Sunday Science Lesson: association vs. causation + knowing the difference between good news and bad

by Carl V Phillips

This article in JAMA, which shows an association between smoking and a cancer-causing HPV infection (oral infection with HPV-16), is not really about THR. So I will post it under my broader and underused “Sunday Science Lesson” series. There is a minor bit of anti-THR lying in it: They bundle in the smokeless tobacco users with the (10 times as many) smokers in their analyses, pretending that those exposures are similar. As a result, they refer to what is really smoking as “tobacco use” (and even worse, “nicotine use”) in the title, press release, and body of the article. But that is not my point today.

The research letter (i.e., short paper), by Carole Fakhry (Johns Hopkins med school), Maura L. Gillison (Ohio State), and Gypsyamber D’Souza (Johns Hopkins School of Public Health), shows a correlation between tobacco use (again, really smoking) and oral HPV-16. The headline result is that current users had a 2.0% prevalence, compared to 0.6% for others. This correlation was actually already known, and the authors’ excuse for publishing it again was to add some biomarker bells-and-whistles which really do not change anything. The details of the article are not worth breaking the paywall for, or even reading (you can see the points of interest in the above links if you want).

As an aside, it is interesting to note that the authors found that 29% of the respondents in the nationally-representative U.S. survey they used were current tobacco users. This is much higher than the statistics that ANTZ like to use to support their claims that tobacco product use is disappearing, their usual message. While this study was not designed to measure that prevalence, it is still interesting that they do not even find the discrepancy notable.

So what are we to make of the observed association? It would be silly to suggest that smoking triples HPV-16 infection risk. There are obvious lifestyle differences that suggest a huge effect from confounding. In particular, they found that current users had a history of much more oral sex, a particularly good way to catch oral HPV. Does anyone think this means that smoking causes one to have oral sex? Yet the authors still say, “These findings highlight the need to evaluate the role of tobacco in the natural history of oral HPV-16 infection and progression to malignancy.” (If the relationship were causal, the takeaway message would actually be, “Hey, kids, disappointed that you are not getting enough oral action? Take up smoking!”)

Since the association is clearly caused either mostly or entirely by confounding, and there is little hope of sorting out whether there actually is any causation in sight (other than oral sex causing HPV), what was the point of this exercise? I think it is safe to say that it was part of the “hype the dangers of tobacco use” agenda. In their discussion, they emphasize a few reasons that perhaps smoking (notably not tobacco or nicotine use, but smoking) increases risks associated with HPV. The mere fact that the authors put out a press release suggests they are trying to trick people into believing that this rather uninteresting association is causal.

They could have actually tested the causal hypothesis more than they did. They did some multivariate analysis, but it was not designed to address the obvious confounding. They could have looked at whether the relationship held when they stratified by intensity of sexual history, or across cohorts (which would show if it had been true when older smokers were in their prime non-monogamous days before, we think, HPV-16 became so prevalent), or done any number of other things to assess what was going on. That is what they would have done if they had been scientists trying to better understand the world, but “public health” researchers are not known for trying to act like real scientists.

But beyond all that, there is an interesting lesson buried in this. Let us pretend, for the sake of argument, that smoking is causing there to be more HPV infection. Would that mean that smoking is worse for you than previously believed?

No. We estimate how harmful smoking is based on observed outcomes, not individual pathways. If one of the reasons that smoking is a leading cause of head and neck cancers were because it caused HPV infection, then that would already be part of the bottom line. You might recall the hype a few years ago, when it was hypothesized that the radioactive decay of polonium 210 in cigarette smoke within the lungs was a cause of lung cancer. (Po210 is naturally occurring in soil, and thus in plants, and thus in the smoke when you burn those plants.) If this were true, would it mean that smoking was causing more lung cancer than originally believed? No, because we estimate the lung cancer resulting from smoking based on the total outcomes, and if one of the causes of those outcomes is (and always was) Po210, it is already in those statistics. I am also reminded of the hype around the discovery of genes that make women far more likely to get breast cancer — same basic story.

The thing about all such discoveries — the one that is clearly true (breast cancer genes), the one that is plausible though probably not all that important (Po210), and the one that is a hypothetical claim for present purposes (HPV) — is that they are good news, not bad. It is only the “public health” tendency to try to spin everything as bad news that makes it seem otherwise.

Finding out that population breast cancer risk is concentrated in particular women is, of course, bad news for them, but it is good news for everyone else. That bit is zero-sum; the total incidence is what it is, so portioning out the risk is a wash. But if the risk is concentrated, those women can pursue intensive screening, or even go so far as to have preemptive mastectomies if they so choose, reducing their risks, whereas someone at average risk should not do so. Thus, total population risk can be reduced thanks to the discovery. It would be even better for reducing total risk if we could know exactly who was destined to get breast cancer if aggressive action were not taken to prevent it. Again, while a really bad revelation for those individuals, the knowledge would reduce the number of cases.

Similarly, if Po210 is really causing a measurable portion of smokers’ lung cancers, its concentration in tobacco leaf could be reduced, lowering risk. If causing HPV were really part of what made smoking cause head and neck cancers, then the push to vaccinate against HPV will reduce the risks from smoking for the next generation (though not by nearly as much as encouraging switching to smoke-free alternatives, of course).

Discoveries like this represent good true public health: Figure out a way to intervene to reduce a risk without doing anything that makes the public worse off (like trying to re-engineer humanity to behave puritanically). The widespread tendency to react exactly backwards is, I think, symptomatic of a few of the problems I have discussed about “public health” here lately. The preference for lecturing and pontificating about sins means that “public health” people do not actually want to find technical solutions. Their war on e-cigarettes is a perfect example of that. Their autocratic medicalized way of thinking means that finding systematic solutions — something that should be at the core of public health — is actually dispreferred in favor of wanting to “fix” individuals and impose punishment on those who cause risks.

However, applying the analysis to the most common topic of this blog, e-cigarettes, yields very different implications. Unlike with smoking, where we basically know the total toll (setting aside the efforts to exaggerate it), any discovery about a potential pathway of harm from e-cigarettes is bad news. There have been no discoveries to date of anything that poses an apparent measurable hazard (other than nicotine itself, which is widely believed to be slightly harmful on net, though this might not be true). But if there were such a discovery, it would not be like the case of smoking, where it would just represent an explanation for some of the known problem. Since we have no estimate of how much total harm e-cigarettes cause, a discovery of an apparently harmful pathway would increase our estimate of the total harm.

Of course, discovering it would be a good thing in a rational world, because then we can try to reduce it. Unfortunately, in the real world the discovery would mostly be used to convince smokers not to switch, so the results could be negative. (This would not be the fault of the fact of the risk or those who discovered it, of course, but of those who used it as a basis for their lies. Make no mistake: I am not falling into unethical “think of the chiiildren” behavior and suggesting that true and useful information be hidden because it might hurt someone. I am all for discovering everything we can.)

The application to smokeless tobacco is mostly like the smoking examples. Since we have ample epidemiology, finding an apparent pathway does not cause us to believe there is a measurable risk. The epidemiology fails to show any risk for any disease — more precisely, nothing that is measurable given the limits of detection of epidemiology — and finding a possible pathway does not change that. So, for example, all the research on TSNAs in smokeless products can basically be interpreted as answering the question, “if, contrary to the evidence, smokeless tobacco did cause a measurable risk of cancer, what might have been the causal pathway for that?” Kind of an odd question.

On the other hand, since the limits of detection of epidemiology mean that we can never know whether there was a single case of disease caused by an exposure, nor even (barring the creation of larger and better datasets) whether there is a 10% increase in risk for a particular cancer, such findings could be slightly good news from a real public health perspective: They cannot tell us how we might reduce a substantial or even measurable hazard (because there is no such), but it could tell us “it is possible that this pathway might pose some very low-level hazard, so we might as well do something about it if we can.” And, indeed, smokeless tobacco manufacturers have dramatically reduced TSNA concentrations as a result of that possibility. That is basically the same as the “might as well reduce the Po210 if it is easy to do — it sure couldn’t hurt” message about cigarettes.

In short, “public health” people do not seem to understand the difference between good news and bad news. I am reminded of a story from years ago, in which a friend of mine — one of the smartest epidemiologists I have known (the only person who has ever written a journal review of one of my papers that fundamentally improved the paper) — was laughing at a study where the authors openly lamented about failing to find a link between exposure to gasoline fumes and bladder cancer. The authors lamented because their “public health” bias was that they hope to find things to complain about in the world. My friend laughed at them because we should obviously prefer to live in a world where this common exposure does not cause cancer, rather than one where it does.

But the double-take here is that it would actually be better if that exposure caused bladder cancer, and if it caused a lot of it. That is because we are in a world where we already know how much bladder cancer there is, and that discovery would not create more cases. But if gasoline fumes did cause of a lot of cases, we could dramatically reduce that exposure and thus the disease incidence.

My friend had a seemingly compelling point, and I enjoyed the joke at the time. It was only in retrospect that I realized that the original authors’ lament was right, though almost certainly for the wrong reason.

Advertisements

4 responses to “Sunday Science Lesson: association vs. causation + knowing the difference between good news and bad

  1. Wonderfully done as usual Carl!

    Slightly off from your mainstream, three of your minor observations spurred some thought/reaction from me:

    1) “They bundle in the smokeless tobacco users with the (10 times as many) smokers in their analyses, pretending that those exposures are similar. ”
    Yes, similar to the game played by Klein et al in bundling in the far more numerous restaurant workers (I believe that may have been 5 to 10 times as many as well!) with the bar workers. That bundling of two separate sets of data allowed them to make the “accurate” statement that “Bar AND Restaurant employment is not significantly hurt by smoking bans.”
    The word “and” of course only worked in the combinatorial sense: bars on their own had their employment decimated in the years after the ban, but the much smaller dip in the much larger pool of restaurant employees hid that decimation nicely. It all depends on how you define “and” — sort of like Bill Clinton defining “is.”

    2) “As an aside, it is interesting to note that the authors found that 29% of the respondents in the nationally-representative U.S. survey they used were current tobacco users. This is much higher than the statistics that ANTZ like to use to support their claims that tobacco product use is disappearing, their usual message. While this study was not designed to measure that prevalence, it is still interesting that they do not even find the discrepancy notable.”
    LOL! And THIS reminds me of the embarrassing release by one of the state antismoking groups (Minnesota? Michigan?) of a table showing how the air quality had improved in bars after the ban. Why was it embarrassing? For precisely the same sort of reason: the table data ALSO had a column that inadvertently showed customer counts — and the customer counts after the ban were a FRACTION of what they’d been beforehand… totally counter to the Antis’ claims about how business hadn’t been hurt!

    3) “a friend of mine … was laughing at a study where the authors openly lamented about failing to find a link between exposure to gasoline fumes and bladder cancer. The authors lamented because their “public health” bias was that they hope to find things to complain about in the world.”
    This sort of lamenting when a relationship is not found is almost universal in secondhand smoke studies that come up with such results. What’s even funnier though is when you have situations like the 1998 WHO Boffetta study where they DID make one truly significant finding, i.e., that childhood exposure to secondary smoke significantly correlated with a REDUCTION in adult lung cancer among matched groups. While that’s what the numbers said, the authors described this one significant finding of their research as having found “no association”!!! See the bottom of http://bit.ly/ETSTable for the full reference.

    • Carl V Phillips

      I will accept that further science lessons are not off-topic for a Science Lesson post, even if tangential :-).

      I have added “conjunction” to the tags for this post, which is my abbreviation for the common disinformation tactic of bundling together tobacco products. You cite another great example of that one.

      It is true that the lament is almost universal. It is certainly politically motivated some of the time — getting a politically dispreferred answer. But not always. I suspect that the researchers in that story had no animosity toward gas stations. They were just disappointed that they failed to discover that something was bad. And, yeah, the game of dismissing something as “no association was found”, when in fact the “wrong” association was found, is a huge problem. It is part of the “publication bias in situ” problem I have written about extensively. It is very difficult to make sense of the literature as a whole when all the results that randomly skewed in the “right” direction get published while the ones that were randomly too far in the wrong direction are suppressed.

      I am reminded of a PhD dissertation I was at the periphery of recently. The student (who was actually already clinical faculty or some such) hypothesized that PAHs were causing a particular mysterious disease. But there was nothing that pointed that way particularly, and there was a striking lack of association with smoking, the dominant PAH exposure source. She kept dancing around that “no evidence was found…” claim. I asked why she did not just say “there is no association with smoking, so it cannot be PAHs”. She had no answer for that, but clung to the hypothesis as if it were supported. I do not know whether her advisors eventually forced her to admit that. Typical “public health” behavior that you just would not see in real sciences.

  2. “Does anyone think this means that smoking causes one to have oral sex?” As a tobacco smoker for 45 years, I can safely say – I certainly don’t think so.

  3. The JAMA link that you provided had this link in the right hand sidebar:

    http://jama.jamanetwork.com/article.aspx?articleid=208418

    It seems to have been a well-conducted study into the possibility that a HPV vaccine might help ‘clearance’ of HPV infection. It appears that not. The reason that I mention it is that, included in the study, was a ‘test’ for smoking relevance. I don’t know why, but no relevance was found, which is not surprising since the study concerned mostly sexually transmitted HPV.

    Reading that study led me to try to find out more about HPV. Wikipedia was helpful. As I read it, there are over 150 types of HPV, some of which are carcinogenic. They seem to be inherent in and on the human body. Thus, a person who has an infection on the soles of his feet can leave ‘spores’ on a carpet by walking barefoot on the carpet. Those ‘spores’ can infect another person, but only of those ‘spores’ can get inside the skin via ‘abrasions’, but these ‘abrasions’ need only be ‘skin-deep’. That is, they need only to expose the basal cells of the skin. Thus, sexually transmitted HPVs can result from, shall we say, ‘vigorous’ sexual activity.

    ——-

    The point of the above relates to the study which you talked about in that the study is ridiculously simplistic. Was it conducted by students for a particular purpose connected to Tobacco Control? It shows nothing other than that people who engage in oral sex are more likely to be smokers than non-smokers. It does not show that there is any causal relationship at all, since the invasion of the virus does not depend upon the smoke but upon ‘abrasions’ in the mouth.
    The real question that needs to be investigated is whether or not tobacco smoke in the mouth causes ‘abrasions’. That question also applies to ST in all its forms, and to e-cig vapour.
    If it (they) does not, then something else must cause the ‘abrasions’. I suppose that excessive consumption of alcohol, alone or combined with tobacco smoke and anything else, could cause ‘abrasions’, and thus enable HPV to gain access to the basal cells.

    On and on it goes.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s