by Carl V Phillips
Stanton Glantz recently published a paper, Electronic Cigarettes and Conventional Cigarette Use Among US Adolescents; A Cross-sectional Study, whose conclusions do not even remotely follow from the analysis. That is hardly news, of course. In fact, it is probably sufficient to end the sentence with “published a paper”, since the rest is pretty much a given. But it is interesting to see that this time even some of the semi-respectable anti-THR liars are pushing back against how blatant it is. I wish I could say that this reflects a new era of tobacco control people consistently calling for honest science, but I seriously doubt that is the case. Still, it is something.
This is a long post (by the standards of this blog – it is what is needed to do a serious scientific analysis), so I outline it so that you can know what you want to skip if you are in a hurry: 1. The real reason why Glantz’s statistics do not support his conclusion. 2. Addressing a common red herring claim about the ordering of events. 3. Delving deeper into exactly what Glantz is claiming and why it is even worse than the simple headline claim. 4. Coming back to the ACS reference in the title and related press coverage. 5. Some further random technical observations.
The paper was published in one of JAMA’s journals, which some people found shocking — but only those who do not know that most of the population epidemiology studies published in clinical journals are junk. Glantz and Lauren Dutra (who is listed as the first author, but I am going to presume, based on historical evidence, that Glantz was the primary actor behind anti-scientific garbage and phrase this post as such, but note that coauthor Dutra is equally guilty) looked at cross-sectional survey data about cigarette and e-cigarette use by US teens in 2011-2012, and claimed that it showed that e-cigarettes are a gateway to smoking, though it actually showed no such thing.
Basically what Glantz claimed is that because there is a positive association with some measures of cigarette use and some measures of e-cigarette use, that e-cigarettes are causing people to smoke. Yes, really, that is basically what he is saying. This is the same anti-scientific garbage that has been the basis for tobacco gateway claims for most of two decades. The authors of this entire literature willfully ignore the obvious fact that liking one particular tobacco product means you are more likely than the average person to like another tobacco product, and the more you like one the more you like another. Well, duh!
To show just how absurd his conclusions are, let’s assume we are in a world where every single person who ever tried an e-cigarette is a smoker who was using them or considering using them to stop smoking. That is, they are all about THR and smoking cessation, always. Now consider what Glantz considers to be his main findings from the data (by quoting from the abstract): “Current e-cigarette use was positively associated with ever smoking cigarettes and current cigarette smoking.” That would be true in our scenario. “In 2011, current cigarette smokers who had ever used e-cigarettes were more likely to intend to quit smoking within the next year.” Certainly true in the scenario. “Among experimenters with conventional cigarettes [someone who had ever tried one puff], ever use of e-cigarettes was associated with lower 30-day, 6-month, and 1-year abstinence from cigarettes.” That would be true in the scenario.
“Current e-cigarette use was also associated with lower 30-day, 6-month, and 1-year abstinence.” That would not necessarily be true in the scenario, but it certainly might be: at a particular point in time, especially early in the history of e-cigarettes use as this is, many people using e-cigarettes to try to quit are likely to still be smokers or very recent switchers, whereas a lot of one-time smokers who are not using e-cigarettes would have already quit smoking. “Among ever smokers of [>100] cigarettes, ever e-cigarette use was negatively associated with 30-day, 6-month, and 1-year abstinence from conventional cigarettes. Current e-cigarette use was also negative associated with 30-day, 6-month, and 1-year abstinence.” Same point as the one prior.
So, every statistic that Glantz considers the to be the best support for his claims is either inevitably true in a world where e-cigarettes are used exclusively to quit smoking, or is very consistent with such a world. And yet he claims “Use of e-cigarettes does not discourage, and may encourage, conventional cigarette use among US adolescents.”
It is difficult to imagine more glaring scientific illiteracy: Propose a hypothesis. Find evidence that is consistent with your hypothesis, but also consistent with the most extreme version of the competing hypothesis. Declare your hypothesis to be right. The University of California and JAMA apparently do not expect their researchers to rise to the standards of a middle school chemistry class. Let’s hope the students who were the subject of the survey understand science better than the authors, or Americans will all be sewing shoes and toys for the Chinese within a generation.
As an illustrative comparison, consider this bit of “science”: Hypothesis: it is nighttime. Alternative hypothesis: it morning and very cloudy. Observation: no one I can observe is wearing sunglasses. Conclusion: it is nighttime. Obviously the observation has done nothing to resolve which hypothesis is true. There are a lot of observations that could help resolve this question, like looking at a clock or observing what meal restaurants are serving. But instead of seeking data that would address the question, Glantz sought data that resolved nothing, presumably to avoid the risk (indeed, based on what we know, the inevitability) that real science would not support his daft claims.
Notice that the reason Glantz is glaring wrong has nothing to do with the data being “cross-sectional”, which means that the subjects were not followed over time but just surveyed once. The cross-sectional nature of the data has been highlighted as the major weakness of the study, though the real scientific problems are those noted above. A claim that has come out in most of the criticism of the study is that because it is cross-sectional data, you cannot determine whether e-cigarette use preceded smoking or not. The popular claim is that unless you can show that e-cigarette use preceded smoking, then there cannot be a gateway, and it is not possible to determine the order of use. This is actually not right for a couple of reasons.
First, a cross-sectional study can determine ordering of past events (subject to the accuracy of subject recall) by asking the right question — e.g., “Which of the tobacco products you have tried did you try first?” It turns out that the U.S. government survey on which the paper is based did not ask a question like that. However, it did ask about age at first use, although only for cigarettes and smokeless tobacco, not for e-cigarettes. It also recorded the current age of the subject. Since almost all first use of e-cigarettes occurred within a year or two of the survey date, it would be easy to show a large percentage of product users who had already smoked before they had even heard of an e-cigarette. Glantz pretended that this data did not exist, even as he made claims that using that date would help inform.
But the second problem with this line of reasoning is rather more important: Ordering does not actually provide evidence of causation, or rule it out. Activists who want to claim that there is a gateway effect like to pretend that if consuming X preceded consuming Y, then X caused the use of Y. Obviously that is not true. The classic counter-observation is that almost every heroin user consumed milk before taking up heroin.
But it is also not the case that if first use of cigarettes preceded any use of e-cigarettes then e-cigarettes could not be causing smoking. It is entirely possible that someone tried a cigarette but would not have been a smoker had e-cigarettes not existed, but she became a user of e-cigarettes and that caused her to become a smoker. I am not saying this happens. It is quite far-fetched, but this is because of empirical evidence and analytic reasoning that have nothing to do with the order. The point is that merely observing that most e-cigarette users were already smoking when they tried e-cigarettes does not rule out the gateway claim. Claiming this is not grossly scientifically illiterate like Glantz consistently is — it is a rather more subtle point, and it is an easy honest error for someone who is not expert in causal reasoning to make — but it is still clearly wrong.
Previous commentators have fixated on Glantz’s concession, in the discussion and conclusions, that the cross-sectional nature of the data does not allow conclusions of ordering or causation. But even this concession reflects further lack of understanding about data and epidemiology on the part of the authors: Cross-sectional data can be used to determine ordering, as noted above. But the authors chose to ignore what data they had that addresses that and the survey was not designed to do any better. It is also quite possible to draw causal conclusions from cross-sectional data, you just have to understand scientific inference a little bit. (More on this in section 5.)
Going back to the quoted conclusion statement from the abstract, notice that it is even stronger than merely claiming that there is a gateway effect. It phrases the claim in terms of “discourage” and “encourage”, which draws conclusions about motivations that not only do not follow from the statistics presented, but could not be addressed at all based on the types of questions in the survey.
It gets worse. I am going to skip over all the random rhetoric and assertions in the introduction (as every good researcher I know always does), and just look at the discussion/conclusions. The first claim in the discussion of the results is that adolescent use of e-cigarettes is increasing rapidly, something that study did not show, and indeed the dataset it is based on could not show. Glantz then concludes that the “results call into question claims that e-cigarettes are effective as smoking cessation aid.” But even if the results had suggested that e-cigarettes were, for some reason, causing more teen smoking to occur than otherwise would have, that would tell us nothing about whether they might also be effective cessation aids.
The discussion then wanders off into an irrelevant discussion of nicotine, addiction, and animal experiments. It is a good sign that someone is not really trying to do a scientific analysis — but rather is reporting some statistics as an excuse to editorialize about their political beliefs — when they wander off like this. All of us who write about political scientific topics sometimes try to present a big-picture analysis of the implications of all the science. That is fine. But it should not be tacked onto a research report.
And, finally, “The results of our study together with those from [Glantz’s equally mis-concluded] study in Korea suggest that e-cigarettes may contribute to nicotine addiction and are unlikely to discourage conventional cigarette smoking among youths.” But these results are perfectly consistent with the teens in the study population only using e-cigarettes for THR, as noted above. Moreover, there is absolutely nothing in the study that addresses the question of addiction (whatever that means).
Notice also that the phrasing is forward-looking (“are unlikely to discourage” rather than “did not, in 2011, discourage”), though there is nothing in the analysis that speaks to predictions about the future. The introduction of mobile phones did not, at first, discourage the use of landlines or even cause the decline of payphones, but that was the very predictable future. Similar predictions should be made about e-cigarettes even if youth use to date does not yet show that pattern (not that we can conclude that it does not from the present analysis). That is, even if e-cigarettes were not reducing youth smoking in 2011 does not mean that they will not be doing so soon.
Basically none of the conclusions of this paper follow from the analysis.
And I am not the only one who thinks so. The New York Times article about this paper, following a summary of what Glantz claims, notes:
But other [sic] experts said the data did not support that interpretation. They said that just because e-cigarettes are being used by youths who smoke more and have a harder time quitting does not mean that the devices themselves are the cause of those problems. It is just as possible, they said, that young people who use the devices were heavier smokers to begin with, or would have become heavy smokers anyway.
“The data in this study do not allow many of the broad conclusions that it draws,” said Thomas J. Glynn, a researcher at the American Cancer Society.
When the ACS says you are going too far in your anti-THR lies, you know you have not just jumped the shark, but have flown clear off the planet.
As is usually the case with study authors who are not really trying to contribute to science, the claims to the press were even more egregious than those in the paper:
Dr. Glantz says that his findings show that use of e-cigarettes can predict who will go on to become an established smoker. Students who said they had experimented with cigarettes — that is, taken at least one puff — were much more likely to become established smokers if they also used e-cigarettes, he said.
“One of the arguments that people make for e-cigarettes is that they are a way to cut down on the smoking of cigarettes, but the actual use pattern is just the opposite,” he said.
But, of course, there is nothing at all in the data or the observation quoted there that supports the conclusion. E-cigarettes might be used entirely to cut down on smoking, and still be associated with smoking. Indeed, it would almost be inevitable: those who are not inclined to smoke much or at all will not use a substitute to cut down.
But David Abrams, executive director of the Schroeder Institute for Tobacco Research and Policy Studies at the Legacy Foundation, an antismoking research group, said the study’s data do not support that conclusion.
“I am quite certain that a survey would find that people who have used nicotine gum are much more likely to be smokers and to have trouble quitting, but that does not mean that gum is a gateway to smoking or makes it harder to quit,” he said.
Yup. Moreover, it almost certainty the case that those who smoke weed, drive unsafely, perform poorly in school, etc. are much more likely to be smokers. As I have previously pointed out, it is theoretically possible to control for propensities (use covariates like these) to estimate whether one behavior is in fact increasing another, rather than just being a shared propensity. But such real science is waaaay over the heads of the tobacco control “researchers”.
The study did have a bright spot: Youths who used e-cigarettes were more likely to plan to quit smoking. Dr. Abrams highlighted that finding, but said it was impossible to tell whether students who planned to quit actually did, because the data did not track this.
Funny, I did not notice that in the conclusions that Glantz wrote. This is actually the only observation from the data that relates to the THR points that Glantz emphasized. Abrams is, of course, right that we do not know whether the intentions were acted upon. But the fact that e-cigarettes are associated with the goal of quitting is strong support for the claim — contra Glantz — that even among teens they are being used for harm reduction.
Finally, for those who are interested (and as some notes for myself), a random collection of more technical observations.
One additional observation about cross-sectional studies I did not want to belabor above: It turns out that some of the most effective and convincing observational epidemiology studies, in terms of being able to draw causal conclusions, are cross-sectional. I speak of disease outbreak studies, where a group of people come down with some relatively uncommon infectious disease and the CDC’s EIS (a good unit of CDC, in contrast with their nanny state units) interviews people to determine their recent exposures. They discover things like, “everyone with the disease visited Africa recently” or “almost everyone who got the disease recalls eating a particular brand of salsa in the past month” or, even better, “the people at the gathering who got sick mostly ate some of the chicken that was served, while those who did not get sick did not.” These cross-sectional studies lead to very solid causal conclusions. There is nothing magic about one study design versus another — all can be misinterpreted in terms of drawing unsupported causal conclusions and all can be correctly interpreted to draw valid causal conclusions.
Further on the topic of temporality: I will be a little more emphatic to the pro-THR activists who keep repeating the claim that if someone smoked a cigarette before puffing an e-cigarette, she cannot be a gateway case: Please quit it! That is not legitimate reasoning. If our side is going to traffic in junk science like that, how are we any better than the tobacco control industry? Obviously it is true that if someone tried a cigarette first, then e-cigarettes did not cause her to smoke her first cigarette. But you are letting the simple concept “cause must precede effect” confuse you about what the effect in question is. We are not interested in what causes someone to smoke her first cigarette, we are interested in whether it causes her to become a smoker.
Does this make that point obvious enough?: Just because someone had a case of salmonella before ever eating chicken does not mean that eating chicken did not cause the salmonella she has now. How about this?: I smoked some cigarettes long before I ever tried snus; but if I were to take up smoking now, it would clearly be because I acquired a taste for tobacco use thanks to snus, and I clearly would be a gateway case. (And why would that never happen? That is the logic I have argued for more than a decade, most recently here: Since I did not become a smoker because I preferred abstinence to enduring the costs of smoking, why would preferring snus use to abstinence cause me to switch to my least-preferred option?)
There are some rather embarrassing flaws in the survey itself. The questions about trying e-cigarettes (asked in 2012) give as the examples “…such as Ruyan or NJOY”. Seriously? The last time I saw a Ruyan-branded product was several years before that. This probably did not bias the answers, but it tells us a lot about how out of touch our government “experts” are about e-cigarettes. But it is the case that by 2012 the question should have indicated that the category of interest included disposables, kits, and mods, since it is possible that the local teen lingo in some places would consider the question as asked to not include all of those. Of course, more useful still would have been to ask about those subcategories separately rather than wasting time in the “other tobacco products” asking about kreteks and other insignificant products. (And that is to say nothing about the utter chaos created by including “snus” in the “other” grouping rather than in with the rest of smokeless tobacco, but that is a disaster for other analyses, not this one.)
One sentence in the discussion subtly betrays the authors’ weak thinking and inability to separate their own rhetoric: After pointing out that e-cigarettes contain nicotine, they state, “The adolescent human brain may be particularly vulnerable to the effects of nicotine because it is still developing.” But they claim the paper is about e-cigarettes causing teens to smoke, not about whether they are harmful in themselves. The standard practice in tobacco control is to throw every argument they can think of against the wall to see what sticks, regardless of whether a claim is accurate or even if the collection of claims are internally inconsistent with one another. They try to pull back on this lying when they pretend to be doing science, but even when they are mostly successful, they just cannot help themselves.
The authors’ use of covariates is naive and simplistic. This is typical for people who do epidemiology without understanding epidemiology (i.e., most people who do epidemiology), and is definitely not unique to tobacco control activists. Most notably, they controlled for age as a covariate. This simply makes no sense. If there is a relevant real difference across ages in this analysis (as there almost certainly is), it does not call for a half-hearted age standardization, but for stratification into age groups so we learn what is different about importantly different groups. On the other hand stratifying by race has no apparent justification, and is just arbitrary window-dressing (that is, assuming they corrected for the oversampling of some racial groups as they claimed to). They did not stratify where it would be useful and did stratify where it served no purpose.
There are other variables in the dataset that would be interesting to stratify on, but the authors presumably have no idea what they are, or just did not care enough about even pretending to do useful science to bother with them. Most notably, there are questions about perception of the risk from e-cigarettes. They are not very good, asking just less, equal, or more compared to smoking, rather than dividing out “less” enough to make sure someone understood the difference between 99% less and just a little less. Still, any differences in behavior between those who think e-cigarettes are as bad as smoking and those who know better could tell us something about conscious THR efforts.
The authors also seem to have no idea how to properly interpret or report random error test statistics (p-values and such). Yawn. That is pretty much the rule in public health and medical journals. The remark in the discussion about the difference in power between the Korean and U.S. studies should elicit a smirk from those who understand statistics, but I am not going to try to explain the joke here.
In the inappropriate discourse into nicotine in the discussion, the authors refer to it as “highly addictive” without ever suggesting what they mean by “addictive”, much less what constitutes “highly”. To pretend that they are doing science, they attach a reference after this, but it is just to a political commentary piece, not something that could be a basis for a scientific claim (which is pseudo-scientific, in reality). Such anti-scientific behavior in the public health literature is so ubiquitous that it might seem not worth mentioning (it is not the only such example in the paper), but it is important to at least try to push back against anti-science like this occasionally. I would certainly never let a student get away with either meaningless rhetorical claims or misuse of references like this. (I sometimes wonder if authors like these just never had a decent teacher, or if someone out there is mortified by his failure.)
One tantalizing observation that is not analyzed (surprise!) is the very low portion of so-called former e-cigarette users who were currently heavy smokers. That is, very few people tried e-cigarettes, gave up on them, and remained heavy smokers at the time of the survey. This could just mean that the heavy smokers did not bother to try e-cigarettes, but it could also mean that large portion of those who were heavy smokers and tried became non-heavy smokers. That is, THR worked! (Indeed, if I were as unconcerned with good scientific reasoning as is Glantz, I would declare that this is what it shows, and would be standing on firmer ground than he is with his conclusions.) This is a place where the cross-sectional nature of the study (and the absence of enough retrospective questions) really is the problem: If the data followed people over time, or if the survey asked whether someone used to be a heavy smoker, this could easily be resolved and the causal role of e-cigarettes in reducing heavy smoking could be fairly effectively inferred.