Implementation of a Cluster Randomized Controlled Trial: Identifying Student Peer Leaders to Lead E-cigarette Interventions

Abstract

E-cigarette use has been increasing among middle school students. Intervention programs to prevent e-cigarette initiation administered by authority figures are met with more resistance from youth compared to peer-led programs. Therefore, this study aimed to assess the feasibility, acceptability, and implementation process of using social network analysis (SNA) to identify student peer leaders in schools and train them to deliver e-cigarette prevention programming to their peers. Nine schools were recruited to participate in the study during the 2019–2020 school year. Schools were assigned to one of three conditions: (1) expert; (2) peer-random (selected peer-leaders would teach to random students); and (3) peer-fixed (selected peer-leaders would teach to assigned students based on nominations). Study participation varied by day due to school attendance, with 686 participants at baseline and 608 at posttest. Almost all students who did not complete the study resulted from the interruption of schools being closed due to COVID-19. Implementation issues fell into three categories: (1) scheduling, (2) day-of logistics, and (3) student group dynamics. Overall, the results showed positive satisfaction among teachers, who unanimously found the program appropriate for the grade-level and that peer-leaders worked well within their groups. Peer-led students—both random and assigned—reported having more fun and willing to tell friends to try the program compared to expert-led students. This study demonstrated the feasibility of implementing a peer-led e-cigarette prevention program for 6^th grade students, using SNA to provide intervention rigidity and validity.

1. Introduction

Electronic cigarette (e-cigarette) use has been rapidly growing among adolescents. The 2019 National Youth Tobacco Survey found current e-cigarette use among middle school students increased from 4.9% to 10.5% from the previous year (Cullen et al., 2019). Studies have consistently found that e-cigarettes may facilitate the uptake of traditional cigarette smoking among otherwise never-smoking youth (Barrington-Trimis et al., 2016; Primack et al., 2015, 2018; Soneji et al., 2017; Spindle et al., 2017).

Nationally representative data regarding perceptions, knowledge, and attitudes surrounding e-cigarettes suggest a simultaneous need and opportunity to conduct school-based education programs for youth. Approximately 80% of youth do not think e-cigarette use is harmful (Miech et al., 2018), and many have limited awareness of nicotine content. The e-cigarette company JUUL, which owns 70% of the e-cigarette market, advertises enhanced nicotine delivery; yet in a sample of 15–24 year-olds who used a JUUL product in the past 30 days, only 37% knew that its cartridges always contain nicotine (Willett et al., 2019). Awareness is also lacking for parents and teachers, increasing the risk of continued misinformation being provided to youth. Although 88% of high school teachers and administrators reported being somewhat or very concerned about student e-cigarette use, 34% of schools reported no formal communication from the school to parents about e-cigarette use (Truth Initiative, 2019).

Previous research suggests that higher rates of school smoking prevalence are associated with an increased risk of adolescent smoking (Alexander et al., 2001; Ali & Dwyer, 2009). This influence may occur through a variety of mechanisms, including within school norms regarding tobacco, peer networks, socioeconomic characteristics, misconceptions about usage, and school-specific tobacco control policies (Cavazos-Rehg et al., 2016; Lipperman-Kreda & Grube, 2009; Sheikh et al., 2017). However, schools also present an opportunity for intervention, as smoking-prevention programs can be administered in a comfortable learning environment (Taylor et al., 2016). Simulation models have found that school-based interventions can reduce long-term nicotine dependency (Chu et al., 2020).

Intervention programs that are administered by central authority figures are met with more resistance when compared with peer-led programs (Schillinger et al., 2017). Friends within social networks can be leveraged to produce behavioral changes (Schillinger et al., 2017). This is particularly true for youth, as they tend to emulate behavior of those whom they consider as a friend (Valente, 2012). Further, network-based interventions can be provided in group settings, which help to reinforce adoption of behavioral change (Schillinger et al., 2017). Peer leaders are able to help modify social norms, which can change perceptions surrounding tobacco products (Lipperman-Kreda & Grube, 2009). These changes can be more resilient because the individual experiences changes in the behaviors that they perceive as normal (Gesell et al., 2013).

Social network analysis (SNA) can effectively identify peer leaders, where various methods can be used to identify, select, and train peer leaders for health behavior interventions (Rogers, 2003). Identifying peer leaders through SNA has the benefits of being informed by the communication and social relational structure of the community as well as implementing a strategy that pairs leaders with those closest to them.

1.1. Study objectives

This primary objective of this project was to assess the feasibility, acceptability, and implementation process of using SNA to identify student peer leaders in schools and train them to deliver e-cigarette prevention programming to their peers. The study was conducted as a cluster randomized controlled trial with three conditions. As an exploratory objective, we examined responses to three student survey questions by time and study arm to demonstrate feasibility of analyzing student e-cigarette outcomes longitudinally under our trial design. Specific discussions regarding the implementation issues and resolutions during each step in the development and execution of this intervention are detailed. This study was prospectively registered on ClinicalTrials.gov (NCT04083469) and approved by the IRB at the lead author’s university.

2. Methods

2.1. Participants and recruitment

Nine schools were recruited in the Pittsburgh area to participate in the study during the 2019–2020 school year. The schools—a convenience sample based on relationships with team members in past research studies—comprised of students with a heterogenous mix of gender, racial, and economic statuses representative of the region (Table 1). At each school, a member of the study team conducted an informational meeting to describe the study background and protocol. The meetings typically included the principal, the guidance counselor, and the teacher of the class in which the intervention would occur. Following the meeting, the principal or classroom teacher presented the proposed intervention to the school board. Upon approval from the school board, the research team worked with the schools to schedule the intervention.

Table 1:

School demographics

	Schools
	S1	S2	S3	S4	S5	S6	S7	S8	S9
Female (%)	50	50	53	43	50	50	50	47	50
Minority (%)	7	12	7	15.3	16	10	37	33.6	68
White (%)	93	88	93	84.7	84	90	63	66.4	32
Black (%)	<1	2.8	2	8.5	6.8	4	32	22	65
Hispanic (%)	<1	2.5	1	<1	<1	<1	<1	1.3	2
Asian (%)	<1	1.7	<1	<1	<1	1	<1	<1	<1
Free lunch eligible (%)	49	29	39	30	50	45	98	57	100
6th grade students	93	194	92	299	231	126	98	174	266

Parents were sent a letter explaining the research, questionnaires, and assessments. Parents were also be provided with the principal investigator’s contact information and encouraged to call with questions or concerns. The letter included a form that parents could complete and return to the PI to opt their child out of participation.

2.2. Randomization

An independent statistician generated an allocation table using Stata’s ralloc function, with schools stratified by demographic composition (delineated by greater or less than 67% white). That allocation table was provided to a research assistant, who assigned schools into the allocation table based on the order of date of agreement to participate in the study. Schools were assigned to one of three conditions: (1) expert; (2) peer-random (selected peer-leaders would teach to random students); and (3) peer-fixed (selected peer-leaders would teach to assigned students based on nominations). Generally, school-based interventions are randomized at the school level (Graham et al., 1984). The cluster randomized design was chosen based on its simplicity for our pilot study, suggestions from school administrative staff, and abatement of comparisons and contamination between conditions.

2.3. Data collection

There were three student survey assessments conducted at each school: social network, baseline e-cigarette pre-test, and e-cigarette post-test. The post-test survey also included questions to help assess the students’ opinions on the acceptability of the intervention. This manuscript reports the results of the social network and acceptability post-test survey. A member of the research team conducted each survey during regular school hours in a standard class period. Surveys were provided to students on paper or electronically depending on the preference of the school. The network and baseline pre-test surveys were administered to students during the first session. The final post-test survey was administered the day following the last intervention day. After program completion at each school, teachers were asked to complete an anonymous online survey of on how the program was conducted. Likert-scale questions included whether the program fit into their class schedule, if the level of instruction was appropriate for the grade, and, in the peer-led arms, whether the groups functioned well.

2.4. Social network analysis

Every student—regardless of school assignment—completed a social network survey. The survey asked students to list five people they consider to be their friends. Degree centrality, specifically weighted in-degree, was the primary metric to identify peer leaders who were most likely to be influential change agents or promote attitudinal and behavioral change regarding e-cigarette use. First order selections were weighted higher than subsequent choices. Both the size of each assigned peer group and the number of peer leaders were informed by SNA and varied in each class; approximately 4–6 students were targeted to each peer-leader. Each of the nomination categories were evaluated to ensure the robustness of identification.

2.5. Intervention program development

The intervention combined components from three separate programs: (1) CATCH My Breath, an extension of the CATCH (Coordinated Approach to Child Health) program (Arkin et al., 1981; Luepker et al., 1996), an e-cigarette prevention program designed for middle and high school students (Franks et al., 2015); (2) Stanford’s Tobacco Prevention Toolkit, an evidence-informed resource that can be adapted to fit the needs of educators and students in elementary, middle, or high school (Stanford School of Medicine, 2019); and (3) on-going work from UPMC Hillman Cancer Center’s Healthy Choices for Students School Program. This study adapted and combined modules most effective in reducing youth tobacco use, including counter-marketing techniques, developing confidence, refusal skills, self-regulation, and normative education (CATCH, 2019; Franks et al., 2015; Leon et al., 2011; McKenzie et al., 2001).

The program consisted of 3 components. First, an introduction was provided about e-cigarettes, nicotine, and flavors. This included a discussion focused on the addictive nature of nicotine and how flavoring can appeal to youth. Second, a hands-on activity provided students with examples of real e-cigarette advertisements. Students were asked to critically analyze the advertisements, identify overarching themes, discuss how the advertisements were designed to appeal to youth, and identify the negative consequences of e-cigarette use that were being left out of the advertisements. Students were then asked to create an alternative advertisement to counter these marketing techniques. Finally, students discussed refusal skills, focusing on different contexts such as family use of e-cigarettes and peer pressure.

2.6. Intervention delivery

There were five members of the research team that were responsible for delivering the intervention. After the development of the intervention curriculum, all members of the research team were trained on the content. School teachers did not perform any instruction or intervention delivery.

In each school, regardless of condition, experts administered the initial sessions. All students received one introductory session, during which the pre-test and social network surveys were administered. In the expert-led condition, students received two 45-minute informational sessions, and one final session during which the post-test was administered. The intervention was delivered by a member of the study team to each classroom in a traditional manner, with students in their normal arrangement and setting. In the peer-led schools, the experts trained the selected student peer leaders, which consisted of two additional 45-minute training sessions between the introductory session and the program. The peer-leaders then administered the two 45-minute informational sessions. The intervention for both peer-led arms was delivered with students seated with their small groups. Both the experts and teachers served a supporting role to assist students if they had questions. On the last day when the program was completed, a celebration party was held.

2.7. Data analysis plan

Student responses were reported for each acceptability survey question, separated by condition. A logistic regression analysis was conducted to determine if program appeal varied by condition, adjusting for fixed effects to control for school clusters. From the SNA data, a multivariable linear regression analysis was conducted to compare the number of friends selected in each of the three program conditions while accounting for the fixed cluster effect of schools. Results of the teacher questionnaire were reported descriptively. Implementation fidelity and barriers to implementation are discussed in 3.4.1. Lastly, we analyzed three survey items longitudinally: (1) intent to try e-cigarettes, (2) intent to try e-cigarettes if offered by a friend, and (3) opinion whether e-cigarette companies try to get young people to vape. Responses were dichotomized as appropriate. Logistic regression models were fit as functions of condition, time, and condition-by-time interaction; models were augmented with random intercepts to account for repeated measures. We estimated odds ratios comparing pre-post improvements between conditions with Bonferroni adjustments for two comparisons per question. All analyses were conducted using Stata version 14.2. A 5% significance level was assumed for all analyses, and no further adjustments were made for multiplicity.

3. Results and Discussion

In response to the COVID-19 pandemic, all Pennsylvania state schools closed for in-class instruction on March 13, 2020 for the remainder of the academic year. Thus, we were able to partially or fully complete the program in seven of the nine schools.

3.1. Feasibility metrics

Table 2 describes how the program was implemented in each school. There were 686 students in attendance on the first day of the program, and 608 on the final day; 17 students chose to opt-out (2%). Almost all students who missed the final day resulted from the interruption of schools being closed; retention was nearly 100% in schools that completed the study before closures.

Table 2:

Program timeline showing who delivered program or training on each day

	Intro	Training 1	Training 2	Program 1	Program 2	Final
Expert	Expert	N/A	N/A	Expert	Expert	Expert
Peer-random	Expert	Expert	Expert	Random student	Random student	Expert
Peer-fixed	Expert	Expert	Expert	Nominated student	Nominated student	Expert

3.2. Acceptability metrics

Acceptability items from the student survey found high overall satisfaction with both peer-led and expert-led conditions (Table 4). However, statistically significant differences were found between students in different conditions for questions about having fun and willing to tell their friends to complete the program. Similarly, a post-intervention survey was sent to all teachers that participated in the study to assess their satisfaction with the study. Each question was measured on a 5-point Likert scale. Overall, teachers responded positively to the program (Table 5). The primary results are presented with schools as-treated (i.e., schools/students are defined by how the intervention was presented). To account for a class that was conducted in the incorrect arm, sensitivity analyses with intention-to-treat (i.e., incorrect class was categorized by their originally assigned arm) as well as per-protocol (i.e., incorrect class was removed from analysis) were included to compare the impact on the as-treated results (Brown et al., 2015; Yelland et al., 2015).

Table 4:

Student post-test acceptability survey

Question	Expert	Peer-random	Peer-fixed
Surprised by what I learned (%)	65	70	80
ITT (Intention-to-Treat)	70	59	80
PP (Per-Protocol)	70	70	80
Lessons were helpful (%)	90	95	93
ITT	90	92	93
PP	90	95	93
Change how I view e-cigarettes (%)	78	85	77
ITT	78	81	77
PP	78	85	77
Fun (%)	80**	90**	88**
ITT	Q2**	Q2**	88**
PP	Q2**	90**	88**
Tell friends to do program (%)	81*	88*	90*
ITT	82*	83*	90*
PP	82*	88*	90*
Help get rid of e-cigarettes (%)	89	96	91
ITT	89	92	91
PP	89	86	91

^*<0.05

^**<0.01

Table 5:

Teacher acceptability survey

Question	Avg
Satisfaction with program (n=12)	4.25/5
Program accommodated class schedule (n=12)	4.58/5
Would run program again (n=12)	4.83/5
Would recommend program to others (n=12)	4.83/5
Program appropriate for grade level (n=12)	5/5
Peer-leaders worked well with groups (n=6)	5/5

3.3. Social network survey

In the social network survey, students selected an average of 3.9 friends (SD=1.4). Results of the MVLR (as-treated) showed that program condition had a significant effect on the number of friends selected [F(2,656)=7.28, p<0.001]. For comparison, the results for intention-to-treat [F(2,656)=7.28, p<0.001] and per-protocol [F(2,560)=7.63, p<0.001].

3.4. Implementation fidelity and barriers to implementation

During the intervention, we encountered several real-time implementation issues: (1) scheduling, (2) day-of logistics, (3), student group dynamics. Most relevant to the reported results, one class from a peer-random school incorrectly began as an expert-led class; this error was caught and the remaining classes in the school were correctly taught as peer-led.

3.4.1. Implementation barriers: participants

Although the heterogenous mix of students allowed for a diverse study with more generalizable results, the differences in the schools presented potential barriers to program implementation. Pittsburgh area schools include 6^th grade instruction at the elementary (Kindergarten through 6^th grade) and middle (6^th through 8^th grade) levels. Our study included two elementary schools and seven middle schools. The program was administered in different courses—including gym, health, or science classes—and settings—such as auditoriums or traditional classrooms. Most importantly, it was necessary to adapt to different school schedules; some schools allowed for a consecutive schedule where the study team engaged with students in consecutive days from beginning to end; other schools allowed for weekly meetings; other variations existed, depending on the scheduling logistics of the school calendar. To reduce scheduling burden, the program was condensed to the minimum necessary classroom lessons.

3.4.2. Implementation barriers: data collection

There were several issues with the ability of students to read the survey. In most cases, these students were supported by a teacher’s aide. As the school year progressed, rosters changed due to student relocations which may not have been reflected on the original rosters sent to the study team. To ensure no duplication in roster IDs within a school, the students that were not on the rosters were given large unique numbers (e.g., 1000) as their roster ID. Students that were on the roster but had moved were removed from the roster. Schools that electronically administered the survey encountered issues with technology, using the paper survey as backup.

3.4.3. Implementation barriers: social network analysis

When the initial class network graphs were developed, groups were initially defined based on network modularity (Newman, 2006). Modularity analysis applies a community detection algorithm to identify sub-groups within a larger network. From each subgroup, the person with the highest in-degree could be selected. However, to account for the possibility of a peer-leader being absent or choosing not to participate, a backup peer leader was chosen for each group. This process could not be automated; while the highest in-degree nodes in each sub-group would inherently be connected to many members within the same group, this was often not the case with other nodes in the group. We chose backup peer-leaders based on several potential criteria, including (listed in descending priority): (1) within-group in-degree, (2) proximity to and strength of tie with the peer-leader, and (3) balance of group sizes. These criteria also served as the basis for group assignment in the peer-fixed condition. As a result, there were instances where group learners did not directly nominate their group leader. However, we strived to minimize the network distance between leader and learner while maintaining practical group sizes (Valente et al., 2003).

3.4.4. Implementation barriers: intervention development

The creation of the intervention curriculum involved multiple rounds of review with the study team and health education experts. The health education experts aided in selecting a curriculum that was age appropriate (e.g., reading level) and helped form the schedule of daily activities. Based on recommendations from school administrators and teachers, the program was designed to be taught in two 45-minute classes, in addition to the introduction pre-test and post-test sessions. To ensure fidelity of the intervention, the contents of the intervention were maintained across peer- and expert- led interventions.

3.5. Longitudinal e-cigarette outcomes

Analysis of longitudinal survey responses demonstrated our ability to formally examine treatment effects under this 3-arm design. Though between-condition comparisons were statistically nonsignificant, results suggest that the peer-fixed condition produced at least as much improvement over time than other conditions (Table 7).

Table 7:

Longitudinal Outcomes: Intent to Use E-Cigarettes, Opinion of E-Cigarette Companies

Question	Expert		Peer-random		Peer-fixed		Peer-fixed vs Expert	Peer-fixed vs Peer-random
	Pre	Post	Pre	Post	Pre	Post	Odds Ratio (97.5% CI) a	Odds Ratio (97.5% CI) b
Q1: Try e-cigarette soon (%)	3	3	3	3	9	6	0.31 (0.03, 3.59)	0.34 (0.01, 13.09)
ITT (Intention-to-Treat)	2	2	4	5	9	6	0.42 (0.03, 6.30)	0.25 (0.02, 3.71)
PP (Per-Protocol)	2	2	3	3	9	6	0.47 (0.04, 6.30)	0.39 (0.01, 12.07)
Q2: Try e-cigarette if offered by friend (%)	3	5	3	3	11	12	0.63 (0.09, 4.42)	1.22 (0.04, 33.17)
ITT	3	3	3	7	11	12	1.23 (0.14, 10.77)	0.35 (0.04, 3.48)
PP	3	3	3	3	11	12	1.24 (0.14, 11.07)	1.23 (0.04, 34.90)
Q3: E-cigarette companies target young people (%)	84	93	93	97	74	94	3.39 (0.65, 17.76)	3.76 (0.26, 54.12)
ITT	82	93	92	95	74	94	2.80 (0.51, 15.19)	6.12 (0.83, 45.23)
PP	82	93	93	97	74	94	2.80 (0.51, 15.30)	3.79 (0.26, 55.20)

^*<0.05

^**<0.01

^aCompares odds of yes on post vs pre-test, between Peer-fixed and Expert groups (lower values favor Peer-fixed for Q1 & Q2)

^bCompares odds of yes on post vs pre-test, between Peer-fixed and Peer-random groups (lower values favor Peer-fixed for Q1 & Q2)

^a,b97.5% confidence intervals (CIs) are presented to adjust for two pairwise comparisons within each survey question.

4. Conclusions

This study demonstrated the feasibility of implementing a peer-led e-cigarette prevention program for 6^th grade students. Overall, the results showed positive satisfaction among teachers and students. In the primary as-treated analysis, peer-led students—both random and assigned—reported higher satisfaction scores than expert-led students on 5 of the 6 questions, although only two were at statistically significant levels (Table 4). The exploratory longitudinal results were nonsignificant, although this pilot was not powered to detect meaningful differences. However, these data served to show the feasibility of our analysis plan and will provide the template for a larger trial.

Figure 1 is an example of a classroom social network from this study. The community detection algorithm identified five clusters, denoted by five separate colors in the figure. The peer leaders and backups were 111(96), 110(98), 99(87), 88(102), 90(86). In each sub-group, there is an obvious peer-leader that received the most nominations; selecting backups was more complicated. Students 98 and 102 were clear choices based on in-degree; students 96, 87, and 86 were selected based on their direct ties to the peer-leader and more structurally central positions. There were 2 students (93 and 91) that were peripherally at the edge of multiple groups and assigned to a group outside their algorithm-defined sub-group to balance group sizes. Our decision to identify backups was immediately merited when a peer-leader was absent for the first instructional day.

Figure 1:

Example of one class network. Each node is a student, connected by directed ties that represent a close friend nomination. The thickness of each tie represent strength, where 1^st order nominations are weighted twice as much as 2^nd order nominations. Larger node sizes denote in-degree, or more nominations. The color represents clusters, as determined by a community detection algorithm.

Using SNA provided both intervention rigidity and validity to the study. In several instances, teachers brought up concerns about the assigned peer-leaders and questioned the methodology. Their worries were alleviated when research team members described the relationship between the student surveys and the network graphs, demonstrating both its power and our agnostic approach. Anecdotally, teachers were all pleasantly surprised that the students they were concerned about performed above expectations.

The significant difference in the number of friends nominated between conditions could be a result of the small sample size, as only seven schools were able to participate in the study due to Covid-19 closures. Another possibility is that 6^th grade student networks could differ between elementary and middle schools. This should be explored as the project is expanded to more schools, especially as the sensitivity analyses showed significant differences. The results also suggest that it might be worthwhile to explore randomizing at the classroom level rather than the school level.

Other implementation issues required changes by the study team. Concerns during the development and setup of the intervention were resolved prior to the conduct of the program. During the intervention, the most severe implementation issues were difficulties in student interactions, incorrect arm assignment, and problems understanding the survey instruments (Table 3). However, each problem was mitigated through monitoring, immediate reactions, and collaborative resolutions. While student interactions cannot be predicted, particularly in the peer-random arm, the issue resulted in more closely monitoring group activities in order to keep students on task, and to solicit more input from teachers on which groups could be potentially problematic. Through regularly scheduled data monitoring meetings, a class that had started in the wrong arm was quickly identified. The class was completed in the incorrect arm to minimize disruption, and all additional classes at that school were subsequently in the correct arm. Student difficulty in understanding the survey instruments required an immediate study team meeting. The issues were discovered to be based on the layout and the formatting was quickly adjusted. Overall, each issue was identified due to a constant evaluation processes that had been put in place before the intervention started. These monitoring efforts were successful, as the research team convened quickly to discuss problems, make appropriate changes, and resolve all cases before the intervention continued in order to maintain the integrity of the study.

Table 3:

School characteristics

School	Assignment	Schedule	Survey	Room	Class	Attendance at Day 1	Attendance at final day
S1	Expert	4 classes with sessions 1-day after another	Paper	Classroom	Health	64	67
S2 ^	Peer-random	1-week straight	Paper	Classroom	Science	81	76
S3	Peer-fixed	1-week straight	Paper	Classroom	Science	75	76
S4	Expert	4 classes 1-day after another	Digital	Classroom	Science or history	233	203b
S5	Peer-random	A/B pattern across 2-week spans	Paper	Auditorium	Gym	98	102
S6	Peer-fixed	1-week straight	Paper	Classroom	Science	94	41b
S7*	Expert	4 classes with sessions on consecutive Wed	Paper	Classroom	Health	41	43
S8*a	Peer-fixed
S9*a	Peer-random

Note: in some cases, more students completed program than started due to absentees during introduction

^*at least 33% minority

^acancelled due to Covid-19

^bsome students did not complete full program due to Covid-19

Table 6:

Protocol deviations

Category	Event	Severity	Corrective action
Group	Students note that no one in the group gets along. Interactions between group members continue to cause issues causing disciplinary action from the classroom teacher.	High	The group members remained the same to maintain fidelity of the intervention. The group was more closely monitored in subsequent days and study team members or school staff intervened if necessary.
	3 students dropped out of being peer leaders with no explanation. One student dropped out due to anxiety.	Low	Groups were formed with a back-up leader to prevent issues if peer leaders decided not to administer the intervention.
	A student was not assigned to any group despite completing the network survey.	Low	The student was randomly assigned to a group and their responses were excluded from the final analysis.
Schedule	Training Day 1 was modified to a block schedule resulting in a shorter period.	Medium	Teachers were able to slightly modify the block schedule to ensure there was enough time for the students to be trained.
	The celebration party was rescheduled multiple times due to close proximity to the holidays. Due to this, no study team member could be present.	Low	Study team members submitted a statement to the classroom teacher for them to read thanking the students.
	There were multiple reschedules due to schoo! events (dodgeball tournament and dance recital).	Low	The study team was able to reschedule the sessions and consulted the school events calendar when doing so.
	There was a fire drill that cut into class time.	Low	Discussion time was cut down to ensure that there was enough time to complete all activities.
Logistical	A school randomized to be peer-led incorrectly began 1 class with expert-led instruction.	High	Students in the incorrect class completed as expert-led. Additional classes in the school were correctly taught as peer-led.
	The network survey proved too difficult for students to complete in the first version. The students were unable to retake it resulting in a loss of baseline data for the first group.	High	The issues were specifically based on the layout (rather than reading level) and the formatting was quickly adjusted.
	Students had opted out of participating in the study but school staff called home to explain the study to parents and ask them to allow their children to participate.	Medium	No corrective action was taken since the actions were outside of the scope of the study team.
	Students were excluded in the network analysis because they were absent on the day of the network survey.	Medium	A study team member administered the intervention to the students and their responses were excluded from the final dataset.
	Students were taking the survey on their Chromebooks. One Chromebook would not work.	Low	The teacher allowed the student to take the survey on his computer. Steps were taken to ensure that the Chromebooks were working prior to starting the survey in subsequent rounds.
	A technical error occurred with creation of Roster IDs in Excel causing duplicates.	High	A study team member wrote student names and new roster IDs on the chalkboard for the students to use. In subsequent rounds, the Roster ID assignments were checked by at least two members of the study team.
	On intervention day, a teacher was absent without a substitute. The 4 classes needed to be divided into 3.	Low	The teachers worked with study team member to ensure that classes were divided to keep group members within the same class section.

• Social network analysis can identify peer leaders for e-cigarette interventions.

• Students in the peer-led intervention had more fun and were more willing to share their experiences than those in the authority-led program.

• Students enjoyed learning about e-cigarettes from their peers.

• Teachers found the peer-led program engaging and age-appropriate.

• Peer-led e-cigarette prevention program was feasible and acceptable.