Our analysis focused on the variation in current use prevalence estimates of four categories of tobacco products (cigarettes, other combustibles, smokeless tobacco and e-cigarettes) overall and distinguished by exclusive, dual and polyuse. We compared the estimated current use prevalence from the TUS-CPS, NHIS and PATH national surveys using three frequency of use thresholds (1+, 10+, and 25 + days in the past 30 days).
The relative differences in current use prevalence using the 1 + day compared to the 10 + days or the 25 + days thresholds were generally consistent across the three surveys. We observed that the prevalence reduction in current cigarette and smokeless tobacco users as the threshold increased was minimal (less than 20%), indicating that the frequency of use in the past 30 days for these users was relatively stable regardless of the threshold used. This lower variability for cigarette estimates might be due in part to the adoption of the “100-cigarette lifetime" criteria for established use. The largest variations were observed for other combustibles, followed by e-cigarettes (around 60%). Previous studies of tobacco product frequency of use [7, 8, 30, 35, 36] have generally observed important differences in prevalence within nondaily users. For example, a study of smokeless tobacco prevalence found that prevalence using a 1 + day threshold was about 17% higher than that using a 10 + days threshold,[30] and another study found that e-cigarette prevalence doubled when using a 1 + day compared to a 20 + days threshold [21].
In assessing multiproduct use, we found that the proportions of exclusive use for cigarettes and smokeless tobacco users tended to be 63% or higher, even when using the 1 + day threshold. In contrast, other combustibles and e-cigarette users were more inclined towards dual and less frequent use with the proportion of exclusive use of these products at around 30% with the 1 + day threshold, reaching 70% when using the 25 + days threshold. Other studies have also found relatively high rates of multiproduct use among e-cigarette and other combustible users, especially when examining use at low frequency thresholds [37–39].
In terms of the most appropriate threshold, we do not prescribe a particular measure, since the choice of threshold should depend on the purpose for which the definition is applied. In gauging some types of public health impacts, a stable measure of regular use is likely more appropriate. While cigarette and smokeless tobacco use was generally more frequent and stable, the large differences between the 1 + and the 10 + days thresholds for e-cigarettes and other combustibles suggest that these products have a less stable usage pattern, consistent with evidence that suggests these products are used more by experimental and social users [19]. In addition, e-cigarettes are the newest tobacco product, with use patterns still to be understood. Dual and polyuse prevalence rates also generally fell substantially when the threshold increased, signaling that multiproduct usage patterns may be less frequent and less stable. However, polyusers have shown greater nicotine dependence than single users [38, 40], suggesting a potential tendency towards future regular use. In addition, while more frequent use of tobacco products is generally associated with more harmful health effects than less frequent use, less frequent use may be relevant in terms of the pathways of usage of current users towards long-term pattern and also has important health implications [41–43]. In particular, one study found that lifelong non-daily smokers who reported 11–30 cigarettes per month had a 34% higher mortality risk compared to never smokers [41].
In determining patterns of initiation and transitions from experimental to regular use, a more sensitive measure of use might be more appropriate. Infrequent use may be particularly relevant in assessing transitions to regular use, especially among e-cigarette and other combustible users [44] but also among cigarette and smokeless tobacco users [45, 46]. Focusing on a lower frequency of use may also be especially important in capturing transitions in dual and polytobacco use patterns. For example, dual e-cigarette users have shown greater cessation intentions compared with exclusive cigarette smokers [47]. Also cigarette smokers who use smokeless tobacco have been found to be more likely than exclusive smokers to attempt quitting cigarette smoking using other tobacco products [48].
In choosing the appropriate measure, different thresholds may be needed for different product categories. For e-cigarettes and other combustibles, lower thresholds (e.g.,1 + day) may be more relevant in estimating potential transitions between product use, and the 10 + days threshold may be more useful in assessing more frequent and stable use. To assess the relevance of different thresholds, it will be important to develop evidence on the stability of use and transitions over time using longitudinal data such as PATH.
Our results also suggest that the appropriate frequency measure may depend on the survey used. Comparing across surveys, the prevalence of current use was higher in PATH followed by NHIS and TUS-CPS, regardless of the product or frequency of use threshold. Despite using the standard definition of established cigarette use (i.e., the 100-cigarettes lifetime criteria), we observed inter-survey variability in cigarette prevalence estimates. There was also variability in the estimated smokeless tobacco prevalence. However, we observed greater variability in e-cigarettes and other combustibles use prevalence. Other studies comparing different surveys have also reported similar variations in current use prevalence for cigarettes [49], smokeless tobacco [50], and e-cigarettes [21]. Some of the variation across surveys may be due to differences in the current use definition (in previous studies) or the sampling procedure, method of interview, and design of each survey. For example, PATH has a longitudinal and more complex design than TUS-CPS and NHIS surveys and this cohort characteristic may be a limitation as population subgroups may change over time as respondents drop out from the survey. The failure to understand the variation in prevalence estimates across surveys is a gap in the literature [20, 51]. With the increasing use of convenience or crowdsourced surveys (e.g., Mechanical Turk) [52], it becomes important to validate the prevalence from these surveys against a larger nationally representative survey. However, our analysis suggests that the validation may depend on the survey used. It may also be important to consider trends over time in the measures used. For example, while we found substantial differences in the measure for 2015, trends over comparable time periods may be similar across surveys.