Longitudinal associations of e-cigarette use with cigarette, marijuana, and other drug use initiation among US adolescents and young adults: Findings from the Population Assessment of Tobacco and Health study (Waves 1–6)

Abstract

Introduction:

Research examining prospective links of e-cigarette use with cigarette, marijuana, and other substance use has been limited largely to 1–2-year follow-up periods and focused on younger adolescents. This study examined longitudinal associations of e-cigarette use with cigarette, marijuana, and other substance use initiation among U.S. adolescents and young adults (AYAs) across an eight-year period.

Methods:

Adolescent (ages 12–17) and young adult (ages 18–25) data from waves 1–6 of the nationally representative Population Assessment of Tobacco and Health study (2013–2021) were used. Discrete time survival models with time-varying weights were employed to examine the risk of cigarette, marijuana, and other drug use initiation over an eight-year follow-up period among AYAs with no lifetime use of e-cigarettes/other tobacco, lifetime but no past 30-day use of e-cigarettes/other tobacco, past 30-day e-cigarettes only, other tobacco use only, or past 30-day e-cigarette/other tobacco use. We compare our time-varying weighting approach to a traditional time-invariant/complete case weighting approach.

Results:

Across six follow-up waves, all three past 30-day nicotine/tobacco use groups, including e-cigarettes only, had greater risk for cigarette, marijuana, and other drug use initiation relative to those not using nicotine/tobacco. The three past 30-day nicotine/tobacco use groups did not differ from each other in risk for marijuana use initiation. Associations were smaller in magnitude for young adults compared to adolescents, but significant for both age groups.

Conclusions:

Substance use initiation risks persist beyond 1–2 years for U.S. AYAs using e-cigarettes. Prevention strategies to reduce AYA e-cigarette use are needed to reduce cancer-related risk.

INTRODUCTION

E-cigarette use remains one of the most prevalent substance use behaviors among adolescents and young adults (AYAs) in the U.S., with 39% of adolescents reporting a history of vaping nicotine (ie, e-cigarette use) by 12^th grade (Miech et al., 2023). Current use is also high among AYAs, with 20% of 12^th graders (Miech et al., 2023) and 14% of young adults ages 18 to 25 (Substance Abuse and Mental Health Services Administration [SAMSHA], 2022) reporting e-cigarette use in the past month. Increasingly, e-cigarette use has been linked to adverse health consequences for AYAs, including risk for other tobacco use and other substance use (Evans-Polce et al., 2020; O’Brien et al., 2021; Unger et al., 2016).

Extant research has now demonstrated a strong and consistent link between e-cigarette use and cigarette initiation (Adermark et al., 2021; Barrington-Trimis et al., 2016; Hammond et al., 2017; Kelly et al., 2023; Miech et al., 2017; O’Brien et al., 2021; Soneji et al., 2017). A meta-analysis across 21 prospective longitudinal cohort studies found that adolescents who used e-cigarettes had three to four times greater odds of cigarette use initiation compared to adolescents who did not use e-cigarettes (O’Brien et al., 2021). Another meta-analysis that included studies of adolescents and adults had similar findings; however, only three of the 30 studies included both adolescents and adults and the longest follow-up time of any study was two years (Adermark et al., 2021). A smaller but growing number of prospective studies have found associations of e-cigarette use with subsequent marijuana use (Evans-Polce et al., 2020; Han et al., 2023; Ksinan et al., 2020; Staff et al., 2022), marijuana initiation (Audrain-McGovern et al., 2018; Dai et al., 2018; Seidel et al., 2022; Sun et al., 2022; Wong et al., 2020), and marijuana vaping specifically (Lee et al., 2021). All but two studies (Ksinan et al., 2020) focused on adolescent samples; one had a follow-up period of two years and the other examined use rather than initiation specifically (Han et al., 2023). Much less research has examined associations of e-cigarette use with risk for other substance use. The few studies that have examined other substance use initiation have found associations of e-cigarette use with prescription drug misuse (Evans-Polce, Patrick, et al., 2020; Harton et al., 2023), cocaine use (Silva et al., 2023), and other illicit drug use (McCabe et al., 2018; Silveira et al., 2018) were all limited to a one- to two-year follow-up period. Other studies have found cross-sectional associations of e-cigarette use and other substance use (Boccio & Jackson, 2021; Curran et al., 2018; Gilbert et al., 2021).

Most studies examining prospective links of e-cigarette use with cigarette, marijuana, and other substance use initiation have focused on one- or two-year follow-up periods. This relatively short window of time has the potential to underestimate the risk that may continue beyond one or two years. Whether risk for other tobacco and substance use initiation among those using e-cigarettes wanes or if groups continue to diverge in their risk over time is not well understood. Understanding this risk over a longer period is important for understanding the potential long-term adverse health consequences of e-cigarette use, especially among AYAs. Studies that have used a longer follow-up period (up to 6 years in one instance) with PATH data have typically used a complete case analysis strategy, which removes data that could inform longitudinal estimates (Si et al., 2022; West et al., 2024).

One difficulty with examining the risk of e-cigarette use over longer periods of time is loss to follow-up (or attrition), which can limit sample sizes and may introduce attrition bias. While survey weighting methods can be used to adjust for attrition bias, weights designed for longitudinal analysis in large national survey data sets are often constructed to be time-invariant and restrict the sample to those who have provided complete data across all waves of interest (Heeringa et al., 2017). Recent comparisons of alternative weighting approaches for the analysis of longitudinal survey data have suggested that analyzing all available observations from study participants and incorporating time-varying weights for each observation into the analysis generally results in the best estimates of trajectories with minimal bias (Si et al., 2022; West et al., 2024). The current study considered and evaluated both of these approaches to adjusting for attrition.

Transitions to other nicotine/tobacco products and other substance use are particularly important to understand among AYAs, when significant initiation occurs (Azagba et al., 2020; Cho et al., 2021; Kasza et al., 2020; U.S. Food and Drug Administration, 2021) and targeting of tobacco advertising is prominent (U.S. Food and Drug Administration, 2021). Currently, most studies examining e-cigarette use links with other tobacco and substance use initiation focus on adolescents only or do not examine adolescents separately from young adults. Research combining AYAs may miss important differences in risk for younger vs. older age groups. Additionally, understanding risk for initiation among young adults is increasingly important as recent research has shown the age of initiation of cigarette use is increasing with as many as 42% of young adults initiate cigarette use at age 18 or older (Barrington-Trimis et al., 2020).

In this study, we used discrete time survival models to examine the risks of initiating cigarette use, marijuana use, and other drug (eg, nonmedical prescription drugs, cocaine, heroin) use over an eight-year follow-up period among adolescents and young adults using e-cigarettes, compared to those not using e-cigarettes. We further differentiated risk for those using e-cigarettes only and those using e-cigarettes combined with other tobacco products. To maximize the data available, we utilized time-varying longitudinal weights, which allowed us to include individuals who provide only partial data across the six waves. We compared our findings using time-varying weights to those using a more traditional time-invariant/complete-case weighting approach.

METHODS

Study population

This study used the Population Assessment of Tobacco and Health (PATH) study restricted data, waves 1 through 6 (2013/2014–2021) (Hyland et al., 2017; PATH Study [United States] Restricted-Use Files, 2024). The PATH study is a nationally representative longitudinal cohort study of U.S. adolescents (12–17 years) and adults (18 years and older). The Wave 1 surveys were conducted between September 2013 and December 2014. Waves 1 through 4 were conducted approximately one year apart and waves 4 to 5 and 5 to 6 were approximately two years apart. Household screenings were conducted with computer-assisted personal interviewing (CAPI) and surveys were conducted using audio computer-assisted self-interviews (ACASI), with the option of being administered in Spanish or English. Wave 6 surveys used both in-person and telephone data collection due to the COVID-19 pandemic. The weighted response rates at wave 1 were 78.4% and 74.0% for youth and adults, respectively. Weighted response rates at follow-up ranged from 87.3% at wave 2 to 56.6% at wave 6 for adolescents and 83.2% at wave 2 to 57.5% at wave 6 for adults. The study was approved by the Westat IRB, and this secondary data analysis study was determined exempt by the first author’s university IRB. Additional details regarding interviewing procedures, sampling, and weighting are available in the PATH Study User Guide (PATH Study [United States] Restricted-Use Files, n.d.).

The current study was focused on adolescents (12–17) and young adults (18–25) in the wave 1 cohort (i.e., representative of the U.S. population at the time of wave 1) who were at risk for the outcomes of interest (i.e., had no lifetime use at baseline). Analyses using time-varying weights included all follow-up observations available from the wave 1 cohort of AYAs for those individuals who remained at risk for a specific outcome of interest (n=15,062 at risk for cigarette initiation, n=16,219 at risk for marijuana initiation, and n=19,417 at risk for other drug use initiation). Individuals were grouped into age groups based on their age at wave 1; individuals did not move across adolescent vs. young adult subgroups in the analysis as they aged (i.e., our comparisons were based on ages at baseline). By the end of the follow up period, those not yet censored by wave 6 were ages 19 to 26 in the adolescent sample and ages 26 to 33 in the young adult sample. However, these remain two separate younger and older age cohorts. Analyses using longitudinal all-wave weights (USDHHS et al., 2020) were restricted to individuals who responded at baseline and all five follow-up waves (n=7,481 at risk for cigarette initiation, n=7,824 at risk for marijuana initiation, and n=9,095 at risk for other drug use initiation), and therefore only included observations from individuals with data at all six waves. The analytic sample for the all-wave weights utilized 49.7%, 48.4%, and 46.8% of observations compared to that used with time-varying weights.

Measures

The exposure variable of interest was a time-varying five-category variable indicating e-cigarette and other tobacco use. Based on lifetime and past 30-day use measures, individuals were categorized as: 1) NO lifetime e-cigarette or other tobacco use, 2) lifetime e-cigarette and/or other tobacco use, NO past 30-day use, 3) past 30-day e-cigarette use and NO other tobacco use, 4) past 30-day other tobacco use and NO e-cigarette use, or 5) past 30-day use of both e-cigarettes and other tobacco. Other tobacco use included past 30-day use of cigars/little cigars/cigarillos, pipe, hookah, smokeless tobacco, snus, or dissolvable tobacco. For marijuana initiation and other drug initiation analyses, other tobacco use also included past 30-day cigarette use. The exposure was allowed to vary over time.

Three different outcomes were examined, namely time to cigarette initiation, marijuana initiation, and other drug use initiation (nonmedical Ritalin^®/Adderall^® use, nonmedical painkiller/sedative/tranquilizer use, cocaine/crack, other stimulants such as methamphetamine, or other drugs including heroin, inhalants, solvents, or hallucinogens). Initiation was determined by the first wave at which lifetime use was reported. We measured time from baseline (wave 1) when all individuals included in the analysis were at risk for (had not yet experienced) a given outcome.

We also adjusted for the following covariates measured at the baseline wave: age (years), race (white, black, Asian, and American Indian or Alaska Native [AIAN]/Native Hawaiian or Pacific Islander [NHPI]/Multirace [those who reported identifying with more than one racial group]), Hispanic ethnicity, U.S. region (Northeast, Midwest, South, West), past 6-month exposure to tobacco marketing, past-year internalizing symptoms (0–4), and past-year externalizing symptoms (0–7); the latter two covariates used items from the Global Appraisal of Individual Needs – Short Screener (GAIN-SS) (Dennis et al., 2006). We also included time-varying controls for wave, past 30-day alcohol use, past 30-day marijuana use (in analyses of cigarette and other drug use initiation), and past 30-day other drug use (in analyses of cigarette and marijuana initiation).

Analysis Plan

Discrete time hazard models were fitted to the PATH data to examine the associations of e-cigarette and other tobacco use with cigarette initiation, marijuana initiation, and other drug initiation. In preparation for this analysis, we created a “long”-format data file with a row for each person-wave of data contributed. By using a person-wave data format we were able to use standard logistic regression models to estimate the discrete time hazard function (Singer & Willett, 1993; Suresh et al., 2022). For each outcome, analyses were restricted to the subpopulation of individuals who were at risk at baseline (wave 1). That is, analyses examining risk for marijuana use initiation included only those who had not yet initiated marijuana use at wave 1, and analyses examining risk for other drug use initiation included only those who had not yet initiated other drug use at wave 1. Analyses accounted for the complex sample design of the PATH using Stata’s suite of svy commands and used balanced repeated replication weights with a 0.3 Fay adjustment. All analyses were stratified into two age groups: an adolescent group for those ages 12–17 at wave 1 and a young adult group for those ages 18–25 at wave 1. Appropriate subpopulation options within the svy commands were used to implement unconditional subpopulation analyses for sampling variance estimation (Heeringa et al., 2017; West et al., 2008).

Two different weighting approaches were used to compare results using the standard PATH longitudinal weighting approach (a single weight for each case providing complete data across all six waves) with results using a time-varying weighting approach (using all available observations). The first weighting approach used the standard wave 6 all-wave weights for the wave 1 cohort provided by the PATH Study Team (USDHHS et al., 2020). We refer to this as the wave 6 all-wave sample. These weights account for non-response across waves (ie, AYAs who did not provide complete data) but are time-invariant and provided only for individuals who responded at every wave of the study. For the second weighting approach, we created time-varying weights that are allowed to vary within a person across waves. These weights account for non-response across waves and allow for the inclusion of all observations from waves in which an individual provided data. For example, individuals who provided data at waves 1–4 would have their data included for those four waves and be right-censored at wave 4 in analyses using time-varying weights. In the all-wave weighting approach, this person would not have a weight and none of their data would be included in the analysis. As a result, the time-varying weight approach allowed us to include more than double the number of observations compared to the all-wave weight approach.

RESULTS

Descriptive Analyses

Table 1 provides descriptive characteristics of those at risk for cigarette initiation, marijuana initiation, and other drug initiation at wave 1. We provide estimates for the full AYA sample at risk for these outcomes and for the subset that responded at all six waves and therefore had an all-wave longitudinal weight available. At wave 1, those with no lifetime use of e-cigarettes or other tobacco ranged from 57.6% for those at risk for other drug use initiation to 82.5% for those at risk for cigarette initiation. In the full sample, those at risk for other drug use initiation at wave 1 were slightly older (Mean[M]=18.2 years) than those at risk for marijuana initiation (M=17.6 years) and at risk for cigarette initiation (M=17.2 years). Other sociodemographic characteristics were similar across the analytic samples.

Table 1.

Weighted baseline descriptives for U.S. adolescents and young adults ages 12–25 with: (a) no history of cigarette use at wave 1, (b) no history of marijuana use at wave 1, (c) no history of other drug use at wave 1.

	Adolescents and young adults at risk for cigarette initiation (a)		Adolescents and young adults at risk for marijuana initiation (b)		Adolescents and young adults at risk for other drug use initiation (c)
	Full sample (n =15,062)	Wave 6 all-wave sample (n= 7,481)	Full sample (n = 16,219)	Wave 6 all-wave sample (n = 7,481)	Full sample (n = 19,417)	Wave 6 all-wave sample (n = 9,095)
	%/M	%/M	%/M	%/M	%/M	%/M
E-cigarette and other tobacco use
No lifetime use	82.47%	83.06%	71.54%	73.20%	57.57%	59.14%
Lifetime use, no past 30-day use	12.66%	12.12%	16.01%	15.57%	20.81%	20.66%
Only other tobacco use	3.65%	3.55%	9.32%	8.20%	15.51%	14.33%
E-cigarette use only	0.77%	0.87%	0.86%	0.95%	1.28%	1.30%
E-cigarette use + other tobacco use	0.47%	0.39%	2.49%	2.08%	4.83%	4.56%
Male	48.96%	49.30%	49.63%	49.16%	50.94%	51.01%
Age (M, SD)	17.22 (.06)	17.22 (.08)	17.63(.05)	17.57(.07)	18.21(.04)	18.1(.06)
Race
White	67.19%	67.34%	69.36%	69.20%	68.49%	68.97%
Black/African American	17.01%	16.87%	15.32%	15.28%	16.27%	15.88%
Asian	8.23%	8.31%	7.86%	8.11%	7.36%	7.30%
AIAN/NHPI/Multiracea	7.58%	7.48%	7.45%	7.42%	7.89%	7.84%
Hispanic ethnicity	18.79%	18.61%	19.33%	19.52%	19.46%	19.35%
Region
Northeast	16.99%	17.60%	16.30%	16.53%	17.22%	17.55%
Midwest	21.35%	20.73%	22.12%	21.82%	21.87%	21.63%
South	37.73%	37.55%	39.48%	39.48%	38.30%	38.34%
West	23.93%	24.11%	22.10%	22.18%	22.61%	22.48%
Internalizing symptoms (0–4)	1.55 (.02)	1.60 (.03)	1.49 (.02)	1.55 (.03)	1.53 (.01)	1.59 (.03)
Externalizing symptoms (0–7)	2.10 (.02)	2.16 (.03)	1.94 (.02)	2.04 (.04)	1.97 (.02)	2.08 (.03)
Past 6-month exposure to tobacco marketing (0–4)	.71 (.01)	.73 (.02)	.78 (.01)	.79 (.02)	.87 (.01)	.88 (.01)
Past 30-day cigarette use	--	--	8.47%	7.80%	14.90%	13.75%
Past 30-day marijuana use	3.56 %	4.00%	--	--	8.14%	8.24%
Past 30-day other drug use	3.05%	3.28%	2.83%	2.80%	--	--
Past 30-day alcohol use	21.84%	23.10%	22.82%	22.56%	32.17%	32.22%

^aAmerican Indian or Alaska Native/Native Hawaiian or Pacific Islander/Multirace

Adolescents aged 12–17

Table 2 provides estimated discrete time hazard models examining the associations of e-cigarette and other tobacco use with all three initiation outcomes among those 12–17 years old at wave 1. We provide results using time-varying weights as well as results using the standard wave 6 all-wave weights for comparison. Regarding cigarette initiation, individuals in all three past 30-day nicotine/tobacco use groups were more likely to initiate cigarette use compared to those who did not use e-cigarettes or other tobacco, in analyses with time-varying weights. Associations were largest in magnitude for those using both e-cigarettes and other tobacco (adjusted odds ratio [aOR]=53.65; 95% confidence interval[CI]=42.21–68.20) compared to those with no lifetime use of any nicotine/tobacco. Compared to those using e-cigarettes only, those using e-cigarettes and other tobacco products were more likely to initiate cigarette use (aOR=2.48; 95% CI=2.03–3.03). The results were similar in magnitude for analyses using all-wave weights, although confidence intervals were larger when using the all-wave weights compared to models using time-varying weights.

Table 2.

Discrete time survival models: Associations of e-cigarette use and other tobacco use with cigarette, marijuana, and other drug use initiation among 12–17 year olds at baseline.

	Cigarette use initiationa		Marijuana use initiationb		Other drug use initiationc
	Time-varying weights, nobs =39,077 nind =10,449	All-wave weights, nobs =26,061 nind =5,813	Time-varying weights, nobs =33,251 nind =9,802	All-wave weights, nobs =22,648 nind =5,595	Time-varying weights, nobs =37,430 nind =10,619	All=wave weights, nobs =25,337 nind =5,912
	aOR (95% CI)	aOR (95% CI)	aOR (95% CI)	aOR (95% CI)	aOR (95% CI)	aOR (95% CI)
E-cigarette and other tobacco use (Reference = No tobacco use in models 1–6)
No lifetime use	REF	REF	REF	REF	REF	REF
Lifetime use, no past 30-day use	7.72 (6.54, 9.13)	7.97 (6.51, 9.75)	4.22 (3.76, 4.75)	3.95 (3.42, 4.56)	1.41 (1.27, 1.57)	1.39 (1.21, 1.60)
Only other tobacco use	19.43 (15.73, 24.01)	18.81 (14.27, 24.79)	6.49 (5.49, 7.67)	6.48 (5.16, 8.14)	1.98 (1.68, 2.33)	1.95 (1.56, 2.43)
E-cigarette use only	21.66 (18.07, 25.95)	23.17 (18.44, 29.11)	6.90 (5.84, 8.15)	6.46 (5.35, 8.23)	1.72 (1.46, 2.04)	1.48 (1.17, 1.88)
E-cigarette use + other tobacco use	53.65 (42.21, 68.20)	54.92 (41.14, 73.31)	8.16 (6.63, 10.05)	6.95 (5.43, 8.90)	3.10 (2.65, 3.64)	3.09 (2.53, 3.78)
E-cigarette and other tobacco use (Reference = E-cigarette use only in models 7–12)
No lifetime use	0.05 (0.04, 0.06)	0.04 (0.03, 0.05)	0.15 (0.12, 0.17)	0.15 (0.12, 0.19)	0.58 (0.49, 0.68)	0.67 (0.53, 0.86)
Lifetime use, no past 30-day use	0.36 (0.31, 0.41)	0.34 (0.29, 0.41)	0.61 (0.51, 0.73)	0.60 (0.48, 0.74)	0.82 (0.67, 0.99)	0.94 (0.73, 1.22)
Only other tobacco use	0.90 (0.74, 1.09)	0.81 (0.63, 1.05)	0.94 (0.76, 1.16)	0.98 (0.73, 1.31)	1.15 (0.94, 1.40)	1.32 (0.99 1.73)
E-cigarette use only	REF	REF	REF	REF	REF	REF
E-cigarette use + other tobacco use	2.48 (2.03, 3.03)	2.37 (1.81, 3.10)	1.18 (0.92, 1.52)	1.05 (0.79, 1.39)	1.80 (1.46, 2.22)	2.09 (1.57, 2.77)

Note: n_obs represent the total number of observations for the person-wave datasets constructed for the discrete time survival models that include multiple observations per person. n_ind represent the number of unique individuals included in each model. All models control for: wave, age, sex, race, Hispanic ethnicity, region, internalizing symptoms, externalizing symptoms, tobacco marketing exposure, and past 30-day alcohol use.

^amodels also control for past 30-day marijuana use and past 30-day other drug use;

^bmodels also control for past 30-day other drug use;

^cmodels also control for past 30-day marijuana use. aOR=adjusted Odds Ratio. Bold = pvalue<0.05

Similarly, regarding marijuana use, individuals in all three nicotine/tobacco use groups were more likely to initiate marijuana use compared to those not with no lifetime use of nicotine/tobacco products (aOR range: 6.49–8.16). When comparing those using e-cigarettes only to other past 30-day nicotine/tobacco use groups, there were no significant differences in risk for marijuana use initiation. Similar to the findings for cigarette initiation, the magnitudes of the associations were similar in the time-varying and all-wave weight analyses, but confidence intervals were smaller for time-varying weights.

Regarding other drug use initiation, individuals in all three nicotine/tobacco use groups also had greater odds of initiating other drug use compared those with no lifetime use of nicotine/tobacco (aOR range: 1.98–3.10). Compared to those using e-cigarettes only, those using e-cigarettes and other tobacco products had greater odds of other drug use initiation (aOR=1.80; 95% CI=1.46–2.22). Associations for all three past 30-day nicotine/tobacco use groups (compared to no lifetime tobacco use) with other drug use initiation were similar in magnitude in the time-varying weight and all-wave weight analyses. We note that all three past 30-day nicotine/tobacco use groups had significantly greater odds of initiation outcomes when compared to those with a history of use but no past 30-day use (results not shown).

Figure 1a shows the marginal hazard estimates for cigarette initiation at each wave by exposure group among adolescents. The probability of cigarette initiation was stable across waves for each of the exposure groups, with the probability consistently highest for those using both e-cigarettes and other tobacco products. Figure 1b shows the marginal hazard estimates for marijuana initiation at each wave by exposure group. The probability of marijuana initiation increased for waves 4, 5, and 6 compared to waves 2 and 3 across exposure groups. This increase was larger in magnitude for adolescents in the three past 30-day nicotine/tobacco groups compared to adolescents not currently using tobacco products. Figure 1c shows the marginal hazard estimates for other drug use initiation at each wave by exposure group. Marginal hazard estimates decreased in later waves (5 and 6) compared to earlier waves.

Figures 1a-1c.

Marginal predicted probabilities for (a) cigarette initiation, (b) marijuana initiation, and (c) other drug initiation at each wave by exposure group (12–17-year-olds), including 95% confidence intervals.

Figures 1d-1f. Marginal predicted probabilities for (d) cigarette initiation, (e) marijuana initiation, and (f) other drug initiation at each wave by exposure group (18–25-year-olds), including 95% confidence intervals.

Young adults aged 18–25

Table 3 provides estimated discrete time hazard models for examining the associations of e-cigarette and other tobacco use with all three initiation outcomes among those 18–25 years old at wave 1. Regarding cigarette initiation, individuals in all three past 30-day nicotine/tobacco use groups were more likely to initiate cigarette use compared to those with no lifetime use of e-cigarettes or other tobacco (aOR range: 9.89–22.01). Compared to those using e-cigarettes only, those using e-cigarettes and other tobacco had greater odds of cigarette initiation (aOR=1.93; 95% CI=1.13–3.33); this difference was only significant in the time-varying weights approach.

Table 3.

Discrete time survival models: Associations of e-cigarette use and other tobacco use with cigarette, marijuana, and other drug use initiation among 18–25 year olds at baseline

	Cigarette use initiation		Marijuana use initiation		Other drug use initiation
	Time-varying weights, nobs =10,047 nind =2,757	All-wave weights, nobs =6,831 nind =1,519	Time-varying weights, nobs =11,365 nind =3,357	All-wave weights, nobs =7,823 nind =1,830	Time-varying weights, nobs =19,731 nind =5,732	All-wave weights, nobs =13,199 nind =3,015
	aOR (95% CI)	aOR (95% CI)	aOR (95% CI)	aOR (95% CI)	aOR (95% CI)	aOR (95% CI)
E-cigarette and other tobacco use (Reference = No tobacco use in models 1–6)
No lifetime use	REF	REF	REF	REF	REF	REF
Lifetime use, no past 30-day use	3.70 (2.46, 5.58)	3.39 (1.90, 6.04)	2.67 (2.00, 3.56)	2.56 (1.88, 3.48)	1.25 (0.98, 1.59)	0.98 (0.72, 1.34)
Only other tobacco use	9.89 (6.30, 15.53)	11.08 (6.00, 20.47)	4.08 (3.02, 5.52)	4.44 (3.13, 6.31)	1.92 (1.49, 2.48)	1.67 (1.21, 2.30)
E-cigarette use only	11.37 (6.72, 19.25)	13.29 (5.80, 30.45)	5.96 (3.29, 10.78)	5.34 (2.75, 10.38)	1.96 (1.27, 3.02)	1.54 (0.85, 2.79)
E-cigarette use + other tobacco use	22.01 (13.44, 36.05)	19.69 (9.86, 39.33)	8.54 (5.49, 13.26)	9.12 (5.03, 16.55)	2.52 (1.84, 3.45)	2.10 (1.37, 3.24)
E-cigarette and other tobacco use (Reference = E-cigarette use only in models 7–12)
No lifetime use	0.09 (0.05, 0.15)	0.08 (0.03, 0.17)	0.17 (0.09, 0.30)	0.19 (0.10, 0.36)	0.51 (0.33, 0.79)	0.64 (0.36, 1.17)
Lifetime use, no past 30-day use	0.33 (0.21, 0.49)	0.26 (0.13, 0.48)	0.45 (0.26, 0.78)	0.48 (0.26, 0.89)	0.64 (0.41, 0.98)	0.64 (0.38, 1.05)
Only other tobacco use	0.90 (0.56, 1.36)	0.83 (0.43, 1.64)	0.69 (0.41, 1.15)	0.83 (0.46, 1.50)	0.98 (0.64, 1.50)	1.08 (0.65, 1.80)
E-cigarette use only	REF	REF	REF	REF	REF	REF
E-cigarette use + other tobacco use	1.93 (1.13, 3.33)	1.48 (0.65, 3.38)	1.43 (0.81, 2.52)	1.71 (0.82, 3.56)	1.29 (0.84, 1.97)	1.36 (0.78, 2.38)

Note: n_obs represent the total number of observations for the wide datasets constructed for the discrete time survival models that include multiple observations per person. n_ind represent the number of distinct individuals included in each model. All models control for: wave, age, sex, race, Hispanic ethnicity, region, internalizing symptoms, externalizing symptoms, tobacco marketing exposure, and past 30-day alcohol use.

^amodels also control for past 30-day marijuana use and past 30-day other drug use;

^bmodels also control for past 30-day other drug use;

^cmodels also control for past 30-day marijuana use. aOR=adjusted Odds Ratio. Bold = pvalue<0.05.

Similarly, regarding marijuana use initiation, individuals in all three past 30-day nicotine/tobacco use groups were more likely to initiate marijuana use compared to those not using tobacco products (aOR range: 4.08–8.54). When using e-cigarette use only as a reference, the three past 30-day nicotine/tobacco use groups were not significantly different in risk for marijuana use initiation. Results were similar in magnitude but with larger confidence intervals in analyses using the all-wave weights.

Regarding other drug use initiation, individuals in all three past 30-day nicotine/tobacco use groups were more likely to initiate other drug use, including those who used only e-cigarettes (aOR=1.96; 95% CI=1.27–3.02) compared to those with no lifetime use of nicotine/tobacco products, in analyses with time-varying weights. In the all-wave weight models, those using e-cigarettes only were not significantly different in risk for other drug use initiation compared to those with no lifetime nicotine/tobacco use (aOR=1.54; 95% CI=0.85–2.79). When using e-cigarette use only as a reference, the three past 30-day nicotine/tobacco use groups were not significantly different from each other. We note here that all three past 30-day nicotine/tobacco use groups had significantly greater odds of initiation outcomes when compared to those with a history of use but no past 30-day use with one exception: those using e-cigarettes only were not significantly different from those with lifetime but no past 30-day use in risk for other drug use initiation (results not shown).

Figure 1d shows the marginal hazard estimates for cigarette initiation at each wave by exposure group among young adults. The probability of cigarette initiation was stable across waves for each of the exposure groups with the probability consistently highest for those using both e-cigarettes and other tobacco. Figure 1e shows the marginal hazard estimates for marijuana initiation at each wave by exposure group. The probability of marijuana initiation increased for waves 4, 5, and 6 compared to waves 2 and 3 across exposure groups. This increase was larger in magnitude for young adults in the three nicotine/tobacco groups compared to young adults not using tobacco. Figure 1f shows the marginal hazard estimates for other drug use initiation at each wave by exposure group. Marginal hazard estimates decreased in later waves (5 and 6) compared to earlier waves. While associations were somewhat smaller in magnitude for young adults compared to adolescents, associations with cigarette initiation and marijuana use initiation remained robust for young adults.

DISCUSSION

This study showed that AYAs who use e-cigarettes and other tobacco products are more likely to initiate cigarette use, marijuana use, and other drug use over eight years of follow-up. This corroborates previous research showing a significant risk for other tobacco initiation and other substance use initiation, even among those using e-cigarettes only (Evans-Polce et al., 2020; Hair et al., 2021; National Academies of Sciences, Engineering, and Medicine, 2018; O’Brien et al., 2021; Soneji et al., 2017). This study adds to the literature by showing that this association held for both adolescents and young adult age groups and remained strong when examined over a much longer follow-up period.

Regarding cigarette initiation, we found that risk for initiation was strongest for those using both e-cigarettes and other tobacco products. This is in line with previous research showing that dual use of e-cigarettes and other tobacco products is associated with the greatest risk for other substance use initiation compared to single product use (Evans-Polce et al., 2020; McCabe et al., 2017). In contrast, regarding marijuana use initiation, those using e-cigarettes only had an equally strong risk for marijuana initiation compared to those using only other tobacco or those using both e-cigarette and other tobacco products—this was true both during adolescence and young adulthood. Moreover, we found that the link with marijuana initiation increased over time. This increase in risk for marijuana use initiation may be due to a combination of historical effects (the increasing prevalence of marijuana use among AYAs) and age effects (individuals being more likely to initiate as they get older), among other factors. However, it is critical to note that there was a larger increase in risk for marijuana use among those using nicotine/tobacco compared to those not using nicotine/tobacco, suggesting that the link of nicotine/tobacco use and marijuana initiation increased over time. This growing link is particularly concerning for AYAs whose brain development is ongoing and who are in an age period of peak risk for substance use disorders (Azagba et al., 2020; Cho et al., 2021; Volkow et al., 2018).

We found significant associations not only among adolescents, a group often studied for e-cigarette risks (O’Brien et al., 2021), but also among young adults. While associations were somewhat smaller in magnitude for young adults compared to adolescents, associations with cigarette and marijuana use initiation were robust for young adults. Additionally, differences in initiation risk between those using both e-cigarettes and other tobacco products and those using e-cigarettes only were more pronounced among adolescents than among young adults. Even though initiation of cigarette use, marijuana use, and other drug use occurs during adolescence for many individuals, these results show that risk continues among those who have not yet initiated by young adulthood, including for those using only e-cigarettes in young adulthood. Thus, prevention programs that target young adults who have not yet initiated cigarette, marijuana, or other substance use remain important. Due to differences in the context and norms in which use occurs, different intervention tailoring may be needed. Similarly, rigorous studies that consider differential risk for other tobacco and substance use initiation by the age at which e-cigarettes are initiated are an important area for future research.

This study employed time-varying survey weights to efficiently allow for the use of more observations than traditional complete-case approaches would allow. With the use of time-varying weights, we were able to include data from more than double the number of respondents compared to the number that a complete case / all-wave weighting approach allowed. In most cases, we found that the magnitudes of the associations were similar when using time-varying weights and all-wave weights; however, confidence intervals were consistently smaller when using time-varying weights. This led to e-cigarette use having a significant association with other drug use initiation only in time-varying weight analysis; the association became nonsignificant in an all-wave weighting approach. For less prevalent outcomes like other drug (eg, nonmedical prescription drugs, cocaine) use initiation, it is particularly important for researchers to make use of all the data available. These types of changes in inference about the associations of e-cigarette use with initiation of these other outcomes underscore the importance of using all available information when analyzing these types of longitudinal data, particularly if the time-varying weights are effectively adjusting for selection bias arising from attrition.

This study had important strengths, including the use of nationally representative longitudinal data, consistent measures spanning eight years, and large samples of AYAs. The study also had important limitations. First, it relied on self-reported survey data which may be subject to social desirability bias. Second, unmeasured confounders such as genetic predisposition to nicotine/tobacco use and other substance use may have impacted our findings. Third, despite the large overall sample, the study did not pursue examining associations with specific drugs due to small subsample sizes.

In summary, the results of this study suggest that e-cigarette use among AYAs carries important risks for engaging in cigarette, marijuana, and other drug use that persist for at least eight years. In fact, we found that all non-cigarette types of nicotine/tobacco product use are associated with increased risk for cigarette, marijuana, and other drug use initiation. While risk for cigarette smoking initiation was strongest for those using e-cigarette alongside other tobacco products, risk for marijuana use initiation was equally strong across all three tobacco use groups, including those using e-cigarettes and no other tobacco products.