by Carl V Phillips
I ran across a tweet touting a press release out of the Global Forum on Nicotine (GFN) meeting (a networking meeting, mostly of e-cigarette boosters) that made the claim that snus is 95% less harmful than smoking. This was variously described as being based on “new data”, “new data analysis” and “the latest evidence”, but with no further explanation of where the number came from. Since the presenter was Peter Lee, those of us who know who’s who can surmise that it is a statistical summary of existing published studies, because that is what Peter does. There is nothing necessarily wrong with that (though for reasons I will explain in an upcoming post, it is potentially suspect in this context). but it is certainly not new data or the latest evidence.
Oh, and it is clearly wrong.
(Aside: I seem to recall a few instances when the GFN types screamed bloody murder about the science-by-press-release game after an attack on e-cigarettes was released to the press before the scientific community had a chance to review the evidence. Go figure. Well, you know, maybe I am misremembering. Also interesting is that most of those attacks on e-cigarettes at least had a sketch of the methodology in the press releases, coming from the press offices of scientific rather than networking meetings. So there was possible to partially assess what was actually being claimed and why, unlike the present case.)
For snus to be 5% as harmful as smoking, the cardiovascular disease risk from nicotine would have to be a bit worse than the worst-case estimates of it. (Contrary to widespread misperception, any plausible level of cancer risk is just a rounding error in comparison to CVD risk.) It is sorta maybe kinda possible this estimate is correct. But I would put the probability of the risk being a negative percentage of the risk from smoking (i.e., that snus use is net beneficial) as solidly higher than probability that it is 5% as harmful.
This was my reply on Twitter:
But after that tweet I found myself wondering whether the metric makes any sense whatsoever as a way to think about the risk of low-risk tobacco products, especially snus. This if for several different reasons I list below. And in case the significance of this point might need mentioning: I am saying this as the guy who pushed that way of thinking as much as anyone over the last couple of decades.
First, consider the absurdity of me having just written “a negative percentage of the risk from smoking”. The “fraction of the risk from smoking” unit has an Orwellian control-of-thought-via-vocabulary effect, constraining our perception: it tends to force us to assume the effects are negative. But it is quite plausible that the net health effects from smokeless tobacco are positive. There is a possibility that the same is true for e-cigarettes. The next-most similar exposure, drinking coffee, once believed to be unhealthy, is looking more and more net-beneficial. This is a simple one-paragraph point, but is as important as any of what follows.
Second, the scalar unit required to make such a comparison is not useful. What is actually being claimed in the press release? What was I saying in suggesting the effects might be positive? It is reasonable to expect that nicotine, like any stimulant, could trigger an incipient stroke or heart attack. But there is good reason to believe that it (perhaps in combination with other aspects of the tobacco exposure) is protective against neurodegenerative disease. If both of those are true, how does one decide which side of the 0% point these two — let alone the myriad other possible effects — net out to?
The simplest statistic to calculate is to compare the risk of mortality attributed to smoking and to smokeless tobacco (taking the former from existing calculations and calculating the latter). This is what I did in what is, as far as I know, the only attempt to calculate the comparative risk (and, yes, I say that having read that press release; it is possible that Peter actually did this, but it is far from clear). This is also the unit that Brad Rodu, who was also a coauthor on my analysis, used in the first cut at this from a decade earlier. (His analysis was based a “what if” for a highly exaggerated risk for oral cancer from smokeless tobacco, with no other risks, so it did not really do the job. Also there was a double-counting error in the calculation.) For what it is worth, my analysis put the risk at 1% that from smoking, while Brad’s came up with 2%. Note that these are gross-risk estimates, ignoring the benefits that should be counted on the other side of the ledger.
However, even that seemingly clean unit is pretty iffy. A death attributed to a particular disease event, that in turn is attributed to an exposure, is not actually all that meaningful. It is not even terribly meaningful if we distill it down to deaths attributable to a single disease. It obviously does not make sense to just chalk up a fatal stroke in a 20-year-old alongside one in a 90-year-old.
Moreover, for several reasons I have explained on this page previously, the meaning of “exposure E caused a death” does not mean what casual readers think it means. (See, for example, the first subsection here, which illustrates how fraught such statements are.) A few days ago I saw tweets ridiculing headlines about an exposure causing a 100% increase in risk of death. Such a figure is perfectly legitimate (setting aside whether the evidence really supported the claim in this case); it clearly means age-specific mortality rates are doubled. But the common misinterpretation leads to jokes about people dying twice.
If we want to measure deaths, we should at least be measuring years of potential life lost (YPLL). We would want to use as the denominator not the deaths among smokers that are declared attributable to smoking — the 1/3 of them, or 1/2 or 9/10 or whatever made-up number tobacco controllers are claiming these days — but the 5 or 10 years by which smoking shortens life expectancy. The “fraction of smokers who die from smoking” statistic is not actually meaningful (due to those complications I just alluded to) but YPLLs is. Calculating YPLLs for estimated or posited disease risks from a low-risk tobacco product is perfectly straightforward, but it is a lot of work and no one has ever done it.
But wait. What about that neurodegenerative disease thing I mentioned? That has minimal impact on the YPLL result. But most people would count five years living with severe dementia as worse than no dementia but dying one year sooner from a heart attack. And that is to say nothing of how to factor in more subtle health effects like reduction in depression or stress, or the simple acute increase in mental acuity and perkiness.
I spent a fair bit of my career doing cost-benefit calculations in the medical context. By far the most suspect part of that effort was creating a conversion rate between goods where there was no natural way to do so. It is hard enough to just put resource expenditures (i.e., money) and one health outcome (e.g., number of cases of a disease cured) in the same equation, though that is a well-worn path and there is a good way to finesse it. But as soon as there was a third consideration (e.g., Treatment A is more expensive than B; it saves more lives; but it is also causes a substantial period of serious pain), it became very difficult to be comfortable with the conversion rates used, let alone to defend them against criticism. One obvious good reason for this: Different people convert between pain and risk of death differently, so any summary average is not relevant to many actual decisions.
If you want to see a cartoonishly absurd example of trying to throw incommensurate considerations into a scalar measure, you need only look to the Nutt et al. paper about the impacts of various tobacco products (see my dismembering of it here) that is the apparent basis for the popular claim that e-cigarettes are 5% as harmful as smoking. It appears that the 5% number traces back, via a telephone game, to a misinterpretation of that paper’s silly aggregation of health, social, financial, and other factors as just being about health (as well as the little matter of ignoring that the paper is complete junk science).
Bottom line: Actually assessing “percentage of the health impact from smoking” would require collapsing the impacts to a scalar, which really does not work. The alternative is looking at only one particular outcome, measured in just one way (usually not a good way), which is inadequate.
Third. is the problem of anchoring bias. This is the common phenomenon of people not adjusting sufficiently from a starting estimate, or even a quantity that should really have no influence at all, in the face of evidence. This tendency is so bad that if you just show someone a random number between 1 and 100, and tell them it is just a random number, before asking them to estimate a percentage, their estimates will be measurably influenced by the random number (i.e., those who draw a bigger random number will systematically give higher estimates). A comparison to the risk from smoking, which causes more YPLL than any other common consumption choice, creates an anchor which will inevitably bias perceived risk upward.
Tobacco controllers love anchoring bias. It permeates their propaganda and makes it easy for them to collect useful idiots. By intentionally anchoring perceptions of all tobacco products to the risks from smoking, they actually prevent people from being able to assimilate the evidence about smokeless tobacco or e-cigarettes. But “tobacco products”, when used as a category of healthfulness, makes no more sense than “corn products”, which includes both soda and whole-grain health foods. Indeed, in terms of the magnitude of the differences in healthfulness, it is perhaps more like referring to the risks from “mushrooms” or “titanium products” (which include both military jets and cardiac pacemakers).
The risks from smokeless tobacco, e-cigarettes, and pharmaceutical nicotine products should not be anchored to the risk from smoking. As noted, the net health effect is quite plausibly positive. But even just looking at the negative side of the ledger, we are looking at a minor everyday-level health impact, a risk that should be compared to that from, say, exercising five fewer minutes or eating twenty more grams of meat per day. Of course few people have an intuition about the quantity of those risks. But the reality is that people who think they have an intuition about the risk from smoking are mostly wrong. The point is those everyday risks should be the anchor point. Indeed, the ignorance itself is really the right anchor. The perception, “I know that a bit more exercise is a little healthier, but I really do not know how much and am not going to worry about it”, is pretty much how someone should think of low-risk tobacco products.
Fourth, and related to the previous point, though a matter of real numbers rather than subconscious bias, is that the risk from smoking is a unit of the wrong magnitude. It is like quantifying the ingredients for a recipe in tons instead of ounces. Saying that an exposure causes no more than 5% the risk of smoking is like saying the nearest star is five-light years away while Pluto is less than 5% of a light-year away. True, yes, but it grossly understates the difference.
The point is one that I have discussed at length (example). Saying something is “only 5% as much!!!” sounds great out of context. But in this case, the unit is like a light-year. Something that really were 5% as harmful as smoking would still be the second-greatest risk of any common consumption choice (assuming, of course, that it were reasonably common, as smokeless tobacco use is in northern Europe and North America, or e-cigarettes are in the UK, the USA, and other places). That leads to the springing of this trap (also from GFN):
The problem is that starting with this wrong unit means that useful measurement must take place way down below the decimal point. But instead of that happening, people tend to treat all small fractions as the same. It is much the same problem with risks that are expressed in terms like “this pollutant you are being exposed to causes a .001 chance you will get cancer”, which gets basically the same reaction if the number is .01 or .00001.
It is a big difference, a genuinely big deal, whether the gross risks from smokeless tobacco are 99% lower than smoking, or are 98%, or 99.9%, let alone merely 95%, but those numbers get thrown around as if they are all interchangeable. When looking at a choice between smoking and using smokeless products, this difference is a rounding error. As I have tried to emphasize throughout my work on this topic, the risks from smokeless products are within an unimportant rounding error of 100% harmless. This is why the cynical “continuum of risk” myth is so harmful: it obscures the fact that on a scale that includes the risk from smoking, there is no continuum, only two points: 100% and ~0%. But “to smoke or to snus (or vape)?” is not the ultimate question. Even if it is the question for someone sometime, there is later the question, “should I now quit snus/ecigs too?” The difference among those rounding error level estimates, let alone between them and a net benefit, would cause many people to make different decisions if they were making a rational assessment.
Fifth, smokeless tobacco use is mostly not about “to smoke or snus?” In the 2000s, when those of us working on tobacco harm reduction thought smokers could be encouraged to switch for risk reduction, this was the right question and so the “percentage of the risk from smoking” meme was the best way to think. This perhaps turned out to be the useful view for Norway, the only place where widespread switching actually took place. But in the other major legal markets — Sweden, USA, Canada — smokeless users are smokeless users, for the most part, not would-be smokers. The dominance of snus in Sweden is a cultural phenomenon, not an anti-smoking phenomenon (despite often being portrayed that way). Despite efforts to promote switching in North American, smokeless use is overwhelmingly concentrated in subcultures where it is the cultural norm, not an volitive substitute for smoking. I do not want to overstate this: Many smokeless users in the USA and Sweden switched from smoking, and we can predict that many more would be smokers in a world that had just cigarettes and not smokeless. There are many dual users who could be persuaded to settle on smokeless if they knew the rough comparative risk. Nevertheless, smokeless use should not be thought of as a variant of smoking.
So, given that most smokeless tobacco users should not be modeled as would-be smokers, a measure where actual quantification of the risk they face is lost in the rounding error is not useful. Should these consumers consider giving up their consumption choice for health reasons? The fact that the risk is some tiny fraction of that from smoking is not a useful answer.
Note that this argument is not quite the same for e-cigarettes. Setting aside the over-hyped but really rather trivial issue of experimentation by teenagers, e-cigarettes are all about substituting for an existing smoking habit. That is a large part of why a faction of tobacco controllers have embraced them (and, indeed, pretty much taken over the mindspace around them) despite continued hostility toward smokeless tobacco. Still, after giving up cigarettes and settling in to never going back, there is still the next question of whether to give up e-cigarettes because it might reduce health risk. So long as the estimates of that risk are imprecise fractions of the risk from smoking, the needed information is missing.
In conclusion, I am sure it is safe to say that everyone will ignore these concerns and keep using the same flawed phrasing. The obvious comparison is Karl-Olaf Fagerstrom declaring, five or six years ago, that the “Fagerstrom Test for Nicotine Dependence” is really only a measure of cigarette dependence, that he had it wrong, and it should be relabeled and not used outside that context. This has been almost universally ignored, presumably mostly because ignoring that advice is much easier than heeding it. Indeed, it is probably safe to say that I will keep using the same flawed phrasing sometimes, though I will try to nudging things in the right direction. Perhaps I will even live long enough to be able to say “who could have predicted that it would become apparent that this was the wrong way to describe the risk from low-risk tobacco products?”