respond to an arrow displayed in the centre of the screen either pointing left or right within 500 ms with the ‘b’ and ‘n’ keys, respectively. If the arrow is surrounded with a white circle (go-signal), participants should respond, however, if a red circle (stop-signal) is presented around the arrow the participant must refrain from responding. The SST started with a block of practice trials where participants did not move onto the critical trials until they either completed 20 practice trials without mistake or completed 50 practice trials. Following this, participants completed 50-critical trials. The dependent measures for this task included response time (ms), errors of omission (i.e. failing to respond to a ‘go-signal) and errors of commission (i.e. responding on a ‘stop-signal’). The SST has been shown to have good validity when discriminating between clinical (e.g. attention deficit hyperactivity disorder) and non-clinical (i.e. normative) populations (Lipszyc and Schachar 2010; Solanto et al. 2001).
The Balloon Analogue Risk Task (BART; Lejuez et al. 2002) was used as an assessment of real-world risk taking. In this task, participants are required to inflate an onscreen balloon by pressing the space bar. Each space bar press equates to an increase of £0.05 of virtual currency which can be ‘banked’ by pressing the return key. Each balloon has a randomly allocated tolerance and over inflation will cause the balloon to burst—losing the amount ‘earnt’ in that trial. Due to each balloons threshold being withheld, we could analyse early (pre-experience) responses, as well as learnt responses. The dependent variable is the mean number of pumps on each trial where the balloon did not burst. There was a total of 20 trials in this task.
Decision making was assessed using the Iowa Gambling Task (IGT; Bechara et al. 1994). Here, participants were shown four on screen choices: ‘A’, ‘B’, ‘C’ or ‘D’ and start with a ‘loan’ of £2000 of virtual currency. After each choice, participants are given feedback about their profits and/or losses. Choices ‘A’ and ‘B’ always yield £100, whereas, choices ‘C’ and ‘D’ always yield £50; moreover, for each choice, there is always a 50% chance of having to pay a penalty—the penalty for choices ‘A’ and ‘B’ is always £250, whilst the penalty for choices ‘C’ and ‘D’ is always £50. There was a total of 100 trials in this task, 50 of which are coded to give the participant a ‘fee’. Therefore, this task has two dependent variables: ‘A-tendency’ (preferring to choose ‘A’ and ‘B’) and ‘B-tendency’ (preferring to choose ‘C’ and ‘D’). Research suggests that more impulsive individuals will have a greater A-tendency due to their tendency to discount the value of delayed rewards (Burdick et al. 2013; Wittmann and Paulus 2008).
Both explicit (assessed via questionnaire) and implicit craving (assessed via computer task) levels were measured in this study. Explicit craving was assessed using a 14-item version of the Desires for Alcohol Questionnaire (DAQ; Kramer et al. 2010). Here, a 9-point Likert scale (1 = ‘Disagree completely’: 9 = ‘Agree completely’) was used by participants to rate a series of statements relating to their desire to consume alcohol at the point in time that the questionnaire was administered. The scores attained from the DAQ provide a single measure of craving for each participant, where greater scores specify a greater desire for alcohol consumption. The literature surrounding the psychometric properties of the DAQ has shown that alcoholic patients have a score of ~40, whereas healthy non-alcoholic drinkers score around ~20 (Kramer et al. 2010).
Implicit craving was assessed using an approach-avoidance task (AAT; Rinck and Becker 2007; Wiers et al. 2010; see Fig. 1). Previous versions of this task have used a different model, where ‘pulls’ on a joystick result in the image being moved closer to the participant, and ‘pushes’ making the image move away. In the current version, we required participants to move an image of a hand located at the bottom of the screen either towards (approach) or away from (avoid) an image positioned at the top of the screen using a joystick. On each trial a fixation cross was presented in the centre of the screen for 1 s. There were two conditions ‘approach alcohol’ and ‘avoid alcohol’, with a total of 64 trials in each, of which 32 were critical trials (pictures related to alcohol, e.g. a pint of beer) and 32 trials were control trials (pictures not-related to alcohol, e.g. pint of water). In the ‘approach alcohol’ condition, participants were required to approach alcohol-related images by moving the joystick towards the screen and vice versa in the ‘avoid alcohol’ condition. The order that participants completed conditions was counterbalanced. The dependent variables are response time (ms) and number of errors made. It is hypothesised that participants with a greater craving for alcohol have a lower response time when approaching alcohol-related pictures vs non-alcohol-related pictures and a slower response time when avoiding alcohol-related pictures vs non-alcohol-related pictures. As this version of the AAT was, as yet, unvalidated as a measure of implicit craving, this study represented an initial validation of the method.
Incentive motivation for alcohol was assessed using a progressive ratio schedule task (PRS; Vezina 2004; Ward et al. 2006). Participants pressed the space bar on a computer keyboard to ‘earn’ a up to a total of 12 5-ml shots of 37% ABV vodka diluted in 20 ml of mixer (e.g. coke, lemonade, tonic). For each subsequent drink, participants’ response requirement doubled; i.e. the first shot was delivered following one spacebar press; the second shot, following two spacebar presses; and the third following four presses. After each drink, participants rated pleasantness on a 15-point Likert scale (1 = ‘Very unpleasant’: 15 = ‘Very pleasant’). The dependent variable in this task was the number of drinks earned by each participant and their subjective enjoyment of consuming each drink earnt.
Heart rate (HR) and inter-beat interval (IBI) data were collected throughout the procedure using a Polar A300 Activity Tracker and a Polar H7 Heart Rate Sensor (Polar Electro, Finland). Research suggests that unbound sC is highly correlated with serum cortisol (Daniel et al. 2006; Dorn et al. 2007; Eatough et al. 2009), thus provides a reliable measure of hypothalamic pituitary adrenocortical axis (HPA) activation. In addition, research suggests that sAA release has a strong relationship with noradrenaline release (Chatterton et al. 1996; Thoma et al. 2012), which allows for a non-invasive bio-marker of sympathetic adrenomedullary axis (SAM) activation. Saliva samples (2 ml of passive drool) were taken twice. In the stress condition, they were taken before and after the Trier Social Stress Test (TSST; Kirschbaum et al. 2008). In the control condition, saliva samples were taken before and after sitting quietly for 15 min in the lab. Samples were placed on ice, then centrifuged (3000 × 15 min) and the supernatant was split into two 1-ml aliquots and frozen (− 20 °C) until assay. Samples were analysed in the laboratory using Salimetrics sC and sAA enzyme-linked immunosorbent assay (ELSIA) kits (Stratech, Ely, UK).
Questionnaires and the SST were programmed using PsyToolKit (Stoet 2010, 2017), and the BART, AAT and PRS were programmed and executed using PsychoPy software (Peirce 2007, 2008).
Phase 1: baseline assessments
Once participants suitability was confirmed via the pre-screening questionnaire, participants were randomly placed into either the experimental or control group and invited to attend a laboratory session lasting approximately 90 min. All study sessions took place between 11:00 and 15:30 to minimise the effects of the diurnal slopes of cortisol levels on our observations (Stone et al. 2001). Upon arrival, participants were given the opportunity to re-read the previously e-mailed information sheet and ask any questions. Two identical consent forms were then signed by the participants, one they could keep and the other was kept by the principal experimenter in a secure master file. The Polar A300 Activity Tracker and H7 Heart Rate Sensor was then fitted and recording was started. Participants then completed the unit/week questionnaire, AUDIT, BDS, BIS-11, AISS and (pre-intervention) DAQ on the computer. Subsequently, participants completed three computer tasks: SST, BART and (pre-intervention) AAT and provided the pre-intervention saliva sample. The order by which questionnaires and computer tasks were administered were counterbalanced to eliminate order effects.
Phase 2: the stress challenge
Stress group (TSST)
The procedure for participants in the stress group was made up of two stages: (A) preparation/anticipation and (B) the stress challenge. A shortened version of the TSST was used because participant feedback and HR data obtained through preliminary studies suggested that participants found the math portion of the TSST to not be as stressful as the speech. At the beginning of stage A, participants were told that they would have 10 min to prepare a 5-min speech about their dream job and what made them an ideal candidate. During this time, participants could plan their speech and makes notes on a piece of paper; however, they were informed that the speech must be performed without notes. During stage B, participants were led to a room containing a panel of three strangers wearing lab coats, sat behind a table and a video camera. The participant was then asked to deliver their speech to the panel. During this time, one member of the panel made notes using a clipboard. If the participant ceased talking for more than 20 s, they were asked to continue and reminded of how much time remained of their 5-min slot. At the end of stage B, following the 15-min stress-test procedure, a post-intervention saliva sample was taken.
Participants in the control group did not complete a stress challenge. Therefore, during stages A and B, participants had a 15-min break in the waiting room. After the 15-min break, participants provided their post-intervention saliva sample.
Phase 3: post-stress
Participants completed a post-intervention DAQ and AAT. They were then invited to take part in the PRS. Any participant who could not take part or did not wish to take part moved directly to debriefing.
Participants were fully debriefed and informed that the video recordings of their speech were to be destroyed and that their saliva samples would be rendered acellular within a 24-h period in accordance with the Human Tissues Act. Any participant that had consumed alcohol was advised to remain in the waiting area for 15 min to allow time for any intoxicating effects to wear off. Finally, participants had another opportunity to ask any questions.
Data preparation and statistical analysis
All data were analysed using R (version 3.4.4) and IBM SPSS (version 24) for Windows. IBI data acquired from the Polar Heart Rate monitor were converted standard deviation of NN intervals (SDNN) using Kubios (version 3.1) for Windows (Tarvainen et al. 2014). For all parametric analyses, studentised residuals were examined and outliers (> 1.5*IQR) removed prior to analysis (< 1% observations). All variables were initially compared for sex differences using a series of independent sample t tests. ‘Reactivity’ (reactivity = speech − pre) and ‘Recovery’ (recovery = speech − post) variables were calculated for HR and SDNN data. ‘Change’ (change = post − pre) variables were calculated for DAQ, AAT score (where AAT score = median avoid reaction time – median approach reaction time), sC and sAA. A single score for IGT was also calculated (IGT score = a tendency − b tendency, where higher scores = more advantageous decision-making). For repeated measures that were relevant to stress (pre- and post-TSST), we refer to the within-subject effects as ‘time’; this refers to pre- and post-TSST for the stress group and pre-vs post-non-stress (sitting quietly in the waiting room) for the control group.
As a manipulation check, physiological responses to the TSST were examined using 2 × 2 mixed-design analyses of variance (ANOVA; two-level within-subject factor = time [pre, post]; two-level between-subject factor = group [stress, control]). The effects of the TSST on change in craving (DV) were examined using independent t tests on both AAT change (post-pre) and DAQ change (post-pre) using group (stress, no stress) as the IV in both tests. In order to examine variables that predict alcohol consumption in the PRS task, we grouped covariates into several categories: (1) physiological responses to stress (HR, SDNN, sC, sAA, group); (2) craving (DAQ change, AAT change); (3) risk-taking, sensation seeking (BART, AISS, IGT) and impulsivity (BIS and SST); (4) prior alcohol use (AUDIT, units/week, BDS); and (5) drink enjoyment. To determine the effects of stress on the number of drinks consumed in the PRS task, we fitted these groups of covariates to several negative binomial regression models (PRS data were overdispersed count data). In addition to the covariates, ‘group’ (stress vs no stress) and gender were added to all models as fixed factors. To simplify regression models we applied backwards elimination, sequentially removing non-significant terms with the largest p values (> 0.05) until only significant effects remained (< 0.05). All significant covariates from the subcategories were then entered into a final negative binomial regression model to determine the best model for predicting alcohol consumption. Descriptive statistics are reported as mean ± SEM.