Failure To Quit Smoking Causes Vaping, Not Vice Versa
In the 1960s, epidemiologists created lists of factors to consider when trying to infer if an observed association between an exposure and outcome is causal, as opposed to being a product of confounding. There was no magic to these lists; they were basically just scientific common sense, sketching out for the young science some of the thought process that proper scientists use in the inference process. (This way of thinking is completely antiquated; any time you see someone “apply” these considerations today, especially if they incorrectly call them “causal criteria,” you can be sure you are reading the work of someone who does not understand scientific inference.) The lists included only one genuinely necessary condition, temporality. For something to be a cause, it must occur before the outcome. This was usually seen as a throw-away, since no one is ever stupid enough to claim that an outcome was caused by an exposure that occurred later. Authors who presented these considerations struggled to find any example of someone making this error. It would have been easy, though, if they could have cited recent anti-vaping research.
One example was previously reported, in which Stanton Glantz and colleagues used European survey data to compare former smokers and current smokers, looking at whether they ever vaped. The data showed that current smokers were more likely to have vaped than former smokers. From this the authors concluded — in what would have been an obvious parody of doing bad science if it were not coming from tobacco controllers — that vaping caused smokers to keep smoking rather than quit. Among the several flaws in this reasoning was a failure to consider temporality: Most of the former smokers quit years before vaping ever happened (and most of them were, presumably, in a content equilibrium, with no desire to find a substitute product), while the current smokers obviously did not quit already. In other words, the outcome (quitting smoking or not) mostly occurred before the supposed cause (vaping).
Taking this one step further, it is easy to see that the association in the data is partially confounding and partially causal in the opposite direction. It was not that vaping caused people to fail to quit smoking, as the authors claimed. Rather, it is clear that not having quit smoking already caused some subjects to try vaping.
The latest example of a failure to understand temporality is an article out of Massachusetts General Hospital that followed discharged hospital patients who indicated they wanted to quit smoking. The researchers recorded whether the subjects were vaping one month and three months after discharge, and whether they were still smoking six months after discharge. They found that the quarter of the subjects who had tried vaping in the first three months after discharge (as part of their effort to quit smoking) were more likely to still be smoking at six months than those who did not try vaping. Based on this, they issued a press release that claimed “e-cigarettes hampered smoking cessation.”
As with any observational study of these behavioral choices there is the problem of confounding by propensity. That is, subjects who get more benefits from using tobacco products are more likely to seek substitutes (because just quitting entirely is not an appealing option) and are, for the same reason, less likely to quit. To their credit, the researchers made some attempt to control for this, though it was inadequate, and conceded in their paper that it might be a problem. That did not stop them from issuing a press release that pretended that there was no such problem.
But the real problem here is a temporality error unicorn, the rare case when researchers fail to realize that the “effect” is preceding the “cause.” Presumably many of these hospitalized smokers who quit did so immediately after they were discharged, or did not smoke when they were hospitalized and never started again. Hospitalization is an interruption in routine, just what is needed as a focal event for smokers who have been intending to quit, like these subjects, but had just not found the right day for it. Moreover, depending on what they were hospitalized for, it can be a strong boost to motivation. Subjects who just walked away from smoking immediately (and thus were not smoking six months out, when it was measured) were very unlikely to have tried vaping at three months. The supposed effect occurred before those supposed cause. For those who did not just walk away, it was their continued smoking after they left the hospital that caused them to try vaping.
These study methods seem almost designed to create a temporality error. If the researchers had measured smoking abstinence before first measuring vaping, the data would have made the problem clear. It presumably would have showed that those who quit smoking mostly did so immediately, before they could have vaped post-hospitalization. This error may not have been intentional. Medical researchers are notoriously bad at doing epidemiology studies that are more complicated than a simple randomized trial. Indeed, most of the time those antiquated causal considerations are recited today are attempts to prevent medical researchers from making stupid mistakes. But that does not excuse them from signing off on a press release that included an absurd conclusion that they did not actually endorse in their paper. It certainly does not excuse population researchers who must know better (unsurprisingly, this would include Glantz again) from intentionally ignoring the error.
It is truly a sad state of affairs when it is necessary to explain that cause must precede effect. But anti-vaping claims are just that bad.
Follow Dr. Phillips on Twitter