Testing the Efficacy and Mechanisms of an SMS Intervention focused on Motivating Peer Network Support to Reduce Alcohol Consumption in Non-Collegiate Young Adults: Protocol for a Randomized Trial

Abstract

Background:

Non-collegiate young adults engage in high rates of heavy drinking but are less likely to access alcohol-related counseling or treatment. Peers play a significant role in shaping drinking behavior, yet few interventions target close peer influence in this population.

Methods:

This two-arm randomized controlled trial will enroll 300 young adults aged 18–25 who report 2+ heavy drinking days (HDD; defined as 4+ drinks for a woman and 5+ drinks for a man) in the past 30 days and are not enrolled in college. Participants are randomized (1:1) to receive either (1) an SMS intervention focused on self-monitoring and goal-related strategies (i.e. Commitment-based Binge drinking prevention Intervention: CBI) or (2) ASPIRE, which adds feedback on peer influences and encouragement for positive support. Outcomes are assessed via REDCap-administered surveys at baseline, 3, 6, and 12 months. Additional data include twice-weekly SMS assessments during the 3-month intervention period and GPS-based mobility tracking with simultaneous ecological momentary assessment (EMA) during 7-day windows following each assessment time point. The primary outcome is number of heavy drinking days (HDD) in the past 30 days. Mediation analyses will assess cognitive, social, and environmental mechanisms of change using multilevel structural equation modeling.

Discussion:

This trial advances the science of digital interventions for alcohol harm reduction in a high-risk and underserved population. ASPIRE is among the first interventions to integrate encouragement for positive peer network support into a scalable mobile platform. The use of high-frequency behavioral and geolocation data allows for novel insights into mechanisms of change.

1. Background

Young adults have high rates of unhealthy alcohol consumption such as heavy episodic (i.e. binge) drinking^1,2 often leading to negative immediate^3,4 and long-term⁵ consequences. Despite these risks, fewer than 20% of young adults engaging in unhealthy alcohol consumption receive any form of counseling or treatment.⁶ These challenges are especially pronounced among the roughly 38% of U.S. high school graduates who do not pursue higher education⁷ given reduced access to health insurance, preventive programs offered through colleges⁸, and often practical barriers—such as time, cost, and transportation—that make traditional in-person interventions less feasible.

Digital interventions offer a promising alternative. Meta-analyses support the effectiveness of SMS-based interventions in reducing alcohol consumption among young adults.⁹ Our prior randomized trials of SMS-based interventions focused on supporting self-monitoring and drinking limit goal processes showed reductions of approximately one heavy drinking day (HDD: 4+ drinks for a woman and 5+ drinks for a man) per month in mixed samples of college and non-college young adults.^10,11 However, these effects may be enhanced by addressing social processes, which strongly shape drinking behavior in this age group.

Consistent with Social Learning Theory¹², peers play a powerful role in shaping alcohol-related behaviors. There is evidence for bidirectional influence: young adults are drawn to peers who model similar drinking behaviors, while peer drinking itself can lead to increases in personal consumption through social modeling and misperceived norms.^13,14 Social network interventions have been found to be effective in behavior change across a range of behaviors.¹⁵ However, peer-led interventions have shown mixed effects on reduce alcohol consumption in young adults^16,17. Among individuals with more severe alcohol use disorders, social network interventions that help participants shift toward sobriety-supportive peers (i.e. alteration approach) have produced improved outcomes.^18,19 However, comparable interventions tailored for young adults with less severe alcohol use—where harm reduction rather than abstinence is the goal—remain underdeveloped.

A related gap in alcohol intervention science is the limited understanding of mechanisms of behavior change in young adults. While the strongest empirical support exists for self-efficacy, social support, and craving as mechanisms in general substance use behavior change²⁰, few studies have directly examined how these mechanisms operate in young adult populations. Our prior work has shown that desire (as a proxy for craving), goal setting, and goal confidence mediate intervention effects in this group.^21,22 As well, descriptive norms—young adults’ perceptions of how much their peers drink—constitute a well-supported mechanism of behavior change.²³

Environmental context could also play a role as a key mechanism linking our intervention to drinking outcomes. Neighborhood alcohol availability and location-specific drinking norms²⁴ influence both drinking behavior and related harms.²⁵ Environment and peers also interact to amplify substance use in young people, underscoring that who one is with and where one goes are intertwined drivers of risk.²⁶ Recent GPS-based ‘activity-space’ methods make this pathway testable by quantifying person-level exposure to alcohol-related settings (e.g., bars, high–outlet-density areas) and transitions into those contexts.²⁷ On this basis, we hypothesize that our intervention may reduce drinking in part by shifting time away from high-risk places and toward lower-risk contexts.

Objectives and Overview of This Trial

To address the need for interventions that more directly target peer influences, our team developed an SMS-based program that, in addition to self-monitoring and drinking limit goal support features, aims to influence close peer networks (i.e. promote seeking support for help limiting drinking). Specifically, we developed ASPIRE: Accountability Support Through Peer-Inspired Relationships and Engagement. ASPIRE uses weekly check-ins to motivate individuals to seek out peer support to help them reduce drinking and provides feedback on past week peer influences. In a pilot randomized trial, ASPIRE showed promising effects on both alcohol use and perceived social support.²⁸ Based on these results, we were funded by the National Institute on Alcohol Abuse and Alcoholism (NIAAA) to conduct a fully powered randomized controlled trial registered under NCT06617702 to compare ASPIRE to a text-message intervention without any peer network components.

In this methodology-focused paper, we describe the trial’s design and highlight four key implementation and measurement challenges addressed: 1) Recruitment of a nationally-recruited non-college sample: We detail our use of targeted online advertisements to identify and enroll eligible participants across the U.S. 2) Verification of authentic participants: We outline our multi-modal identity verification procedures, developed in response to growing concerns around fraudulent enrollment in digital trials.^29,30 3) Assessment of peer network change and influence: We use egocentric network data collected at baseline and follow-up to assess changes in the composition of participants’ close social networks. Additionally, 12 weeks of SMS-based assessments allow us to track week-to-week fluctuations in perceived peer pressure and support for reducing alcohol use. 4) Integration of passive and active behavioral data: We utilize the Effortless Assessment of Risk States (EARS) mobile sensing platform to collect GPS and ecological momentary assessment (EMA) data during one-week sampling windows at baseline, 3, 6, and 12 months, allowing us to explore how changes in daily activity and mobility patterns relate to changes in drinking behavior.

2. Methods

2.1. Study design

This study is a two-arm, parallel-group randomized controlled trial enrolling 300 young adults aged 18–25 years from across the United States. Recruitment is conducted using targeted internet advertisements. Following informed consent and baseline assessment, participants are randomized (1:1) into one of two 12-week SMS-based intervention arms: (1) a text message intervention uses weekly check-ins to motivate self-monitoring and drinking limit goal setting (CBI: Commitment-based Binge drinking prevention Intervention), or (2) ASPIRE. Primary and secondary outcomes will be assessed via self-administered, web-based surveys at baseline, 3-, 6-, and 12-months post-randomization. To evaluate whether ASPIRE differentially influences daily activity and social contact patterns, passive GPS data will be collected in 1-week bursts at baseline and at each follow-up time point (3, 6, and 12 months). The CONSORT diagram depicting participant flow is presented in Figure 1.

Figure 1.

CONSORT Diagram

2.2. Study procedures

2.2.1. Recruitment

We employ a multi-platform digital recruitment strategy, using paid social media advertisements and identity verification protocols previously shown to be effective in alcohol-focused studies involving young adults.^31,32 Recruitment and campaign management is conducted in collaboration with BuildClinical, a vendor specializing in participant recruitment for behavioral health research. BuildClinical deploys targeted ads and creates a custom study landing page. The advertisements are primarily placed on Facebook and Instagram, which continue to have high reach among young adults—70% and 71% usage rates, respectively, as of 2021.³³ Interested individuals are directed to the study website, where they complete a brief eligibility screening survey that includes questions on inclusion/exclusion criteria. We selected digital recruitment to enhance reach and efficiency while improving representativeness among young adults.³⁴ In our pilot study using similar procedures, we successfully enrolled 92 young adults from 31 states. To further diversify the sample, particularly with regard to rural representation, we adjusted ad targeting parameters to enrich recruitment from rural zip codes.

2.2.2. Participant eligibility

Inclusion criteria were as follows: 1) Self-report of 2+ heavy drinking days (HDD; defined as 4+ drinks for a woman and 5+ drinks for a man) in the past 30 days in the past month; 2) Aged 18 to 25 years; 3) Endorsement of “at least sometimes” thinking about changing their drinking behavior on the Contemplation Ladder³⁵; 4) Residency in the United States; 5) Fluency in English.

Exclusion criteria included: 1) Current enrollment in a 4-year college or graduate/professional school; 2) Highest level of education completed is a Bachelor’s, Master’s, or Doctoral degree; 3) Active military service; 4) Current incarceration; 5) Prior treatment for alcohol or other substance use disorders; 6) Drinking most frequently on Mondays through Wednesdays; 7) For individuals assigned female at birth: current pregnancy or plans to become pregnant during the study period.

2.2.3. Identity Verification and Validation Procedures

To ensure participant authenticity and data integrity, we implemented a multi-step verification protocol. Individuals can only access the pre-screen by directly clicking on the advertisement link. During pre-screening, we used IP address tracking and CAPTCHA to prevent multiple entries. After electronic consent, participants completed a four-item comprehension quiz; only those who answered all items correctly proceeded. To confirm phone ownership, a unique SMS password was required to log in—serving as two-factor authentication. Enrollment required a U.S. mobile phone number capable of receiving SMS from a major telecommunication carrier. We verified the number via SMS one-time passcode and carrier/line-type lookup (e.g. country = US; line type = mobile; carrier=Verizon). Numbers identified as VoIP or non-U.S. are ineligible. Each participant was issued a unique ID to access the EARS mobile sensing platform, and successful app installation was confirmed. We did not require video-based identity verification, as prior research indicated it is not desired in potentially stigmatizing conditions³⁶, and may introduce sample bias. We also check GPS coordinates to ensure device is being within the U.S. geofence (≥60 consecutive minutes of valid GPS within U.S. boundaries). Cases with no GPS indicating U.S. location in the first 7 days were removed from the study.

2.2.4. Randomization procedures

If all conditions above are met, eligible participants are assigned to study arms based on block randomization stratified by sex at birth (male/female) and AUDIT score (<8 vs. 8+) in 1:1 ratio to the ASPIRE and CBI study arms via REDCap³⁷. Participants are not informed of their treatment condition.

2.3. Study interventions

2.3.1. Overview

Both study arms deliver fully automated, two-way SMS interventions over a 12-week period using a platform hosted by Stanford University Research IT, integrated with REDCap, and delivered by Twilio, a cloud communications platform that provides application programming interfaces (APIs) to programmatically send and receive SMS through mobile carrier networks. Text messaging was selected for its broad accessibility and popularity among young adults.¹⁷ Participants receive assessments and prompts twice weekly—on Thursdays and Sundays—timed to coincide with typical binge drinking patterns.^38,39 Figure 2 demonstrates the logic models of CBI and ASPIRE. Message content was developed by behavioral scientists, written at a 6th-grade reading level for accessibility⁴⁰, and refined with input from young adults and expert reviewers. Libraries were finalized before enrollment and locked for trial duration to ensure rigor.

Figure 2.

ASPIRE vs. CBI Intervention Logic Model

2.3.2. CBI

All participants randomized to CBI (Commitment-based Binge drinking prevention Intervention) receive a fully automated, evidence-based 12-week SMS program designed for young adults with unhealthy alcohol use.^11,41 Grounded in cognitive and goal-setting principles, the intervention prompts participants each Thursday to report weekend drinking intentions. Those with no plans receive abstinence reinforcement; those planning to drink report their desire to get drunk.²¹ Based on desire level, they receive either reinforcement or messages incorporating motivational reflection. Participants planning to drink are asked whether they are willing to set a weekend drinking limit. Those who agree receive a protective behavioral strategy and rate their confidence in meeting the goal. Low-confidence responses trigger additional self-efficacy support. Those unwilling to set a goal receive a nonjudgmental message addressing ambivalence. On Sundays, participants report their maximum alcohol consumption. Those who drank receive follow-up questions on duration and peer influence. Participants who set goals receive feedback tailored to whether they succeeded—either reinforcing achievement or constructively reframing failure.^{42 43} No peer network feedback is provided.

2.3.2. ASPIRE (CBI + Peer Network Support)

Participants assigned to the ASPIRE (Accountability Support Through Peer-Inspired Relationships and Engagement) condition receive all components of CBI, along with additional messaging designed to promote reflection on peer influence, encourage social accountability, and strengthen relationships that support harm reduction. On Thursdays, participants who commit to a drinking limit goal are asked if they are willing to reach out to a trusted friend for support. Those who confirm willingness receive a peer communication strategy; those who decline receive a message encouraging reflection on their peer support network. On Sundays, additional Network Support messages are delivered. If participants report peer pressure, they receive a message prompting reflection on whether those peers align with their goals. If no pressure is reported, a reinforcing message emphasizes the value of supportive environments. Participants are also asked if anyone helped them adhere to their drinking goal. If so, they receive a message reinforcing that support; if not, they are encouraged to consider cultivating friendships that align with their behavior change efforts.

2.4. Data collection

We collect research data from three primary sources: 1) Web-based questionnaires administered via REDCap at baseline, 3-month, 6-month, and 12-month time points; 2) Text message responses from twice-weekly (Thursday and Sunday) assessments delivered over the 12-week intervention period; 3) Passive and active mobile sensing data collected via the EARS app. At each wave (immediately after baseline, and at 3, 6, and 12 months), participants completed 7 consecutive days of web-based EMA consisting of two same-day evening prompts and one next-day recap. The 8:00 PM prompt assessed the prior 3 hours (5–8 PM) and the 11:00 PM prompt assessed the prior 3 hours (8–11 PM). The Assessment Schedule and Measurements are outlined in Table 1. Participants will receive: $10 for completing the baseline assessment, $30 for the 3-month follow-up, $40 for the 6-month follow-up, and $50 for the 12-month follow-up. Additionally, participants will earn $6.25 per day of valid GPS data submission during the 7-day collection periods at each time point, for up to 28 days total, amounting to a maximum of $175. The maximum total compensation for completing all study components is $305.

Table 1.

Measures by Modality, Domain, and Time Point

Modality	Domains	Assessments (item count)	BL	Weekly × 12 Wks.	3 Mo.	6 Mo.	12 Mo.
Web-based	Demographics	Age (1), zip code (1), sex at birth (1), race/ethnicity (2), education (2), employment (2), living situation (1), biometrics (2)	◆
	Past 30-day alcohol consumption	TLFB Calendar (30)	◆		◆	◆	◆
	Past 30-day alcohol consequences	B-YAACQ (20)	◆		◆	◆	◆
	Alcohol use severity	AUDIT (10)	◆				◆
	Close Peer Network	IPI (12)	◆		◆	◆	◆
	Resistance to Peer Influence	RPI (21)	◆		◆	◆	◆
	Drug use	NM-ASSIST (7)	◆		◆	◆	◆
Text Message	Cognitive processes	Desire to get drunk scale (1)		◆
	Behavioral processes	Drinking limit goal commitment & confidence (2)		◆
	Peer support & pressure	Peer support (1); Peer pressure (1)		◆
EARS app	Location traces	Time at drinking locations (GPS)	◆		◆	◆	◆
	Semantic location labels	EMA (hourly from 5p-11p)	◆		◆	◆	◆
	Alcohol consumption	EMA (hourly from 5p-11p; next day recall)	◆		◆	◆	◆
	Friends drinking	EMA (hourly from 5p-11p)	◆		◆	◆	◆
	Alcohol consequences	EMA (next day recall)	◆		◆	◆	◆

Abbreviations: BL: Baseline; SMS=text messaging; Wks: Weeks; Mo=Month; S-UPPS-P: Short Urgency, Premeditation, Perseverance, Sensation seeking Positive Urgency; NM-ASSIST: NIDA-modified Alcohol Smoking and Substance Involvement Screening Test; AUDIT-C: Alcohol Use Disorder Identification Test-Consumption; TLFB: TimeLine Follow-Back calendar; B-YAACQ: Brief Young Adult Consequences Questionnaire

2.5. Measures

2.5.1. Alcohol Consumption

We will assess alcohol consumption at three distinct temporal resolutions: monthly, weekly, and hourly, using validated and technology-enabled tools to capture both retrospective and real-time drinking behavior.

1. Monthly: We will administer the web-based 30-day Timeline Followback (TLFB) to assess self-reported alcohol use at baseline and each follow-up. The TLFB has demonstrated strong psychometric properties in young adult populations⁴⁴ and has been validated for online administration.⁴⁵ The primary trial outcome is the number of HDD in the past 30 days. This threshold aligns with blood alcohol concentrations of ≥0.08 g/dL^46,47, a level at which the risk of alcohol-related harm increases significantly⁴⁸ and is widely used in alcohol intervention trials with young adults, facilitating cross-study comparisons. Secondary outcomes derived from TLFB include: Days with any drinking, and Drinks per drinking day (DPDD).

2. Weekly: During the 12-week intervention period, participants will receive weekly text message prompts asking them to report the maximum number of drinks consumed on any single day over the weekend (from Thursday through Sunday). These data will enable us to examine within-person changes in drinking patterns during the active intervention phase and may offer insights into temporal trends and treatment response.

3. Hourly: We will use the EARS platform to collect hourly alcohol use data during four 7-day sampling windows. We will calculate the number of standard drinks of alcohol consumed over each hour from 5p to 11p. When an hourly EMA drink count window is missing, we will use the next-day total-drinks recall to interpolate hour-level values. These high-resolution data will allow us to assess time- and location-specific drinking behavior in conjunction with GPS data and examine how drinking varies by context.

2.5.2. Alcohol-associated Consequences

We will assess alcohol-related consequences at two temporal resolutions: monthly and daily. At the monthly level, participants complete the 24-item Brief Young Adult Alcohol Consequences Questionnaire (B-YAACQ)⁴⁹ to capture the number of negative alcohol consequences in the past 30 days. At the daily level, we will use EMA via the EARS app to assess prior-day consequences which included a fixed checklist of alcohol-related consequences⁵⁰ for ‘yesterday’ (check all that apply): verbal argument/physical fight; injury; trouble with police; trouble with parents/other adults; unprotected sex; drove after drinking; rode with a drinking driver; none of the above. The EMA checklist complements—rather than replaces—our broader consequences assessment (via standard questionnaires at the scheduled survey waves). Participants are additionally asked to rate the perceived role of alcohol in each event, with response options: not at all, somewhat, very much, and completely. We will derive day-level outcomes including: (a) any consequence (yes/no), (b) count of consequences, and (c) an alcohol-attributed consequence indicator. This approach allows us to quantify both the occurrence and alcohol-attributable burden of acute harms, and to aggregate to wave-level summaries that can be linked to EMA drinking, peer context, and GPS-derived place exposure.

2.5.3. Other Outcomes

To assess alcohol use severity, we will administer the 10-item Alcohol Use Disorders Identification Test (AUDIT)⁵¹ at baseline and 12-month follow-up. To assess drug use, participants will complete the NIDA Modified Alcohol, Smoking, and Substance Involvement Screening Test (NM-ASSIST)⁵², designed to identify substance use disorders. Drug use will be examined both as a covariate and as a secondary outcome to evaluate whether participants may be substituting alcohol with other substances⁵³.

2.5.3. Close Peer Network and Influences

To assess how close peer networks are differentially influenced by ASPIRE versus CBI, we will administer a web-based modified Important People Inventory (mIPI)⁵⁴ at baseline, 3-, 6-, and 12-months. Participants are asked to list their three closest friends (using nicknames), based on evidence that close friends exert greater influence on alcohol use behaviors than broader peer networks.⁵⁵ Specifically, they are prompted as follows: “We are going to ask you questions about your three closest friends that you currently spend time with, at least once a month. Think of your best/closest friend, then your next closest, then third closest.” For each nominated friend, participants report: 1. In-person contact frequency in a typical month (0=about once/month; 1=about once/week; 2≈three times/week; 3=daily; 4=roommates/housemates). 2. Friend’s binge-drinking frequency (how often this person consumes ≥4 drinks when drinking) using an 8-level scale (0=not in past 6 months; 1=once in past 6 months; 2=less than monthly; 3=about monthly; 4=about every other week; 5=1–2×/week; 6=3–6×/week; 7=daily). 3. Pressure to drink (“How often does this person encourage or pressure you to drink more alcohol?”; 0=never, 1=some of the time, 2=most of the time, 3=every time). 4. Support if reducing drinking (“How supportive would this person be if you told them you wanted to reduce your drinking?”; 0=not at all, 1=somewhat, 2=very, 3=extremely).

Scoring.

Because influence may be concentrated in a small number of close ties, we summarize the peer environment using summative (“dose”) and peak metrics. For each nominated friend k, we map contact frequency to approximate monthly contact counts (1, 4, 12, 30, 60) and rescale influence variables to 0–1 (pressure: ÷3; comfort: ÷4; binge frequency: ÷7). We then compute: 1. Dose Peer Pressure= ∑_kContact_k × Pressure’_k; 2. Dose Support= ∑_kContact_k × Comfort’_k; 3. $\text{Peak}\text{Peer}\text{Pressure}=\underset{k}{\text{max}}\left({\text{Contact}}_k\times {\text{Pressure’}}_k\right)$; 4. $\text{Peak}\text{Support}=\underset{k}{\text{max}}\left({\text{Contact}}_k\times {\text{Pressure’}}_k\right)$. We will also similarly calculate peak and dose of descriptive norms using contact frequency and friend’s binge drinking frequency.

In addition to the mIPI data, we are collecting weekly SMS-based measures of peer influence throughout the 12-week intervention. Each Sunday, participants responded to: “How much did friends encourage you to drink this weekend? 0=Not at all, 1=Somewhat, 2=A lot.” This was followed by: “Did anyone help you limit or reduce your alcohol consumption? Yes or No.”

To capture trait-level peer susceptibility, we are administering the Resistance to Peer Influence (RPI) scale^56,57, which assesses the degree to which an individual is influenced by peer norms and behaviors. This measure will be used to explore whether more resistance to peer influence moderates effects of peer pressure on alcohol use.

2.5.4. Activity and Locations

To examine how participants’ activity spaces and movement patterns are influenced by ASPIRE versus CBI, we are collecting GPS coordinates from participants’ personal smartphones via the EARS app. Despite growing interest in the application of GPS technology in alcohol research^24,58, relatively few studies have leveraged these data, and none to our knowledge have linked real-time location tracking to behavioral change following digital intervention exposure. We will apply standard preprocessing procedures to derive “staypoints,” defined as clusters of consecutive GPS samples within a 300-meter radius for at least 10 minutes. To contextualize activity patterns, we will integrate EMA data, collected during the hours of 5:00 p.m. to 11:00 p.m., to assign semantic labels to each staypoint. Each hour, participants will be asked to “choose all the places you were between X and Y” (Home, Work, Friend’s house, Arts & Entertainment, Food, Nightlife spot, Outdoors & Recreation, Professional or Medical office, Spiritual, Shop or Store, Travel or Transport, Other). To capture the influence of alcohol-related environments, we define alcohol-related staypoints using three tiers of evidence: 1) Universal (Objective): GPS staypoints that spatially intersect with alcohol-serving establishments, as identified via external APIs (e.g., Foursquare, Google Places). 2) Idiosyncratic (Self-Referenced): Staypoints at locations where the participant has previously reported alcohol consumption (via EMA) on any prior or current day. 3) Secondary (Peer-Inferred): Staypoints where the participant has previously reported that friends were drinking. These classifications will be used to generate daily binary indicators and cumulative summary variables reflecting alcohol-exposed activity space.

3. Analytic plan

3.1. General approach

All analyses will be conducted following the intent-to-treat principle, i.e. all randomized participants will be analyzed according to their assigned intervention. Arms will be compared for differences in demographics (age, birth sex, race, ethnicity, educational level, work status) and pre-enrollment alcohol use and severity (i.e. AUDIT score). Our threshold for statistical significance will be α=0.05. Our data and supporting documents will be submitted to the National Institute on Alcohol Abuse and Alcoholism's National Data Archive (NDA) and will be available within three years of completion of the study.

3.2. Treatment efficacy

To leverage the longitudinal follow-up data, our analytic strategy will involve examining the data as a panel and estimating time effects (at 3, 6, and 12 months) by intervention arm (ASPIRE vs. CBI) using generalized estimating equations (GEE) to account for the correlation present in repeated measures. We specify the planned link function to use in the generalized linear models fit with GEE based on the expected distribution for each outcome; however, we will evaluate the distribution and model for poor fit. We will include a time by intervention interaction term to identify intervention differences at each time point. In all models, we will assume an exchangeable correlation structure and will adjust for the randomization stratification variables. All GEE models will adjust for a prespecified minimal covariate set: the baseline value of the outcome, randomization/stratification variables, and key baseline prognostic covariates (e.g., age and sex). This approach follows guidance that covariate adjustment with strong pre-treatment predictors increases precision without inflating type-I error and is compatible with marginal GEE estimation.⁵⁹

3.3. Mediation Analyses of Cognitive, Social, and Environmental Pathways

To examine mechanisms through which ASPIRE affects alcohol outcomes, we will conduct mediation analyses across cognitive, social, and environmental domains (see Table 2). We adopt the counterfactual causal-mediation framework to estimate natural indirect, natural direct, and total effects. For each mediator–outcome pair, we fit regression models appropriate to the distributions (linear for approximately normal, logit for binary, Poisson/negative binomial for counts) and compute effects via parametric regression + simulation with nonparametric/Monte-Carlo bootstrap confidence intervals; when treatment–mediator interaction is present, effects are reported with interaction-aware formulas. Guided by Social Learning Theory¹² and empirical evidence, we will model both within-person (e.g., day/weekend) and between-person (e.g., cumulative change) mediation pathways using multilevel and structural equation modeling (SEM). See example Mediation will follow the framework outlined by Kraemer et al.⁶⁰. For each pathway, we will estimate direct, indirect, and total effects, using posterior credible intervals to determine significance.

Table 2.

Planned Mediation Analyses

Mediator domain	Mediation Pathway	Mediator Type	Outcome	Level of Analysis	Statistical Model
Cognitive	Goal confidence Weekend-Level, scale)	SMS (scale, weekly)	Weekend Heavy Drinking Day (SMS, binary)	Within-person (weekends)	Multilevel SEM (2-level logistic)
	Cumulative goal confidence (12 Weeks)	SMS (proportion over 12 weeks)	TLFB Heavy Drinking Days (3-month, continuous)	Between-person	Path Analysis
Social	Peer Pressure/Support (Weekend-Level, continuous)	SMS (binary, weekly)	Weekend Heavy Drinking (SMS, binary)	Within-person (weekends)	Multilevel SEM (2-level logistic)
	Cumulative Peer Pressure/Support (12 Weeks, continuous)	SMS (proportion over 12 weeks)	TLFB Heavy Drinking Days (3-month, continuous)	Between-person	Path Analysis
	Change in Close Peer Pressure/Support, continuous	IPI-derived score (baseline to 3 months)	TLFB Heavy Drinking Days (3-month, continuous)	Between-person	Parallel Mediation Model
Environm ental	Alcohol Location Exposure (Daily)	GPS-based location (daily binary)	Total Daily Drinks (EMA)	Within-person (days)	Multilevel SEM (2-level linear)
	Change in Alcohol Location Exposure	GPS-based alcohol location days	TLFB Heavy Drinking Days (3-month, continuous)	Between-person	Path Analysis

Cognitive domain:

By increasing positive peer support, ASPIRE may increase drinking limit goal confidence, with those changes mediating effects on alcohol consumption. We will examine whether goal confidence mediates ASPIRE’s effects on weekend binge drinking. Weekly SMSreported goal confidence will serve as a within-person mediator of weekend binge drinking, while cumulative goal confidence over 12 weeks will be tested as a between-person mediator of 3-month TLFB-based HDD.

Social Domain:

ASPIRE may alter exposure to peers (i.e. reducing drinking by decreasing exposure to peer pressure and increasing supportive peers), with those changes mediating effects on event-level and monthly alcohol outcomes. Mediation models will assess SMS-based weekend-level peer pressure on weekend binge drinking. Additionally, change in Social Support for Drinking and Social Support for Help (from baseline to 3 months) will be tested as parallel mediators of 3-month HDD.

Environmental Domain:

ASPIRE may reduce drinking by decreasing daily exposure to alcohol-related staypoints, with those changes mediating effects on event-level and monthly alcohol outcomes. Daily alcohol-related staypoints will serve as a within-person mediator of daily drinking (EMA), while change in days at alcohol-related locations (baseline to 3 months) will be tested as a between-person mediator of 3-month HDD. Within-person predictors will be person-mean centered to isolate temporal effects. All models will adjust for baseline levels, treatment arm, and relevant covariates (e.g., age, sex).

We will probe robustness of the “daily exposure to alcohol-related staypoints” mediator along four dimensions: 1. Staypoint detection parameters: We will re-estimate exposure varying the dwell-time and distance thresholds used to detect stays (e.g., 5, 10, 20 min; 100–200 m). Shorter thresholds can capture transient pauses; longer thresholds capture more stable visits. 2. Place-exposure operationalization: We will re-estimate exposure based onTime-weighted spatial averaging (TWSA) as an alternative to simple buffers (minutes at risk weighted by spatial intensity), and kernel density estimation (KDE) surfaces of alcohol outlets to form continuous exposure fields and compare buffer-based vs TWSA/KDE mediators.⁶¹ 3. Temporal alignment and lags: We vary the mediator window relative to drinking: concurrent (same 3-h window), lead/lag ±1 window, and day-level (minutes at alcohol-related places) to evaluate temporality and displacement. We also test majority-place vs proportional-allocation of exposure when windows include multiple place types. 4. GPS data quality and missingness: We will handle missing GPS with (i) complete-case (primary), (ii) inverse-probability weighting for missing-by-design/availability, and (iii) multiple imputation sensitivity assuming MAR/MNAR, and (iv) substitution with EMA place categories to label evening windows lacking GPS, as recommended⁶².

3.4. Missing data

For intent-to-treat analyses, following current recommendations when outcomes are missing^63,64 and following best practices for alcohol trials⁶⁵ we will use multiple imputation procedures to generate accurate estimates of uncertainty, including checks for the missing completely at random (MCAR) assumptions using Little’s test⁶⁶, and value of invoking missing at random (MAR)⁶⁷ assumptions. After identifying potential predictors in the imputation of variables associated with missing outcomes, we generate N sets of imputed data, based on the highest fraction of missing information, per recommendations,⁶⁸ and engage checks for how well the imputation model fits the observed outcome data by inspecting their distributions in each treatment arm.

3.5. Sample Size

We have designed the study to determine whether ASPIRE is more effective than CBI in reducing HDD at follow-ups at the 3-month follow-up. Based on an expected pooled standard deviation of 4.17 and an expected difference between arms of 2.2 points (effect size d=0.53), we calculated that 106 participants per arm (212 total) would be needed to detect this difference with 90% power at α=0.05. We plan to enroll a total sample size of 300 to account for an expected attrition rate of 25%.

4. Discussion

This randomized controlled trial evaluates the efficacy of ASPIRE, a fully automated text message intervention targeting unhealthy alcohol use in non-college young adults—a population with high rates of binge drinking and limited access to traditional treatment. ASPIRE integrates cognitive strategies with peer network reflection and support to promote harm reduction. By comparing ASPIRE to a validated SMS intervention (CBI), the trial will be able to examine the unique effects of the peer-focused intervention components from other cognitive components.

Our trial has several key strengths. First, we are recruiting a national sample of non-college young adults using targeted social media ads, with additional outreach to enhance rural representation. Second, we are employing a multi-step identity verification protocol—including IP filtering, CAPTCHA, two-factor authentication, comprehension checks, and GPS-based de-duplication—to protect against fraudulent enrollment and ensure data integrity. Third, we will be assessing dynamic peer network influences using a modified Important People Inventory and 12 weeks of weekly SMS reports to capture shifts in perceived social support and peer pressure.

Fourth, we have integrated GPS and EMA data via the EARS platform to examine how real-time activity patterns relate to alcohol use. Finally, we use multilevel SEM to test cognitive, social, and environmental mediation pathways at monthly, weekly, and daily time scales. Together, these features represent a comprehensive approach to ensure findings are robust and reproducible.

5. Limitations

Although we employ a national recruitment with attention to diversity, participants must have smartphone access and be open to digital communication, which may limit generalizability. Self-reported alcohol use, while collected via validated tools, is subject to recall and social desirability bias. While EMA and GPS data improve ecological validity, compliance and data completeness may vary. The ASPIRE intervention encourages peer engagement; however, peer-support messages are intentionally generic and do not ‘pipe in’ or reference specific named peers from the mIPI roster. This design decision prioritized privacy and feasibility (avoiding misidentification and unintended disclosure) but may reduce personalization for some users; future versions will evaluate opt-in peer tagging and network-informed tailoring. Also, we do not independently verify peer behaviors or mutual support. The 12-week intervention duration may not sustain effects long-term. Lastly, ASPIRE emphasizes harm reduction, which may not align with abstinence-oriented treatment goals appropriate for individuals with more severe alcohol use disorders.

6. Conclusion

This trial is among the first to test the added value of motivating young adults to interrogate and alter close peer networks to help support alcohol consumption reductions, advancing the science of scalable, personalized intervention for an underserved, high-risk population. Results will provide insight into both efficacy and mechanisms, informing future efforts to reduce alcohol-related harms in young adults.

Figure 3.

Example Multilevel SEM Model