Does brushing your teeth lower cardiovascular disease?

Yes, says a new article in the British Medical Journal:

Toothbrushing is associated with cardiovascular disease, even after adjustment for age, sex, socioeconomic group, smoking, visits to dentist, BMI, family history of cardiovascular disease, hypertension, and diagnosis of diabetes.

…participants who brushed their teeth less often had a 70% increased risk of a cardiovascular disease event in fully adjusted models.

The idea is that inflamed gums lead to certain chemicals or clot risks.

In the past five days I’ve seen this study reported in five newspapers, half a dozen radio news shows, and several blogs. These researchers know how to use a PR firm.

Sounds convincing. What could be wrong there?


I’m consistently blown away by what passes for causal analysis in medical journals.

Here’s a cruel and simplified guide to prevailing opinion and practice:

Medical research: Correlation never implies causation.

Epidemiological research: My correlation implies causation, because I controlled for socioeconomic status.

Economics: Confused and conflicted, but sometimes correlation implies causation. If you can find a cool instrumental variable we’ll surely publish you.

I side with the economists on this one.

As punishment, I suggest the people who ran (and published) the BMJ study be made to eat nothing but Krispy Kremes while brushing their teeth four times a day.

P.S. If interested, my causal inference syllabus is here.

32 thoughts on “Does brushing your teeth lower cardiovascular disease?

  1. Thank you for posting this! Even my husband, whose only training in research design is what he hears from me, knew what was wrong with this study as soon as we heard it reported! Oh news media …

  2. Did you bother to read the paper?!? They presented two models of confounding variables.
    Model 1 adjusts for: age, sex, socioeconomic group, smoking, physical activity, and visits to dentist.
    Model 2 additionally adjusts for: BMI, family history of cardiovascular disease, hypertension (diagnosed by doctor or clinic blood pressure >140/90 mm Hg), diabetes diagnosed by doctor.

    These confounds are a combination of basis socioeconomic factors and a good set of markers that should tease out the people who simply care for themselves better.

    The tooth brushing frequency relationship to CVD holds even when accounting for all those confounds. It’s possible there are other issues with this paper and perhaps they don’t use the best possible models for this confounds, but your pat critique of the paper is wrong.

  3. I had the same initial reaction after hearing about this, but I read through the BMJ article (the best I could with my non-medical background) and didn’t feel there was any assertion of a causal relationship.

    Also, from the conclusions: “Future experimental studies will be needed to confirm whether the observed association between oral health behaviour and cardiovascular disease is in fact causal or merely a risk marker.”

    I’ve also found some moderately interesting reading that seems to link the plaque on teeth to the plaque build-up in arteries but I can’t find them now… but maybe someone else has seen this?

  4. That’s even in the quote I list. The point is that these are inadequate controls; the key omitted variables (like eating habits, physical activity and fitness level, etc) are unaccounted for, and probably at least as important as the current controls. Plus their socioeconomic controls look pretty inadequate.

  5. You didn’t even mention that tooth-brushing is one of hundreds of x-variables that researchers have tested for correlation with heart disease, and that the 19/20 negative findings were never even submitted, much less published…

  6. So this seems like a pretty extensive list of controls for other proxies for how people take care of themselves:

    “…, smoking, visits to dentist, BMI…hypertension, and diagnosis of diabetes.”

    The only obvious thing missing from that list is exercise (given that BMI and diabetes are pretty good proxies for diet).

    Certainly seems like something to build on at the very least.

  7. So the comment above was written before the other responses and replies came in addressing the same points. Forgive the repetitiveness.

    The only point that hasn’t been addressed is whether or not BMI and diabetes are good proxies for diet. I would contend, when also controlled for some of the other factors, that they are as close as is practicable.

    On the other hand, I would point to this article from the WSJ this weekend as a far more egregious misuse of data and trampling upon quality research:

    In summary: “Look we found two kids from the same neighborhood who were friends as middle schoolers. The one who went to a charter school is going to college and the one who went to public school isn’t! Go charter schools!”

  8. I agree, and if this were the tone taken in the article, then it would be better. In any analysis, one should (1) discuss the unobservables and the limitations of the proxies, (2) discuss and possibly even model the (predictable) bias, and (3) make the limitations glaringly obvious in the summary bits.

    On the other hand, making grand causal claims in the international media… not such good practice.

  9. I’d also say we can do better than BMI and diabetes, such as questions about regularity and level of exercise, actual diet (even that day or week’s diet, resting heart rate.

    Here’s where a smaller sample with more in-depth information would be a better design.

    Even then the unobservables would be challenging. What one wants here is an RCT.

  10. I was totally with you on this…then my wife told me that her grandfather, who was incredibly fit and took care of himself, developed a cardiovascular condition which led to his death due to his rotten teeth. Not that one case proves it…but it is interesting!

  11. Chris,

    As an economist myself, I agree with all of your basic points regarding causal inference, though in fairness to the study, they do not make a causal claim. It seems like the problem lies in the media coverage. What I’m curious about is your opinion on the trade-offs. I was having a conversation about this with a friend who is a media professional, on hiatus to complete an MPH in hopes of bringing the two together. She made the following points: 1) There is a plausible mechanism through which this could work; 2) In her experience at the NPR science desk, the scientists themselves want to include details and caveats, but the public does not want to hear them, they want quick summaries, conclusions. Including details and caveats drives down reader/listener-ship, diminishing the ability to convey any information at all; and 3) The news stories likely made many people think about brushing their teeth, and CVD, arguably a public health success. On point 2: in some cases no information is better than misinformation; is this one of them? On points 2 and 3: in the event that a causal claim is going to influence behavior in such a way as to reallocate significant resources away from more efficient public health activities, much caution is essential. Is this one of those cases? What are the costs of people brushing their teeth more? To the extent that people additionally engage in other harmful practices, it might be a problem, but in fairness to the media, they just told people to brush their teeth more, not that if they did they could use that as an excuse to also smoke more or eat more salty foods.

  12. Reminds me of my Epi professor telling the class that regular shaving (or beard-trimming) has often been found to be associated with a longer life. He was careful to say it wasn’t causal, though: it’s not necessarily an Epi thing to overlook endogeneity!

  13. I’ve found over the years that hanging out with epidemiologists has been enormously valuable to my own work as an empirical social scientist. At our worst, we are obsessed with causal identification, and have absolutely no respect for mechanism. At their worst, they are the other way around.

    The periodontal disease -> inflammation -> epithelial dysfunction -> cardiovascular disease chain involves some pretty fascinating biology, all of which has been much better tied down in animal models than in statistical ones. The empirical finding here is striking, because it suggests that all those animal models *might* be relevant in people.

    Now, let’s all get down to the hard work of identifying a source of exogenous variation in periodontal disease (or, at least, chronic inflammation). It’s true there are epi people who still roll their eyes at me when I propose ideas to work this last bit out. I find that infuriating.

    In fact, I find it just about as infuriating as social scientists who put some biomarker on the RHS of a regression without the first clue what it measures or where it comes from, and then roll their eyes when you start thinking through the relevant biological mechanisms.

    Corollary to all this being that yet again, there is a Great Division, and yet again, I find myself deeply disagreeing with both “sides.” Sigh.

  14. Sorry, quick-but-naive-question: couldn’t a lot of the OVB been mopped up with an individual-level fixed effect? The researchers *must* have known that there were a large number of biographical, medical, etc. variables for which they couldn’t control- surely coefficients from FE estimates should have been included somewhere? (Caveat: I haven’t read the paper).

  15. Your article reflects the issue people are concerned about. The article provides timely information that reflects multi-dimensional views from multiple perspectives. I look forward to reading quality articles that contain timely information from you.

  16. It’s fun to read the debate about something unusual like this, wherein where cross-polls create something new from the little things that happen every day.