by Carl V Phillips
This new paper, by Marewa Glover and me, is just out in Harm Reduction Journal. In it, we review the available epidemiology evidence about the effects of nicotine-sans-smoke (NRT, snus, vape) on pregnancy outcomes. It was a bit of a challenge to get it published because we wrote the paper we needed to, rather than a “typical review”. As you might know, the journal publication process is rather …well, let’s just say conservative.
(I should note that we finalized the review before this new contribution to the literature by Igor Burstyn et. co., “Smoking and use of electronic cigarettes (vaping) in relation to preterm birth and small-for-gestational-age in a 2016 U.S. national sample” (note that this link bypasses the paywall, but does not seem to work in all browser configurations). Igor’s paper is higher quality than anything we reviewed.)
A typical review of epidemiology looks at the results that are reported in journal articles and then just naively believes them, suggesting that What We Know consists of a vague summary of whatever results the previous authors chose to publish. Or even worse — so much much worse — suggesting that a calculated average of those results is the best estimate. That is never a legitimate assessment of existing knowledge, and less so in our case.
In this case, we know a lot without looking at this epidemiology: We know that these other products have some of the pregnancy-affecting potential of smoking (which mostly means they have nicotine, with only trivial doses of some of the other exposures from smoking). We know from toxicology that nicotine itself apparently affects some pregnancy outcomes. We have little doubt that the other exposures from smoking that are absent when using the other products (particulate matter, carbon monoxide, etc.) affect the fetus and the mother. And, thanks to the billions of dollars that have been squandered besieging every possible question about the effects of smoking, we know that smoking causes an approximately X% change in the risk of Y, or an average X change in the measure of Y, for the various outcomes Y that we looked at.
From this knowledge we can infer, with near certainty, that any smoke-free product causes somewhere between zero and X for the various Y. It would take some extraordinary epidemiology evidence to dissuade us from this prior belief. So what we would ask of the epidemiology is to tells us where within [0,X] each of these measures lies.
The verdict is, without the slightest doubt, that the available epidemiology is grossly inadequate to do that. It is so limited, fatally flawed, and uncertain that it cannot help us refine our estimates. So after reviewing the evidence, we are left with nothing but our prior, which is based on the chemistry of the products and toxicology:
Pregnant women and those who advise them would like to be able to assess the risks of using smoke-free nicotine and tobacco products, not just have vague notions of what is better and worse. “There is probably a cost” is not terribly useful for decisions that must be weighed against other costs and benefits. But that is where we are. The use of smoke-free nicotine products almost certainly has less effect than smoking on pregnancy outcomes (most of which are negative, but there are some positive effects), but any use of nicotine is probably worse for the fetus than none. This review reinforces both the validity of that advice and the fact that more precise advice cannot be offered.
You can see the details of all this in the paper. As I noted, we had to resist some pressure, but we were able to publish the paper we wanted to. So all this is covered there. We refused to yield to the pressure to highlight the specific estimates from the previous literature, let alone pretend that they were informative. We presented them as the trivia that they effectively are. In particular, we dismissed the reported results that would suggest that snus or NRT are more harmful than smoking (the claims to that effect about snus were, of course, published by anti-snus fanatics; the results about NRT were based on patently flawed methodology). Sometimes you should look at a research result that is contrary to prior beliefs and say “wow, perhaps I should doubt my prior beliefs.” In this case, though, the conclusion can only be, “there was something clearly flawed in that research.”
You can also find in the paper a very simple quantitative bias uncertainty analysis, of the type I pioneered 20 years ago and that made my name in epidemiology. I have not done much of that lately. I will try to do more. There is no excuse for everyone not doing it, given that my successors have created some simple online tools, like the one I used.
One thing that is particularly interesting about this analysis is a rare reversal of the usual roles of epidemiology and of toxicology, chemistry, and such. Normally the latter sciences can only give us a vague reality-check on the effects of an exposure on people. They can show that a realistic level of an exposure probably will not cause any wild unpredicted effect. They can tell us that massively high doses of the exposure produce particular effects so maybe(!) those effects occur in real life, and we should look for them. But these methods are completely incapable of offering a quantitative estimate of effects. That is what epidemiology does, by design.
However, in this case, with the quantitative anchoring that the research on smoking gives us, the lab sciences let use estimate a range of true values (the [0,X]) with a very high degree of confidence. Meanwhile, the epidemiology is only good enough to offer a reality check: None of the results were so wildly outside of this range that they give us reason to question our existing estimates. However, they are inadequate to do anything to refine the quantification that the lab sciences offer.