Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2022 Mar 1.
Published in final edited form as: Addict Behav. 2020 Dec 8;114:106770. doi: 10.1016/j.addbeh.2020.106770

Electronic cigarette use and risk of cigarette and smokeless tobacco initiation among adolescent boys: A propensity score matched analysis

Brittney Keller-Hamilton a,*, Bo Lu b, Megan E Roberts c, Micah L Berman d,e, Elisabeth D Root f,g, Amy K Ferketich g
PMCID: PMC7811243  NIHMSID: NIHMS1659743  PMID: 33316588

Abstract

Introduction:

Electronic cigarette (e-cigarette) use among adolescents is associated with increased risk of subsequent cigarette smoking initiation in observational research. However, the existing research was not designed to answer causal questions about whether adolescent e-cigarette users would have initiated cigarette smoking if they had never used e-cigarettes. The current study used a causal inference framework to identify whether male adolescent e-cigarette users were at increased risk of initiating cigarette smoking and smokeless tobacco (SLT) use, compared to similar boys who had never used e-cigarettes.

Methods:

Boys from urban and Appalachian Ohio (N = 1220; ages 11–16 years at enrollment) reported use of e-cigarettes, cigarettes, and SLT at baseline and every six months for two years. A propensity score matching design was implemented, matching one e-cigarette user to two similar e-cigarette non-users. This analysis was completed in 25 multiple imputed datasets to account for missing data. Risk ratios (RRs) comparing risk of initiating cigarettes and SLT for e-cigarette users and nonusers were estimated.

Results:

Compared to non-users, e-cigarette users were more than twice as likely to later initiate both cigarette smoking (RR = 2.71; 95% CI: 1.89, 3.87) and SLT (RR = 2.42; 95% CI: 1.73, 3.38). They were also more likely to become current (i.e., past 30-day) cigarette smokers (RR = 2.20; 95% CI: 1.33, 3.64) and SLT users (RR = 1.64; 95% CI: 1.01, 2.64).

Conclusions:

Adolescent boys who used e-cigarettes had increased risk of later initiating traditional tobacco products when compared to similar boys who had never used e-cigarettes.

Keywords: Adolescents, Electronic cigarettes, Cigarettes, Smokeless tobacco

1. Introduction

The prevalence of electronic cigarette (e-cigarette) use among adolescents in the United States (U.S.) continues to increase. In 2019, an estimated 27.5% of high school students and 10.5% of middle school students in the U. S. reported e-cigarette use in the past 30 days (Cullen et al., 2019). In addition to the growing evidence of the negative health consequences of vaping (Cho et al., 2016; Skotsimara et al., 2019), several longitudinal studies have identified that non-smoking youth who had used e-cigarettes at baseline were more likely than other youth to start smoking cigarettes at follow-up (Soneji et al., 2017). Such associations were found regardless of whether the exposure was ever use (Barrington-Trimis et al., 2016; Wills et al., 2017; Watkins et al., 2018; East et al., 2018; Primack et al., 2015) (i.e., having puffed an e-cigarette, even once or twice) or past 30-day use (Barrington-Trimis et al., 2016; Watkins et al., 2018; Bold et al., 2018; Dunbar et al., 2018; Hammond et al., 2017). To our knowledge, only one study has measured the association between e-cigarette use and subsequent initiation of other tobacco products (Barrington-Trimis et al., 2016). This study identified that baseline e-cigarette use was associated with increased odds of initiating hookah, cigars, and pipes one year later (Barrington-Trimis et al., 2016). We are not aware any research that has evaluated whether e-cigarette use is associated with risk of initiating smokeless tobacco (SLT).

A critique of the argument that e-cigarette use increases the risk of cigarette smoking initiation is that adolescents who progressed to cigarette smoking after using e-cigarettes might have started using cigarettes anyway (Etter, 2018). In other words, these youth may have a “common liability” for tobacco use, which drives the observed associations (Etter, 2018; Vanyukov et al., 2012). This argument is worth consideration due to the overlap in some risk factors for e-cigarette use and cigarette smoking, including use of other substances (Dunbar et al., 2018), impulsivity, older age, male sex, and peer and family tobacco use (Hanewinkel and Isensee, 2015). Moreover, recent cross-sectional research that used causal inference methods concluded that the association between e-cigarette use and current cigarette smoking in a nationally representative sample of adolescents was due to shared risk factors (Kim and Selya, 2019). However, other research has identified divergent risk factors for e-cigarette use and cigarette smoking: E-cigarette users tend to have less severe mental health symptoms than cigarette smokers (Leventhal et al., 2016), and the association between e-cigarette use and subsequent cigarette smoking appears to be strongest among adolescents who were otherwise at low risk of cigarette smoking (e.g., lower rebelliousness and intentions to smoke) (Barrington-Trimis et al., 2016; Wills et al., 2017). Other studies have assessed whether the association between e-cigarette use and cigarette smoking is bidirectional (e.g., cigarette smoking increases risk of later e-cigarette initiation), which would suggest that e-cigarette users and cigarette smokers share common liability for tobacco use; however, results of these studies have been inconclusive (East et al., 2018; Bold et al., 2018; Dunbar et al., 2018).

In addition to the possibility of a common liability driving the association between e-cigarette use and cigarette smoking among adolescents, existing studies are limited by relatively short follow-ups periods (with a few exceptions (Barrington-Trimis et al., 2016; Bold et al., 2018; Dunbar et al., 2018)), which may not be enough time to observe the extent of the transition from e-cigarette use to cigarette smoking. Another limitation, due to the observational, longitudinal design necessary to answer this research question, is that studies experience attrition that is often related to risk factors for tobacco use. Though a few studies addressed item or unit nonresponse using maximum likelihood estimation, weighting, or multiple imputation (Watkins et al., 2018; Bold et al., 2018; Dunbar et al., 2018), most conducted a complete case analysis. Finally, although existing research in this area has at times attempted to make causal claims, the analytic plans of most studies were not designed to answer the causal question of whether adolescents who started smoking cigarettes after they had used e-cigarettes would have done so anyway. The goal of the current study, then, was to overcome these limitations by using a propensity score matching design, long-term follow-ups, and multiple imputation to address non-response to determine whether e-cigarette use is causally related to ever and current (i.e., past 30-day) cigarette or SLT initiation in a cohort of adolescent males.

2. Methods

2.1. Design

Participants were enrolled into the Buckeye Teen Health Study from January 2015 through June 2016 and completed surveys every six months for two years. All participants were 11- to 16-years-old at baseline and male, as an aim of the parent study was to measure predictors of SLT use (which is more prevalent among males) (Gentzke et al., 2019). At baseline, participants lived in urban Franklin County, Ohio (N = 708) or one of nine rural Appalachian Ohio counties (N = 512). Male youth were 14 years old on average at study enrollment and 69.2% were non-Hispanic white. Additional details about the cohort’s demographics can be found elsewhere (Friedman et al., 2018). A majority of participants were recruited using address-based probability sampling (N = 991), and the remainder were recruited using convenience sampling methods (N = 229), including newspaper advertisements, recruitment at community events, and flyers posted in the community. Up to two parents or guardians (hereafter referred to as parents) could enroll in the study.

2.2. Procedures

The Ohio State University’s Institutional Review Board approved all study procedures. Male youth provided assent, and their parents provided permission for them to enroll in the study. Boys who turned 18-years-old during the study period were re-consented to the study as adults. Parents who enrolled in the study provided informed consent.

The baseline and two-year follow-up sessions were completed in person at participants’ homes or a mutually agreed-upon public location. Sensitive items (e.g., substance use) were administered via audio computer-assisted self-interviewing (ACASI), and the remaining items were interviewer administered. When permitted, youth were separated from their families when completing the ACASI portion of the questionnaire. The six-, twelve-, and eighteen-month follow-up surveys were administered over the phone. For sensitive items, youth were instructed to say the number associated with their response rather than stating their full response to protect their privacy while completing phone surveys (e.g., answering “B” instead of “menthol”).

Parents completed a self-administered survey at baseline to provide additional information about the family’s socioeconomic status and tobacco use in the home.

2.3. Measures

2.3.1. Dependent variables

Cigarette and SLT use were assessed at each survey using items from the Population Assessment of Tobacco and Health (PATH) study (Hyland et al., 2017). Participants were classified as ever cigarette smokers if they answered “yes” to the question: “Have you ever tried cigarette smoking, even one or two puffs?” They were classified as current cigarette smokers if the last time they smoked a cigarette, even one or two puffs, was in the past 30 days. Similarly, participants were classified as ever SLT users if they selected “yes” for: 1) snus pouches; or 2) loose snus, moist snuff, dip, spit, or chewing tobacco in response to the following question: “Have you ever used any of the following smokeless tobacco products, even one or two times?” They were classified as current SLT users if the last time they used snus or any other SLT product was in the past 30 days. Participants self-reported their age of first using cigarettes or SLT if they had ever used the product at baseline.

2.3.2. Independent variables

E-cigarette use was assessed at each survey using PATH items (Hyland et al., 2017). Participants were classified as ever e-cigarette users if they answered “yes” to the question: “Have you ever used an e-cigarette, such as Smoking Everywhere, NJOY, Blu, or Vapor King, even one or two times?” (These were popular e-cigarette brands at the time of data collection.) Participants self-reported their age of first using e-cigarettes if they had ever used them at baseline.

2.3.3. Covariates

Covariates used in the propensity score model (described below) included variables that were associated with e-cigarette use in the literature and could reasonably be assumed to precede initiation of e-cigarette use. These variables included birth year (Hammond et al., 2017), race (dichotomized to white non-Hispanic vs. other) (Hammond et al., 2017), region (urban vs. Appalachian), parental education (college degree vs. less), living with at least one adult tobacco user (yes vs. no) (Hanewinkel and Isensee, 2015), impulsivity (Hanewinkel and Isensee, 2015) (mean sensation seeking score (Stephenson et al., 2003) [Cronbach’s α = 0.70] and the natural log of delay discounting (Kirby et al., 1999 k-score), and ever use of the following substances prior to initiation of e-cigarette use: cigarettes (for SLT models only), SLT (for cigarette models only), cigars, little cigars, cigarillos, hookah, pipes, bidis, kreteks, marijuana, and alcohol (Dunbar et al., 2018; Hammond et al., 2017). Baseline values of all sociodemographic and impulsivity measures were included in the propensity score models, as these were expected to be relatively stable. For substance use, all waves of data were used to determine whether the substance initiation occurred prior to e-cigarette use; tobacco use was assessed at each wave, marijuana use was assessed at baseline, and alcohol use was assessed at baseline and the 12-month follow-up.

Susceptibility to using tobacco products was also used in the imputation models and was measured at baseline and the 12-month follow-up. Susceptibility was assessed for each product by asking participants whether they would smoke/use the tobacco product soon and whether they would smoke/use the tobacco product if offered by a friend. Response options included “definitely yes,” “probably yes,” “probably not,” and “definitely not.” Any response other than “definitely not” to either item was coded as being susceptible to using the product. Results were combined across products to determine whether each participant was susceptible to any tobacco product (yes/no).

Male youth reported all covariate values except for region (assessed upon sampling), parental education, and tobacco use by adults in the household (both parent-reported).

2.4. Statistical analyses

2.4.1. Missing data

There was a small proportion of item non-response for tobacco use and sociodemographic items (≤5%). These values were imputed using hot deck single imputation prior to the current study.

Item non-response was also observed for baseline alcohol use, marijuana use, and impulsivity measures (<9% missing). Moreover, item non-response and implausible values were observed for self-reported age of first using substances at baseline. As hot deck single imputation had not already been performed, these missing values were imputed using multiple imputation by chained equations (MICE) (Lee and Carlin, 2010). Unit nonresponse as a result of study attrition was also handled using MICE, as retention was 92.7% at 6 months, 85.7% at 12 months, 82.5% at 18 months, and 73.4% at 24 months. Ever use of e-cigarettes, cigarettes, SLT, other tobacco products, alcohol, and marijuana across study waves were multiply imputed. Current use of e-cigarettes, cigarettes, and SLT were also imputed for all waves, conditional on participants being an ever user of each respective product at each wave. The imputation model included all covariates used in the propensity score model (von Hippel and Lynch, 2013). Susceptibility to using tobacco products was also added to the imputation model due to its association with initiating tobacco use (Hammond et al., 2017). Logistic regression models were used for the imputation, and 25 datasets were created. MICE was completed using Stata version 14.2 (StataCorp., 2015).

Propensity score estimation and matching, and all statistical analyses, were completed in the 25 imputed datasets. Results were combined following Rubin’s method (Rubin, 1987).

2.4.2. Propensity score model and matching

Prior to the estimation of propensity scores, participants were classified into one of two exposure groups, which represented the dependent variable in the propensity score model: e-cigarette ever users and e-cigarette never users. Because we were estimating whether e-cigarette use increased the risk of initiating cigarette smoking and SLT use, e-cigarette use had to precede initiation of cigarettes or SLT for one to be included in the exposed group. Thus, the e-cigarette users group included participants who reported e-cigarette use at a time point when they had not already initiated cigarette smoking (for the cigarette incidence models) or SLT use (for the SLT incidence models); participants who had initiated both e-cigarettes and cigarettes/SLT at baseline were classified as e-cigarette users if their self-reported age of first using e-cigarettes was younger than that of cigarettes/SLT. The e-cigarette non-users group included participants who never reported e-cigarette use or reported e-cigarette initiation after they had already begun using cigarettes/SLT. Participants who initiated e-cigarettes and cigarettes/SLT at the same age or survey wave were excluded from analyses and not assigned propensity scores, as temporality of product initiation could not be established. Multivariable logistic regression models including the covariates described above were used to estimate the probability of each participant being an e-cigarette user (e.g., their propensity score).

After propensity scores were estimated, the propensity score match was completed, with each e-cigarette user being matched to two e-cigarette non-users. An optimal matching algorithm and linear propensity score distance were used to complete the match (Austin, 2014). After the match was completed, absolute standardized differences (ASDs) for each covariate were calculated, stratified by e-cigarette user status. A priori, post-match ASD values < 0.10 for all covariates indicated an acceptable match (Morgan, 2018). Propensity score estimation and matching were completed using RStudio version 1.1.383 (R Studio Team, 2016). The packages “MatchIt” and “optmatch” were used for the propensity score match.

2.4.3. Analyses

Generalized linear models with log links were used to estimate risk ratios (RRs) comparing the risk of: 1) ever cigarette use; 2) current cigarette use; 3) ever SLT use; and 4) current SLT use according to e-cigarette user group. Models were run in the 25 matched datasets; participants who were not matched were excluded. An alpha value of 0.05 was used to assess statistical significance. Statistical analyses were completed using SAS version 9.4 (SAS Version 9.4, 2013).

2.4.4. Sensitivity analysis

We conducted a complete case analysis using participants who were compliant at all five time points and were not missing data for variables in the propensity score model. A 1:2 propensity score match was also completed in this dataset and generalized linear models with log links were used to estimate RRs as in the analysis using multiple imputation.

3. Results

The analytic sample size across the imputed datasets after the propensity score match ranged from 348 to 387 for cigarette incidence analyses (N = 116 to 129 e-cigarette users; N = 101 to 126 initiated cigarette ever use) and from 435 to 486 for SLT incidence analyses (N = 145 to 162 e-cigarette users; N = 108 to 131 initiated SLT ever use). The difference in sample sizes between cigarette and SLT analyses was due to more participants initiating e-cigarettes and cigarettes at the same age and therefore being removed from the analysis.

All e-cigarette ever users in each dataset were matched to two e-cigarette never users. Pre-match, there was notable imbalance in the distribution of several covariates between e-cigarette user groups: use of other tobacco products, use of marijuana, sensation seeking mean score, and birth year according to e-cigarette use (all ASD values > 0.2; results not shown). Post-match, acceptable covariate balance was achieved in both the cigarette and SLT incidence analytic samples (Table 1).

Table 1.

Characteristics of Ohio adolescent boys included in propensity score matched analysis, 2015–2018.a

Cigarette incidence analyses SLT incidence analyses
E-cigarette ever users E-cigarette never users Post-match ASD E-cigarette ever users E-cigarette never users Post-match ASD
Appalachian region (%) 45.2 45.2 0.03 35.3 37.8 0.05
Birth year (mean) 2000.4 2000.4 0.03 2000.4 2000.4 0.03
Adult tobacco user in household (%) 39.7 38.3 0.05 38.0 37.1 0.03
Parent graduated college (%) 54.4 54.4 0.04 48.1 49.8 0.04
Minority race/ethnicity (%) 20.0 20.4 0.03 30.8 29.0 0.04
Sensation seeking score (mean) 3.37 3.35 0.03 3.35 3.31 0.04
Natural log of K-score (mean) −4.03 −4.04 0.05 −4.12 −4.14 0.03
Alcohol use (%) 23.7 23.1 0.04 25.4 26.0 0.03
Marijuana use (%) 7.3 8.1 0.04 16.3 13.6 0.08
Other tobacco use (%) 20.8 20.7 0.03 33.4 28.8 0.09
Open in a new tab

Abbreviations: ASD = Absolute standardized difference.

a

Data were collected from adolescent males in urban and Appalachian Ohio. Hot deck single imputation and multiple imputation by chained equations were used to impute missing data due to item and unit nonresponse. Summary statistics and ASD values reported in this table were first calculated individually within each of the 25 imputed datasets, and then were averaged across the datasets. Due to rounding of descriptive statistics, some ASD values are greater than 0 when there appears to be no difference in balance of covariates.

Although covariates were well balanced by e-cigarette user group, there were some differences in characteristics of the final analytic samples for cigarette and SLT incidence analyses (Table 1). For example, the proportion of participants living in Appalachia was greater in the cigarette incidence analytic sample. The proportions of participants who were of minority race/ethnicity, marijuana users, and other tobacco product users were higher in the SLT incidence analytic sample.

3.1. Risk of cigarette and SLT initiation

Compared to male youth who never used e-cigarettes, those who used e-cigarettes were more than twice as likely to subsequently initiate ever cigarette smoking (RR = 2.71; 95% confidence interval [CI]: 1.89, 3.87) and SLT use (RR = 2.42; 95% CI: 1.73, 3.38; Fig. 1). They were also at increased risk of initiating current cigarette smoking (RR = 2.20; 95% CI: 1.33, 3.64) and SLT use (RR = 1.64; 95% CI: 1.01, 2.64).

Fig. 1.

Fig. 1.

Open in a new tab

Risk of cigarette and SLT initiation for e-cigarette ever users compared to never users estimated from a propensity score matched analysis in 25 imputed datasets, Ohio, 2015–2018 (Data were collected from adolescent males in urban and Appalachian Ohio. Hot deck single imputation and multiple imputation by chained equations were used to impute missing data due to item and unit nonresponse. RRs were estimated following propensity score matching in 25 imputed datasets using a generalized linear model with a log link. RRs and standard errors used to calculate confidence intervals across the 25 imputed datasets were obtained using Rubin’s method (Rubin, 1987)).

3.2. Complete case analyses

All e-cigarette ever users were matched to two never users in the complete case analyses. The total sample size was 135 in cigarette incidence analyses and 156 in SLT incidence analyses. Acceptable balance was achieved for all covariates (results not shown), with the exception of living with an adult tobacco user in the cigarette incidence analytic dataset (ASD = 0.12) and natural log of K-score in the SLT incidence analytic dataset (ASD = 0.11). Because these values were close to the a priori ASD cut-off of 0.10, and to more directly compare results between the complete case analyses and imputed analyses, these covariates were not controlled in the final models.

Compared to the results from the imputed analyses, point estimates for all complete case analyses were in the same direction but attenuated toward the null (Table 2). This, combined with the reduced power driven by study attrition, led to no statistically significant findings.

Table 2.

Risk of cigarette and SLT initiation for e-cigarette users compared to non-users estimated from a propensity score matched, complete case analysis, Ohio, 2015–2018.a

Incident Cigarette Smoking Incident SLT Use
Ever use Current use Ever use Current use
RR 95% CI RR 95% CI RR 95% CI RR 95% CI
E-cigarette use
Never Ref Ref Ref Ref
Ever 2.22 0.90, 5.47 1.25 0.41, 3.82 1.67 0.72, 3.86 1.56 0.58, 4.18
Open in a new tab

Abbreviations: RR = Risk ratio; CI = confidence interval.

a

Data were collected from adolescent males in urban and Appalachian Ohio. RRs and standard errors for confidence intervals were estimated following propensity score matching using a generalized linear model with a log link.

4. Discussion

E-cigarette use increased the risk of initiating both cigarettes and SLT in a cohort of adolescent boys who were balanced on risk factors for e-cigarette use. Additionally, e-cigarette use increased the risk of initiating current cigarette smoking and SLT use, indicating more regular tobacco use. These findings therefore add to the growing literature that demonstrates that e-cigarette use increases the risk of using other tobacco products among adolescents in the U.S.

A causal link between e-cigarette use and onset of cigarette smoking or SLT use is plausible physiologically and behaviorally. Physiologically, nicotine addiction and dependence as a result of exposure to nicotine in e-cigarette liquid is a mechanism through which e-cigarette use could plausibly lead to cigarette smoking and SLT use (Vogel et al., 2018). This assertion is supported by evidence that use of other tobacco products has also been associated with onset of cigarette smoking among adolescents, including hookah, other non-cigarette combustible tobacco products, and SLT (Watkins et al., 2018). Behaviorally, the actions involved in vaping e-cigarettes are similar to those involved in smoking cigarettes. Thus, hand-to-mouth movement, puffing, and inhalation/exhalation habits can be established among e-cigarette users, even in the absence of nicotine, due to consistent repetition of these behaviors that becomes automatic (Verplanken, 2006). Moreover, the establishment of these behaviors could lead to more positive expectancies related to cigarette smoking, which mediates the association between e-cigarette use and cigarette smoking onset (Wills et al., 2016).

Our findings that e-cigarette-using adolescents, who were otherwise similar to e-cigarette non-users, were more likely to progress to cigarette smoking and SLT use have serious public health implications. The U.S. has made steady progress in reducing the prevalence of cigarette and SLT use among adolescents since 2011 (Gentzke et al., 2019). The health consequences of using both products have been well established for decades, and cigarettes remain the leading preventable cause of death in the U.S. (U.S. Department of Health and Human Services, 2014). As e-cigarettes appear to lead more adolescents to using these harmful tobacco products, the burden of tobacco on the public’s health will remain high. Even absent the association we observed between e-cigarette use and later initiation of traditional tobacco products, the high prevalence of e-cigarette use among adolescents in the U.S. is concerning. Although e-cigarette use is generally associated with a lower health risk than cigarette smoking, e-cigarettes do appear to increase risk of respiratory, cardiovascular, and dental health problems among people who never smoked cigarettes (Bozier et al., 2020). Finally, youth who become addicted to nicotine will be burdened with the chronic condition of addiction (U.S. Public Health Service, 2008).

5. Strengths and limitations

One strength of the current study was the use of multiple imputation to address attrition, which retained adolescents at high risk of tobacco use in the analyses. As demonstrated in the complete case analyses, excluding these participants biased the estimated RRs toward the null—likely because baseline ever and current tobacco use, as well as predictors of tobacco use among adolescents like substance use and living with an adult tobacco user (Dunbar et al., 2018; Hanewinkel and Isensee, 2015), were strongly associated with study attrition. A second strength of our study was the use of a propensity score matched analysis to estimate risk of cigarette and SLT initiation by e-cigarette exposure group, which allowed us to estimate the causal effect of e-cigarette use on initiation of traditional tobacco products. A third strength was our large pre-match sample size, which supported our ability to match every e-cigarette user to two non-users in all 25 imputed datasets and achieve good post-match covariate balance.

There were also several limitations of the current study, which can be addressed in future work to strengthen the evidence base on this topic. The first limitation was that girls were excluded because the objective of the parent study was to examine SLT use, which is more prevalent among adolescent males (Gentzke et al., 2019). Given differences in the prevalence of and risk factors for tobacco use according to sex (Gentzke et al., 2019; Audrain-McGovern et al., 2015), we cannot assume the same results would be found among girls. The second limitation was that our participants were sampled from Ohio, and so results might not generalize to male youth in the rest of the U.S. A third limitation was coarseness of measuring age of product use, which resulted in excluding some e-cigarette users from analyses due to our inability to establish temporality of product use.

A fourth limitation was that we could not include every important predictor of e-cigarette use in the propensity score model (e.g., peer use of tobacco) because we could not discern whether those exposures occurred prior to e-cigarette use for participants who entered the study as e-cigarette users. For other exposures in our propensity score model (i.e., region, parental education, tobacco use by adults in the household, and impulsivity) we had to assume that the values of these variables were the same as they were prior to the participant’s initiation of e-cigarette use. Finally, because this was an observational study rather than a randomized controlled trial, we could not control unmeasured confounders; we conducted our analysis under the commonly-used ignorability assumption (Rosenbaum, 2010).. The direction of this bias is unclear but it is possible that it led to inflated point estimates or spurious associations.

6. Conclusion

This study identified that e-cigarette use increased the risk of ever and current cigarette smoking and SLT use initiation among adolescent boys. E-cigarettes pose a public health threat, not only due to the health risks associated with their use, but also because they may lead more adolescents to using cigarettes and SLT. Future work that uses a causal framework to evaluate the relationship between specific characteristics of e-cigarettes (e.g., nicotine content, device type, or flavors) and risk of initiating other tobacco products, as well as transitions to established tobacco use in adulthood, is warranted.

Acknowledgments

Funding

This work was supported by grant P50CA180908 from the National Cancer Institute and Food and Drug Administration’s Center for Tobacco Products. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH or the Food and Drug Administration.

Footnotes

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

References

  1. Cullen KA, Gentzke AS, Sawdey MD, Chang JT, Anic GM, Wang TW, Creamer MR, Jamal A, Ambrose BK, & King BA (2019). e-Cigarette use among youth in the United States, 2019. JAMA, 322(21), 2095 10.1001/jama.2019.18387. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Cho JH & Paik SY (2016). Association between electronic cigarette use and asthma among high school students in South Korea. Fehrenbach H, ed. PLOS One 11(3), e0151022. doi: 10.1371/journal.pone.0151022. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Skotsimara G, Antonopoulos AS, Oikonomou E, Siasos G, Ioakeimidis N, Tsalamandris S, Charalambous G, Galiatsatos N, Vlachopoulos C, & Tousoulis D (2019). Cardiovascular effects of electronic cigarettes: A systematic review and meta-analysis. European Journal of Preventive Cardiology, 26(11), 1219–1228. 10.1177/2047487319832975. [DOI] [PubMed] [Google Scholar]
  4. Soneji S, Barrington-Trimis JL, Wills TA, Leventhal AM, Unger JB, Gibson LA, Yang JaeWon, Primack BA, Andrews JA, Miech RA, Spindle TR, Dick DM, Eissenberg T, Hornik RC, Dang R, & Sargent JD (2017). Association between initial use of e-cigarettes and subsequent cigarette smoking among adolescents and young adults: A systematic review and meta-analysis. JAMA Pediatrics, 171(8), 788 10.1001/jamapediatrics.2017.1488. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Barrington-Trimis JL, Urman R, Berhane K & et al. (2016). E-cigarettes and future cigarette use. Pediatrics 138(1), e20160379–e20160379. doi: 10.1542/peds.2016-0379. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Wills TA, Sargent JD, Gibbons FX, Pagano I, & Schweitzer R (2017). E-cigarette use is differentially related to smoking onset among lower risk adolescents. Tobacco Control, 26(5), 534–539. 10.1136/tobaccocontrol-2016-053116. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Watkins SL, Glantz SA, & Chaffee BW (2018). Association of noncigarette tobacco product use with future cigarette smoking among youth in the Population Assessment of Tobacco and Health (PATH) Study, 2013–2015. JAMA Pediatrics, 172 (2), 181 10.1001/jamapediatrics.2017.4173. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. East K, Hitchman SC, Bakolis I, Williams S, Cheeseman H, Arnott D, & McNeill A (2018). The association between smoking and electronic cigarette use in a cohort of young people. Journal of Adolescent Health, 62(5), 539–547. 10.1016/j.jadohealth.2017.11.301. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Primack BA, Soneji S, Stoolmiller M, Fine MJ, & Sargent JD (2015). Progression to traditional cigarette smoking after electronic cigarette use among US adolescents and young adults. JAMA Pediatrics, 169(11), 1018–1023. 10.1001/jamapediatrics.2015.1742. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Bold KW, Kong G, Camenga DR, Simon P, Cavallo DA, Morean ME, & Krishnan-Sarin S (2018). Trajectories of e-cigarette and conventional cigarette use among youth. Pediatrics, 141(1), e20171832 10.1542/peds.2017-1832. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Dunbar MS, Davis JP, Rodriguez A, Tucker JS, Seelam R, D’Amico EJ (2018). Disentangling within- and between-person effects of shared risk factors on e-cigarette and cigarette use trajectories from late adolescence to young adulthood. Nicotine & Tobacco Research Published online. doi: 10.1093/ntr/nty179. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Hammond D, Reid JL, Cole AG, & Leatherdale ST (2017). Electronic cigarette use and smoking initiation among youth: A longitudinal cohort study. Canadian Medical Association Journal, 189(43), E1328–E1336. 10.1503/cmaj.161002. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Etter J-F (2018). Gateway effects and electronic cigarettes. Addiction, 113(10), 1776–1783. 10.1111/add.13924. [DOI] [PubMed] [Google Scholar]
  14. Vanyukov MM, Tarter RE, Kirillova GP, et al. (2012). Common liability to addiction and “gateway hypothesis”: Theoretical, empirical and evolutionary perspective. Drug and Alcohol Dependence, 123, S3–S17. 10.1016/j.drugalcdep.2011.12.018. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Hanewinkel R, & Isensee B (2015). Risk factors for e-cigarette, conventional cigarette, and dual use in German adolescents: A cohort study. Preventive Medicine, 74, 59–62. 10.1016/j.ypmed.2015.03.006. [DOI] [PubMed] [Google Scholar]
  16. Kim S, & Selya AS (2019). The relationship between electronic cigarette use and conventional cigarette smoking is largely attributable to shared risk factors. Nicotine & Tobacco Research. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Leventhal AM, Strong DR, Sussman S, et al. (2016). Psychiatric comorbidity in adolescent electronic and conventional cigarette use. Journal of Psychiatric Research, 73, 71–78. 10.1016/j.jpsychires.2015.11.008. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Gentzke AS, Creamer M, Cullen KA, et al. (2019). Tobacco product use among middle and high school students — United States, 2011–2018. Morbidity and Mortality Weekly Report, 68(6), 157–164. 10.15585/mmwr.mm6806e1. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Friedman KL, Roberts ME, Keller-Hamilton B, et al. (2018). Attitudes towards tobacco, alcohol, and non-alcoholic beverage advertisement themes among adolescent boys. Substances Use & Misuse, 53(10), 1706–1714. 10.1080/10826084.2018.1429473. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Hyland A, Ambrose BK, Conway KP, et al. (2017). Design and methods of the Population Assessment of Tobacco and Health (PATH) study. Tobacco Control, 26(4), 371–378. 10.1136/tobaccocontrol-2016-052934. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Stephenson MT, Hoyle RH, Palmgreen P, & Slater MD (2003). Brief measures of sensation seeking for screening and large-scale surveys. Drug and Alcohol Dependence, 72(3), 279–286. 10.1016/j.drugalcdep.2003.08.003. [DOI] [PubMed] [Google Scholar]
  22. Kirby KN, Petry NM, & Bickel WK (1999). Heroin addicts have higher discount rates for delayed rewards than non-drug-using controls. Journal of Experimental Psychology: General, 128(1), 78–87. [DOI] [PubMed] [Google Scholar]
  23. Lee KJ, & Carlin JB (2010). Multiple imputation for missing data: Fully conditional specification versus multivariate normal imputation. American Journal of Epidemiology, 171(5), 624–632. 10.1093/aje/kwp425. [DOI] [PubMed] [Google Scholar]
  24. von Hippel P, & Lynch J (2013). Efficiency gains from using auxiliary variables in imputation. Cornell University Library. [Google Scholar]
  25. StataCorp. (2015). Stata Statistical Software Version 14.2. StataCorp LP. [Google Scholar]
  26. Rubin DB (1987). Multiple imputation for nonresponse in surveys. John Wiley & Sons, Inc. [Google Scholar]
  27. Austin PC (2014). A comparison of 12 algorithms for matching on the propensity score. Statistics in Medicine, 33(6), 1057–1069. 10.1002/sim.6004. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Morgan CJ (2018). Reducing bias using propensity score matching. Journal of Nuclear Cardiology, 25(2), 404–406. 10.1007/s12350-017-1012-y. [DOI] [PubMed] [Google Scholar]
  29. R Studio Team (2016). RStudio: Integrated Development for R. RStudio, Inc. http://www.rstudio.com/. [Google Scholar]
  30. SAS Version 9.4. The SAS Institute; 2013. [Google Scholar]
  31. Vogel EA, Ramo DE, & Rubinstein ML (2018). Prevalence and correlates of adolescents’ e-cigarette use frequency and dependence. Drug and Alcohol Dependence, 188, 109–112. 10.1016/j.drugalcdep.2018.03.051. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Verplanken B (2006). Beyond frequency: Habit as mental construct. British Journal of Social Psychology, 45(3), 639–656. 10.1348/014466605X49122. [DOI] [PubMed] [Google Scholar]
  33. Wills TA, Gibbons FX, Sargent JD, & Schweitzer RJ (2016). How is the effect of adolescent e-cigarette use on smoking onset mediated: A longitudinal analysis. Psychology of Addictive Behaviors, 30(8), 876–886. 10.1037/adb0000213. [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. U.S. Department of Health and Human Services (2014). The Health Consequences of Smoking—50 Years of Progress: A Report of the Surgeon General. U.S. Department of Health and Human Services, Centers for Disease Control and Prevention, National Center for Chronic Disease Prevention and Health Promotion, Office on Smoking and Health; 1–978. https://permanent.access.gpo.gov/gpo45352/PDFversion/Fullreport/full-report.pdf. [Google Scholar]
  35. Bozier J, Chivers EK, Chapman DG, et al. (2020). The evolving landscape of e-cigarettes. Chest, 157(5), 1362–1390. 10.1016/j.chest.2019.12.042. [DOI] [PubMed] [Google Scholar]
  36. U.S. Public Health Service. (2008). A clinical practice guideline for treating tobacco use and dependence: 2008 update. American Journal of Preventive Medicine, 35(2), 158–176. 10.1016/j.amepre.2008.04.009. [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Audrain-McGovern J, Rodriguez D, & Leventhal AM (2015). Gender differences in the relationship between affect and adolescent smoking uptake: Gender differences. Addiction, 110(3), 519–529. 10.1111/add.12797. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Rosenbaum PR (2010). Design of observational studies (1st ed.). Springer-Verlag. [Google Scholar]

RESOURCES