Thursday, September 20, 2007

Gary Taubes Discusses Epidemiology

Gary Taubes, a science journalist whose 2002 article in the New York Times Magazine, "What If It's All Been A Big Fat Lie" shined a spotlight on low-carbohydrate diets, has written another piece. This one is on epidemiology. It was also published in the New York Times Magazine this past Sunday.

The article is, in essence, a summary of some very basic tenets of the field of epidemiology - the study of the distribution of diseases in populations. I enjoyed it.

If you'd like a taste of what Epidemiology 101 is like, read Mr. Taubes article:

Do We Really Know What Makes Us Healthy?

If you'd rather hear Mr. Taubes tell you what it said, listen to an interview he did with Brian Lehrer of New York Public Radio the day after his article was published:

Weird Science

The site includes some terse listener comments.

Some Background On Epidemiological Studies

Large epidemiological studies (observational as opposed to an intervention), such as the Nurses Health Study (120,000 participants followed since the 1970s), can be used to generate hypotheses, e.g.:
Study 1 (Observational): In the Nurses Health Study, we found that women who took hormone replacement therapy (HRT) also had fewer heart attacks. We hypothesize that HRT can lower the risk for heart attacks.
What we derive from epidemiologic studies are associations and hypotheses. We don't derive cause and effect, so these associations cannot be applied to an individual. For that you need a follow-up clinical trial which tests the hypothesis, e.g.:
Study 2 (Clinical Trial): We gave 1000 women HRT, and another 1000 women a placebo. After 5 years we found that those taking HRT had more heart attacks than those taking placebo.
As you can see, the hypothesis generated from the observational study can be refuted in a clinical trial. Just because it was observed that women who took HRT in Study 1. had fewer heart attacks does not rule out other possible unobserved causes, such as ... are women who take HRT healthier? wealthier with better access to healthcare? These other possible causes, or confounders, may in fact be the true mechanism, not HRT.

To narrow down a mechanism for the observed association you need an intervention, like Study 2. But Study 2 has its own limitations. Note the smaller population size. Clinical trials are also expensive, usually shorter in duration, and come with their own set of confounders (only certain types of people sign up for clinical trials). They're also utterly uselss if you're trying to test a hypothesis that would require exposing one group to a strongly suspected pathogen or other dangerous therapy.

Bad Science?

Mr. Taubes goes into a lot more detail, with many examples. Alas, he takes a dimmer view of epidemiology than I do: "What if it's just bad science?"

Science is a tool. It can be useful, e.g. epidemiological studies first discovered the link between smoking and lung cancer. But it can also be abused, especially when used to advance the agenda of profit-seeking drug companies, reputation-protecting political entities, or fund-seeking academics. So, I can understand Mr. Taubes' cynicism. The media will grab the association from an observational study and run with it, implying that it has ramifications for an individual's health. (Even some study authors will do this.) That is not and never was the case. But you'll see it written up in an Associated Press article that way: "HRT found to protect against heart disease!" Millions of women will start, or continue to take costly hormones with the hope of protecting their heart only to find they've increased their risk for cancer.

Gary Taubes In Quotes

To close, I'll list three quotations from that Gary Taubes' WNYC interview that made me perk up:
Taubes: "Exercise makes you hungry. It does not make you lean."
That was interesting.
Lehrer: "Do you have a hypothesis [as to what causes breast cancer]?"
Taubes: "Yes, I do."
Lehrer: "Which is?"
Taubes: "Which is ... the refined carbohydrates and the sugars in the diet."
Now that's a hypothesis begging to be tested.
Lehrer: "You can't prove your hypothesis either, which is that carbohydrates cause metabolic syndrome."
Taubes: "You cannot prove any hypothesis true. All you can do is refute the hypotheses and see which ones you have left."
This is an elementary fact of science. You cannot prove something is true. You can only disprove, and then refine your hypotheses.

Note: Gary Taubes has a book due out in a few weeks: Good Calories, Bad Calories: Challenging the Conventional Wisdom on Diet, Weight Control and Disease.
Photo from WNYC's The Brian Lehrer Show

No comments: