If anything is supposed to be certain in medicine, it’s that people with high cholesterol levels should be treated. But should they? Sifting through the underlying science reveals that the way in which scientists and drug companies describe the benefits of many medications—by framing the question in terms of “relative risks”—systematically inflates their value. The result is that patients frequently buy and consume medicines that do very little good. An alternative way of describing the benefits of medical therapy could help change that—if doctors and nurses would start using it.
Take cholesterol-lowering drugs. In 1995, the prestigious New England Journal of Medicine published a study strengthening the case that otherwise-healthy men with high cholesterol should take cholesterol-lowering drugs called statins. Researchers in Scotland reported a 31-percent reduction in the risk of heart attacks among men taking the statin pravastatin, sold by Bristol-Myers Squibb under the brand name Pravachol. Due in part to this study, Pravachol became one of Bristol-Myers’ most profitable drugs and now grosses more than $2 billion in sales per year.
A 31 percent reduction in heart attacks, after all, seems impressive. Yet this pervasive way of describing clinical trials in medical journals—focusing on the “relative risk,” in this case of heart attack—powerfully exaggerates the benefits of drugs and other invasive therapies. What, after all, does a 31 percent relative reduction in heart attacks mean? In the case of the 1995 study, it meant that taking Pravachol every day for five years reduced the incidence of heart attacks from 7.5 percent to 5.3 percent. This indeed means that there were 31 percent fewer heart attacks in patients taking the drug. But it also means that the “absolute risk” of a heart attack for any given person dropped by only 2.2 percentage points * (from 7.5 percent to 5.3 percent). The benefit of Pravachol can be summarized as a 31 percent relative reduction in heart attacks—or a 2.2 percent absolute reduction.
There’s another instructive way to consider the numbers. Suppose that 100 people with high cholesterol levels took statins. Of them, 93 wouldn’t have had heart attacks anyway. Five people have heart attacks despite taking Pravachol. Only the remaining two out of the original 100 avoided a heart attack by taking the daily pills. In the end, 100 people needed to be treated to avoid two heart attacks during the study period—so, the number of people who must get the treatment for a single person to benefit is 50. This is known as the “number needed to treat.”
Developed by epidemiologists in 1988, the NNT was heralded as a new and objective tool to help patients make informed decisions. It avoids the confusing distinction between “relative” and “absolute” reduction of risk. The NNT is intuitive: To a savvy, healthy person with high cholesterol that didn’t decrease with diet and exercise, a doctor could say, “A statin might help you, or it might not. Out of every 50 people who take them, one avoids getting a heart attack. On the other hand, that means 49 out of 50 people don’t get much benefit.”
But drug companies don’t want people thinking that way; whenever possible, they frame discussions of drugs in terms of relative risk reduction. That’s why the package insert for Pravachol highlights the 31 percent reduction and mentions the NNT not at all. In Pfizer’s 2005 press release promoting the Food and Drug Administration’s approval of Lipitor for patients with diabetes and other risk factors for heart disease, the company said the drug “reduced the relative risk of stroke by 26 percent compared to placebo.” In its 2002 press release promoting an anti-osteoperosis drug, Actonel, Aventis exulted that treated women were “75 percent less likely to experience a first vertebral fracture.” It’s standard for such promotions to make no reference to NNT and to bury information about absolute risks or leave it out entirely.
The reason is simple: Big numbers encourage people, even those who should know better, to prescribe drugs. In 1991, researchers performed a survey of faculty and students in epidemiology at Harvard Medical School—a group that should understand health statistics. When they were presented with identical information about a drug in different formats, almost half had a “stronger inclination to treat patients after reading of the relative change,” or risk reduction, as opposed to the NNT.
When a therapy is extremely effective—like surgery for acute appendicitis or insulin for juvenile diabetes—no one worries about NNTs. But most interventions aren’t home runs, and so NNTs are often the only way to tell if they may be worthwhile, medically and economically. Is your shoulder painful and stiff? The NNT for a cortisone shot is three, which is pretty good, but that also means two out of three patients won’t feel any better after the needles. Does your child have an ear infection? Your pediatrician obliges with a bottle of amoxicillin, but the NNT for antibiotics to shorten the duration of fever is more than 20; thus, at least 19 out of 20 parents force the stuff down their toddlers’ throats for no reason. Is your prostate enlarged? The NNT to avoid surgery is 18 if you take Proscar for four years. The drug costs $100 per month per person, so an insurer spends $86,400 to prevent a single surgery for enlarged prostate. Are you thinking of taking aspirin to help avoid a heart attack? The NNT is a lousy 208. Keep in mind that none of these figures include the risks of side effects.
In some cases, drug companies aren’t the only ones with an incentive to exaggerate a drug’s benefit. Consider statins again. Though an individual person with high cholesterol has little reason to take them (since 49 out of 50 get no benefit), when millions of at-risk people consume the drug, the numbers of averted heart attacks add up. Well-meaning public-health authorities may tolerate the exaggerations of relative risks because they scare people into taking drugs of very slight individual benefit, and if huge numbers of people comply, a few lives may be saved.
These kinds of not-entirely-honest messages about public health aren’t necessarily a problem, even when the NNT for a treatment is very high and thus the likelihood of individual benefit of treatment is very low—that is, as long as the touted intervention is cheap, painless, and accessible. Two examples: wearing a seatbelt and eating a healthy diet. But the calculation is different if we’re talking about an expensive drug. Or something people are supposed to do themselves that’s highly polemical, rife with guilt, and sometimes extremely stressful. Like breast-feeding.
In June, the New York Times ran an article headlined, “Breast-Feed or Else.” It suggested that experts believe that “breast-fed babies are at lower risk for sudden infant death syndrome and serious chronic diseases later in life, including asthma, diabetes, leukemia and some forms of lymphoma.” Yet, the article never mentions the NNT for breast-feeding to prevent these scary diseases. Neither does any general-interest press article in LexisNexis, a database. There’s a reason for this omission: The NNTs are astronomically high. Reasonable women might think that breast-feeding isn’t worth the trouble—a conclusion that you don’t want drawn if you’re promoting breast-feeding at any cost.
In the end, the excuse that it’s OK to promote largely ineffective drugs and interventions for the sake of the greater good doesn’t really wash. Nor does the excuse that NNTs are difficult to understand conceptually or that the math is too hard. Patients look to doctors to translate and interpret complex, often-conflicting information from drug companies, medical journals, and the media. NNTs are a tool for doing that. Doctors must keep faith that patients can get it together to understand their choices. If we make their use standard, patients and public discussion will follow suit.
Correction, Sept. 26, 2006: The sentence originally said that the “absolute risk” of a heart attack for a person taken Pravachol dropped by 2.2 percent, rather than 2.2 percentage points. (Return to the corrected sentence.)