Will "third of UK teenagers who vape... go on to start smoking"?

The overstated conclusions of Mongilio et al. 2025

The above study, just published in the journal Tobacco Control — one of the top journals in tobacco/nicotine research by traditional metrics (impact factor 4.7) — has gotten news coverage, including by the Guardian, with the following headline:

Let’s take a look at what the study did in the analysis, how the authors interpreted the results and what the issues are, and how the media interpreted the authors’ narrative.

1. What Did the Study Analyze?

1. Compared current smoking prevalence in 16-17-year-olds across 3 eras (1974, 1986, and 2018) in the UK, each using data from a separate national survey.

   1. Current smoking prevalence decreased from 33% to 25% to 12%, respectively.

   2. Current e-cigarette use prevalence was 11% in 2018.

2. Looked at risk factors associated with smoking (e.g. parent smoking, other substance use, sociodemographics, etc.).

3. Used the above risk factor associations to predict who was likely to smoke.

4. Estimated that 43.4% of e-cigarette users (vs. 1.4% of non-users) in 2018 were likely to smoke (based on risk factors):

[Red markup is mine]

2. How Did the Authors Interpret the Results?

• Celebrating the decline in smoking:

  • “we achieved great success in reducing cigarette smoking among youth”

• … But implied that e-cigarettes could be stalling or even reversing this decline:

  • “the anticipated reduction in risk of cigarette smoking was not uniform for all youth… those who reported current use of e-cigarettes had a probability of smoking that was, in fact, slightly higher than that of youth born over 40 years earlier in 1958”

  • “the success of previous tobacco control efforts… may be mitigated when adolescents use e-cigarettes”

3. Critique

1. The results do not at all disambiguate the “gateway” explanation from other explanations such as “common liability.” If I were doing a “cold reading” of the Results, I might interpret them as supporting common liability, because they show that participants with situational or personal characteristics that predispose them to cigarette smoking are concentrated in the e-cigarette use group. It’s not a strong case for common liability due to the other flaws below, but if I can point to the same results to make the opposite case to the authors, it goes to show how ambiguous the results are.

2. The concerns are based on estimated smoking rates (based on risk factors), not actual smoking, which continued to decline (33% in 1974 to 25% in 1986 to 12% in 2018). To give the benefit of the doubt, I wondered if focusing on estimated rather than actual smoking rates might be a way of estimating a counterfactual scenario, but that doesn’t seem to be the case in the narrative. Rather than use actual smoking (which is directly available in the 2018 data set), the authors instead use a circuitous process of indirectly estimate smoking rates from risk factors. Is this because the rates of estimated smoking are higher than actual smoking (since estimated smoking might theoretically capture not only the youth who already stated smoking, but the ones who will start in the future)?

3. Speaking of estimated smoking, the accuracy of this estimation is not shown. It’s assumed that it’s valid because it’s based on risk factors that are correlated with smoking, but correlation alone doesn’t always mean you can use one variable to accurately predict another. As I showed in a recent methodological article (Selya et al. 2025), risk factor models actually don’t predict smoking status very well: they predict non-smoking much more accurately than smoking, meaning the false alarm rate is high (i.e. estimated smoking is higher than actual smoking).

4. Shifting group comparisons: e-cigarette users in 2018 are found to have similar estimated smoking as the whole population in 1974. Estimated smoking is similar in these two comparisons (not “slightly higher” as it’s described, given the wide error bars), but the e-cigarette use group is only 11% of the whole population, so overall there’s still a population-level decline in both actual smoking and estimated smoking.

5. Speaking of differences, “estimated smoking” is calculated differently over the 3 eras. The authors acknowledge that the associations with risk factors are different across studies, which means that it’s not sensible to compare the different bars across years in Figure 1 above. Moreover, e-cigarette use is a predictor of smoking in 2018 but not in the prior (pre-e-cigarette) waves, which makes sense on its face but exacerbates the incomparability across eras, in a way that I think would further inflate “estimated smoking” in 2018 specifically (since many other risk factors were less common then).

Also see the Science Media Centre, which has some excellent comments from experts that I completely agree with (and they raise additional points, such as not knowing the order of smoking and vaping).

4. Further Distortion in the Media

The media article further adds implications that are not supported by the actual results:

• “will go on to start smoking tobacco” refers to the estimated smoking from models based on risk factors. This wording sounds like smoking is an objectively documented outcome, but it’s a speculative, uncertain, and probably overestimated (as I explain above).

• The media subtitle has explicit causal and directional claims that are not supported by results: “e-cigarettes increasingly act as ‘gateway’ to nicotine for children, undermining falling smoking rates”.

5. The Verdict

This study would have really benefited from preregistration because the results are a Rorschach test that can be interpreted in opposite ways to support either gateway or common liability. Objectively, the declining rates of actual smoking is a favorable result.

A more accurate, neutral title would be “youth who used e-cigarettes in 2018 are similar to those who smoked in 1974 with respect to personal and situational factors.”