Thursday, 16 June 2016

Selection gateways

Suppose you observed that kids who tried some risky-looking novel activity X were more likely to go on to try some other risky activity Y. Should you conclude that X leads to Y, or that kids who are risk-, sensation-, or novelty-seeking will exhibit that on more than one margin? Cohort selection effects matter: the ones who try X aren't a random sample from the population.

A few years ago, the Dunedin Longitudinal Survey found that early use of marijuana correlated with problems down the track; Ole Rogeberg argued, and further evidence later showed, that that was all selection effects. The kinds of kids who try marijuana early are not the same as other kids and would have different outcomes even if they didn't use marijuana.

Out this week in Pediatrics (supplementary materials here): a study showing that kids who use e-cigarettes are more likely to go on to try tobacco than kids who didn't use e-cigarettes as baseline.

Adjusting for gender, ethnicity, grade and highest parental education, those who'd used e-cigs at baseline were 6.2 times as likely to go on to try tobacco. I expect this will be painted as e-cigarettes being a gateway drug for tobacco. But kids who tried e-cigarettes at baseline will be different from other kids: even in the absence of any gateway effect, you would expect to find correlations between exhibiting novelty-seeking behaviour on one margin and exhibiting it later on another margin.

They note that controlling for number of friends who smoke, friends' attitudes toward smoking, and others in the household smoking had no appreciable effect on results, but that having tried cigars, pipes or hookah prior to initial evaluation reduced the odds ratio on e-cigarettes from 6.2 to 5.5. It would be interesting to see what would happen with other controls for baseline risk-seeking: age of initiation of sexual activity, any other alcohol or drug use, or even number of days' detention in the past year. It can be tough though to get additional controls through IRBs though. Some assessment of time preference could have been interesting too.

They do try to control for a measure of susceptibility to tobacco use. They asked the kids a series of questions like "At any time in the next year, do you think you will use these products?" or "If one of your best friends were to offer you these products would you use them?" A "definitely not" answer to 3 questions was taken to indicate that the person was not susceptible to smoking. Restricting the sample to those deemed susceptible reduced the odds ratio to 2.12.

Among those who'd answered "definitely not" to all three questions, those who used e-cigarettes at baseline were 9.7 times more likely to try cigarettes.

Again, you can tell two stories here.

You could say "Among those most emphatically saying that they wouldn't try cigarettes, those who tried e-cigarettes were at much greater risk of smoking".

Or, you could say "There's a strong social desirability bias around smoking; few people are willing to admit they might ever try cigarettes. 79 kids didn't completely rule out ever wanting to try cigarettes, and most of them used e-cigarettes at baseline. 216 kids gave emphatic no answers, and of those, far more of the e-cigarette users went on to try tobacco. If admitting you might try cigarettes is shameful, is it surprising that those kids who were more sensation-seeking to begin with, as demonstrated by using e-cigs at baseline, were more likely to smoke despite having said they never would?"

It would have been interesting to see whether a set of susceptibility questions around risky drinking would have revealed similar patterns between e-cigarette and non-users on follow-up.

Mike Siegel has a bit more to say on it here:
This study is virtually meaningless in terms of its evaluation of the “gateway” hypothesis.

Baseline e-cigarette use was defined as ever having taken even one puff of an e-cigarette. And smoking initiation was similarly defined as ever having taken even one puff of a cigarette. So the study did not document that even one subject in the study was evera regular vaper. It is entirely possible (and in fact likely) that the majority of these kids had experimented with e-cigarettes, failed to become vapers, and then turned to regular cigarettes. In fact, it’s entirely possible that had these kids been able to stick with vaping, they would never have become smokers.

In addition, the study counted anyone who had even puffed a cigarette as being a smoker. So theoretically, a subject could have had a single puff of an e-cigarette and hated it, and then had a single puff of a cigarette and hated it, and they would be considered someone who initiated smoking because of first becoming addicted to vaping.
However, they did find that e-cigarette users were 5.5 times as likely to have initiated smoking but not to have smoked in the past 30 days, and 7.5 times as likely to have reported cigarette use in the past 30-days. If regular smokers are more likely to report having smoked in the past 30 days, then it isn't just the single-puff problem driving things.

I'm more worried about underlying cohort heterogeneity that is far from adequately controlled by asking kids whether they think they might ever take up smoking.

No comments:

Post a Comment