A Personalized Normative Feedback Text-Message Intervention to Reduce 21st Birthday Alcohol Use and Problems

Objective: Twenty-first birthdays are associated with extreme levels of heavy drinking and alcohol-related harm. However, few 21 birthday preventive interventions have been tested, and even fewer have been supported as effective. The current study was designed to 1) test a text-message intervention to reduce 21 birthday estimated Blood Alcohol Content (eBAC) and alcohol problems, and 2) examine potential mediators and moderators. Method: College students (n=200; 69% female, 87.5% White/Caucasian) with an upcoming 21 birthday completed a baseline assessment and were randomized to a text-message intervention or an assessment-only control condition. Participants in the intervention group were sent, and were asked to reply to, a text-message the day before and day of their planned 21 birthday celebration focusing on 21 birthday specific personalized normative feedback (PNF) and protective behavioral strategies (PBS), respectively. All participants were sent a follow-up assessment the day after their birthday celebration (92.9% retention rate). Results: Among participants in the intervention group, 95.8% responded to textmessage 1, and 95.8% responded to text-message 2. Regression analyses did not reveal an overall treatment effect for eBAC or alcohol problems. However, there were indirect effects with perceived norms as the mediator and eBAC (-.175 [SE=.060]; 95% CI [-.292, -.080]) and alcohol problems (-.124[SE=.044]; 95% CI [-.245, -.057]) as the outcome such that the intervention was associated with lower perceived norms, which was, in turn, related to less alcohol involvement. Conclusions: Although no main effect of treatment was observed, this study provides further evidence that changing perceived norms is a promising strategy for preventive interventions with event-level alcohol use.


In Memoriam
This project is dedicated to Dr. Mark D. Wood (1960Wood ( -2015)).Dr. Wood was my Major Professor until his death, and the sponsor on my initial F-31 submission.He was enthusiastic about the study, and encouraged me to apply for the grant despite some initial reluctance.Mark spent countless hours reviewing major sections of my application and offering detailed feedback.This will come as no surprise to anyone who knew him.Mark was instrumental in helping me ultimately obtain the F award.
More generally, I thank Mark for his four years of dedicated mentorship.
Leading by example, he taught all his students the importance of personal responsibility, diligence, integrity, and work-life balance.These are lessons I hope to carry forward, and often still find myself thinking "What would Mark do?" vi PREFACE This dissertation is prepared in manuscript format.
Even among these very high-risk events, 21 st birthday drinking stands out as especially risky.In one study, average Blood Alcohol Content (BAC) among college drinkers on their 21 st birthday was .186,which was 116% higher than St. Patrick's Day, 74% higher than peak consumption over spring break, and 47% higher than New Year's Eve (Neighbors et al., 2011).In a large and rigorous descriptive study of 21 st birthday celebratory drinking (N=2,518), 83% reported some alcohol consumption, with drinkers consuming an average of 12.65 (SD=8.5)alcoholic beverages (Rutledge, Park, & Sher, 2008).Nearly half (48%) of those who drank to celebrate consumed more alcohol that day than at any previous point in their life.Of particular concern, 34% of men and 24% of women who drank to celebrate reported consuming 21 or more alcoholic beverages, consistent with the "21run," or "21 for 21" ritual.Most people experience one or more alcohol-related problems as a result of 21 st birthday drinking (Neighbors et al., 2014), with 41% blacking out, 45% having a hangover, and 35% vomiting (Brister, Sher, & Fromme, 2011).

st Birthday Preventive Interventions
Effective preventive interventions focusing on this event are needed.(Hembroff et al., 2007;Neighbors et al., 2005;Smith et al., 2006).More recent trials have included an array of empirically-based intervention approaches, with more promising results.
For example, Neighbors et al. (2009) assigned college students who intended to consume at least two drinks to celebrate their 21 st birthday to an assessment only control group, or to an emailed intervention.The email contained a link to a 9-page personalized feedback report that was sent one and two days before their 21 st birthday.

Text Message Interventions
One promising communication modality for disseminating alcohol prevention interventions with young adults is Text Messaging (TM) (Suffoletto, 2016).A 2011 Pew Research study (Smith, 2011) (Bacigalupo et al., 2013), low-density lipoprotein (LDL) cholesterol reduction (Chow et al., 2015), and smoking cessation (Whittaker et al., 2012).In spite of the general popularity for using TM to manage health behaviors, alcohol researchers are only beginning to utilize the technology as an intervention modality.TM interventions for health promotion are more effective when they are tailored to the recipient (Head et al., 2013).For example, in one study (N=765), young adults being discharged from an Emergency Department who received TMs that included feedback on willingness to set drinking moderation goals reported less alcohol use and fewer alcohol-related injuries than participants in control groups up to nine-months post-intervention (Suffoletto et al, 2015).The current study will also tailor TM to recipients in an effort to reduce heavy drinking.

Current Study Overview
The goal of the current study was to test a 21 st birthday intervention delivered through text-messaging.Since individual texts can only be 160 characters, the intervention needed to be brief.Therefore, PNF was initially chosen as the basis for the intervention and subsequently shaped through a pilot study to include PBS (see below).A large body of research indicates that PNF is associated with reductions in alcohol consumption in stand-alone interventions (Dotson, Dunn, & Bowers, 2015), and Neighbors et al (2009) demonstrated its effects as part of a larger treatment package to reduce alcohol use via e-mail delivery.
Hypothesis one was that a TM intervention for 21 st birthday drinking, relative to an assessment-only control group, would be associated with a reduction in estimated Blood Alcohol Content (eBAC) and alcohol-related problems for 21 st birthdays.Hypothesis two was that this main effect would be moderated by 21 st birthday drinking intentions, such that the treatment effect would be greater for participants who anticipated a high (versus low) 21 st birthday eBAC.Hypothesis three was that perceived 21 st birthday drinking norms would mediate this association.
Specifically, it was hypothesized that a) the intervention (relative to control) would be associated with a reduction in perceived 21 st birthday drinking norms, and b) lower drinking norms would be related to less alcohol involvement.Protective behavioral strategies (PBS) was also examined as a potential mediator, although no a priori hypothesis was established because it did not mediate intervention effects in the Neighbors et al. (2009) study.

Method Participants & Recruitment
Participants were recruited on a rolling basis between May 2016 and November 2016.In the approximately 7 days prior to their 21 st birthday, students (n=1,283) received up to three recruitment emails for a study that ostensibly examined how young adults celebrate their birthday.Those who were interested completed a very brief online questionnaire to verify eligibility (n =280).Students were eligible if they a) intended to consume at least two standard drinks to celebrate their 21 st birthday, and b) had a mobile phone from which they could send and receive text messages.Of people who took the eligibility survey, 85.0% were eligible.Sample characteristics at baseline are presented in Table 1, and recruitment is depicted in Figure 1.The racial composition of this sample is similar to the racial composition at the University of Rhode Island (http://web.uri.edu/ir/files/RIBOE-Official-Enrollment--2002- -2016- .pdf, accessed Jan 24, 2017)), and the sample is diverse with respect to gender.

Procedure
Upon completing the eligibility survey, eligible participants were presented with an informed consent document.Those who consented were asked to immediately complete a short baseline survey.Next, participants were urn randomized by gender and drinking intentions (strata of standard drinks: 2-5, 6-10, 11-15, 16-20, 21+) to the intervention or control condition (Stout, Wirtz, Carbonari, & Del Boca, 1994).As discussed below, those assigned to the intervention condition were texted the day before and the day of their anticipated birthday celebration. 1 Participants assigned to the control condition were not contacted during this period.At 4 PM the day after their anticipated celebration, participants received a link via email for an online follow-up questionnaire.Non-responders received up to four reminders (one by email, two by text message, and one by phone call, in that order).Participants were entered into a raffle for an iPad mini for completion of the baseline survey, and received $10 in cash or towards an Amazon Gift card for completion of the follow-up survey.All policies and procedures were approved by the University of Rhode Island Institutional Review Board (IRB).

Intervention Development
Focus Groups.To refine the wording and delivery of text messages (described below), six 30-60 min.focus groups were conducted with drinkers 19-20 years old (n=23).To start, general feedback was solicited about information that may be helpful for students to receive proximal to their 21 st birthday; next, structured questions were asked to elicit specific information.In the structured question portion of the focus groups, participants were asked their opinion of receiving two nearly identical text messages presenting PNF.Of n=13 responses 2 , 69% thought the texts should be 1 To check accuracy of the anticipated birthday celebration date, which was assessed at baseline, participants were also asked to indicate the date of their birthday celebration at follow-up (henceforth called actual birthday celebration date).Since the date TMs were sent occurred according to the planned birthday celebration, it was possible that participants in the intervention condition would receive the first TM the day of their actual birthday celebration, and receive the second TM the day after their actual birthday celebration (if the actual birthday celebration date was exactly one day before the planned birthday celebration date).It was also possible that participants could receive no TM prior to their actual birthday celebration (if the actual birthday celebration date was two or more days before their planned celebration date).Among those in the intervention condition, participants who received one TM (n=5) were retained but participants who received no TM (n=1) prior to their actual birthday celebration were excluded.
Other than one participant who was erroneously sent the follow-up too early and therefore excluded from the study, no one indicated their birthday celebration occurred after completing the follow-up assessment.
2 Due to the nature of the focus groups, not every participant answered each question, and n=10 did not reply to this particular question."Thanks!We will check in tomorrow."Participants who replied to text message two by 7 PM immediately received: "OK thanks!We'll be emailing you in the next day or two."Although a reply of "OK" was requested at both time points, any response (e.g. "yes," "sounds good," etc.) was treated as indicative of compliance.All text messages were sent from a secure program built for this study at the University of Pittsburgh.

Measures
Eligibility Survey.This survey consisted of two items: 1) "Do you have a mobile phone that you use to send and receive text messages?"2) "How many standard alcoholic drinks do you intend to consume on your 21 st birthday celebration?"Standard drink definitions were given.
Demographics.At baseline, participants were asked their gender, weight, ethnicity, and race.Greek status was assessed with one item adopted from Capone, Wood, Borsari, and Laird (2007).
Intended Birthday Celebration Day.At baseline, participants were shown a calendar and asked to "Indicate the day you intend to celebrate your 21 st birthday.If you plan on celebrating two or more days, please choose the day that you anticipate will be the 'largest' or 'primary' celebration." Intended/Actual 21 st Birthday BAC.At baseline and follow-up, participants were asked: 1) "In total, how many standard drinks do you plan on consuming/did you consume during your 21 st birthday celebration?" and 2) "Over how long a period of time do you plan on drinking/were you drinking?"Standard drink estimates were provided.Intended/actual 21 st birthday BAC were calculated using these data, as well as gender and weight (from the demographics survey) with the formula provided by Matthews and Miller (1979).format (e.g."use a designated driver").We observed a co-efficient alpha of .77.

Intervention Satisfaction. At follow-up, participants in the intervention
condition were asked the following questions on a 5-point Likert Scale from 1 (not at all) to 5 (extremely): "The feedback was useful to me," "I thought about this information over the course of my birthday celebration," "I would have preferred receiving this information through other means, such as email."

Power
The one intervention most similar to the present study observed an effect size (ES) of d=.33 (Neighbors et al., 2009).Since the current study includes fewer intervention components than did Neighbors et al., a lower ES might be anticipated.
On the other hand, by using TM to deliver the intervention, participants may read the information more carefully and/or be more receptive to it.Therefore, d=.33 was utilized to develop an adjusted ES (ESadj) based on controlling for intended eBAC as a covariate.Using the formula provided by Rossi (2013) Regression Analyses.To examine intervention efficacy, two regression models were run where treatment condition, anticipated 21 st birthday eBAC, and the condition by anticipated 21 st birthday eBAC interaction were entered as independent variables.Dependent variables were actual 21 st birthday eBAC, and 21 st birthday alcohol consequences.Regression analyses assumed a negative binominal distribution with the GENLIN command in SPSS v. 24.For these analyses, results from the Wald χ 2 test were reported, which accounts for the log-transformations used in negative binomial tests.

Mediation.
To examine mediation, four models with bootstrapping were run in MPlus version 7also assuming a negative binomial distribution for the outcomes.
Specifically, mediation models were run with actual eBAC and alcohol consequences as the outcome, and T2 Norms and PBS as the mediator.In all models, intended 21 st Birthday eBAC was included as a covariate for the outcome.In models with T2 Norms as the mediator, T1 Norms was included as a covariate for the mediator.
Indirect effects were computed as the a path (independent variable to mediator) multiplied by the b path (mediator to dependent variable).A significant indirect effect indicates mediation.

Intervention Efficacy
Results from the negative binomial regressions suggest actual 21 st birthday eBAC and 21 st birthday alcohol consequences were not significantly different between the treatment and control condition.The Treatment by Anticipated 21 st Birthday eBAC interactions were also not significant.Full results are presented in Table 4.

Mediation
Full mediation results are presented in Figure 2

Discussion
The current study was designed to test the efficacy and mechanisms of a TM intervention for 21 st birthday drinking.Counter to expectations that participants receiving the intervention would exhibit less 21 st birthday alcohol involvement than those in a control group, no direct intervention effects were observed for either eBAC or alcohol-related consequences.There was also no interaction between eBAC intentions and treatment (hypothesis two), failing to replicate findings in Neighbors et al. (2009).However, in support of hypothesis three, perceived drinking norms (but not PBS) functioned as a mediator for both 21 st birthday eBAC and 21 st birthday alcohol consequences.Specifically, participants in the intervention condition had lower drinking norms at follow-up (after controlling for baseline drinking norms), and follow-up drinking norms were positively associated with 21 st birthday eBAC and 21 st birthday alcohol consequences.Furthermore, the magnitude of these effects was rather large.For example, after accounting for baseline norms, the intervention (relative to control) was associated with a 2.256 unit reduction in the perceived number of 21 st birthday drinks consumed by a same-sex University of Rhode Island student.Then, for every one unit decrease in perceived norms, there was an 8% decrease in 21 st birthday eBAC.6 Counter to early theorists (i.e.Baron & Kenny, 1986) who argued that a total effect (i.e.X and Y are associated) is a necessary prerequisite for an indirect effect, recent experts have argued otherwise (e.g.Hayes, 2013;MaKinnon, 2008).As discussed by Hayes (2009), a total effect can be conceptualized as the sum of all indirect effects and a direct effect.Then, "two or more indirect effects with opposite signs can cancel each other out, producing a total effect… that is not detectably different from zero, in spite of the existence of specific indirect effects that are not zero" (Hayes, 2009;p. 414).Although many indirect effects might exist, there are limitations to what data can be reasonably collected, and then tested statistically.It is very possible that the intervention exerted additional effects on the outcome through one or more unknown mediators that went in the opposite direction of the Treatment -> Norms -> alcohol involvement indirect effect.One potential factor is enhanced recall of alcohol consumed on 21 st birthdays.In other words, it is possible that the intervention led participants to have a more accurate memory of how much they drank, which was then associated with the number of drinks and/or consequences reported.If true, the intervention may actually be efficacious, but we observed null results due to differential reporting bias in the intervention versus control condition.
However, this is just one of several possibilities.That the indirect effect may have been "cancelled out" by some other unknown effect should be treated as preliminary since this is a relatively new area of study and findings should be replicated elsewhere.
Although a main effect of treatment was not observed, the existence of an indirect effect through drinking norms is promising and consistent with two recent reports.A systematic review by Reid and Carey (2015), which included 61 trials where a mediator was tested for intervention effects in college student samples, identified descriptive norms as the most widely supported mediator.In another review of mediators for technology-delivered psychosocial treatments for substance use, seven studies tested perceptions of peer drinking as a mediator, and it was supported in six (87.5%) trials (Dallery, Jarvis, Marsch, & Xie, 2015).The current study provides further evidence that changing perceptions of drinking norms is a promising tool for preventive interventions, and likely applicable to event-level drinking in addition to typical patterns of alcohol use (e.g.Rodriguez et al., 2015).

Strengths
The major strength of the present experiment is that a high level of study integrity was achieved.Among the 196 participants who were randomized and not excluded, 92.9% were retained at follow-up, with no evidence of differential attrition.
Furthermore, nearly all participants assigned to the treatment group responded to the text messages (95.8% for TM 1 and 95.8% for TM 2), which suggests they received and read the information.This finding points to the general promise for using textmessaging as a means of intervening with young adults, consistent with recent reviews (Orr & King, 2015;Suffoletto, 2016).Although no total intervention effect was observed, another strength of the current study is that the text-messages were carefully created based on focus group pilot testing, and on PNF, which is widely supported in the literature.Based on the descriptive results, participants reported moderate satisfaction with the intervention.
increased by sending several messages over the course of a night or intervening on friends with whom participants plan to celebrate.Also, it might be beneficial to send initial messages 1-2 weeks prior to an event to capture the period of time when students are likely planning the night's activities.

Conclusion
The efficacy of a PNF and PBS text-message intervention for 21 st birthday drinking.Nearly all students assigned to the treatment condition received and read the intervention, and participants were reasonably satisfied with the messages.In spite of this, there was no main effect of treatment on eBAC or alcohol problems.There were, however, indirect effects through perceived 21 st birthday drinking norms.Future   Matthews and Miller (1979) and Rutledge et al. (2008).BAC=Blood Alcohol Content, a 0=Control Condition, 1=Intervention Condition.Results for the main effects were not substantially changed when separate models were run without the interaction term .

For
instance, Mason et al. (2015) conducted a meta-analysis of RCTs examining the efficacy of a TM intervention for adolescent or young adult substance use.Overall, they observed a weighted effect size of d=.25.Eleven of these trials used TM for tobacco prevention, whereas only three focused on alcohol.
(T2 Norms as the mediator) and Figure 3 (with PBS as the mediator).For each model, the a (Independent Variable to Mediator), b (Mediator to Dependent Variable), and c' (Direct Effect) paths are reported in the diagram, and the c (Total effect) and a*b (Indirect effect; also equivalent to c-c') paths are reported in the Figure note.For example, in Figure 2a, the a path equals -2.256, the b path equals 0.078, the c' path equals 0.130, the c path equals -0.045 and the a*b (or indirect effect) equals -.175.For the purpose of establishing mediation, it is the indirect effect that is most important.A significant indirect effect through T2 Norms was observed with both actual 21 st birthday eBAC and alcohol problems as the outcome (-.175[SE=.060],95% CI [-.292, -.080] and -.124[SE=.044],95% CI [-.245, -.057], respectively).An indirect effect through Protective Behavioral Strategies was not observed for either outcome.

Finally, since perceived
norms, but not PBS, mediated the treatment -> alcohol outcomes path, future studies might benefit from removing PBS and further focusing on norms.One possibility would be correcting both descriptive (quantity of actual behaviors, as addressed in this study) and injunctive (approval of certain behaviors) norms.As discussed by Krieger et al. (2016), injunctive norms are more strongly associated with behavior when the two are closely connected.Perhaps the total intervention effect could be enhanced by including the following feedback: "The average student at [study site] believes XX drinks is the maximum one should consume on a 21 st birthday."Another option might be adopting injunctive norms for Protective Behavioral Strategies (e.g."X% of students at [school/university] think people should adopt strategy A on their 21 st birthday) 7 .

Figure 3 .
Figure 3. Mediation Models with Protective Behavioral Strategies

Table 4 .
Effect of Treatment and Anticipated 21 st Birthday eBAC on Actual 21 st Birthday eBAC and Alcohol Consequences………………………………………….32 control group, this intervention reduced 21 st birthday estimated BAC (eBAC), d=.33.Moderation analyses revealed that the intervention was efficacious for participants with a high intended eBAC (d=.42 at one SD above the mean), calculated as the estimated BAC participants thought they would achieve on their 21 st birthday during a baseline assessment.The intervention was not efficacious for participants with a low intended eBAC (d=.04 at one SD below the mean).Mediation analyses showed that the intervention worked as a result of reducing perceived 21 st birthday-specific Personalized Normative Feedback (PNF).Relative to the assessment-only birthday drinking norms.No indirect effect was observed through use of PBS.This study indicates that PNF is a promising intervention for 21 st birthday drinking, although replication is needed, particularly across modalities that are used more frequently than email.
different and 31% liked the redundancy.In the unstructured portion of the focus group, when participants were asked what information might be helpful to receive on a 21 st birthday, responses almost exclusively entailed reminders about Protective Hi [participant name].Here are some tips to stay safe from the URI Young Adult Birthday Study: Keep track of how many drinks you have and space them out with water, eat beforehand, and have a sober driver ready.Enjoy your time with friends and make it a night to remember!PLEASE RESPOND "OK" so we know you got this.For both text messages, participants who did not reply received up to two follow-ups at 5PM and 6PM.The 5PM follow-up to text message one was: "We missed your response.Did you see our earlier message?If so, reply 'OK'."The 6PM follow-up to text message one was: "Just checking in again.PLEASE RESPOND 'OK' if you got our earlier message."The 5PM follow-up to text message two was: "We missed your response.Please respond 'OK' if you got today's message."The 6PM follow-up to text message two was: "Because we did not hear back from you, we assume you did not get our message.PLEASE RESPOND 'OK' so we know you got it."Participants who replied to text message one by 7 PM immediately received: Hi [participant name].Happy almost birthday from the URI Young Adult Birthday Study!Earlier, you said you would have W drinks on your 21st birthday celebration 3 .This is more than what X% of URI [males/females] drink on their 21st birthday.If you drink this much over Y hours, you will have a blood alcohol content of Z.This may result in [effect from Table 2] PLEASE RESPOND "OK" so we know you got our message.The number of anticipated drinks (W), time spent drinking (Y), and gender were taken from participants' replies on the baseline survey, and used to calculate Blood Alcohol Content (Z).The normative feedback component (X) was based on gender-specific 21 st birthday drinking data collected from 961 undergraduate students 21 years or older in April 2015 at the same study site.
Neighbors et al. (2009)ual 21 st birthday eBAC variables, scores greater than .50,whichwereconsidered likely impossible, were recoded to .50 4 as done inNeighbors et al. (2009).Data checking revealed that 21 st birthday eBAC, and 21 st birthday alcohol consequences were overdispersed and non-normal.Therefore, primary analyses utilized a negative binomial distribution where standard general linear model assumptions do not apply.

Table 1 .
-level preventive intervention studies could further refine the normative feedback component of this treatment to potentially enhance the overall intervention effect.Sample Characteristics Note.Based on the n=200 who were enrolled in the study.
7 I thank Dr. John Stevenson for this suggestion.event

Table 3 .
Perceived 21 st Birthday Drinking Norms

Table 4 .
Effect of Treatment and Anticipated 21 st Birthday eBAC on Actual 21 st Wald χ 2 for the Test of Model Effects are displayed.Degrees of freedom (df)=1 for all analyses.Anticipated 21 st Birthday eBAC and Actual 21 st Birthday BAC were calculated according to