The logic is simple. Low levels of high-density lipoprotein (HDL) cholesterol are a risk factor for coronary disease. Torcetrapib raises HDL levels; therefore, it should prevent coronary disease. However, Pfizer Inc. recently withdrew its licensing application for torcetrapib after phase 3 trials showed that there was an increase in cardiovascular events and overall mortality in patients taking the drug in combination with atorvastatin (Lipitor) compared with those taking atorvastatin alone (although the increase was not statistically significant).1 Moreover, a study of patients with familial hypercholesterolemia showed that adding torcetrapib to atorvastatin resulted in a large increase in HDL cholesterol levels and a large decrease in low-density lipoprotein cholesterol and tri-glyceride levels, but was paradoxically associated with progression of atherosclerotic disease.2 What are the lessons for practicing physicians, patients, and the U.S. Food and Drug Administration (FDA)?
The short answer is that modifying risk factors does not necessarily modify risk, but the longer answer is more interesting and more instructive. Risk factor intervention may fail because the science is simply wrong. Not all risk factors are causal. Risk factors are simply statistical associations; causation requires different proof.3 For example, genital herpes is a risk factor for cervical cancer. However, it does not cause cervical cancer. Genital herpes may be a marker of multiple sex partners and, therefore, a marker of human papillomavirus, which is the causal factor.4 Although genital herpes is a risk factor for cervical cancer, treating the infection will not prevent cervical cancer. However, efforts aimed at preventing genital herpes (e.g., abstinence programs, condom use) may also prevent cervical cancer. In other words, even if the science is wrong, the intervention might be effective.
It is never acceptable to simply advocate altering risk factors; the intervention used to alter the risk factor and the population to which the intervention is applied must always be specified. For example, there is good evidence that lowering cholesterol levels with statins is beneficial in high-risk persons,5 but this does not translate to a general statement that lowering cholesterol with other drugs (e.g., ezetimibe [Zetia], fenofibrate [Tricor]) is beneficial, or that lowering cholesterol levels in low-risk populations is beneficial. Because our knowledge is almost always incomplete, we can anticipate that many interventions will have unknown harmful effects that counteract their beneficial effects (but occasionally will have unknown beneficial effects as well). This is a particular problem with new drugs in which there is no information available about long-term side effects.
Another reason that interventions might not work is because biology is generally nonlinear. For example, a blood pressure of 110/65 mm Hg does not significantly increase the risk of stroke compared with a blood pressure of 100/55 mm Hg. However, an increase from 140/95 to 150/105 mm Hg increases the risk substantially. On the other hand, an increase from 280/165 to 290/175 mm Hg probably does not cause much extra harm. Even in the midrange, where an association appears relatively linear, risk factor reduction may not result in the benefit we might predict by looking at the risk factor curve. This is because the degree of nonlinearity going up the curve will probably not be identical to the degree of nonlinearity going down the curve. For example, an increase in blood pressure from 140/90 to 160/110 mm Hg increases the risk of myocardial infarction. However, lowering the blood pressure from 160/110 to 140/90 mm Hg does not completely reverse this risk.6 Predicting decreases in risk from decreases in risk factors is a task fraught with difficulty; this is even more true when predictions are extrapolated outside the near-linear, middle part of the curve.
So how do we overcome these problems? Perform randomized trials and do not extrapolate. If a trial was performed in men, do not assume that the results will apply to women. If a trial was performed in older persons, do not assume the results apply to younger persons. If a trial used chlorthalidone (Thalitone), do not assume doxazosin (Cardura) will give the same results.7
From a public health perspective, there are two lessons to be learned. First, I believe that the FDA should not license drugs based solely on their ability to alter risk factors, but should demand proof of clinically important benefits. Although the risks of torcetrapib were discovered before the drug was approved, we have not always been that lucky. Consider recent reports that rosiglitazone (Avandia) increases the risk of myocardial infarction and cardiovascular death.8 Far too many drugs have been licensed solely on their ability to alter risk factors without any evidence that they produce clinically important benefits. Remember: hypercholesterolemia and high blood glucose levels are not diseases in themselves; strokes and heart attacks are. I also think there should be better regulation of drug advertising to patients and doctors. So much of the information we are given is, if not simply false, at least misleading. For example, the FDA-approved patient information handout on ezetimibe conspicuously fails to mention that the drug has no benefits on the important clinical outcomes of morbidity (e.g., strokes, heart attacks), mortality, or quality of life.9
For all these reasons, torcetrapib can teach us a lot.