The study design is described in greater detail elsewhere (33, 34), including more information on sample recruitment and selection as well as study procedures. All study protocols, including obtaining informed consent to participate, were carried out in accordance with relevant guidelines and regulations, and approved by PIRE’s Institutional Review Board (federal-wide assurance: FWA00003078). As participants were all under 18, we obtained parental consent and participant assent for all 50 participants.
Recruitment. Adolescents were recruited to participate in surveys that asked about opinions, perceptions, marketing exposures, and use of e-cigarettes and tobacco through a study website and recruitment flyers. Inclusion criteria were: (a) being between ages 14 and 17 years old, (b) living within a 100-mile radius of Louisville, Kentucky, and (c) self-report past two-week ENDS use. To maintain confidentiality and avoid potential response bias, neither parents nor the adolescents were informed that past two-week ENDS use was an inclusion criterion. Parental consent and youth assent were obtained electronically.
Participants. The first 50 eligible adolescents to complete the screening were enrolled. Of those, three did not complete the initial online survey and did not progress to the EMA surveys, and one dropped out of the study on the second day of EMA. Those four adolescents were replaced by the next four eligible adolescents on our recruiting list to obtain a sample of 50 adolescents (to ensure adequate statistical power). Participants were asked to complete EMA surveys over two weeks (14 days). On average, each participant provided 13.4 days of EMA data, resulting in 670 of 700 possible observations for a 96% completion rate.
Study Design
Data were collected January through October 2018. Participants were sent a link via text or e-mail to a 30-minute initial online survey that included demographics, individual characteristics, and ENDS and tobacco use behaviors. Next, an online text messaging platform was used to send survey links and reminders to a cell phone provided to participants by the study team. On each of 14 consecutive days, participants were asked to respond to the EMA surveys to report vaping and tobacco use behaviors, reasons for vaping, and contexts of vaping (where, with whom, vaping exposures). Incentives for study participation included: $15 for completing the initial survey, $5 for each completed EMA survey, a $20 bonus if they completed all EMA surveys, and $15 for returning the study phone.
On Mondays through Thursdays, participants were surveyed once daily, asking them about the past 24 hours (i.e., from 4pm the prior day to 4pm that day). On Fridays, they reported at two timepoints (i.e., from 4pm the day prior to 4pm that day, and from 4pm to 8pm that day). On Saturdays and Sundays, they reported at three timepoints (i.e., from 8pm the prior day to 11am that day; from 11am to 4pm that day, and from 4pm to 8pm that day). Data collection was set up this way to capture more data when the adolescents were not in school when we suspected there would be more use (i.e., the weekend). We aggregated multiple datapoints into one daily measure representing each 24-hour period so that all observations represented the same period of observation.
Initial Survey Measures
Individual characteristics. Participants were asked to respond to questions on individual and family characteristics including age, sex, and race/ethnicity. Socioeconomic status was assessed through one item asking how much money they had in a typical week to spend on whatever they want, not including basic necessities. The nine response categories were coded and rounded to the next integer to approximate an interval measure by taking the midpoints (<$5 = $4; $5-$10 = $8; $11 to $25 = $19; $26 to $50 = $39; $51 to $75 = $64; $76 to $100 = $89; $101 to $125 = $114; $126 to $150 = $139; and >$150 = $151).
Daily Measures
ENDS use and dual use with tobacco cigarettes. Each survey asked adolescents to report if they vaped nicotine during the survey window. If they said yes, they were asked how many times they vaped nicotine (i.e., number of occasions) during the survey window and how many total puffs they typically vaped on each occasion. To create daily vaping measures on weekend days, we summed use occasions across all survey time windows to equal a 24-hour period. The total number of vaping puffs per day was estimated as the product of the sum of use occasions and the mean of typical puffs per day. At the 4pm survey each day, participants were also asked if they smoked tobacco cigarettes during the survey window, and a dual use outcome was created representing use of tobacco cigarettes and ENDS on the same day. We coded outcome measures to represent (a) dual use, (b) ENDS use only, (c) cigarette use only, or (d) neither. We coded cigarette use only occasions as missing data in our substantive analysis, due to the small number of observations (n = 56).
Contextual factors of ENDS use. If adolescents said yes to vaping nicotine during a survey window, they were asked a series of questions about the last time they vaped nicotine, including: who they vaped with (recoded to by myself versus with others), where they vaped (recoded to someone’s home, my home, outdoors, or school), and why they vaped on the last occasion (recoded to it’s easy to get, I like the flavors, it feels good, trying to quit tobacco cigarettes, no odor, or because tobacco is prohibited). Since weekend data were aggregated across multiple observations to the daily level, we selected context responses from the last vaping occasion to make these observations consistent with weekday observations.
Community-level factors of ENDS use. At the 4pm survey each day, participants were asked if, in the last 24 hours (i.e., 4 pm to 4pm), they were exposed to others’ use, warning messages about vaping, and advertisements about vaping. For exposure to others’ use, participants were asked if they saw any adults (yes, no) or people their age (yes, no) using e-cigarettes or vape devices. For exposure to warning messages, they were asked whether they saw or heard any health warning messages about vaping (yes, no). For exposure to ads, they were asked whether they saw e-cigarettes or vape devices in various types of media (i.e., inside or outside of a store or on a billboard in or near your neighborhood, inside or outside of a store or on a billboard in or near your school, online or social media, and in a magazine/TV/movie). Responses for ad exposure were re-coded into an overall daily exposure to ENDS advertising variable (0–4 exposures).
Analysis
Given the focus of our study on contextual and community factors related to ENDS use and dual use of ENDS with tobacco cigarettes, our analyses excluded days on which only tobacco cigarette use occurred. We then excluded from all analyses days on which participants reported exclusive tobacco cigarette use (n = 56). Our primary analyses were random intercept, multilevel, binomial and multinomial logit regressions, accounting for daily observations nested within individuals. Binomial models were used for contextual factors since these measures were only asked when adolescents reported ENDS use (either exclusive or dual use), and multinomial models were used for community-level factor predictors. Specifically, binomial models examined exclusive ENDS use (referent category) vs. dual use, and our multinomial models examined exclusive ENDS use, dual use of ENDS and tobacco cigarettes, and no ENDS or tobacco cigarette use (referent category). There was a great deal of variability among participants (i.e., daily changes in use patterns occurred for participants), as evidenced by large intraclass correlation coefficients (ICC) (see Table 1). Substantive models regressed outcome on predictors, where predictors were grouped into topical areas (i.e., where vaped, why vaped, who with when vaped, exposure to adults vaping, exposure to peers vaping, warnings, and advertising) and we ran separate models for each topical area. We also explored whether polynomial growth patterns (linear, quadratic, and cubic) within a periodicity of one week significantly explained our dependent measures; however, as they did not, they were excluded from our models. Odds ratios and their accompanying 95% confidence intervals were calculated for models. Binomial models were fit using the lmer library (35) in the R environment for statistical computing (36) and multinomial models were fit using Stata 15 (37).
Table 1
Percentages, means (and standard deviations) of sample characteristics.
Sample Sizes | |
Individuals | 50 |
Observations | 670 |
Demographics | |
Age | 16.22 (.86) |
Male | 42% |
White race | 90% |
Spending money / week | $52.08 ($49.47) |
Type of Use (EMA) | |
Exclusive ENDS use days (ICC = .51) | 44% |
Dual use days (ICC = .52) | 8% |