Transition 1: Among Adolescents and Young Adults with No History of Combustible Tobacco Use, Does E-Cigarette Use Affect Risk of Combustible Tobacco Cigarette Ever Use?

The extent to which factors implicated in ever smoking impact public health is qualified by the fact that a subset of those who become ever users progress to smoke at greater levels of smoking frequency, intensity, or duration (Path 3a, Figure 16-1) and hence become a regular smoker, whereas others may smoke infrequently, smoke very little on each smoking day, and discontinue use shortly following initiation (Path 3b, Figure 16-1) (HHS, 2014). Studying the effect of e-cigarette use on the likelihood of ever smoking is important because ever use is a necessary precursor to progression to regular use. Hence, if e-cigarette use impacts ever smoking, it may have downstream effects on the prevalence of health-damaging courses of smoking. Studying ever use is a particularly useful outcome measure at this time because e-cigarettes have been widely available in the United States only for a short period, leaving the majority of studies lacking sufficient duration of follow-up to study the naturalistic cigarette smoking progression sequence, which can involve a lengthy period between ever use and reaching daily smoking (Chassin et al., 2009; HHS, 2012).

There are three potential ways in which e-cigarette use may impact ever use of combustible tobacco use in young populations. These ways are discussed in the paragraphs below.

Transition 2: Among Adolescents and Young Adults, Does E-Cigarette Use Affect Risk of Progression to Combustible Tobacco Use Patterns of Greater Frequency, Intensity, or Duration?

Existing evidence on typical smoking trajectories indicate that an appreciable portion of ever users never become frequent smokers, will discontinue, and are temporarily experimenting with smoking (i.e., Path 3b, Figure 16-1) (Dutra et al., 2017; HHS, 2012; Sargent et al., 2017). Other ever users progress to become frequent, heavy, and chronic smokers (i.e., Path 3a, Figure 16-1)—a trajectory that is increasingly more likely the longer someone continues to smoke (Dutra et al., 2017; HHS, 2012; Sargent et al., 2017). A key question is whether adolescents and young adults who become ever smokers after e-cigarette use versus those who become ever smokers with no prior history of e-cigarette use differ in risk of progression in frequency (i.e., days used in the past 30), intensity (i.e., cigarettes per day on smoking day, sometimes termed “heaviness”), or duration of smoking. In Figure 16-1, this question is depicted by whether the probability of progression (versus temporary experimentation) in Path 3a differs as a function if one started smoking through vaping (2a) or not (2b). If e-cigarette use affects post-initiation progression of smoking, the impact of e-cigarette use among never smoking adolescents and young adults on the tobacco-related public health burden will be dependent on e-cigarette effects on both ever use and progression to more frequent, heavy, and chronic smoking.

There are three ways in which e-cigarette use may subsequently impact post-initiation adolescent and young adult smoking trajectories.

Considerations for Observational Data on the Association of E-Cigarette Use with Combustible Tobacco Cigarette Ever Use and Progression

Most evidence addressing this question comes from observational studies of the association between e-cigarette use and ever use and progression of smoking. Although randomized controlled trials are generally considered the strongest study design, randomized studies examining the effects of e-cigarette exposure on never-smoking youth are unethical. A challenge of observational studies is potential confounding, because without randomization, certain factors (confounders) may be systematically and unequally distributed across e-cigarette users and non-users. If these factors are associated with both e-cigarette use and combustible tobacco cigarette smoking initiation and are not in the causal pathway, they could statistically bias the study results and create a spurious effect of e-cigarette use on a combustible tobacco cigarette use outcome. For example, if e-cigarettes attract youth who already have a strong interest in smoking, then e-cigarette use may serve as an indicator of an underlying proclivity to smoke rather than playing a causal role in cigarette initiation. Studies that control for known confounders would be stronger evidence than studies that do not.

In the evidence review of the observational data, studies varied in the breadth of covariates included as confounders for which to statistically control. While residual confounding from unmeasured factors is always possible, the committee considered studies that adjusted for a more comprehensive set of covariates as stronger evidence. These plausible confounders include (1) sociodemographic factors that may address nonspecific shared risk factors; (2) environmental factors that may increase the opportunity, willingness, or interest to use both products (e.g., low monitoring of youth by parents, use of tobacco products by peers or family, permissiveness of tobacco product use by peers or family, exposure to tobacco product advertising); and (3) an endogenous propensity to engage in risk-taking behavior assessed via measures of psychological disposition (e.g., sensation-seeking personality traits, rebelliousness, depression) or risky behaviors (e.g., use of products other than e-cigarettes or combustible tobacco cigarettes, use of non-tobacco drugs of abuse, delinquent behaviors). See Tables 16-1, 16-2, and 16-3 for a listing of the covariates adjusted for in studies that were included in the evidence review.

TABLE 16-1

INITIATION: Summary of Prospective Cohort Studies of the Association Between Ever Use of E-Cigarettes (Versus Never Use) and Subsequent Risk of Ever Smoking of Combustible Tobacco Cigarettes Among Youth/Young Adults Who Were Non-Smokers at Baseline.

TABLE 16-2

PROGRESSION: Summary of Prospective Cohort Studies of the Association Between E-Cigarette Use and Subsequent Risk of Recent Smoking/Heavier Smoking of Combustible Tobacco Cigarettes Among Youth/Young Adults.

TABLE 16-3

DOSE–RESPONSE: Summary of Prospective Cohort Studies of the Association Between E-Cigarette Use Frequency and Subsequent Risk of Smoking of Combustible Tobacco Cigarettes Among Youth/Young Adults.

Among observational study designs, longitudinal studies were considered stronger evidence compared with cross-sectional studies. Given the high plausibility of reverse causality such that combustible tobacco cigarette smoking may also impact e-cigarette use, longitudinal cohort studies that assessed e-cigarette use at baseline and smoking at a future follow-up assessment would provide the strongest evidence to rule out potential reverse causality. Removal of ever smokers at baseline assessment rules out the possibility of reverse causality; however, this approach selects a portion of the population and therefore may not generalize to the entire population of youth and young adults, including those who start smoking at an early age. Alternatively, statistical control of baseline smoking can also address this issue to some extent. The strongest design would follow an entire population of youth beginning at an age at which risk of use of any product is negligible (e.g., 10 years old) and investigate time-varying associations between e-cigarette ever use and later combustible tobacco cigarette use at multiple developmental stages throughout the entire period of risk (e.g., up until age 29), while using multiple methods to establish temporal precedence of vaping relative to smoking. If consistent results are observed across all of the time-varying associations across development through adolescence and young adulthood, a stronger conclusion can be made with confidence in both internal validity and generalizability. Given the brief period in which e-cigarettes have been available, studies with such designs are not available. Hence, the committee considered the body of evidence and whether results were consistent across studies following youth from different regions (e.g., Europe versus North America), youth of different age ranges (e.g., early adolescence versus late adolescence versus emerging adulthood), and those applying different methods (e.g., eliminating baseline smokers versus statistical controls).

Among the longitudinal observational data that are currently available, for the first research question—Among adolescents and young adults, does e-cigarette use impact risk of ever smoking?—the committee considered studies that removed ever smokers at baseline from the analytical sample and assessed ever use of smoking at follow-up to reflect the strongest evidence of temporal precedence. For the second research question—Among adolescents and young adults, does e-cigarette use impact risk of progression in smoking frequency, intensity, and duration?—the strongest study design for establishing temporal precedence would include at least three time points (see Figure 16-1): a wave 1 in which ever smokers would be eliminated from the sample and e-cigarette use would be assessed as the primary exposure variable. The outcome would address trends in smoking status over time to capture the persistence of use (e.g., smoked since the previous wave of assessment or in the past 30 days leading up to the assessment [yes/no]) as well as trends in frequency (e.g., days smoked in the past 30) or intensity (e.g., cigarettes smoked per day on smoking days). In the case of multiple waves of follow-up data, evidence of a positive baseline e-cigarette use by time interaction term for either smoking status, frequency, or intensity would indicate that e-cigarette use is associated with increased likelihood or speed of progression in the respective outcome, whereas a negative interaction term would indicate that e-cigarette use is associated with reduced likelihood or speed of progression.

Other strategies employed for addressing whether e-cigarette use was associated with progression include two time points, but by contrast with evidence addressing initiation, studies with only two waves would not merely assess ever use at follow-up. Given that youth who ultimately progress to become frequent, heavy, and chronic smokers in adulthood are of greatest public health importance, the weakest evidence of progression to regular use in the two time-point design includes studies that measure recent (e.g., past 30-day) smoking status (yes/no) at follow-up. While past 30-day use typically involves a higher level of smoking than the smoking ever use outcome, the past 30-day outcome is considered a weak indicator of likelihood of progression. Among U.S. high school student past 30-day smokers in 2014, 37.0 percent smoked only 1 or 2 days per month, 22.6 percent smoked every day, and 40.4 percent smoked intermittently (Neff et al., 2015), which underscores the uncertainty of past 30-day use status. Instead, studies that include more detailed measures of smoking frequency that distinguish among monthly, weekly, and daily use and smoking intensity that identify the number of cigarettes smoked per smoking day were considered to provide stronger evidence to address smoking progression. Among studies including only two time points that addressed smoking progression, those that did not eliminate baseline ever smokers and used an alternate strategy for addressing reverse causation (e.g., statistical adjustment of baseline combustible tobacco use, elimination of baseline current combustible tobacco users without eliminating baseline past combustible tobacco users) were considered to provide less robust evidence of temporal precedence of an association of e-cigarette use with smoking progression (Etter, 2017).

Studies in which more frequent or chronic use of e-cigarettes is associated with proportional differences in combustible tobacco use initiation or progression in frequency, intensity, or duration provided further support of causation. Studies using e-cigarette ever use as an assessment of e-cigarette exposure amalgamate youth who may have used e-cigarettes on only one occasion with those who are regular vapers. Such studies were deemed to provide weaker evidence than studies that provide more fine-grained differentiation along a continuum of e-cigarette exposure (e.g., temporary use versus recent use on a monthly basis, recent weekly use versus recent daily use) capable of testing dose–response effects.

Supportive Evidence

While the major emphasis of the evidence review focused on the association of e-cigarette use with smoking initiation and progression in observational data, supplementary lines of evidence were used to provide additional information to address the research questions. Just as studies on the effects of e-cigarettes on health outcomes drew upon in vivo animal and in vitro studies as evidence supporting the biological plausibility of a hypothesized disease pathway, study results that address the specificity of hypothesized biological and psychological mechanisms proposed by the diversion and catalyst hypotheses, respectively, were considered by the committee as supportive lines of evidence of the plausibility that an association identified in the primary review was causal.

Tests of whether the association of e-cigarette use with ever smoking and smoking progression differs by baseline smoking risk status captured by the moderator variables of psychological traits (e.g., rebelliousness) and cognitive susceptibility to smoking (e.g., reported interest and willingness in trying smoking in the future; see Figure 16-1) reflect one line of evidence. That is, evidence of a statistical interaction between e-cigarette use and a moderating variable indicative of a general liability to smoking that is outside of the putative causal pathway from e-cigarette use to changes in smoking was considered. In such research, results consistent with the diversion hypothesis that e-cigarette use prevents ever smoking and smoking progression would demonstrate that there would be a negative association between e-cigarette use and smoking that becomes stronger in higher (versus lower) risk youth in stratified analyses whereby the e-cigarette–combustible tobacco cigarette association is estimated in subgroups differing in level of liability for smoking captured by other moderator variables outside of the causal pathway (e.g., rebelliousness). Results consistent with the catalyst hypothesis would show a positive association between e-cigarette use and smoking initiation that becomes stronger in youth who score lower (versus higher) on such moderator variables.

Also considered were studies examining whether e-cigarette use is associated with changes in intermediate psychosocial mediators that putatively link e-cigarette use with subsequent changes in smoking risk. The mediator variables would include those that are based on concepts proposed in either the diversion or catalyst hypothesis. For instance, evidence that a positive association between baseline e-cigarette use and later combustible tobacco cigarette smoking is mediated by increases in beliefs that smoking is enjoyable would be interpreted as evidence consistent with the catalyst hypothesis and that e-cigarette use increases risk of smoking initiation. Studies demonstrating that e-cigarette use is associated with a subsequent reduction in curiosity or interest in trying cigarettes would support the diversion hypothesis and that e-cigarette use reduces smoking risk. As enjoyment of nicotine's pleasurable effects are hypothesized mechanisms in the catalyst model, studies of whether e-cigarette nicotine concentration is associated with smoking were also considered as relevant to determining plausibility. Qualitative research involving direct testimonials of youth e-cigarette users who explained why they did or did not start smoking were also considered in determining plausibility.

Finally, the committee drew upon additional forms of evidence that provided relevant data to help draw inferences, including ecological studies and studies of other non-cigarette tobacco products. Ecological studies of whether the slope of changes in the prevalence of e-cigarette use and smoking over time are in parallel or opposing directions would cohere with interpretations that e-cigarette use increases or reduces smoking, respectively, and were considered by the committee. As randomized controlled experiments addressing the research questions are unavailable, naturalistic experiments comparing smoking in communities with restrictive e-cigarette use policies for youth (e.g., restrictions against sales to minors) compared with those with permissive youth policies (e.g., no sales restrictions) were considered to provide indirect evidence of the causal effect. Analogous evidence of associations among use of tobacco products popular among youth that are similar to e-cigarettes and cigarette smoking could also provide indirect evidence of causal mechanisms. Like e-cigarettes, hookah and cigarillos are available in characterizing flavors preferred by youth, have not been subject to federal regulation, and have increased in popularity among youth over the past decade. Thus, the committee considered whether analogous associations of relevant, non-cigarette tobacco product use and cigarette smoking initiation existed to interpret the causal effect of e-cigarettes on smoking. Furthermore, the committee reviewed evidence of analogous associations between e-cigarette use and uptake of non-cigarette combustible tobacco products. The committee also reviewed evidence regarding whether e-cigarette use was associated with other risk behavior outcomes (e.g., cannabis use). Among the putative psychosocial and biological mechanisms linking e-cigarette use and changes in smoking proposed by the diversion and catalyst hypotheses are non-specific mechanisms that would also be expected to change risk of other risk behavior outcomes; other mechanisms would be expected to be specific to nicotine and tobacco products. For example, e-cigarette use resulting in satiation of the desire for novelty in the diversion hypothesis and in feeling emboldened to take risks in the catalyst hypothesis are non-specific mechanisms. The enjoyment of the pharmacological effects of nicotine via e-cigarettes in the catalyst hypothesis is a specific mechanism expected to directly increase risk of nicotine and tobacco products more strongly than other risk behaviors. Thus, evidence that the (positive or negative) association between e-cigarette use and subsequent combustible tobacco cigarette use is stronger than the corresponding association of e-cigarette use with another risk behavior, such as cannabis use, would suggest specificity of the association and further strengthen the conclusion from the primary review.

For this supportive evidence, the committee considered study results consistent with the epidemiological data on the association and plausibility as confirmatory evidence to strengthen the weight of conclusions. However, because a causal effect can exist in the absence of supporting ecological or analogous evidence due to potential methodological or conceptual limitations that restrict causal inferences from such studies, causal conclusions could be made in the presence of inconsistent evidence across epidemiological studies and these additional forms of evidence.

Systematic Review

Results

Nine studies met inclusion criteria (n = 16,621): seven provided data to address the association with smoking initiation and two addressed the association with past 30-day smoking.

Ever-use analysis The ever-use analysis in Soneji and colleagues (2017) can be used for evaluation of the first research question in this section: Among adolescents and young adults, does e-cigarette use affect risk of ever smoking? Each of the seven ever-use studies eliminated ever smokers of combustible tobacco cigarettes at baseline, providing strong evidence of temporal precedence and eliminating some reverse causation explanations (i.e., past smokers seek out e-cigarette use by baseline and then subsequently return to smoking at follow-up). The combined unadjusted and adjusted ORs for the association of e-cigarette–ever use with ever cigarette use across the seven studies were OR = 3.83 (95% CI = 3.74–3.91) and OR = 3.50 (95% CI = 2.38–5.16), respectively, which reflect strong magnitudes of association. Results are shown in Table 16-4. The seven studies were performed by four independent research groups, included a variety of methodologies (i.e., paper-and-pencil surveys, the Internet, phone), sampling regions and strategies (i.e., three national samples, two from southern California, one from Hawaii, and one from Virginia), and age ranges (mean age at baseline range = 14.1–23.5 years). Given the differences across studies and that each study individually found a statistically significant positive association between e-cigarette ever use and smoking initiation, strong evidence in consistency of the association was determined.

TABLE 16-4

Meta-Analysis of Unadjusted and Adjusted Odds of Ever Smoking Combustible Tobacco Cigarettes Among Combustible Tobacco Cigarette–Never Smokers at Baseline and E-Cigarette–Ever Users at Baseline Compared with E-Cigarette–Never Users (more...)

The probability of transition from never use to use of combustible tobacco cigarettes during the follow-up period for baseline–e-cigarette–ever users ranged from a low of 7.9 percent in a sample of 9th-grade students (Leventhal et al., 2015) to 40.4 percent in a sample of 11th- and 12th-grade students (Barrington-Trimis et al., 2016a). These estimates of ever use are high compared to previous results for smoking over 12-month follow-up periods in youth and young adults (HHS, 2014) and the comparison groups of e-cigarette–never users in each study for which use over the follow-up period ranged from 3.0 percent (Leventhal et al., 2015) to 10.6 percent (Spindle et al., 2017) across studies.

The associations in each of the seven studies were statistically significant and there was heterogeneity in the combined estimate, suggesting variation in effect magnitude across the studies. The authors attempted to explain heterogeneity via subgroup analysis, which found non-significant heterogeneity estimates when the set of studies was limited to the six with young adults only, three studies conducted after 2014, and three nationally representative studies. A limitation of this finding is that a formal interaction test was not conducted to determine whether age, time of publication, or regional versus nationally representative sampling significantly moderated the combined OR estimate across the two groups of studies, perhaps because of the small number of studies precluding formal interaction tests.

This review used a standardized assessment of risk of bias of the individual studies, all of which were deemed as moderate due to potential confounding. There was variability in the types and number of confounding covariates included in the primary articles. Some papers included a more comprehensive set of covariates addressing plausible shared risk factors across demographic, environmental, and intrapersonal/endogenous dis-positional domains, providing a more rigorous evaluation of the degree to which an observed association between e-cigarette use and subsequent smoking is direct (i.e., not due to confounding), whereas some included demographic factors and excluded important environmental or intrapersonal factors. The relative difference in ORs between unadjusted and adjusted estimates varied across the seven studies included in the meta-analysis (see Table 16-4). Some studies found larger ORs for unadjusted than adjusted estimates (Leventhal et al., 2015; Miech et al., 2017; Wills et al., 2016b). Others found larger ORs for the adjusted than the unadjusted estimates (Primack et al., 2015, 2016). For two studies, the unadjusted and adjusted ORs did not meaningfully differ (Barrington-Trimis et al., 2016a; Spindle et al., 2017). The net results of this pattern across studies were combined estimates of unadjusted OR = 3.83 (95% CI = 3.74–3.91) and adjusted OR = 3.50 (95% CI = 2.38–5.16) that did not markedly differ from one another as evidenced by highly overlapping confidence intervals for the two estimates. Variation in the disparity in OR estimates between unadjusted and adjusted results across studies could be due to cross-study differences in covariate adjustment. However, inspection of the covariates adjusted for in each study failed to reveal a systematic effect (i.e., the studies with more comprehensive covariate adjustment did not necessarily show reductions in OR estimates from unadjusted to adjusted models). Given this pattern of results, it is difficult to determine the potential influence of residual confounding by unmeasured variables on the associations observed.

An important limitation was the high loss to follow-up in six of the studies (greater than 20 percent). The studies addressed attrition in a number of ways, including (1) comparing associations with complete case analysis to results from analysis using maximum likelihood estimation methods (Wills et al., 2016b), (2) using alternative assumptions that all lost to follow-up are either initiators or non-initiators (Leventhal et al., 2015), and (3) using auxiliary variables to estimate and impute outcome data (Primack et al., 2015). In no cases did the results substantively differ when comparing differing methods for addressing attrition. Because youth and young adults who are lost to follow-up are typically more likely to possess risk factors for cigarette use in previous research (Young et al., 2006) as well as being e-cigarette users at baseline (Barrington-Trimis et al., 2016a; Leventhal et al., 2015; Wills et al., 2016c), the differential loss to followup would have resulted in selection bias that most likely would have suppressed the magnitude of the association of the individual articles reviewed for the meta-analysis.

Past 30-day use analysis The combined unadjusted and adjusted ORs for the association of past 30-day e-cigarette use with past 30-day cigarette use were OR = 5.68 (95% CI = 3.49–9.24) and OR = 4.28 (95% CI = 2.52–7.27), respectively (see Figure 16-2). For the past 30-day use analysis, only two primary literature studies were included (Hornik et al., 2016; Unger et al., 2016), raising questions about the generalizability and stability of the combined OR estimate for past 30-day associations. Additionally, youth with a prior history of smoking were permitted in the past 30-day use analysis, which was not adjusted for in one of the studies, allowing for reverse causation. Furthermore, past 30-day use does not address ever use, per se, and provides an imprecise estimate of progression to more frequent, heavy, or chronic smoking. Hence, the past 30-day use analysis in Soneji and colleagues (2017) did not provide rigorous evidence to evaluate either the ever use or progression research questions.

FIGURE 16-2

Meta-analysis of adjusted odds of current (past 30-day) combustible tobacco cigarette smoking at follow-up among non-current combustible tobacco cigarette smokers at baseline and current e-cigarette users at baseline compared with non-current e-cigarette (more...)

Other Original Studies

Leventhal and colleagues (2015) examined the association between baseline ever use of e-cigarettes with past 6-month smoking status at 6- and 12-month follow-ups among 2,530 baseline never smoker ninth-grade students in Los Angeles. Although this article was reviewed by Soneji and colleagues (2017), they focused on the bivariate association between e-cigarette use and ever-smokers averaged across follow-up, without considering the effects across multiple time points reported in Leventhal and colleagues (2015), which addresses the question of whether e-cigarette use is associated with duration of smoking. Hence, the multiple-time-point analysis in Leventhal and colleagues (2015) was reviewed by the committee. The exposure variable was ever versus never e-cigarette use at baseline in this study. The outcome was smoking status over the previous 6 months (any smoking versus no smoking). The study found no significant interactions between e-cigarette ever use and time in the prediction of combustible tobacco cigarette smoking (interaction adjusted OR = 0.74; 95% CI = 0.34–1.61, p = 0.44), which suggests that the likelihood of persistence (versus discontinuation) of smoking across the 6- and 12-month follow-ups was non-significantly lower among e-cigarette users. Of baseline e-cigarette–ever users compared to e-cigarette–never users, 9.7 percent versus 3.0 percent smoked in months 1 through 6 of follow-up and 7.9 percent compared to 3.3 percent had smoked in months 7 through 12 of follow-up, respectively. An additional analysis found that baseline e-cigarette–ever users (versus never) were more likely to be sustained users of any combustible tobacco product across both follow-ups (10.0 percent versus 3.6 percent). This study had a high follow-up rate (98 percent), adjusted for a comprehensive set of covariates and potential confounders, and removed baseline smokers to provide temporal precedence. The study was limited because dose–response effects could not be evaluated given the e-cigarette ever use exposure variable. In sum, no conclusive evidence of whether e-cigarette use is associated with duration of smoking among initiators was found in this study.

Wills and colleagues (2016b) reported associations of e-cigarette use with smoking in 9th and 10th graders in Hawaii who were surveyed in 2013 (baseline) and followed up 1 year later. While included in the Soneji meta-analysis, this study was individually reviewed by the committee because it included supplementary analyses of e-cigarette and combustible tobacco cigarette use frequency estimates not addressed in Soneji and colleagues (2017). In analyses of dose–response associations with smoking initiation among 1,070 adolescent never smokers at baseline, the probability of ever smoking at 1-year follow-up was 5 percent among baseline never vapers, which was significantly lower than the probability of ever smoking for youth who had vaped one to two times in their life (14 percent versus 5 percent; adjusted OR = 2.88; 95% CI = 1.96–4.22), three to four times in their life (11 percent versus 5 percent; adjusted OR = 2.29; 95% CI = 1.35–3.87), yearly/monthly (19 percent versus 5 percent; adjusted OR = 4.17; 95% CI = 2.03–8.57), or weekly/daily (19 percent versus 5 percent; adjusted OR = 4.09; 95% CI = 2.43–6.88) at baseline after adjusting for demographic-, environmental-, and psychological/personality-related covariates. Although these estimates suggest a possible threshold effect for association between e-cigarette use frequency and smoking initiation for the two higher versus two lower e-cigarette exposure groups, pairwise comparisons with the Tukey-Kramer adjustment indicated that the four frequency levels of e-cigarette use did not differ significantly from each other in smoking initiation likelihood. Pairwise tests may have been limited in power due to smaller sample sizes for individual group pairs. In the same subsample, analyses examined dose–response associations with probability of initiating and progressing to smoking at least a few times per month or more (versus no initiation or initiating, but not progressing to at least a monthly smoking frequency). The probability of initiating and progressing to monthly smoking was not different between adolescents who vaped one to four times versus never in their life at baseline (0.6 percent versus 1.0 to 1.2 percent, adjusted OR < 1.88), but was higher among those who vaped yearly/monthly (4.2 percent versus 0.6 percent, adjusted OR = 7.13; 95% CI = 1.28–39.73) or weekly/daily (9.7 percent versus 0.6 percent: OR = 17.19; 95% CI = 7.24–40.79) versus never vaped at baseline. These estimates suggest a possible threshold effect whereby at higher frequency levels of vaping, the likelihood of initiating and progressing to more frequent smoking is increased, but at low levels of vaping frequency the likelihood of progression in smoking frequency is not changed. However, the small sample size for the individual groups, and resulting wide confidence intervals of the estimates, tempers conclusive inferences regarding possible threshold effects. In sum, Wills and colleagues (2016b) show suggestive evidence that “dose” of e-cigarette exposure (i.e., use frequency) may differentiate likelihood of smoking initiation and progression in frequency.

Leventhal and colleagues (2016) examined the association between baseline e-cigarette use frequency with progression of combustible tobacco cigarette use frequency and intensity at a 6-month follow-up in 3,084 10th-grade high school students in Los Angeles. This paper included the same cohort as Leventhal and colleagues (2015), but used different time points. This study was not included by Soneji and colleagues (2017) because it was published after their analysis. The e-cigarette use exposure variable was defined as a four-level gradient variable (never versus ever use with no use in past 30 days [prior use] versus 1–2 days in past 30 [current infrequent use] versus ≥3 days in past 30 [current frequent use]). The combustible tobacco cigarette use frequency outcome was a three-level variable based on days used in past 30 (0 versus 1–2 days versus 3 days of use or more). The intensity of combustible cigarette use outcome was characterized with a four-level variable based on the amount of smoking per smoking day (no smoking versus less than one cigarette versus a whole cigarette versus two cigarettes or more). Adjusting for baseline smoking and other covariates, ordinal logistic regression found that each increment higher on the four-level baseline vaping level was associated with proportionally higher odds of smoking at a greater level of frequency (OR = 1.37; 95% CI = 1.16–1.61) and intensity (OR = 1.26; 95% CI = 1.07–1.48) by follow-up. Adjusting for baseline smoking, pairwise ordinal logistic regression results suggested a dose–response association for smoking frequency, yielding ORs of 4.61 (95% CI = 2.49–8.53) for prior (versus never) baseline vaping, 6.60 (95% CI = 3.48–12.51) for current infrequent (versus never) baseline vaping, and 10.62 (95% CI = 6.46–17.46) for current frequent (versus never) baseline vaping. This study also demonstrated that the positive association between baseline e-cigarette using and follow-up smoking frequency and intensity was stronger among baseline non-smokers than baseline infrequent and frequent smokers in the form of a statistical interaction. In baseline non-smokers, there was evidence of a dose–response association such that smoking frequency at follow-up increased proportionately across baseline vaping frequency, as follows: baseline never users (infrequent smokers = 0.7 percent; frequent smokers = 0.5 percent); prior vapers (infrequent = 3.9 percent; frequent = 2.3 percent); infrequent vapers (7.1 percent, 3.6 percent); and frequent vapers (5.4 percent, 9.7 percent). In baseline smokers, vaping was not significantly associated with smoking frequency at follow-up. Lower strength of vaping–smoking associations in baseline smokers (versus nonsmokers) has been reported elsewhere (Conner et al., 2017; Miech et al., 2016; Wills et al., 2016b). As reported in Leventhal and colleagues (2016), this result could suggest that e-cigarette use is more strongly associated with the onset (or return) to smoking as well as the progression to higher levels of cigarette use over the follow-up period, but may have less of an effect among those who are already smokers—many of whom may have started vaping after smoking initiation. In several other studies that report results by baseline smoking status (Conner et al., 2017; Leventhal et al., 2016; Miech et al., 2016; Wills et al., 2016b), the association between vaping and subsequent smoking was null or weakly positive among baseline smokers, which suggests that e-cigarette use is not associated with smoking reduction or cessation in youth and young adults. A strength of the study was the comprehensive adjustment of covariates, which included 15 demographic, environmental, and intrapersonal factors, and high participation (80 percent or more) and retention (99 percent) rates. A limitation was the brief follow-up and the inclusion of youth with a history of smoking, which treated ever smoking as a covariate and baseline current smoking frequency as a moderator rather than eliminating baseline ever-smokers from the analysis to prevent reverse causation. Overall, this study provided fairly strong evidence of dose–response associations between e-cigarette use and progression in smoking frequency and intensity for non-smokers at baseline.

In a sample of 391 non-daily combustible tobacco cigarette smokers age 18 to 24 in California, Doran and colleagues (2017) conducted a longitudinal study that examined whether e-cigarette use frequency was associated with changes in use of combustible tobacco cigarettes over a 12-month follow-up period. This study was published after the Soneji and colleagues (2017) analysis. E-cigarette use frequency was initially assessed via a five-level variable indicative of use over the previous 6 months (never; one to three times; one to two times per month; weekly; two to four times per week; and daily/almost daily). At each follow-up evaluation, the investigators assessed the number of days smoked (i.e., frequency) and total amount of cigarettes smoked (i.e., intensity) over the previous 14 days. Negative binomial regression models adjusting for baseline cigarette use and a propensity score based on a prediction model involving a number of demographic, environmental, and intrapersonal covariates was used. There was a positive association between initial e-cigarette use frequency with cigarette use frequency and intensity at follow-up. For smoking intensity, the incidence rate ratio (IRR) was 1.13 (95% CI = 1.06–1.21), indicating that each one-category increase in initial e-cigarette frequency (e.g., from one to three uses in 6 months to monthly use) was associated with a 13 percent greater number of total cigarettes reported at the second assessment. Analyses of trends in smoking over time across multiple follow-up time points showed a significant interaction for smoking quantity, indicating that this gap widened over time (IRR = 1.16; 95% CI = 1.09–1.23), as illustrated by the higher slope in cigarette intensity across the follow-ups for individuals who used e-cigarettes at a greater frequency at the initial assessment. For smoking frequency outcomes, the interaction between e-cigarette use frequency and time was not significant. A strength of this study was the assessment strategy using frequent follow-ups and short interassessment intervals (i.e., once every 3 months for five total assessments), detailed characterization of exposure and outcome variables to show dose–response relations, and high retention rate (95 percent). In addition, the covariates used to generate a propensity score were comprehensive. Because all participants were smokers at baseline, it is possible that most started smoking before using e-cigarettes, leaving the temporal precedence of the association unclear. It is possible that there is a selection bias whereby those most at risk who are likely to escalate their smoking start using e-cigarettes as a means to help with cravings or because of unsuccessful efforts to quit (rather than a causal effect whereby using e-cigarettes intensifies smoking progression). However, the authors included intention to quit in the propensity score that was adjusted for in the analysis. In sum, this study provides suggestive evidence that e-cigarette use may be associated with increased progression in smoking intensity and in frequency to some degree.

Best and colleagues (2017) examined the association of baseline e-cigarette–ever use (versus never) with cigarette use initiation at 1-year follow-up among 11- to 18-year-old baseline never smokers (n = 2,125) enrolled in four schools in Scotland from 2015 to 2016. This study was published after the Soneji and colleagues (2017) analysis. In an unadjusted model, the OR for ever smokers at follow-up in e-cigarette–ever users versus e-cigarette–never users was 4.62 (95% CI = 3.34–6.38). Adjusting for demographics, susceptibility to smoking, and family/peer smoking, the OR remained significant and was 2.42 (95% CI = 1.63–3.60). A limitation of this study was that other intrapersonal covariates indicative of a propensity toward risk-taking behavior were not included. Also, there was a modest retention rate (70.8 percent), which may reduce the magnitude of the association due to the possibility of dropouts being disproportionately likely to use e-cigarettes at baseline and smoke at follow-up. Finally, e-cigarette exposure level was not reported, precluding investigation of dose–response associations. Strengths are elimination of baseline never smokers from the analytical sample, which permits temporal conclusions about the association that e-cigarette use at baseline was related to an increased probability of transition to becoming an ever smoker at followup. In sum, the study methods and results are highly similar to the seven original studies included in the Soneji and colleagues (2017) meta-analysis of initiation and suggests generalization of e-cigarette use–smoking initiation associations to a sample outside the United States.

Conner and colleagues (2017) investigated the association of e-cigarette use at baseline and smoking at a 1-year follow-up among 2,836 adolescents (age 13 to 14 years at baseline) in 20 schools in England from 2014 to 2015. In baseline never smokers (n = 1,726), probability of transitioning from never to ever combustible cigarette use at follow-up was associated with ever (compared to never) use of e-cigarettes at baseline (OR = 5.38; 95% CI = 4.02–7.22), which remained significant when controlling for covariates (OR = 4.06; 95% CI = 2.94–5.60). In adolescents who had tried cigarettes but were not current smokers at baseline (n = 318), ever use was associated with progression in smoking frequency at follow-up (OR = 2.16; 95% CI = 1.01–4.62), which became non-significant when controlling for covariates (OR = 1.89; 95% CI = 0.82–4.33). A unique strength of this study was that smoking was biochemically verified by breath carbon monoxide (CO) readings, and those who had recently smoked or who were defined as having progressed showed significantly higher CO than non-smokers. Due to its half-life, CO is only detected after recent smoking and is thus an insensitive measure for detecting ever use or infrequent smoking. Also, there was a comprehensive set of covariates adjusted for that help to rule out confounding effects. There was a moderate 21 percent attrition rate speculated by the authors to be due to a failure to correctly match anonymous code numbers for students across assessments. Attrition analyses found modest differences between those with and without follow-up data on key variables and no differences in baseline e-cigarette use, suggesting that the impact of attrition on association estimates was small. Adolescents who indicated “I have only tried smoking once” or “I used to smoke sometimes, but I never smoke cigarettes now” at baseline and then selected one of the following, “I sometimes smoke cigarettes now, but I don't smoke as many as one a week,” “I usually smoke between one and six cigarettes a week,” and “I usually smoke more than six cigarettes a week” at follow-up were classified as having progressed in their smoking. This definition is highly circumscribed and fails to differentiate among low-, moderate-, or high-frequency levels. Also, due to a small number of cases of frequent e-cigarette use, the e-cigarette exposure variables were based on ever use, leaving unclear possible dose–response effects. In sum, this study provides evidence that e-cigarette use is associated with smoking initiation in British youth and weak and inconclusive evidence that e-cigarette use is associated with progression to more frequent smoking.

Selya and colleagues (2017) examined the association between past 30-day e-cigarette and combustible tobacco cigarette use frequency in a young adult sample (age 19–23) originally recruited to be enriched with smokers attending high schools in Chicago, Illinois (81.1 percent had a history of smoking). Participants were surveyed annually four times from 2011 to 2015 and a cross-lagged structural equation model was used to examine the average estimate of association of e-cigarette use with combustible tobacco cigarette use at the subsequent wave for all prospective paths (i.e., Wave 1 → Wave 2, Wave 2 → Wave 3, Wave 3 → Wave 4). The effects of combustible tobacco cigarette dependence severity and use at the previous wave was adjusted for and the lowest possible score on the smoking dependence scale was imputed for never smokers. The results showed that e-cigarette use was not associated significantly with later increases in combustible tobacco cigarette smoking (β = 0.02, p = 0.08). This study was not well positioned to address whether e-cigarette use contributed to ever smoking and subsequent progression of smoking behavior because the sample was recruited to be enriched for smoking in 2006 (prior to the availability of e-cigarettes). Thus, it is likely that the majority of the sample began using combustible tobacco cigarettes prior to using e-cigarettes, including those who sought out e-cigarettes as a cessation aid. The authors acknowledge this limitation and that they did not adjust for any possible confounders of the association. In sum, this study provides negligible evidence that can be used to address the research questions posed in this review.

Loukas and colleagues (2018) administered four semiannual surveys to 2,558 never smoking 18- to 25-year-old (19.71 ± 1.61) students from 24 Texas colleges four times over a 1.5-year follow-up period beginning in 2014–2015 with retention rates ranging from 79 percent to 81 percent across waves. Transition probability from baseline never smoking to ever smoking during the follow-up period was 20.1 percent versus 8.4 percent for baseline e-cigarette–never users (versus ever). Multivariable, multilevel discrete-time hazard models indicated that baseline e-cigarette–ever use was associated with subsequent transition to combustible tobacco cigarette ever smoking (adjusted OR = 1.36; 95% CI = 1.01–1.83) after adjusting for demographics, combustible tobacco cigarette use susceptibility, family-of-origin tobacco use, friend cigarette use, and other tobacco product use. Analyses stratified by other tobacco product use at baseline showed that e-cigarette–ever use was associated with greater odds of combustible tobacco cigarette–ever use during follow-up in baseline never users of cigars, hookah, or smokeless tobacco (OR = 2.26; 95% CI = 1.35–3.76) but not among ever users of other products at baseline (OR = 1.13; 95% CI = 0.81–1.58). The strengths of the study include a longer follow-up period than most prior research (i.e., 18 months here versus 12 months) and comprehensive adjustment of interpersonal risk factors for smoking. The statistical adjustment of use of other tobacco products and the estimation of associations among young adults who had never used any other tobacco products is another strength, which suggests that the associations were not explained entirely by a non-specific constellation of poly-tobacco product use behavior that generalizes to any form of nicotine or tobacco use. Limitations include the application of ever use in operationalizing the e-cigarette exposure, precluding investigation of dose–response associations between level of e-cigarette use and odds of becoming an ever smoker, and the omissions of some covariates that may provide a more comprehensive adjustment for endogenous liability to smoking (e.g., sensation seeking, depression). In sum, this study provides further evidence of an association of e-cigarette use with transition to ever smoking in young adult college students.³

Supporting Evidence

Trends in E-Cigarette and Combustible Tobacco Cigarette Use in the Adolescent and Young Adult Population

Trends in the prevalence of e-cigarette use relative to combustible tobacco cigarette smoking over time may also shed light on whether and how e-cigarette use affects smoking. Holding all other factors constant, if e-cigarette use increased risk of smoking, then changes in the prevalence of e-cigarette use should parallel changes in the prevalence of smoking. If e-cigarette use prevented combustible tobacco cigarette smoking, changes in the prevalence of e-cigarette use and smoking over time should oppose each other. Evidence from the National Youth Tobacco Survey (NYTS) show that from 2011 to 2016, the past 30-day use of e-cigarettes increased (non-linearly) and combustible tobacco cigarette use decreased linearly in middle and high school students (Jamal et al., 2017) (see Figure 16-3).

FIGURE 16-3

Past 30-day use of e-cigarettes and combustible tobacco cigarettes among high school and middle school students in the 2011–2016 National Youth Tobacco Survey. SOURCE: Jamal et al., 2017.

The population-level ecological data are more consistent with the diversion hypothesis than the catalyst hypothesis and run counter to individual-level results, which predominately show that youth and young adults who use e-cigarettes are more likely to become ever smokers and past 30-day smokers. A more granular examination of the population-level data indicates that the pattern of trends in use for the two products are not unequivocally concordant. There is a non-linear trend for past 30-day e-cigarette use in NYTS high school students, which increased from 1.5 percent in 2011 to 16.0 percent in 2015 and then decreased to 11.3 percent in 2016. For past 30-day combustible tobacco cigarette use in the NYTS, use decreased from 15.8 percent in 2011 to 9.3 percent in 2015 and was 8.0 percent in 2016. Due to the methodological limitations of interpreting causality from this descriptive population-level trend analysis and the multitude of influences on population-level e-cigarette use (e.g., policy changes, cultural changes, historical cohort effects), the ecological analysis does not provide strong evidence to rule out a possible risk-enhancing effect of e-cigarette use on youth smoking. Some population-based ecological studies report that downward trends in smoking rates in youth across time stem back for decades. The rate of such reductions did not accelerate over the past several years after e-cigarette use was introduced into the U.S. market and became popular among youth (Barrington-Trimis et al., 2016a,b). In an analysis of NYTS data from 2004 to 2014, interrupted time series of ever (1 puff or more) and current (last 30 days) smoking comparing the rate of deceleration in combustible tobacco cigarette use before and after 2009 when e-cigarettes first became available showed no significant change in the overall linear trend in reduction in smoking by year before versus after 2009 (p = 0.57 and 0.23). Hence, the overall concordance in the ecological data is not clear. Overall, the population-based data broadly show opposing trends in e-cigarette and cigarette use prevalence across time among U.S. youth in recent years and thus do not provide confirmatory evidence of the epidemiological person-level positive associations of vaping and smoking.