Thursday, May 17, 2012
The World's Most Popular Fallacy
Or at least I think it probably qualifies. That would be conflating association with causation, and it's a plague (hah!) in public health research. Today's entry is the so-called "good cholesterol" hypothesis. Or at least, it should have been considered a hypothesis all this time, but instead people have treated it as a finding.
The linked article may be a bit esoteric. It has gotten some coverage in the lay media, but I'll offer my own summary. Most people have heard that "high cholesterol" is a risk factor for heart disease and strokes, but that there are actually two kinds, called Low Density Lipoprotein, LDL, and the high density HDL. There's a lot more LDL than HDL in your blood, so the LDL level determines most of your total cholesterol, but higher levels of HDL are actually associated, in observational studies, with reduced risk for heart disease.
Randomized controlled trials of statin drugs, which lower LDL, have shown that they do indeed reduce the risk of atherosclerosis, and consequent heart disease and ischemic stroke, at least for people who have already had heart attacks or are at high risk. (Their value in people at average or even just above average risk is controversial.) So that adds to the evidence that high LDL causes the bad outcomes. (I must caution that it doesn't actually prove it. Statins could simultaneously lower LDL and do something else that reduces risk. But it adds to the plausibility of a direct causal association.) So naturally, drug companies have been working on pills that will raise HDL in hopes of making billions.
I have always been skeptical. (Yeah, that's easy for me to say now, but it's true.) HDL is higher in people who engage in lots of aerobic exercise, for one thing, and it's lower in people who smoke. Observational studies try to isolate its effect by controlling for consequences of exercise -- leanness and lower blood pressure, slower heart rate -- exercise itself, smoking, and all that good stuff. But that's really hard to do. HDL could just be a marker that goes along for the ride with the true protective factors.
So these researchers whose work is published in The Lancet used certain genotypes as what's called an "instrumental variable" -- some characteristic that people happen to have that effectively randomizes them in a sort of natural experiment. Some people have genes that predispose them to higher HDL. It turns out that having these genotypes is not protective against heart disease. As a check on the concept, having genes that predispose to higher LDL is indeed associated with higher risk.
This cautionary tale is important for several reasons, but the most important practical lesson is that we need to be much more cautious about approving drugs based on so-called "surrogate end points." We need to prove that they do what we really want them to do, which is to make us healthier or keep us healthier, or at least make us feel better. Changing some technical indicator that we think has something to do with being healthier isn't good enough.