An Examination of the Variation in Estimates of E-Cigarette Prevalence among U.S. Adults

Introduction: Accurate estimates of e-cigarette use are needed to gauge its impact on public health. We compared the results of online and traditional, large scale surveys and provide additional estimates from the Population Assessment of Tobacco and Health (PATH) survey, with the aim of assessing the extent of variation in prevalence estimates. Materials and Methods: We searched the peer-reviewed literature for nationally representative estimates of U.S. adult e-cigarette prevalence, and developed our own estimates from waves one, two, and three of the PATH survey. We compared estimates by age, gender, cigarette smoking status, and e-cigarette use intensity both between online and traditional surveys and among the traditional surveys. Results: For specific years, online surveys generally yielded higher adult use rates than most traditional surveys, but considerable variation was found among traditional surveys. E-cigarette prevalence was greater for less intensive than for more intensive use. Levels of use were higher among current and recent former cigarette smokers than among former smokers of longer quit duration and never smokers, and by those of younger ages. Conclusions: Considerable variation in e-cigarette use estimates was observed even for a specific year. Further study is needed to uncover the source of variation in e-cigarette prevalence measures, with a view towards developing measures that best explain regular use and transitions between the use of e-cigarettes and other tobacco products.


Introduction
As the leading cause of preventable death [1], the prevalence of cigarette use ("smoking") is commonly used to gauge potential societal health impacts and to set goals for reducing smoking [2][3][4]. An example is the U.S. Healthy People goal to reduce smoking prevalence to 12% by 2020 [5,6]. In addition, studies generally rely on changes in prevalence estimates to evaluate policy effectiveness [7]. While playing a central role in monitoring and evaluating tobacco use, studies have found discrepancies between estimates of U.S. cigarette prevalence in the large, traditional surveys [8][9][10][11][12][13]. Survey administration mode (telephone or in-person), respondent type (self or proxy), and survey response rate have been identified as potential reasons for the variations [11,12,14].
All of the U.S. ongoing, established, and publicly available traditional surveys of tobacco use, except the National Smoking and Drug Use Survey, now ask about vaping. Online surveys also collect vaping information, and generally provide more up-to-date data on use and adoption. Despite being a common survey question, studies have not been conducted to compare vaping estimates between online and traditional surveys or to compare the variability of vaping estimates from traditional surveys. However, estimates of youth e-cigarette prevalence have been found to depend on the specific questions asked [27]. In addition, studies have observed that estimates of adult vaping [28] and youth vaping [29] depend on the specific measure adopted (e.g., ever vs. current use, less vs. more intense use).
The purpose of this study was to compare published adult vaping prevalence estimates from the different U.S. nationally representative surveys. In addition, we provided our own estimates developed from the Food and Drug Administration-sponsored Population Assessment of Tobacco and Health (PATH) surveys. We compared use rates by age, gender, cigarette smoking status, and vaping intensity of use both between online and traditional surveys and among the traditional surveys.

Search Strategy
To collect estimates of e-cigarette prevalence, we conducted a search of online databases (e.g., PubMed, Google Scholar, EconLit, Web of Science, Social Science Research Network, and the Federal Trade Commission website). Our searches were confined to inclusion of at least one product term (e.g., "e-cigarette", "ENDS", or "vaping") and at least one term related to use (e.g., "prevalence", "use", "someday", or "every day"). The search was conducted through January 15 2019. Inclusion was limited to nationally representative studies in peer-reviewed journals, with one study omitted because it was not intended to be nationally representative [30].
With the relatively large number of studies, we limited this study to U.S. surveys in order to be able to make meaningful comparisons. We also limited the survey to studies that include adults, because youth studies often consider different age ranges and are generally less comparable. We considered current use (last 30 day, or every day-someday) rather than ever vaping since ever use can occur in previous years and increasingly represents past use. We included studies that estimate adult e-cigarette use, where use refers to any vaping product with or without nicotine. We also extracted e-cigarette use by age, gender, and cigarette smoking status (never, current, and former, including by years quit). Finally, we considered vaping intensity, distinguishing daily ("every day") vs. nondaily ("someday") use, and the number of days used in the past 30 days. We distinguished "traditional" from "online" surveys. Traditional surveys are considered those that are probability-based, ongoing, and that included data using face-to-face interviews or traditional mail. Online surveys can adopt convenience or probability sampling and primarily use a web survey but may use a mixed mode.
In view of the importance of the PATH survey for its size and longitudinal design, we included our own analysis of PATH as part of the assessed estimates. We replicated the results of two studies using PATH [31,32] and further distinguished e-cigarette use by gender, age, and smoking status in wave one (2013-2014), wave two (2014-2015), and wave three (2015-2016). Additionally, we categorized former cigarette smokers by quit years (<1, 1≤3, 3≤5, ≥5 years). The vaping products listed shifted from only "e-cigarettes" in wave one to both "ENDS (including e-cigarettes, e-cigar, e-hookah, and e-pipe)" and "e-cigarettes (as one sub-type of ENDS)" in wave two and three. For consistency, we used e-cigarettes for the analyses in all waves. We used the suggested pre-processed method from the PATH Codebook [33] for measuring daily/nondaily and past-30-day use, and used the self-response weights provided in PATH for the mean and confidence interval estimates.
To measure the percentage of past-30-day users that were at least 5-and 20-day e-cigarettes, we used the PATH question about the number of days vaping in the past 30 days for someday, experimental every day, and former users (in waves two and three) and assumed other daily users were "frequent (at least 20 days in past month)" users. Since waves one and two only asked about days use by e-cigarette users but no other ENDS users, our analyses for past-30-day estimates were restricted to e-cigarettes in waves one and two but included ENDS use in wave three.

Comparison of Studies
Results extracted from the included papers and our analyses are presented as mean percentages and 95% confidence intervals (CI) when available. Descriptive comparisons were made between estimated means and/or 95% confidence intervals of the studies, focusing on variations in estimates between studies using online and traditional surveys, among studies using traditional surveys, and among studies using different measure of use (e.g., daily-nondaily vs. any past-30-day use). Table 1 provides a summary of the survey methods and measures used in different studies. In addition to our estimates from PATH, we found 28 studies (n = 11 online and n = 17 traditional) with a variety of sampling methodologies. Traditional surveys used data from six national surveys: National Health Interview Survey (NHIS) [34][35][36][37][38][39], National Adult Tobacco Survey (NATS) [40][41][42][43], PATH [31,32,44,45], Tobacco Use Supplement of the Current Population Survey (CPS-TUS) [46], Behavioral Risk Factor Surveillance System (BRFSS) [47], and National Health and Nutrition Examination Survey (NHANES) [48]. Online surveys primarily employed the GfK/Knowledge online panel either on its own (n = 3) [49][50][51], by selecting a custom sample (n = 5) [15,[52][53][54][55], or in combination with other data (n = 1) [56]. One online survey [57] recruited a household random sample using random digit dialing and administered via phone or the web. One survey used a national consumer-based web panel [58]. Heard of and tried an e-cigarette; currently use daily/less than daily, but ≥1 week/less than weekly, but ≥once a month/less than monthly.   Now use every day, somedays, or rarely.

Methods and Measures of E-Cigarette Use
Every day/ some days/ rarely current use In person longitudinal study using audio computer-assisted self-interview, with address-based, area-probability sampling, and in-person household screener.

32,320 (74%)
Have seen or heard, ever used, and now use e-cigarettes every day or somedays.
In-frequent users: 0-2 days; moderate users: use on >2 of days; and everyday use. Measures based on questions about daily and non-daily use only were adopted in 12 studies [31,32,[34][35][36][37][38][39][40]45,47,56]. Two studies using NATS [41,43] and two online studies [55,59] included "rarely use" along with daily and nondaily use. One study [44] distinguished daily and nondaily regular users from current triers who have not regularly vaped. Measures that included any past-30-day use were adopted in seven studies [15,46,49,[51][52][53]58]. Another study [48] used any past-5-day prevalence. One study [32] defined current use as those who ever used "fairly regularly" and now use daily or nondaily, as well as those who used at least once in the past 30 days. One study [57] measured use as those who tried an e-cigarette and generally included those who currently use e-cigarettes, and another [50] considered daily, weekly, or monthly use.
Eight studies distinguished between e-cigarette intensity and duration of use. Two studies [34,35] provided daily vs. nondaily use, one study [44] distinguished regular use from triers and daily vs. nondaily use, and four studies distinguished rarely use [41][42][43]53]. One study [49] defined use as those who vaped at least once in the past 30 days and more than 50 times during their lifetime. Three studies [31,46,53] considered the number of days used in the past month.
. Figure 1. The prevalence of current e-cigarette/ENDS use among the total adult population, online and traditional survey estimates *, 2010-2017. * The estimates from online surveys are indicated by an X, and the estimates from traditional surveys are indicated by a circle. The legend adopts the format "author's name, publication year (survey name and year(s))," and the names of all online surveys are omitted. For surveys that collected data over a two-year period, we plotted the estimates in the midperiod (e.g., the estimate 2.6% from Jaber et al. using NHANES 2013-14 is plotted in 2013.5). ** Authors' estimate one is measured by daily/non-daily ENDS use, and authors' estimate two is measured by past-30-day ENDS use. For surveys that collected data over a two-year period, we plotted the estimates in the mid-period (e.g., the estimate 2.6% from Jaber et al. using NHANES 2013-14 is plotted in 2013.5). ** Authors' estimate one is measured by daily/non-daily ENDS use, and authors' estimate two is measured by past-30-day ENDS use.
Differentiating by length of time quit (not shown in Figure 3), an online survey using a daily-nondaily measure yielded an adult vaping rate of 6.1% among those who quit ≤1 year and 0.2% among quitters >1 year in 2012 [53]. For 2015, use [55] was at 24.7% among those who quit <1 year, 16.5% among those who quit between 1 and 5 years, and 2.0% among >5 years quitters. In 2016, reported daily/weekly/monthly e-cigarette use was 14.3% among former smokers who quit ≤2 years [50]. Two 2014 NHIS studies [34,37] reported daily-nondaily vaping prevalence much lower for those <1 year quit than among those ≥1 year. Using past-30-day, 2013 to 2014 PATH had vaping at 27.9% for quit <1 year, 14.1% for quit 1 to 2 years, 7.3% for 3 to 5 years quit, and 1.2% for quit ≥5 years. 2014 to 2015 CPS-TUS estimates [46] were at 16.3% for quit <1 year, 9.8% for 2 to 3 years, and 0.5% for quit ≥3 years. The 2015 to 2016 PATH past-30-day use was at 21.8% for quit <1 year, 22.3% for 1 to 2 years, 28.7% for quit 3 to 4 years, and 8.5% for quit ≥5 years.  For surveys that collected data over a two-year period, we plotted the estimates in the midperiod. ** Authors' estimate one is measured by daily/nondaily ENDS use, and authors' estimate two is measured by past-30-day ENDS use. For surveys that collected data over a two-year period, we plotted the estimates in the mid-period. ** Authors' estimate one is measured by daily/nondaily ENDS use, and authors' estimate two is measured by past-30-day ENDS use.  For surveys that collected data over a two-year period, we plotted the estimates in the mid-period. ** Authors' estimate one is measured by daily/nondaily ENDS use, and authors' estimate two is measured by past-30-day ENDS use.

Never Smokers
As shown in Figure 4, online surveys using a daily-nondaily adult vaping measure showed an increase from 0.1% in 2010 to 2012 to 1.4% in 2013 [56] and 3.0% in 2015 [55]. Past-30-day use was at 0.2% in 2010 to 2011 [15], 0.04% in 2012 [53], and 0.8% [58] and 0.9% [52] in 2014. Traditional surveys with daily-nondaily measures also showed an increase from 0.4% in 2014 [26] to 0.6% in 2015 [29] to 0.7% in 2016 using NHIS [38]. We estimated use at 1.4% in 2013 to 2014, 1.3% in 2014 to 2015, and 1.3% in 2015 to 2016 with PATH, and BRFSS had use at 1.4% in 2016 [47]. For past 30-day use, never smoker vaping rates were at 0.4% from 2013 to 2014 NHANES [48] and at 2.2% with PATH. For 2014 to 2015 CPS-TUS [46] indicated 0.2%, and our 2014 to 2015 PATH rates were at 1.9%. Figure 4. The prevalence of current e-cigarette/ENDS use among never smokers, online and traditional survey estimates *, 2010-2016. * The estimates from online surveys are indicated by an X, and the estimates from traditional surveys are indicated by a circle. The legend adopts the format "author's name, publication year (survey name and year(s))," and the names of all online surveys are omitted. For surveys that collected data over a two-year period, we plotted the estimates in the midperiod. ** Authors' estimate one is measured by daily/nondaily ENDS use, and authors' estimate two is measured by past-30-day ENDS use.
Finally, established use measured as "used an e-cigarette at least once in the past 30 days and more than 50 times in their lifetime" among current and former (<5years) smokers accounted for 31.0% and 17.5% of current past-30-day use among males and females in 2013, with higher rates among ages 45 to 59 than other age groups [49]. A 2016 online survey [50] had the ratio of daily/(weekly and monthly) e-cigarette use at 4.7%/5.9% among daily smokers, 7.0%/6.2% among nondaily smokers, and 11.7%/2.6% among former smokers (quit ≤2 years).

Discussion
Our study of U.S. adult vaping prevalence shows an increasing trend since 2010 with a possible leveling off since 2014. Unfortunately, the high level of variability found in e-cigarette rates for individual years made the results unsuitable for accurately assessing trends and characterizing the magnitude of use. We detected vast differences in the methodological approaches to collect data not just between but also within surveys.
Comparing online and traditional surveys conducted since 2013, higher estimates were generally observed by online surveys. Most of the online study samples came from the Knowledge Panel (GfK) ( Table 1). Due to its sampling methods, GfK surveys, which are conducted close to one another, may re-survey the same individual, potentially leading to bias from repeated sampling. This potential increases for sub-populations, such as current and former cigarette smokers.
We also found considerable variation in adult vaping estimates from traditional surveys even for a given year. In 2014 and 2015, three major surveys, the NHIS, CPS-TUS, and PATH, provided estimates of vaping prevalence. Estimates from PATH were generally twice or greater than CPS-TUS rates, with NHIS prevalence generally in between. PATH daily-nondaily measures were also less than past-30-day measures.
Similar discrepancies for adult smoking prevalence have also been observed among studies using traditional surveys [8][9][10][11][12][13]60], but with far less deviation. The variation in adult vaping estimates appears to be closer to those observed for smokeless tobacco use [61]. The apparent lack of consistency may be attributed to the variation in survey methodologic approaches observed for data collection, i.e., survey mode, respondent type (self vs. proxy), survey length and timing of questions, response rate, and the specific wording of the questions asked [11,12,14]. For example, the response rates shown in Table 1 varied from 25% to 83%. Some of the surveys were specific to tobacco use (e.g., NATS, most online surveys), while others covered a broader array of topics (e.g., NHIS, CPS-TUS). Surveys were predominantly cross-sectional; only two (PATH and ITC) were longitudinal. The length of the questions also varied, with PATH asking a more extensive series of e-cigarette questions than other surveys.
Of particular importance in gauging the e-cigarette use are the many varieties of products at any point in time and their changes over time [62]. With third-generation e-cigarettes already being sold by 2013 [63] and the term "juuling" becoming popular by 2018 [64,65], the terminology used to describe the different types of devices may play an important role [65][66][67][68]. The descriptions of the products also varied among major surveys. While NATS just asked about e-cigarette use in general, NHIS gave specific examples, such as "vape-pens, hookah-pens, e-hookahs, or e-vaporizers"; CPS-TUS [46] asked about "vape-pens, hookah-pens, e-hookahs, e-vaporizers, e-cigars, or e-pipes"; NHANES [48] asked about "e-cigarette use" and showed examples of in-hand cards; and two surveys [52,58] asked about any of the ENDS, including "electronic or e-cigarettes; electronic hookahs, hookah pens, or vape pens; or some other electronic vapor product, such as e-cigars or e-pipes". PATH described, "E-cigarettes look like regular cigarettes, but are battery-powered and produce vapor instead of smoke. There are many types of e-cigarettes. Some common brands include NJOY, Blue and Smoking Everywhere" in its first wave. PATH then generalized to "electronic nicotine products, such as e-cigarettes, e-cigars, e-pipes, e-hookahs, personal vaporizers, vape pens and hookah pens" in the second wave and further added an open classification of "something else" in the third wave.
The variation in estimates across surveys may also reflect the transient use of e-cigarettes. We considered the intensity of e-cigarette use, including daily vs. nondaily use and the number of days in the past month. In general, the prevalence was higher for measures that required fewer days use in the past month, included "rarely" and infrequent use, and used past-30-day compared to past-5-day use. We found a tendency toward more intensive use over time, and most intensive use by recent quitters followed by current smokers and less recent former smokers. Substantially lower levels of intensity were found among never smokers.
Research examining how the frequency of use affects prevalence estimates is clearly needed. Other measures, such as total use, e.g., at least 50 times lifetime use, used fairly regularly, and the duration of e-cigarette use, should also be considered in the development of suitable measures [62]. However, different measures may be appropriate for different applications. Regular use and longer duration of use [26,69] are more relevant in gauging the health impact of vaping. Shorter-term, less-intensive use may be relevant in gauging transitions from vaping to cigarette use or vice-versa [26,29]. With additional waves of PATH and the ITC surveys, these transitions can be systematically explored, taking advantage of the longitudinal nature of these surveys.

Conclusions
Proper characterization of vaping use and trends is urgently needed to develop effective regulatory strategies. With considerable variation in adult vaping prevalence estimates even for a specific year, it is not clear which measure or methodology should be adopted in future studies. Indeed, different measures may be appropriate for different purposes and different approaches may shed light on different issues. Further exploration of the different measures is needed to determine which best capture tobacco use transitions and which best assess long-term health use. Without better-defined and consistent measures of use, the ability to evaluate the public health impact associated with vaping and other tobacco product use is likely to be increasingly problematic.