Misread signals: a misinterpretation of population-level vaping and smoking trends

Introduction

The debate over the role of e-cigarette use – or vaping – in the subsequent initiation of cigarette smoking, particularly among adolescents and young adults, has been shaped by conflicting interpretations of epidemiological data. Analyses of individual-level data from prospective cohort studies have consistently shown increased risks of cigarette smoking uptake for young people who have previously used e-cigarettes, compared with those who have not (sometimes called a ‘gateway effect’).^1–4 For example, a 2021 systematic review and meta-analysis of 25 prospective cohort studies found that for vapers, the odds of subsequent smoking initiation was 3.19 (95% confidence interval [2.44, 4.16]) times higher than for non-vapers.² Despite the findings of cohort studies, several studies analysing repeated cross-sectional data to examine population-level trends have concluded that vaping may actually decrease the risk of smoking via a ‘displacement effect’, pointing to an increasing vaping trend coinciding with a decreasing smoking trend as evidence.^5–7 For instance, a study of US adolescents observed that between 2014 and 2020 ‘[a]n increase in regular EC [e-cigarette] use was accompanied by a decrease in regular cigarette smoking’, concluding ‘[o]ur results are more consistent with the hypothesis that ECs are displacing young people away from combustible cigarette smoking at the population level’.⁵ Similarly, a 2020 study of New Zealand adolescents concluded that ‘[t]he overall decline in smoking over the past 6 years in New Zealand youth suggests that e-cigarettes might be displacing smoking’.⁶ These conclusions, however, stem from a misinterpretation of coinciding vaping and smoking trends, which we illustrate through a simple hypothetical example.

Example

Table 1 illustrates three hypothetical scenarios with steady-state adolescent populations, each maintaining a constant size of 10,000 individuals per year, balanced by new entrants each year as others age out. In all three scenarios, vaping prevalence starts at 0% in year 1 while smoking prevalence among non-vapers starts at 15%.

In the ‘gateway effect’ scenario, vaping prevalence increases by 3% annually in absolute terms, while smoking prevalence among non-vapers concurrently decreases by 10% each year in relative terms. For the years when the number of vapers is non-zero, smoking prevalence for vapers is set at three times higher than that of non-vapers (which induces a matching 10% annual declining trend in smoking prevalence for vapers). This threefold-higher smoking prevalence among vapers aligns with the pooled relative risk (RR) of 3.0 estimated in a 2021 meta-analysis of 17 prospective cohort studies of adolescents.⁴ Across a 5-year span, we observe vaping prevalence rise from 0% to 12%, while overall smoking prevalence concurrently drops from 15% to 12.2%.

The ‘no effect’ scenario begins with the same initial conditions as the ‘gateway effect’ scenario, including an annual increase in vaping prevalence by 3% in absolute terms, while smoking prevalence among non-vapers decreases by 10% each year in relative terms. However, in this scenario, the RR is set at 1.0, indicating that vaping has no effect on the likelihood of initiating smoking. Consequently, smoking prevalence for vapers equals that of non-vapers, and this lack of effect results in a more pronounced drop in overall smoking prevalence – from 15% to 9.8% over 5 years – when compared with the gateway effect scenario.

The ‘displacement effect’ scenario also starts from the same initial conditions, with an annual increase in vaping prevalence by 3% in absolute terms and a 10% annual decrease in smoking prevalence among non-vapers in relative terms. Here, however, the RR is set at 0.33, the reciprocal of the RR = 3.0 used in the ‘gateway effect’. This RR of 0.33 results in a smoking prevalence among vapers that is one-third that of non-vapers, which leads to a more pronounced decrease in overall smoking prevalence – from 15% to 9.0% over 5 years – when compared with the other two scenarios.

Discussion

Several studies have concluded that an increasing vaping trend coinciding with a decreasing smoking trend suggests vaping is displacing smoking.^5–7 Indeed, in our displacement effect scenario, we observed that as vaping prevalence increased, smoking rates declined. However, these coinciding trends were not unique to the displacement effect scenario. Our analysis also showed increasing vaping coinciding with decreasing smoking when vaping had no association with smoking initiation (the no effect scenario) and even when vaping increased the risk of smoking (the gateway effect scenario). Taken together, these three scenarios demonstrate that the mere co-occurrence of an increasing vaping trend and a decreasing smoking trend does not indicate vaping is displacing smoking, and to conclude this from coinciding trends alone is a misinterpretation of the data. Equally though, our example offers little reassurance to those who might be inclined to view concurrent increasing trends in both vaping and smoking as being indicative of vaping’s culpability.⁸ Just as easily, we could construct a hypothetical example showing vaping and smoking trends concurrently increasing under all three scenarios, including the displacement effect scenario. Hence, the mere co-occurrence of increasing vaping and smoking trends does not indicate vaping is increasing the risk of smoking (i.e. a gateway effect).

Although we have shown that an increasing vaping trend coinciding with a declining smoking trend does not constitute evidence of an association between vaping and smoking, a more nuanced aspect merits attention. Analysis of population-level trends can offer insights into potential associations between vaping and smoking by examining changes in smoking trends. This approach is most commonly employed through interrupted time series analysis by comparing trends in smoking prevalence ‘before and after’ a major change in vaping prevalence, such as the initial introduction of e-cigarettes into a population. For example, a 2018 study of youth in England, Scotland and Wales observed ‘a marginal slowing in the decline in regular smoking during the period following 2010 …’ suggesting ‘… little evidence that renormalisation of youth smoking was occurring during a period of rapid growth and limited regulation of e-cigarettes from 2011 to 2015’.⁹ This essential ‘before and after’ comparison of smoking trends was absent in the repeated cross-sectional studies referred to earlier,^5–7 all of which examined a single trend in smoking from 2014 onwards, after vaping had already become widespread in their respective countries, with approximately 20% of their adolescent populations having already tried vaping.^6,10 By not assessing whether smoking trends changed rather than simply declined, the studies incorrectly attributed the observed declines beginning in 2014 to vaping, without considering whether these declines represented a continuation, an acceleration or a slowing of pre-existing declines.

It is important to recognise that our hypothetical example simplifies reality. For instance, in the gateway effect scenario, we assumed a threefold increase in smoking risk that applies annually for simplicity. In reality, however, the RRs and odds ratios from the 17 prospective cohort studies in the meta-analysis were sometimes measured over – and thus correspond to – periods extending up to 2 years.⁴ Consequently, our annual application of the gateway effect RR of 3.0 may not perfectly align with the pooled estimate of 3.0 from the meta-analysis. Nevertheless, the purpose of our hypothetical example is not to capture every possible scenario but to provide a single counterexample to challenge the assertion that an increasing trend in vaping coinciding with a decreasing trend in smoking necessarily indicates that vaping is displacing smoking.

Although it seems intuitive to think that an increasing trend in vaping coinciding with a decreasing trend in smoking suggests that vaping might be displacing smoking, our example shows that this is a misinterpretation of coinciding trends. Such misinterpretations can have significant implications for public health policies and the general understanding of the impact of vaping on smoking behaviours. For example, the New Zealand paper mentioned earlier⁶ was published in The Lancet Public Health and has been cited 61 times in the 5 years since its publication.¹¹ It was the most frequently cited piece of evidence in submissions to the New Zealand Parliament’s Health Select Committee – including by British American Tobacco – regarding a 2020 Bill that aimed to regulate the sale and marketing of e-cigarettes.¹² Furthermore, British American Tobacco referenced the paper as key evidence to argue against the tightening of e-cigarette regulations in its submission to the Australian Parliament’s Select Committee on Tobacco Harm Reduction.¹³ In their submissions to both committees, British American Tobacco cited the same misleading conclusion from the paper that we quoted earlier (i.e. ‘The overall decline in smoking over the past 6 years in New Zealand youth suggests that e-cigarettes might be displacing smoking’).

Conclusion

Our hypothetical example demonstrates how researchers can inadvertently misinterpret coinciding population-level trends, while serving as a reminder of the need to consider the limitations of repeated cross-sectional and population-level data in public health discourse and policy. By highlighting misinterpretations such as this, we aim to foster a more careful and informed approach among researchers and policymakers. This careful consideration is needed to develop policies that accurately reflect the risks associated with vaping.