Lies, damn lies, and epidemiology

I owe the world a post on the scientific investigation of thimerosal and autism. Although I empathize with the world's painful anticipation, I must first prepare the way with a few remarks about epidemiology and public health science.

People complain all the time, "First they tell us that everything cause cancer, then it doesn't after all, then maybe again it does a little. They tell us that the orange and green pills are good for us, then they tell us they're killing us. They keep changing what we're supposed to eat, what tests we're supposed to get, how fat we should or should not be. Fuhgedaboudit, from now on I'm ignoring it all."

Actually it's not quite that bad. Not quite. After all, people used to think that the earth was at the center of the universe, and the sun and the planets and the stars went around us. Now we know better, but just because we used to believe otherwise doesn't mean there is any doubt that the earth goes around the sun. Science advances, and sometimes ideas are overthrown.

One reason epidemiology is different from cosmology, though, is that people act on current beliefs in ways that have a direct and immediate effect on our health and even our survival. Sometimes those beliefs contain a substantial degree of uncertainty, but decision makers feel compelled to go with their current best estimates.

Epidemiology also happens to present some particular difficulties. Only rarely are there ethical concerns about experiments in particle physics, energy fields, or gravitation (beyond the issues of appropriate allocation of scarce resources), but experimenting on human beings is definitely a thicket of ethical brambles. That means we often have to depend on observing relationships among variables such as toxic exposures and health outcomes under real world conditions that can't be controlled. Conclusions drawn from that sort of research are usually stated in terms of probabilities, but even those probabilities depend on the assumption that we haven't overlooked some factor that is associated with the supposed risk factor we are studying, that is the real cause of the outcome.

New drugs are generally approved only after experimental trials, but there are several difficulties here as well. One of the most important is that in an experiment, you set out to study certain pre-defined, hypothesized outcomes. It is easy to overlook adverse effects that you don't happen to be looking for, and even if you do notice certain adverse effects that the experiment wasn't designed to study, it can be unclear whether they really are caused by the drug. Perhaps they affect only a subset of all people, such as women or men, or people with other pre-existing conditions, or specific genetic endowments. Then they might affect too few of the study participants to be noticed at first, but become clear when more people take the drug, or even when the original data are re-analyzed. By the same token, associations that are noticed can turn out to be spurious.

Uncertainties can ultimately be resolved by new experiments specifically designed for the purpose. Overlooked associations can emerge clearly once we have much larger numbers to analyze. In the meantime, vested interests (such as, obviously, drug manufacturers) may be in a position to pick and choose from the available data in order to create a misleading impression of certainty or uncertainty, or may simply be able to bamboozle the public which has very limited access to the original information and no ability to interpret it.

Nevertheless, the truth is out there, and sometimes we find it. At least we become sufficiently sure of it that it's not worth investigating any further. Maybe we really are just software running on a superduperultracomputer and the whole universe, including the earth and the sun and the stars, doesn't even exist. But that possibliity is not worth worrying about, especially since there doesn't seem to be a whole lot we can about it anyway. The possibility that thimerosal has caused an epidemic of autism is in the same category.