Study I: Validation and feasibility of collecting high frequency consumption data
251 participants were recruited at the Fayre yielding an average FAST score of 3.96 (SD = 3.46). Three participants stated that they never have and never would drink alcohol; they were not invited to participate in the SMS study. Of the remaining 248 participants, 82 (51% male, 92% undergraduates, mean age 20.95 years, SD = 2.90) completed the online survey in response to the initial text (sent October 2010 by SMS).
Three participants completed the initial web survey twice, the second set of responses for these participants were dropped. A logistic regression analysis was used to assess the relationship between FAST score at recruitment and subsequent completion of the initial online survey, no relationship was observed (OR = 1.03, 95% CI = 0.96, 1.12); 69 participants completed the FAST at recruitment, as well as the first online FAST and AUDIT. Pearson product moment correlation statistics indicate that all scores were strongly associated (r > 0.85, p < 0.0001 for all comparisons). Participants were asked to complete three online surveys, once at the beginning and again in December 2010 and March 2011 at the end of the project; 82 participants completed and provided FAST and AUDIT scores in the first online survey, 48 in the second, and 28 completed the third online survey; 84 participants responded at least once to the daily SMS messages (one participant continued to respond beyond 31 March 2011, these data were discarded, and two participants responded to SMS messages despite not completing the first online survey). On average 86.25 (min = 1, max = 156) responses were received from participants replying at least once to SMS text questions. A t-test on initial FAST score collected at recruitment for those who did not respond (n = 175, mean FAST = 3.697, SD = 3.338) to SMS messages against those who responded at least once (n = 84, mean FAST = 4.579, SD = 3.682) indicated responders had higher FAST scores (t = 1.86, p < 0.05). Furthermore, Spearman’s rank measure of association suggested no relationship between FAST scores collected at recruitment and number of responses to SMS messages for those responding at least once or more (ρ = -0.100, p = 0.391). These results suggest that attrition is pronounced for online surveys prompted by SMS, whereas SMS responses remain stable over time (see Figure 2).
We computed the average number of units consumed and the proportion of days participants consumed alcohol. A multiple regression model indicated that a higher baseline FAST score (B = 0.392, p < 0.01, 95% CI 0.180 0.604) was associated with a higher average number of units consumed whereas being female (B = -1.823, p < 0.05, 95% CI -3.360 -0.293) was associated with lower average number of units consumed. A second model indicated that a higher FAST score (B = 0.008, p = 0.302, 95% CI -0.008 0.025) was not statistically associated with the average number of days in which alcohol was consumed whereas being female (B = -0.118, p < 0.05, 95% CI -0.235 -0.001) was associated with lower number of days drinking. These results provide external validity; indicating that the average quantity of alcohol consumed on each drinking occasion, recorded through SMS, is predicted by initial FAST scores, although not the frequency of drinking occasions.
Twenty eight participants completed all three online surveys and provided FAST and AUDIT scores as well as responded to SMS survey messages (min responses = 100). To enable further modelling, a cubic spline interpolation algorithm was used to create smoothed FAST and AUDIT scores across the response period for these 28 participants. As there were fewer FAST and AUDIT readings taken over the course of the study, compared to daily SMS messages, we preferred to maximise the use of the SMS data to increase power. As this algorithm can, in some instances, yield interpolated values less than zero, values less than zero were replaced with zero (44 interpolated AUDIT scores and 30 interpolated FAST scores, out of a total of 3,365 data points for each, were replaced). Neither interpolated FAST scores (B = 0.025, p = 0.834, 95% CI -0.210 0.260) nor interpolated AUDIT scores (B = 0.157, p = 0.059, 95% CI -0.006 0.320) were significantly associated with self-report alcohol units consumed in FE linear models. A drinking frequency variable (0 if no alcohol was consumed, 1 if alcohol was consumed for each day) was derived. In FE logits interpolated AUDIT scores were positively associated with drinking likelihood (B = 0.100, p < 0.01, 95% CI 0.035 0.164) although interpolated FAST scores were not (B = 0.063, p = 0.175, 95% CI -0.028 0.155).
Figure 4 shows average alcohol consumption by day of week. The highest consumption occurred on Fridays (mean = 4.08 units, SD = 7.29), closely followed by Saturdays (mean = 3.86 units, SD = 7.10). Of weekdays (Sunday–Thursday inclusive), Wednesdays saw the heaviest alcohol consumption (mean = 2.57, SD = 5.63), followed by Thursdays (mean = 2.08 units, SD = 5.55). Sundays saw the lowest alcohol consumption (mean = 1.44 units, SD = 3.87).
Fixed-effects Poisson regression was used to determine the significance of this observed weekly variation. Binary (dummy) variables were assigned to each day of the week and Saturday was used as a reference category. Compared to Saturday, significantly fewer units were consumed on all other days of the week (Sunday: B = -1.01, 95% CI = -1.07 -0.95, Monday: B = -0.81, 95% CI = -0.87 -0.76, Tuesday: B = -0.91, 95% CI = -0.97 -0.85, Wednesday: B = -0.41, 95% CI = -0.46 -0.36, Thursday: B = -0.63, 95% CI = -0.68 -0.58) except Fridays, which did not significantly differ from Saturday (B = 0.02, 95% CI = -0.02 0.07). Sundays saw the lowest levels of consumption (B = -1.01, 95% CI = -1.07 -0.95). The exponentiated coefficient on Sunday yields a rate ratio of 0.44 (95% CI = 0.34 0.39) suggesting that on Sundays participants drank less than half the amount they drank on Saturdays.
Figure 5 shows daily alcohol consumption across the study period. Alcohol consumption was highest on New Year’s Eve (mean = 12.51 units, SD = 9.14) followed by the Welsh Varsity sporting event (mean = 7.36 units, SD = 10.70). Alcohol consumption was also elevated on Christmas Eve (mean = 6.40 units, SD = 7.71), Christmas Day (mean = 6.18 units, SD = 5.92) and Boxing Day (mean = 6.70 units, SD = 9.30). The trend line in Figure 5 suggests a peak in alcohol consumption during late December and a trough in early January. Alcohol consumption during the 'twelve days of Christmas’ (defined here as 21st December to 1st January) was greater (mean = 4.07 units, SD = 3.18) compared to the rest of the study period (mean = 2.32 units, SD = 1.49).
Longitudinal fixed-effects Poisson regression analysis confirmed this observation. A dummy variable described whether, for each day of the study, reported alcohol consumption fell within the “twelve days” (= 1) or outside (= 0) and further day of week dummies were included to control for the variation in consumption across a typical week. Significantly more alcohol was consumed during the 'twelve days of Christmas’ compared to the rest of the study, irrespective of the days of the week that fell within the two time periods (B = 0.53, 95% CI = 0.48 0.57, p < 0.001). The corresponding rate ratio of 1.69 (95% CI = 1.62 1.77) suggests that participants drink 70% more during the holiday period.
A logit was used to consider whether the reported units of consumption on the preceding day influenced the likelihood of alcohol use. A derived binary variable described whether reported units consumed was zero (0) or greater than zero (1) and a FE logit indicated that the more units consumed the preceding day the less likely participants were to consume alcohol (B = -0.016, p < 0.01, 95% CI -0.027 -0.005), controlling for day of week effects. Using the time that SMS messages were sent and responses were received we calculated the delay between when the SMS message was sent and participants’ response. A FE linear model indicated that a the more alcohol consumed the longer the response delay on the following day (B = 0.019, p < 0.001, 95% CI 0.010 0.028).
The eight celebratory events were all associated with an increase in alcohol consumption. The highest mean units of alcohol were consumed on New Year’s Eve, Welsh Varsity, Boxing Day, Christmas Eve, Christmas Day, Halloween, St Patrick’s Day and Valentine’s Day (Figure 6).
Each event was described by a dummy variable and day of week dummies were included as above. FE Poisson regression indicated that alcohol consumption was statistically greater during these events compared to other days of the week. The largest increase in drinking was seen on Boxing Day followed by Halloween, New Year’s Eve, Welsh Varsity and Christmas Eve. The rate ratio of Boxing Day suggests respondents drank 6.57 (95% CI = 5.77 7.47, p < .001) times what they would usually drink and on Halloween drank 4.18 (95% CI = 3.59 4.86, p < .001) times what they would usually drink. Similarly, on New Year’s Eve participants drank 3.88 (95% CI = 3.53 4.27, p < .001) times their usual level of consumption and during Welsh Varsity students drank 3.51 (95% CI = 3.12 3.96, p < .001) times usual consumption. Valentine’s Day and St Patrick’s Day had the smallest increases, 1.42 (95% CI = 1.17 1.73, p < .001) and 1.42 (95% CI = 1.18 1.69, p < .001) respectively.
Results are organised according to the principal emergent themes. Interviews lasted between 17 and 55 minutes (average 32 minutes).
With respect to attitudes towards SMS messaging in general, all participants regarded information through this medium positively and preferred SMS messages to telephone calls and electronic mail.
“It’s really good. I think I like texts because sometimes, you know, you don’t check your emails every day or, you know, log on to a computer, bit of an effort. […] You get a text, you read a text” (male, opt in, VN550029).
“Well, [SMS is] certainly the easiest way. Emails don’t always get checked and phone calls are almost too intrusive” (male, opt out, VN550039).
The perceived privacy afforded by SMS messages was explicitly referenced in one case.
“It was confidential, and like, you could guarantee it was confidential” (male, opt out, VN550038).
SMS messaging frequency
Twelve participants stated daily text messages were acceptable: “[the frequency] didn’t really bother me” (male, opt out, VN550045), “daily was fine” (female, opt in, VN550033) and “the frequency was fine” (female, opt out, VN550040). Where participants objected to message frequency this was attributable to misaligned study expectations.
“I don’t think I knew enough when I signed up that it was going to be so, erm, persistent” (female, opt out, VN550041).
Message timing was one of the most significant barriers to participation. Messages were originally sent at 7 am, this was changed to 9 am after the fourth week of the study (end of week 46 in Figure 1), following complaints that the text messages were waking participants up.
“It was a bit off putting, a bit irritating, especially early in the morning. If it was in the day time it would have been fine, but it was always around 7 am. No, I don’t think I wanted texting at 7 am in the morning, especially to be asked about my drinking habits… so the frequency of the texts themselves wasn’t too bad… it was the timing of the texts rather than the number” (male, opt out, VN550039).
As with numerous studies, this study elected to use the UK standard unit (8 g alcohol) as the measure of consumption. Participants described units as not “user-friendly”, “confusing” and hard to work out to any real degree of accuracy.
“…although we were told what units are what…it can be a bit blurry [especially] if you’re drinking cocktails and things like that” (female, opt out, VN550032).
“[… it] is quite hard to track. I mean everyone brings a bottle and pours it in […] it was very hard to judge the amount everyone was drinking” (male, opt in, VN550037).
Intervention delivery and content
The study was conducted at a time where several educational interventions were on-going (safe drinking limits were advertised on posters around campus). Participants were asked their view on alcohol interventions generally, in relation to student culture and for their views concerning SMS delivered interventions. Existing interventions were ignored:
“I wouldn’t say I have [seen intervention material] in [Study Town], […] there’s a little sticker on my mirror that says 'you’re looking at the person responsible for your health […] I’m not sure. There was the odd pamphlet around Fresher’s week […] I probably didn’t pay that much attention to it” (male, opt in, VN550029).
Several participants felt “sensory overload” to traditional posters and similar “mass market” campaigns targeting alcohol use, indicating habituation and they further suggested such media were just ignored. Whereas SMS methods were regarded as more appropriate because they were personal.
“[Mass market campaigns are] written to […] everyone in my age group [they do not reach] out to me personally, so I don’t feel the need to respond” (male, opt in, VN550036).
In respect of content, advertising the negative consequences of excessive consumption were seen as aversive for three participants, beneficial for two and the remaining 13 suggested emphasising the positive with the negatives would likely provide the most effective intervention content. The consensus was that content that “grabbed attention” but did not come across as “parenty” was the best overall approach.
“I think [it is better to emphasise the] benefits […] when [interventions] focus on the negative points it just feels like you’re back at home and it’s your parents telling you off for drinking too much” (male, opt in, VN550036).
“If you do it both in moderation, the negatives and the positives, it could discourage binge drinking […] there’s been lots of adverts [warning against] heavy drinking and it doesn’t seem to be doing much good, students are still drinking as much as ever” (female, opt in, VN550046).
When asked further about potential intervention content, interventions that referred to the financial aspects of alcohol use appealed to the majority of those interviewed, 17 participants, the remaining participants suggested interventions should refer to both health and finance.
“Finance is a big thing for students. I mean it’s the biggest conversation I would have thought. Amongst my friends it’s always money and how they can save money… so I think tapping into finance would be a really good idea for students” (male, opt out, VN550038).
“I think everyone is aware of their health, health and what to do, but most don’t care. Like smoking and drinking and taking drugs…whilst they’re young as well, a lot of people have the mentality that 'I’m going to die of something so I might as well have a good time 'til I do’” (female, opt out, VN550040).
The qualitative study revealed that students are happy with SMS delivered messages, as they are perceived as secure and private. Obstacles included messages that were sent too early in the morning and that emphasised the negative aspects of excessive alcohol consumption. A strong positive theme for intervention content was the financial implications of excessive alcohol use and the opportunity for messages to help students reduce their expenditure. Financial information was therefore used as the content for an SMS delivered simple intervention that would be delivered late morning (Study II).
Study II: A feasibility randomised control trial using SMS text messaging to provide feedback on alcohol expenditure
Participants were recruited from a panel of individuals (N = 493) who had completed research on an unrelated project but had agreed to be contacted again if further research opportunities arose. None had participated in Study I. Of these, 87 participants consented and were recruited onto the current study (58 male, 43 students), yielding an average FAST score of 5.54 (SD = 2.28); 86 participants responded at least once to the daily SMS messages (see Figure 7). One person actively opted out of the study. The average age of students (mean = 22.00 years, SD = 3.66) was greater than that of non-students (mean = 38.49 years, SD = 14.26; t = 7.05, p < 0.001).
A mixed two (within factor, average daily units consumed before and after the intervention) by two (student and non-student) by two (experimental group: intervention and control) ANOVA yielded a non-significant main effect of experimental group (F(1, 70) = 1.56, p = 0.216, partial η2 = 0.02). The means presented in Table 1 indicate that three of the four groups showed a reduction in alcohol consumption from baseline to follow-up: Control Students showed a decrease of 5.5%; Intervention Non-Students showed a decrease of 4%; Intervention Students showed a decrease of 19%. However, Control Non-Students showed an increase of 17.1%. Sample size estimates for the change in mean on intervention students and assuming σ = 1.7, power analysis for a one-sample mean test, taking pre-intervention mean as the null and the post-intervention mean as the alternative (see Table 1), and with a 0.8 power (1-β) to detect significance at α = 0.05 suggests a sample size of 140 would be sufficient to raise this to significance. There was little evidence to suggest that the expenditure based intervention generalises as well to non-student populations.