Of Mice And Menace: Why Rodent Studies Of Vapor Products Are Meaningless
It seems like we are living in a a mad lib. Almost every week sees a new journal paper that reports effects of something associated with vapor products (pick one: e-liquid, proper vapor, dry-hit vapor, super-concentrated vapor, concentrated e-liquid ingredients) causing some outcome (cellular changes that might be a precursor to cancer, other cellular damage, change in blood chemistry, change in blood pressure or other vital signs, etc.) when administered to rodents (mice, rats, hamsters) in megadoses. These studies provide approximately zero information. That fact is widely understood among vaping advocates, though there seems to be some confusion about why the results are meaningless.
It is useful to first understand the few cases in which animal toxicology studies are sometimes useful (setting aside questions of whether it is ethical to experiment on animals even if the results are informative). Other animals can be used to get a rough idea of whether plausible doses of a chemical cause acute harm (toxicity) to humans. It is useful to know whether drinking 30 ml of e-liquid might be dangerous. Of course, there are other ways to figure that out, since we already have better estimates of the toxicity of the various ingredients and byproducts than any new study is going to provide. There is always a chance that a cocktail of chemicals will have dramatically different effects from the added-up effects of the individual chemicals, but that seems rather unlikely in this case.
Another informative use for animal experiments is to understand infectious diseases. Those involve complicated agents interacting with even more complicated mammalian bodies, and thus are usually beyond our technology to model any other way. This has little relevance to vapor products. It is, however, a good reminder that those complications mean that any results from animal experiments are, at best, extremely rough estimates of effects on people. Some infectious agents will sicken us but not our close biological relatives, and vice versa. Animal experiments might predict what might be a good vaccine or cure, but we do not know for sure until it is tried on humans. It turns out that those predictions are usually wrong.
Straying even further from having a good model — a simplification that captures the relevant aspects of a more complicated system — are attempts to predict whether an exposure causes cancer in humans, and how big that risk is. It is easy to understand the temptation to use animal models for this. Until we have good epidemiology data, with a lot of people having similar exposures for many decades (which we may never have for current vapor products), it is difficult to estimate the risk for cancer or other uncommon outcomes caused by long-term exposure. However, just because it is possible to use a particular model does not mean it is useful. As with statistical models in epidemiology, the model will produce a result, but there may be no reason to believe that result is at all accurate.
A common criticism of rodent experiments is that the administered doses of the exposure are enormously higher — typically by orders of magnitude — than real-world exposures. This is a problem, but for reasons more subtle than most critics seem to understand. If an acute toxicity experiment used hundreds of times a realistic dose and came back with “vapor is a deadly poison!”, then the simple criticism would be correct: The dose makes the poison, and that high dose being toxic does not mean that a realistic dose is.
With cancer, however, the whole megadose methodology is based on a theory known as LNT (linear, nonthreshold). Basically it says that each quantity of a possible carcinogen, divided by the weight of the body that is exposed to it, always causes the same probability of cancer. So if a 70 kg human is exposed to X per day for a year, the risk is the same if a .07 kg rat is exposed to X/1000 per day for a year. Moreover, the theory says that to model the risk for that human’s exposure for 50 years, you just need to multiply the rat’s dose by 50. Since no one wants to run the experiment for a full year, though, they cut it down to 2 months and increase the exposure by another factor of 6. Also, they do not want to use 100 rats to represent the risks for 100 people, so they increase the dose by another factor of 10 so each rat has the collective risk of 10 people. According to LNT, that 3000-fold overdose means that each rat, over the course of 2 months, has the same probability of getting cancer as at least one of 10 humans who vapes for 50 years.
If you find yourself thinking, “Are you kidding me?”, then you are showing a greater understanding of cancer risk than animal researchers. The estimates from that model are incorrect not because the dose is unrealistically high (the simplistic criticism) but because the LNT model for converting the effects of that high dose is known to be badly wrong. Interested readers can find a nice and very readable recent paper about that here. To summarize, the risk from cancer (and even more so, for other diseases) is not linear across the dosage, the time over which that dose is accumulated, or body weight. Also there is apparently usually or always a threshold (i.e., until you get up to some minimum dosage level, there is no risk), models might produce some positive estimate of cancer risk when there is actually zero risk. In addition there is the problem that mice are not just little humans. Carcinogenicity in mice is not even a very good predictor of carcinogenicity in rats, let alone in humans. Again, biology is complicated.
This is still not the most absurd use of animal models. As noted, it would be useful to have some way to substitute for hundreds of thousands of person-years of nonexistent epidemiological data. Merely wishing there was a way to create such a model does not, of course, mean that a convenient model provides useful information. But sometimes there is no reason to even wish and guess at the right model. If the question is, “What is the acute effect of vaping on a particular measurable outcome?”, then it is simply insane to run an animal model. The latest mice-and-menace study is an example of that. The measured endpoints were changes in blood biochemistry, which could be measured in vapers to show the effects of the real human exposure. There is simply no reason to perform a megadose experiment on mice. The observed outcomes might or might not occur at all at realistic doses or in humans, and there is a perfectly good way to find out if they do.
To summarize, animal toxicology models are useful for a few applications but they have almost no useful application to vaping exposures. Minor acute effects can be measured in short-term studies of actual human vapors, and if the effects in question only happen with a megadose (e.g., inducing a heart attack) then the results are useless. Whatever is “learned” from megadose studies of acute effect is not based on any theory that says, “for this particular outcome, if we see a change of quantity X in mice exposed to 30 times the realistic dose, then that means the average person experiences risk Y when exposed to a real-world dose.” There is no such theory. In the one case where there is such a theory — LNT for cancer risk — the theory is known to be wrong.
It is possible that some rodent study of vapor product exposures would provide useful health risk information. But none that I have seen do so, nor had any chance of ever doing so, whatever the results. One has to wonder what kind of person chooses to torture animals for no apparent purpose beyond getting more grant money, only to provide the world with results that in the end are actually meaningless.