Developing Consistent and Transparent Models of E-cigarette Use: Reply to Glantz and Soneji et al.

During the coming years, it will be important that those on both sides of the debate on e-cigarettes have a framework to examine the potential effect of e-cigarettes and be open to revising their views as the data unfolds. In a previous article,1 we provided a framework to analyze public health impacts. In this article,2 we applied that framework and we applied the most recent, nationally representative data to make public health projections for a single cohort. Glantz3 has claimed that Kalkhoran and Glantz (K&G)4 is using the appropriate data to model e-cigarette trends, whereas our model is ad hoc. A careful examination of the K&G model reveals that they are using data on ever use to characterize both initial use and regular use. As is well known, only a relatively small portion of ever users (especially among youth) become current users and, among current users, only a portion becomes regular users.5,6 K&G’s implicit assumption that ever use translates into established use is highly questionable and consequently their projections are built on a weak base. In contrast, we explicitly consider how trial use translates to established use. The parameters are based on evidence documented in our paper and follow from the application of our decision-theoretic framework. In addition, we conduct extensive sensitivity analysis to show how public health implications depend on those transitions. Glantz3 also criticizes our paper for using “hypothetical measures.” This charge is not accurate. In our paper we attempt to use the strongest available data to arrive at the appropriate transitions that need to be considered for never-smokers, namely the transitions by those users who would have become smokers in the absence of e-cigarettes as compared to those that would not have become smokers. That distinction is critical to estimating the public health impact of e-cigarettes as recognized in the scientific literature.1,7–10 K&G implicitly and somewhat ironically distinguish otherwise smokers from otherwise nonsmokers in building their model on a framework that divides the population of youth and young adults into smokers and nonsmokers. In addition, implicit assumptions are made about the hypothetical smokers who would have quit or who would remain smokers in the absence of e-cigarettes. Glantz3 also claims that the K&G steady state approach is appropriate, implying that our cohort-based approach is somehow less appropriate, or even inappropriate. But it is important to note that a steady state approach, as implied by its name, makes the assumption that the effects by age group are constant over time. This is highly problematic in today’s rapidly shifting tobacco landscape. As is well known,11–13 youth in 2012 faced a completely different set of e-cigarette product options than same aged youth in 2014, which likely contributed to much of the increase in use over those 3 years.14 In addition, the Monitoring the Future Survey15 reported that the latest rates of past 30-day e-cigarette use levelled off or dropped slightly from 2014 to 2015, thus altering the rapid rates of increase from 2011 to 2014. Another concern with a steady state model is that erroneous estimates arising from failing to consider cohort variations will be compounded over time. For example, a recent study16 found levels of current e-cigarette use nearly as high among 35–54 year olds as among 18–24 year olds. As we have observed from years of data on cigarette smoking uptake, later cohorts are likely to have established patterns of use at earlier ages and thus have lower rates of uptake at later ages. We employ a cohort approach to specifically allow for those age variations, rather than using a steady state approach in which there is no variation by age or cohort. In future modeling and empirical analyses, it will be important to allow for e-cigarette transitions to vary by cohort, as well as age. Our paper was also criticized3,17 for not considering higher levels of the excess risk of e-cigarettes as compared to cigarettes. Given the evidence from the use of smokeless tobacco18 and from our current knowledge on biomarkers of harm and functioning,19,20 estimating the relative harm of e-cigarettes compared to cigarettes at 25% is in all likelihood a conservative upper limit; some reviews and studies of the evidence to date21–23 have estimated risk of e-cigarettes being about 5% of the risk of cigarettes. A coherent, evidence-based case has not been made by Glantz or others for a level of risk of 25%, let alone greater. While there may be some debate about the level of e-cigarette risks, well-designed regulations can be expected to reduce possible risks and promulgate product standards that ensure quality, consistency of ingredients and ensure as low a risk of harm as possible. Contrary to suggestions by Soneji et al.,17 we used standard methods to compute life years lost, and by conducting sensitivity analyses over a reasonable range of risks, we attempt to convey the uncertainty in those calculations. We agree with Soneji et al.17 that validation is important, but disagree that the appropriate data are now available to allow reliable validation. Soneji et al.17 inappropriately compared the NHIS 18-year-old cigarette smoking prevalence to the level in our counterfactual. However, we based our counterfactual on data from 2012 and earlier in order to project what would have occurred in the absence of e-cigarettes. The more relevant validation would be the actual reduction in prevalence among young adults (ages 18–21) incorporating their use of e-cigarettes since 2012. Using the NHIS data, the age 18–21 smoking prevalence declined from 14.4% to 9.6% between 2012 and 2015, a relative decline of 33%. Comparing the 18- to 21-year-old counterfactual from our model to the predicted prevalence with e-cigarettes, we estimate a relative reduction in smoking prevalence of 10%. While it is likely that some of the actual 33% reduction in age 18–21 cigarette smoking prevalence is attributable to population-level interventions (eg, cigarette tax increases, mass media campaigns), this comparison suggests that we may be underestimating the impact of e-cigarettes on cigarette use, and thus underestimating the public health benefits from e-cigarette use. We appreciate the time and effort that others have taken to review our model and discuss strengths and weaknesses. We believe our model represents a fair and balanced effort to apply a comprehensive framework to estimating the impact of e-cigarettes on public health using the best evidence now available—and as new and more reliable data become available, our model can readily be modified to produce updated estimates of the impact of e-cigarettes on important public health outcomes. In our view, a first tenet of model building is to make explicit the underlying assumptions, and show the implications of those assumptions. Indeed, that is particularly important at this early stage of modeling e-cigarette use, so that models can usefully be compared. As is true in model building, researchers will need to make explicit the assumptions that they make in conducting empirical analyses and in interpreting the data, and also to be ready to change those assumptions if refuted by the evidence.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.