by Carl V Phillips
Just a quick note to vent my amusement about the never-ending war of commentaries about whether e-cigarettes are a gateway to smoking. That war apes a scientific debate, but it is not one for several reasons. Most notably, no one (on either side) ever explains what they would mean by “there is a gateway effect.” There are also serious problems about what would constitute useful evidence.
I suppose you don’t vent amusement, do you? You vent frustration. And it is frustrating that I recently spelled most of this out and yet even the ostensible scientists in the debate do not seem to have bothered to read that or any of the other serious scientific analysis on the topic. And they won’t read this either, so it does not seem to merely be a matter of tl;dr. I blame social media and the motivations it creates to write without doing the reading. And the thirty-second news cycle. And blogs. And Twitter. Also, would you kids please be so kind as to get off my lawn.
Aaaaanywaaay…. The problem of definitions is mostly a matter of quantities. This contrasts with the comparison from the title (“addiction” is debated without defining it at all, so there cannot even be a quantitative issue, notwithstanding that people make quantitative claims like “more addictive”). But it is still a problem of arguing about something that is undefined. The problem is worse if someone mis-defines a gateway to be about the mere order of using the products, but that is so clearly wrong I will not bother to say more.
If the claim is that at least one would-be never-smoker was or will be caused to become a smoker because of e-cigarettes, then it is obviously true. Denying that is just innumerate. Millions of would-be never-smokers try e-cigarettes. It is impossible to believe that there is not one in a million of them who discovers he likes nicotine a lot more than he expected, and then gets tricked into believing that cigarettes are no more harmful, or spends time somewhere e-cigarettes are banned, or even just discovers he likes smoking more sometimes. One in a million is a very small number. It is roughly the chance of you dying from a car crash tomorrow. Similarly it is inevitable that there is someone who will die sooner because they switch to e-cigarettes than they would have if they had kept smoking. It is possible, and therefore it occurs. That is what happens when you deal with really big populations.
At the other extreme is a claim is that a large portion (say, half) of would-be never-smokers who try e-cigarettes rapidly become smokers. Such an effect would be impossible to miss in the data even given the problems with confounding. No such results have showed up, so we can pretty confidently rule that out, even though it is what is often implied (although, again, what is being claimed is never actually defined). I qualified that with “rapidly” because time quantification matters too. If the claim were that the gateway effect occurs five years after someone tries an e-cigarette we would not yet have any useful evidence, one way or the other. I lay all that out in the paper.
Thus, any scientific debate about the implications of current empirical findings must be about a fairly rapid transition with a conversion rate that is greater than “at least one” and less than 50%. That does not narrow it down much, and no one ever clarifies. There is no empirical evidence that affirmatively suggests there is a measurable gateway effect. But if you define the conversion rate low enough or the lead time high enough, none of the evidence can be said to suggest there is no such effect either.
The best argument that there is no gateway effect, beyond a trivial “at least one” level, is that (a) it is an extraordinarily unlikely event based on the simple economics of the situation, and (b) there is not even a single published case study where an individual is convincingly identified as a strong candidate for being a real gateway case. (These points are also detailed in my paper.) Thus the existence of a quantitatively non-trivial gateway effect is an extraordinary claim that requires extraordinary evidence before it can even be considered a plausible hypothesis, and there is no such evidence.
The current round of commentary on the topic was touched off by a paper by Primack et al. out of University of Pittsburgh, in which they reported that in a particular (see below) population of teenage non-smokers, those that had tried e-cigarettes were much more likely to try smoking. That paper has been widely criticized because their number of exposed cases was very small, and also because their endpoint was trying smoking and not becoming a smoker. These are definitely weaknesses, but not fatal flaws. A small number of exposed cases is pretty common in epidemiology, and is not a fatal problem as long as there is not model fishing (“publication bias in situ“) happening (which might be happening, but that level of scientific sophistication has not been part of the debate). The endpoint is broader than the real endpoint of interest, but obviously trying smoking is a legitimate predictor of whether someone is more likely to go on to become a smoker. This means that the quantitative estimate will be too high, but still on the same side of the null.
The fatal flaw, as with all other studies that are cited as showing there is a gateway effect, is confounding: The same type of people who are more likely to try/use e-cigarettes are more likely to try/use cigarettes. You need to do a really good job of controlling for that common propensity when trying to sort out any gateway effect. This is particularly critical in this case because the effect of the confounding is clearly much greater than any gateway effect — assuming someone actually defined what gateway quantity they were talking about and it was in the range that was both realistic and consequential. It would be kind of like trying to assess the health effects of e-cigarettes without controlling for the effects of the widely disparate levels of pre-vaping smoking that individuals engaged in; the variation in the latter swamps whatever effects e-cigarettes might have. I cover the why and some of the how of dealing with the propensity in my paper.
But here is the thing: The Pitt researchers actually tried harder to do that than anyone has before. They narrowed the population. Indeed, if you take them at their word, they did a good job of it:
In this longitudinal cohort study, a national US sample of 694 participants aged 16 to 26 years who were never cigarette smokers and were attitudinally nonsusceptible to smoking cigarettes…. (emphasis added)
If you genuinely narrowed your population down to people who were almost certainly destined to be never-smokers, but for the intrusion of e-cigarettes, and then got a result like theirs, you would have solid evidence of a gateway effect. The problem is that in the survey data they used, almost every never-smoker answers the “susceptibility” questions — which are about the possibility they will try a cigarette in the next year and their willingness to do so if encouraged by a friend — with the most emphatic negative response, that there is absolutely no chance it could happen. As I recall, about 90% of them give those answers. This is absurd in itself and says more about them trying to give the “right” answer than about understanding reality; after all, the chance that you might find yourself committing homicide in the next year — let alone trying a cigarette — is not absolutely exactly zero. Still, that is what almost all of them claim.
For present purposes, that means that this attempt to measure propensity is almost worthless. Yes, it removes from the population the small minority who are considering smoking, and are willing to admit it to themselves and then say so on a survey when sitting in their classroom, but that still leaves a huge range of propensities. The confounding is still there. Some of the commentaries correctly note the problem with the confounding, but I have not seen one that recognizes that there was a prima facie legitimate attempt to deal with it, but then explains why it was a fail for empirical reasons.
Ok, I feel better now — at least until next week when a new commentary comes out that again fails to consider any of these observations.