Foxon et al.

BMC Public Health (2022) 22:1940 BMC Public Health

https://doi.org/10.1186/s12889-022-14341-z

RESEARCH Open Access

Population-level counterfactual trend

modelling to examine the relationship
between smoking prevalence and e-cigarette
use among US adults
Floe Foxon1*, Arielle Selya1, Joe Gitchell1 and Saul Shiffman1

Abstract
Background  Studies have suggested that some US adult smokers are switching away from smoking to e-cigarette
use. Nationally representative data may reflect such changes in smoking by assessing trends in cigarette and
e-cigarette prevalence. The objective of this study is to assess whether and how much smoking prevalence differs
from expectations since the introduction of e-cigarettes.
Methods  Annual estimates of smoking and e-cigarette use in US adults varying in age, race/ethnicity, and sex were
derived from the National Health Interview Survey. Regression models were fitted to smoking prevalence trends
before e-cigarettes became widely available (1999–2009) and trends were extrapolated to 2019 (counterfactual
model). Smoking prevalence discrepancies, defined as the difference between projected and actual smoking
prevalence from 2010 to 2019, were calculated, to evaluate whether actual smoking prevalence differed from those
expected from counterfactual projections. The correlation between smoking discrepancies and e-cigarette use
prevalence was investigated.
Results  Actual overall smoking prevalence from 2010 to 2019 was significantly lower than counterfactual
predictions. The discrepancy was significantly larger as e-cigarette use prevalence increased. In subgroup analyses,
discrepancies in smoking prevalence were more pronounced for cohorts with greater e-cigarette use prevalence,
namely adults ages 18–34, adult males, and non-Hispanic White adults.
Conclusion  Population-level data suggest that smoking prevalence has dropped faster than expected, in ways
correlated with increased e-cigarette use. This population movement has potential public health implications.
Keywords  Adult, E-Cigarette, Vaping, Counterfactual, NHIS, Smoking, Trend modelling

*Correspondence:
Floe Foxon
[email protected]
1
PinneyAssociates Inc, 201 North Craig Street, Suite 320, 15213 Pittsburgh,
PA, USA

© The Author(s) 2022. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use,
sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and
the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this
article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included
in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will
need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The
Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available
in this article, unless otherwise stated in a credit line to the data.
Foxon et al. BMC Public Health (2022) 22:1940 Page 2 of 10

Background population level is to model expected trends in smoking

The net effect of e-cigarette use on cigarette smoking at prevalence, and then assess whether the introduction of
the population level is not well quantified. E-cigarette e-cigarettes was associated with a net deviation from the
use could affect cigarette smoking in two main opposing expected smoking prevalence, either an increase (gate-
ways: First, e-cigarettes could act as a catalyst to smok- way) or decrease (substitution). Such modeling studies
ing among non-smokers (former and never) who would have generally found that the introduction of e-cigarettes
not have initiated or re-initiated cigarette smoking had was associated with more rapid declines in smoking prev-
it not been for e-cigarettes; such ‘gateway’ effects have alence [9, 26–29]. The present study uses this approach
been inferred among adolescents and young adults [1, 2]. to assess whether and how much the introduction of
Second, e-cigarettes could displace smoking via substitu- e-cigarettes in the US may be correlated with declining
tion, as smokers ‘switch’ from cigarettes to e-cigarettes, smoking prevalence among adults in the following ways.
and non-smokers are diverted from smoking initiation. To test whether declining smoking prevalence is corre-
The combination of these processes determines the lated with increasing e-cigarette use among adults, analy-
population-level impact of e-cigarette use on smoking ses examine subpopulations in which this correlation is
prevalence. especially likely to be evident. If e-cigarette use is corre-
Concerning the gateway, longitudinal studies have lated with smoking prevalence, the correlation should be
reported significant associations between e-cigarette use greater in populations with higher e-cigarette prevalence.
among non-smoking adolescents and subsequent smok- Use of electronic cigarettes by US adults is particularly
ing initiation [3]. However, Lee et al. [4] and Chan et al. concentrated among cigarette-smoking younger adults
[5] have argued that this effect is not causal, but rather and males [30–32]. Thus, discrepancies in expected ver-
due to common liabilities, that is, shared risk factors for sus actual smoking prevalence are examined in age, race/
both vaping and smoking, such as parental smoking and ethnicity, and sex cohorts whose e-cigarette use preva-
delinquent behavior, which predispose adolescents to lence differ.
both forms of nicotine use and which are not adequately
controlled in such analyses [6–8]. Conversely, it has been Methods
hypothesized that e-cigarette use among non-smoking Sample
adolescents may prevent those who otherwise would Annual smoking prevalence estimates for US adults were
have smoked cigarettes from doing so, as their e-cigarette derived from 29 waves of the National Health Interview
use may replace cigarette smoking, rather than lead to Survey (NHIS), an annual, cross-sectional, population-
cigarette smoking. This so-called ‘diversion’ effect has representative health survey with a geographically clus-
been observed in multiple studies [9–13]. tered sampling design administered by the US Centers
Concerning switching, randomized trials have indi- for Disease Control and Prevention’s National Center
cated potential for e-cigarettes to help adult smokers for Health Statistics [33]. NHIS provides trend data for
switch away from combustible cigarettes [14, 15], but cigarette smoking dating back decades, and has included
some of these studies have been criticized methodologi- data on e-cigarette use starting in 2014, thus providing an
cally [16]. Some cohort studies of individuals purchas- appropriate source of nationally representative data for
ing particular ENDS products in real-world settings have these analyses. NHIS interviewed 17,317–43,732 indi-
demonstrated high switching rates [17], with reduced viduals on their tobacco use behaviors each year between
cigarette consumption among dual users [18], and mini- 1990 and 2019. Data from the 2020 NHIS, while avail-
mal smoking initiation and relapse among baseline never able, were not included in the present study due to seri-
and former smokers using e-cigarettes [19, 20]. However, ous potential confounding by the COVID-19 pandemic
other cohort studies have come to opposite conclusions, both with respect to data collection procedures (shifting
suggesting that e-cigarette use does not prevent relapse from in-person to all-telephone interviews) as well as
to cigarette smoking [21, 22]. Using different analytic COVID-related impacts on cigarette smoking prevalence
techniques, economic studies examining cross-elastici- which are beyond the scope of this study.
ties between cigarettes and e-cigarettes have suggested Current smokers were defined as adults who had
these products are economic substitutes [23–25], which smoked at least 100 lifetime cigarettes and who ‘now’
would suggest that e-cigarette use would reduce the like- smoked cigarettes ‘every day’ or ‘some days’ [34].
lihood of smoking. Agent-based population modeling Similarly, current e-cigarette users were defined from
also suggests that the introduction of e-cigarettes would 2014 (the first year that e-cigarette use was assessed
be expected to reduce smoking prevalence [26]. in NHIS) to 2019 as those respondents who now used
Another useful approach to determining the overall e-cigarettes every day or some days. Cumulative lifetime
impact of e-cigarette use on smoking prevalence (i.e., the measures of e-cigarette use were not surveyed in NHIS,
combination of gateway and substitution effects) at the therefore ‘established’ use could not be defined. In 2014,
Foxon et al. BMC Public Health (2022) 22:1940 Page 3 of 10

e-cigarettes were defined as “electronic cigarettes, often Linear weighted least squares regression models were
called e-cigarettes” without explicit reference to nicotine. fitted to NHIS adult e-cigarette use prevalence from 2014
From 2015 to 2018, e-cigarettes were defined as “vape- (the first year e-cigarette use was assessed) to 2019 (with
pens, hookah-pens, e-hookahs, or e-vaporizers… usually e-cigarette use prevalence defined as zero in 2010). These
contain[ing] liquid nicotine.” In 2019, e-cigarettes were models were used to estimate e-cigarette use prevalence
defined as “Electronic cigarettes (e-cigarettes)… include from 2010 to 2013.
electronic hookahs (e-hookahs), vape pens, e-cigars, and The correlation between e-cigarette use prevalence
others… usually contain[ing] nicotine… These questions from 2010 to 2019 (model-based e-cigarette use esti-
concern electronic vaping products for nicotine use. The mates for 2010–2013; actual NHIS e-cigarette use preva-
use of electronic vaping products for marijuana use is lence estimates for 2014–2019) and cigarette smoking
not included in these questions.” For full definitions, see discrepancies from 2010 to 2019 was then investigated
Additional File 1. by calculating Pearson correlation coefficients with
The prevalence of current smoking and current e-cig- two-tailed p-values (alpha = 0.05). These analyses were
arette use was determined for three age cohorts (18–34 repeated for the three age cohorts (18–34, 35–54, 55+),
years, 35–54 years, and 55+ years, following Axelsson et three race/ethnicity cohorts (Hispanic, NH White, NH
al. [35]), three race/ethnicity cohorts (Hispanic, non-His- Black), and two sex cohorts (female, male). Goodness of
panic (NH) White, and NH Black), and two sex cohorts fit was evaluated with Root Mean Square Errors (RMSE),
(female and male). These cohorts were selected to maxi- which is appropriate for forecasts on means [38].
mize the sample sizes used in each prevalence estimate. NHIS prevalence estimates were calculated in SAS
Analyses were not repeated for any of the other race/ version 9.4 using PROC SURVEY to account for com-
ethnicity categories in NHIS due to very low sample size, plex survey design. All other analyses were performed in
producing coefficients of variation (relative standard Python version 3.7.6 with the packages NumPy version
error) >30% which is standard practice with NHIS data. 1.18.1, Scipy version 1.4.1, Uncertainties version 3.1.5,
[34]. Kneed version 0.7.0, and Matplotlib version 3.1.3.

Analyses Sensitivity tests

A cut-off year was determined from the NHIS data using In 2019, NHIS underwent a questionnaire redesign
the knee (or inflection point) identification algorithm which, among other changes, shortened the survey length
‘Kneedle’ published by Satopaa et al. [36]. This algorithm and changed the e-cigarette question wording (described
identified 2010 as the inflection point in NHIS cigarette above) [39]. As a sensitivity test, correlations were re-run
smoking prevalence data. 2010 was also used as the cut- excluding the 2019 point estimates (the last time point
off year in cigarette and e-cigarette use trend modelling analyzed) to account for possible variation in findings.
studies by Wagner and Clifton [29], and by Foxon and As a sensitivity test, correlations between smoking
Selya [9], and Selya and Foxon [11]. Indeed, data from discrepancies and e-cigarette use prevalence were re-
objective financial analyses by Wells Fargo and Agora calculated excluding the regression-estimated e-cigarette
Financial suggest minimal e-cigarette market presence prevalence estimates. Another sensitivity test examined
prior to 2010 compared to after [37]. the effect of alternative cut-off years centered around the
Linear weighted least squares regression models relat- Kneedle-identified cut-off year of 2010. Finally, a sen-
ing smoking prevalence to year were fitted from 1990 to sitivity test used an exponential decay function instead
2009 (before the cut-off ), and these were used to gener- of a linear function in the regression analyses, following
ate best-fit estimates for 2010–2019 (after the cut-off ) Foxon and Selya [9] (this form allows the change in users
to model the counterfactual: i.e., what would have been across time to depend on the number of users at a given
expected to happen to smoking prevalence in the US in time and is consistent with the hardening hypothesis
each year if e-cigarettes had not been introduced in 2010. [40]).
These projections were compared to the actual NHIS Finally, to consider the effect of major, distinct national
smoking prevalence estimates for 2010–2019. The dif- population interventions, the impacts of the FSPTCA
ference between these two – what we will refer to as the and the CDC’s ‘Tips®’ campaign were considered. This
‘discrepancy’ in cigarette smoking prevalence – can pro- was done by taking quantitative estimates for the associa-
vide information on the effect of e-cigarettes on smoking tion between these two interventions and US adult smok-
prevalence among US adults. The discrepancy is defined ing prevalence from the published literature [41, 42] and
as d = yp − ya , where yp is the projected smoking prev- comparing these estimates to the smoking discrepancy
alence and ya is the actual NHIS smoking prevalence, or prevalence observed in the present study. If the lit-
such that positive values of d mean that actual smoking erature estimates for the decrease in smoking prevalence
prevalence is lower than expected from projections. expected due to Tips® and the FSPTCA do not account
Foxon et al. BMC Public Health (2022) 22:1940 Page 4 of 10

Table 1  Combined Sample Characteristics female, majority never smoking/e-cigarette using, and is
Demographic Percent of sample approximately evenly distributed by age category.
% (n)
Root mean square errors of all models were consistent
Total N = 870,652
and small relative to the y-axis scale, ranging from 0.518
Age 18–34 31.8 (252,410)
to 1.115, at least one order of magnitude smaller than cig-
35–54 36.9 (309,068)
arette smoking prevalence (see Additional File 1, Supple-
55+ 31.3 (309,174)
mentary Table 1).
Race/Ethnicity Hispanic 12.4 (128,597)
Figure 1 shows the results of counterfactual trend mod-
NH White 70.7 (576,919)
elling among all adults. Smoking prevalence declined
NH Black 11.5 (119,210)
NH Other 5.4 (44,869)
steadily from 1990 to 2010. This decline apparently accel-
Sex Female 51.9 (489,143)
erated in the post-2010 period, where actual smoking
Male 48.1 (381,506)
prevalence was as much as approximately 3.4 ± 0.5 (SE)
Cigarette Smoking Status Current 20.6 (180,511) percentage points lower than projected. This smoking
Former 22.4 (196,163) discrepancy coincided with a rise in e-cigarette use prev-
Never 57.0 (486,181) alence to approximately 4.5 ± 0.2% of adults in 2019. The
E-Cigarette Use Status Current 3.5 (6,013) correlation between smoking discrepancy and e-cigarette
Former 11.2 (19,931) use prevalence from 2010 to 2019 was high and statisti-
Never 85.3 (158,470) cally significant (Pearson r = 0.803, p = 0.005).
Figure  2 shows the results by age group. Smoking
prevalence declined steadily from 1990 to 2010 among
for the smoking discrepancy or prevalence observed in 18–34 and 35–54 year olds, while smoking prevalence
the present study, this suggests these interventions alone was more stable among those aged 55+. The discrepancy
do not explain the observed smoking discrepancy, which between projected and actual smoking prevalence was
allows for possible association between e-cigarette use most pronounced among 18–34 year olds, with discrep-
and the observed smoking discrepancy (among other ancies up to 8.0 ± 0.9  percentage points. This age cohort
factors). also had the highest e-cigarette use prevalence, with
approximately 8.2 ± 0.4% of 18–34 year olds being current
Results e-cigarette users in 2019. Smoking discrepancies were
Main results approximately half as pronounced among 35–54 year
Table  1 shows the combined NHIS sample distribution. olds as they were among 18–34 year olds, but were still
The total sample consists of nearly one million observa- substantial (up to 3.5 ± 0.7 percentage points). E-cigarette
tions (N = 870,652) and is majority NH White, majority use prevalence among 35–54 year olds was approximately

Fig. 1  Trends in Smoking and E-Cigarette Use Prevalence among All Adults
Foxon et al. BMC Public Health (2022) 22:1940 Page 5 of 10

Fig. 2  Trends in Smoking and E-Cigarette Use Prevalence by Age Group

4.6 ± 0.3% in 2019, which is also about half the prevalence seen in Fig. 2. This age cohort also had the lowest e-cig-
among 18–34 year olds. Smoking discrepancies were not arette use prevalence at 1.4 ± 0.1% in 2019. Correla-
apparent among 55+ year olds, with most actual smok- tion between smoking discrepancy and e-cigarette use
ing prevalence estimates from 2010 to 2019 falling within was higher among 18–34 year olds (r = 0.869, p = 0.001),
the 95% confidence limits of the counterfactual model, as
Foxon et al. BMC Public Health (2022) 22:1940 Page 6 of 10

Fig. 3  Trends in Smoking and E-Cigarette Use Prevalence by Sex

followed by 35–54 year olds (r = 0.614, p = 0.06), and those Finally, Fig.  4 shows the modelling results by race/
age 55+ (r = 0.115, p = 0.8). ethnicity cohort. From 1990 to 2019, smoking preva-
Figure 3 shows the results by sex cohort. Similar to the lence declined consistently among all three race/eth-
age cohort results, smoking discrepancies were most pro- nicity cohorts. Smoking prevalence discrepancies up to
nounced for the cohort with the highest e-cigarette use 4.2 ± 0.6 percentage points were observed among the NH
prevalence. Among males, smoking discrepancies up to White cohort, whereas discrepancies were less appar-
4.2 ± 0.6  percentage points were observed, while among ent among the NH Black and Hispanic cohorts (up to
females, smoking discrepancies up to 2.5 ± 0.6 percentage 1.9 ± 1.2 and 2.0 ± 0.8  percentage points respectively).
points were observed. E-cigarette use prevalence mean- E-cigarette use prevalence in 2019 was highest among
while was approximately 5.5 ± 0.3% among males and NH White individuals (5.1 ± 0.2%) compared to NH Black
3.5 ± 0.2% among females in 2019. Correlation between (3.4 ± 0.4%) and Hispanic (2.8 ± 0.3%) individuals. Finally,
smoking discrepancy and e-cigarette use was stronger correlation between e-cigarette use prevalence and
among males (r = 0.869, p = 0.001) than among females cigarette smoking discrepancy was greatest for the NH
(r = 0.634, p = 0.05). White cohort (r = 0.804, p = 0.005) followed by the NH
Foxon et al. BMC Public Health (2022) 22:1940 Page 7 of 10

Fig. 4  Trends in Smoking and E-Cigarette Use Prevalence by Race/Ethnicity

Black (r = 0.676, p = 0.03) and Hispanic (r = 0.570, p = 0.09) Sensitivity test results
cohorts. Results from the main analyses were largely robust to the
five sensitivity tests described in the Methods, namely,
(1) excluding the 2019 point estimates due to NHIS
survey changes; (2) excluding the regression-estimated
Foxon et al. BMC Public Health (2022) 22:1940 Page 8 of 10

e-cigarette prevalence; (3) using 2009 as an alternative adults in 2018. Because the 3.3 ± 0.5 percentage point dis-
to the Kneedle cut-off year of 2010; (4) using 2011 as an crepancy observed in the present study is much greater
alternative to the Kneedle cut-off year of 2010; and (5) than the 0.4 percentage point decrease in smoking prev-
using exponential decay functions instead of linear func- alence associated with Tips®, Tips® does not explain the
tions (see Additional File 1, Supplementary Table 2). smoking discrepancy observed.
For the total sample (all adults), the Pearson correla- The association between the FSPTCA and smok-
tion between smoking discrepancy and e-cigarette use ing prevalence is quantified in the literature by a study
prevalence were similar to the main result (main result: which estimated a 0.6% reduction in US adult smoking
r = 0.803, range of r values across sensitivity tests (lowest prevalence each quarter following implementation of
to highest): r = 0.679–0.843; p < 0.05 for 4/5 tests). This the FSPTCA in June 2009 [42]. Cumulatively, this would
was also true for the 18–34 cohort (main result: r = 0.869, result in a 24% reduction in adult smoking prevalence
test range: r = 0.789–0.889; p < 0.05 for 5/5 tests), the Male from mid-2009 to mid-2019 (0.6% times 40 quarters).
cohort (main result: r = 0.869, test range: r = 0.782–0.897; NHIS smoking prevalence among all adults in 2009 was
p < 0.05 for 5/5 tests), the NH White cohort (main result: approximately 20.6 ± 0.4% (present study). Applying the
r = 0.804, test range: r = 0.668–0.840; p < 0.05 for 4/5 tests), 24% reduction associated with the FSPTCA to the 2009
the NH Black cohort (main result: r = 0.676, test range: NHIS smoking prevalence provides a predicted adult
r = 0.572–0.742; p < 0.05 for 3/5 tests), and Female cohort smoking prevalence of approximately 15.7% in 2019, due
(main result: r = 0.634, test range: r = 0.488–0.700; p < 0.05 to the FSPTCA. By comparison, in the present study the
for 3/5 tests). actual NHIS smoking prevalence in 2019 was approxi-
For the 55+ cohort, the main correlation result dif- mately 14.0% (95% CI: 13.5–14.5%), which is statisti-
fered more substantially from the sensitivity test results cally lower than the 15.7% prevalence from the FSPTCA.
(main result: r = 0.115, test range: r=-0.011–0.452), how- Because the actual NHIS smoking prevalence of 14.0% is
ever correlations were consistently low (below r = 0.5) and statistically lower than the 15.7% prevalence predicted
non-significant in all analyses for this cohort. from FSPTCA effects, FSPTCA effects do not explain the
For the 35–54 cohort, while the main correlation result smoking prevalence observed.
was high (r = 0.614), the correlations across sensitivity
tests ranged from low (r = 0.386) to high (r = 0.692). This Discussion
is also true for the Hispanic cohort for which the sensi- The aim of this research was to use population-level
tivity tests also ranged widely (r = 0.175–0.728), but only data to examine the correlation in trends between e-cig-
reached significance in sensitivity tests and not in the arette use and smoking prevalence among US adults
main analysis. from 2010 to 2019. Results suggest that actual smoking
Root mean square errors for non-linear models were prevalence was lower than it otherwise would have been
the same as those for the linear models to between one if trends from 1990 to 2009 (before e-cigarettes became
and three significant figures (see Additional File 1, Sup- widely available) had continued uninterrupted. Further,
plementary Table 1), suggesting little difference between the discrepancy between actual and predicted smoking
the linear and non-linear fits for these data. prevalence tended to be highest in groups with higher
e-cigarette use prevalence, such as among adults age
Other considerations 18–34, adult males, and non-Hispanic White adults.
The effect of the FSPTCA and the CDC’s ‘Tips®’ cam- Overall, the sensitivity analyses largely confirmed
paign, which represent major, distinct national popu- results from the primary analysis. However, sensitiv-
lation interventions, were considered by comparing ity test results ranged more widely for the age 35–54
quantitative estimates for the association between these and Hispanic cohorts. Some of the variations showed a
two interventions and smoking prevalence from the pub- stronger association between e-cigarette prevalence and
lished literature, to the smoking prevalence observed in discrepancy in smoking prevalence, suggesting that the
the present study. main analysis is conservative.
The association between the Tips® campaign and smok- These lower-than-expected smoking prevalences, cor-
ing prevalence is quantified in the literature by a CDC related with e-cigarette use, suggest population-level dis-
study which estimated approximately one million Tips® placement of cigarettes by e-cigarettes, consistent with
campaign-associated sustained quits between 2012 and extant modelling literature [28, 29]. A significant over-
2018 [41]. This equates to a 0.4 percentage point decrease all decline in adult cigarette smoking is observed, above
in smoking prevalence (because one million adults repre- what was otherwise expected, even for younger adults,
sent approximately 0.4% of the US adult population [43]). among whom the reported ‘gateway’ effect is claimed to
By comparison, in the present study, a 3.3 ± 0.5  percent- be strong and e-cigarette use is relatively high [1]. Levy et
age point smoking discrepancy was observed among all al. [27] also identified similar vaping-related reductions in
Foxon et al. BMC Public Health (2022) 22:1940 Page 9 of 10

smoking prevalence, and – consistent with our results - Conclusion

noted that these effects were primarily driven by younger This analysis of nationally representative data supports
adults aged 18–44. an association between the availability of e-cigarettes and
Low smoking initiation and relapse among baseline decreased cigarette smoking at the population level. Con-
never and former smokers using e-cigarettes have also sistent with a substitution effect, the results consistently
been noted in longitudinal cohort studies [19, 20]. How- show that subgroups of US adults reporting higher preva-
ever, use by unintended groups (e.g., nonsmokers) is of lence of e-cigarette use show bigger discrepancies from
high concern and efforts to reduce use in these popula- the expected trend in cigarette smoking prevalence.
tions should continue to be a high priority.
List of abbreviations
The predictions for smoking prevalence in the pres- FSPTCA Family Smoking Prevention and Tobacco Control Act.
ent study may be compared to predictions from pre- NHIS National Health Interview Survey.
vious modelling efforts. A dynamic simulation model RMSE Root Mean Square Error.

developed by Mendez and Warner [44] and similarly cal-

ibrated with NHIS data predicted an overall adult smok- Supplementary Information
ing prevalence of 16.8% for 2020. The smoking trend The online version contains supplementary material available at https://doi.
org/10.1186/s12889-022-14341-z.
among all adults from 1990 to 2009 in the present study
predicts an adult smoking prevalence of 16.3% (95% CI: Supplementary Material 1
15.4–17.2%) when this trend is projected to 2020. This is
statistically consistent with Mendez and Warner which Acknowledgements
validates the modelling methods of the present study. The authors would like to thank Rasmus Wissmann of Juul Labs for insightful
Importantly, neither accurately predict the actual 2020 comments and suggestions.
smoking prevalence of 12.5 ± 0.3% from NHIS. This fur- Authors’ contributions
ther suggests the introduction of some effect on US adult FF and JG conceptualized the study. FF managed, cleaned, and summarized
smoking prevalence circa 2010 which is not accounted the data, performed statistical analyses, interpreted results, and drafted
the manuscript. AS and SS provided conceptual comments. AS, JG, and SS
for by simple extrapolation of prior trends or by models reviewed drafts and contributed to the final manuscript.
based on population dynamics structures. The lower-
than-expected smoking prevalence of this study as well as Funding
The preparation of this manuscript was funded by Juul Labs Inc.
Levy et al. [27], and Wagner and Clifton [29] suggest that
this unaccounted-for effect coincides with the introduc- Availability of Data and Materials
tion and use of e-cigarettes among adults. The datasets supporting the conclusions of this article are freely available in
the National Health Interview Survey (NHIS) repository, https://www.cdc.gov/
Because the analysis used cross-sectional data, the nchs/nhis/.
results presented here are subject to the usual limitations
of such data, including selection bias, response bias, and Declarations
inability to infer causality because changes in behavior
between survey waves may not reflect a trend but dif- Ethics approval and consent to participate
This research uses only publicly available de-identified survey data. Under
ferences between samples [45]. Additionally, these data exemption four of NIH definitions, this research is therefore exempt from NIH
and this methodology cannot precisely parse the effects human subjects research. All methods were performed in accordance with
of e-cigarettes’ introduction and other market/policy relevant guidelines and regulations.
changes that may also have impacted smoking prevalence Consent for publication
declines since 2009. However, we show that two of the Not applicable.
major changes (the ‘Tips®’ campaign and FSPTCA) do not
Competing Interests
explain the observed smoking discrepancy, even when Through PinneyAssociates, FF, AS, JG, and SS provide consulting services
optimistically assuming their impacts are sustained over on tobacco harm reduction on an exclusive basis to Juul Labs Inc. The
the last decade. Other important demographic distinc- preparation of this manuscript was supported by JLI, and JLI reviewed and
provided comments on a draft manuscript. The content and the decision to
tions may exist, for example trends among 18–24 year publish are the responsibility of the authors. JG and SS also own interest in a
olds. However, analyses were limited by low sample size novel nicotine gum that has neither been developed nor commercialized.
for these subpopulations. Lastly, when comparing across
age cohorts over a 10-year period, it should be noted that Received: 9 May 2022 / Accepted: 10 October 2022
part of the population in one cohort would have aged
into the next cohort (e.g., respondents age 18–34 in 2009
will be age 28–44 in 2019, which partially overlaps the
References
35–54 category).
1. Khouja JN, Suddell SF, Peters SE, Taylor AE, Munafo MR. Is e-cigarette use
in non-smoking young adults associated with later smoking? A systematic
review and meta-analysis. Tob Control 2020.
Foxon et al. BMC Public Health (2022) 22:1940 Page 10 of 10

2. Chatterjee K, Alzghoul B, Innabi A, Meena N. Is vaping a gateway to smoking: 23. Cotti CD, Courtemanche CJ, Maclean JC, Nesson ET, Pesko MF, Tefft N: The
a review of the longitudinal studies. Int J Adolesc Med Health 2016, 30(3). Effects of E-Cigarette Taxes on E-Cigarette Prices and Tobacco Product Sales:
3. Soneji S, Barrington-Trimis JL, Wills TA, Leventhal AM, Unger JB, Gibson LA, Evidence from Retail Panel Data. National Bureau of Economic Research Work-
Yang J, Primack BA, Andrews JA, Miech RA, et al: Association Between Initial ing Paper Series 2020, No. 26724.
Use of e-Cigarettes and Subsequent Cigarette Smoking Among Adolescents 24. Pesko MF, Courtemanche CJ, Catherine Maclean J. The effects of traditional
and Young Adults: A Systematic Review and Meta-analysis. JAMA Pediatrics cigarette and e-cigarette tax rates on adult tobacco product use. J Risk
2017, 171(8):788–797. Uncertain. 2020;60(3):229–58.
4. Lee P, Coombs K, Afolalu E: Considerations related to vaping as a possible 25. Zheng Y, Zhen C, Dench D, Nonnemaker JM. U.S. Demand for Tobacco Prod-
gateway into cigarette smoking: an analytical review [version 3; peer review: ucts in a System Framework. Health Econ. 2017;26(8):1067–86.
2 approved]. F1000Research 2019, 7(1915). 26. Wissmann R, Zhan C, D’Amica K, Prakash S, Xu Y. Modeling the Population
5. Chan GCK, Stjepanović D, Lim C, Sun T, Shanmuga Anandan A, Connor JP, Health Impact of ENDS in the US. Am J Health Behav. 2021;45(3):588–610.
Gartner C, Hall WD, Leung J. Gateway or common liability? A systematic 27. Levy DT, Sanchez-Romero LM, Travis N, Yuan Z, Li Y, Skolnick S, Jeon J, Tam
review and meta-analysis of studies of adolescent e-cigarette use and future J, Meza R. US Nicotine Vaping Product SimSmoke Simulation Model: The
smoking initiation. Addiction. 2021;116(4):743–56. Effect of Vaping and Tobacco Control Policies on Smoking Prevalence and
6. Kim S, Selya AS. The Relationship Between Electronic Cigarette Use and Con- Smoking-Attributable Deaths. Int J Environ Res Public Health 2021, 18(9).
ventional Cigarette Smoking Is Largely Attributable to Shared Risk Factors. 28. Selya AS, Wissmann R, Shiffman S, Chandra S, Sembower M, Joselow J, Kim S.
Nicotine Tob Res. 2020;22(7):1123–30. US Sales of Electronic Nicotine Delivery Systems (ENDS) Are Associated with
7. Considerations related to vaping as a possible gateway into cigarette smok- Declines in Cigarette Sales. Journal of Consumer Policy 2022.
ing [https://osf.io/z3st5/]. 29. Wagner LM, Clifton SM. Modeling the public health impact of e-cigarettes
8. Sun R, Mendez D, Warner KE. Is Adolescent E-Cigarette Use Associated With on adolescents and adults. Chaos: An Interdisciplinary Journal of Nonlinear
Subsequent Smoking? A New Look. Nicotine Tob Res. 2021;24(5):710–8. Science. 2021;31(11):113137.
9. Foxon F, Selya AS. Electronic cigarettes, nicotine use trends and use 30. Dai H, Leventhal AM. Prevalence of e-Cigarette Use Among Adults in the
initiation ages among US adolescents from 1999 to 2018. Addiction. United States, 2014–2018. JAMA. 2019;322(18):1824–7.
2020;115(12):2369–78. 31. Syamlal G, Clark KA, Blackley DJ, King BA. Prevalence of Electronic Cigarette
10. Levy DT, Warner KE, Cummings KM, Hammond D, Kuo C, Fong GT, Thrasher JF, Use Among Adult Workers - United States, 2017–2018. MMWR Morb Mortal
Goniewicz ML, Borland R. Examining the relationship of vaping to smoking Wkly Rep. 2021;70(9):297–303.
initiation among US youth and young adults: a reality check. Tob Control. 32. 2018-19 TUS-CPS Data, Table 3: E-Cigarette Use https://cancercontrol.cancer.
2019;28(6):629–35. gov/brp/tcrb/tus-cps/results/2018-2019/table-3.
11. Selya AS, Foxon F. Trends in electronic cigarette use and conventional smok- 33. National Health Interview Survey [https://www.cdc.gov/nchs/nhis/about_
ing: quantifying a possible ‘diversion’ effect among US adolescents. Addiction nhis.htm].
2021. 34. Cornelius ME, Wang TW, Jamal A, Loretan CG, Neff LJ. Tobacco product use
12. Sokol NA, Feldman JM. High school seniors who used e-cigarettes may have among adults—United States, 2019. Morb Mortal Wkly Rep. 2020;69(46):1736.
otherwise been cigarette smokers: Evidence from Monitoring the Future 35. Axelsson GT, Eythorsson ES, Hardardottir H, Gudmundsson G, Hansdottir S.
(United States, 2009–2018). Nicotine Tob Res 2021. [The impact of lung diseases, smoking and e-cigarette use on the severity of
13. Walker N, Parag V, Wong SF, Youdan B, Broughton B, Bullen C, Beaglehole R. COVID-19 illness at diagnosis]. Laeknabladid. 2020;106(12):574–9.
Use of e-cigarettes and smoked tobacco in youth aged 14–15 years in New 36. Satopaa V, Albrecht J, Irwin D, Raghavan B: Finding a “kneedle” in a haystack:
Zealand: findings from repeated cross-sectional studies (2014-19). Lancet Detecting knee points in system behavior. In: 2011 31st international
Public Health. 2020;5(4):e204–12. conference on distributed computing systems workshops: 2011: IEEE; 2011:
14. Hartmann-Boyce J, McRobbie H, Butler AR, Lindson N, Bullen C, Begh R, 166–171.
Theodoulou A, Notley C, Rigotti NA, Turner T, et al: Electronic cigarettes for 37. Electronic cigarette dollar sales worldwide from 2008 to 2017
smoking cessation. Cochrane Database of Systematic Reviews 2021(9). (in million U.S. dollars) [https://www.statista.com/chart/9607/
15. Thomas KH, Dalili MN, López-López JA, Keeney E, Phillippo D, Munafò MR, sales-electronic-cigarettes-dollar-sales-worldwide/].
Stevenson M, Caldwell DM, Welton NJ: Smoking cessation medicines and 38. Hyndman RJ, Athanasopoulos G. Forecasting: principles and practice. 2nd ed.
e-cigarettes: a systematic review, network meta-analysis and cost-effective- Melbourne: OTexts; 2018.
ness analysis. 2021, 25:59. 39. 2019 Questionnaire Redesign https://www.cdc.gov/nchs/nhis/2019_quest_
16. Chan GCK, Stjepanović D, Lim C, Sun T, Shanmuga Anandan A, Connor JP, redesign.htm.
Gartner C, Hall WD, Leung J. A systematic review of randomized controlled 40. Goodwin R, Wall M, Galea S, Zvolensky MJ, Gbedemah M, Hu M-C, Hasin
trials and network meta-analysis of e-cigarettes for smoking cessation. Addict DS. Has nicotine dependence increased among smokers in the United
Behav. 2021;119:106912. States? A new test of the hardening hypothesis. Drug Alcohol Depend.
17. Goldenson NI, Shiffman S, Hatcher C, Lamichhane D, Gaggar A, Le GM, 2017;100(171):e76.
Prakash S, Augustson EM. Switching away from Cigarettes across 12 Months 41. Murphy-Hoefer R, Davis KC, King BA, Beistle D, Rodes R, Graffunder C. Associa-
among Adult Smokers Purchasing the JUUL System. Am J Health Behav. tion Between the Tips From Former Smokers Campaign and Smoking Cessa-
2021;45(3):443–63. tion Among Adults, United States, 2012–2018. Prev Chronic Dis. 2020;17:E97.
18. Selya AS, Shiffman S, Greenberg M, Augustson EM. Dual Use of Cigarettes 42. Rossheim ME, Livingston MD, Krall JR, Barnett TE, Thombs DL, McDonald KK,
and JUUL: Trajectory and Cigarette Consumption. Am J Health Behav. Gimm GW. Cigarette Use Before and After the 2009 Flavored Cigarette Ban. J
2021;45(3):464–85. Adolesc Health. 2020;67(3):432–7.
19. Le GM, Holt NM, Goldenson NI, Smith LC, Hatcher C, Shiffman S, Augustson 43. The US. Adult and Under-Age-18 Populations: 2020 Census https://www.
EM. Cigarette Smoking Trajectories in Adult Former Smokers Using the JUUL census.gov/library/visualizations/interactive/adult-and-under-the-age-of-
System. Am J Health Behav. 2021;45(3):505–26. 18-populations-2020-census.html.
20. Shiffman S, Holt NM. Smoking Trajectories of Adult Never Smokers 12 44. Mendez D, Warner KE. Setting a challenging yet realistic smoking prevalence
Months after First Purchase of a JUUL Starter Kit. Am J Health Behav. target for Healthy People 2020: learning from the California experience. Am J
2021;45(3):527–45. Public Health. 2008;98(3):556–9.
21. Pierce JP, Chen R, Kealey S, Leas EC, White MM, Stone MD, McMenamin SB, 45. Sedgwick P. Bias in observational study designs: cross sectional studies. BMJ:
Trinidad DR, Strong DR, Benmarhnia T, et al. Incidence of Cigarette Smoking Br Med J. 2015;350:h1286.
Relapse Among Individuals Who Switched to e-Cigarettes or Other Tobacco
Products. JAMA Netw Open. 2021;4(10):e2128810.
22. Barufaldi LA, Guerra RL, de Albuquerque RCR, Nascimento A, Chança RD, de Publisher’s Note
Souza MC, de Almeida LM. Risk of smoking relapse with the use of electronic Springer Nature remains neutral with regard to jurisdictional claims in
cigarettes: A systematic review with meta-analysis of longitudinal studies. Tob published maps and institutional affiliations.
Prev Cessat. 2021;29:29.