• How to Measure a Medical Treatment’s Potential for Harm

    The following is co-authored by Aaron Carroll and Austin Frakt. It originally appeared on The Upshot (copyright 2015, The New York Times Company). Click over to that version of the post to see the accompanying charts.

    As we wrote last week, many fewer people benefit from medical therapies than we tend to think. This fact is quantified in a therapy’s Number Needed to Treat, or N.N.T., which tells you the number of people who would need to receive a medical therapy in order for one person to benefit. N.N.T.s well above 10 or even 100 are common. But knowing the potential for benefit is not enough. We must also consider potential harms.

    Not every person who takes a medication will suffer a side effect, just as not every person will see a benefit. This fact can be expressed by Number Needed to Harm (N.N.H.), which is the flip side of N.N.T.

    For instance, the N.N.T. for aspirin to prevent one additional heart attack over two years is 2,000. Even though this means that you have less than a 0.1 percent chance of seeing a benefit, you might think it’s worth it. After all, it’s just an aspirin. What harm could it do?

    But aspirin can cause a number of problems, including increasing the chance of bleeding in the head or gastrointestinal tract. Not everyone who takes aspirin will bleed. Moreover, some people will bleed whether or not they take aspirin.

    Aspirin’s N.N.H. for such major bleeding events is 3,333. For every 3,333 people, just over two on average will have a major bleeding event, whether they take aspirin or not. About 3,330 will have no bleed regardless of what they do. But for every 3,333 people who take aspirin for two years, one additional person will have a major bleeding event. That’s an expression of the risk of aspirin, complementing the fact that one out of 2,000 will avoid a heart attack.

    Granted, one out of 3,333 is a pretty tiny risk. But remember that the chance of benefit is pretty small, too.

    Sometimes, though, the N.N.H. can be much lower, even lower than that of N.N.T., which suggests the chance of harm is greater than the potential benefit. Consider screening mammograms, which are considered so essential that they are the only screening tests specifically mentioned in the Affordable Care Act, and coverage for them with no cost sharing is required by the law.

    If you look at the data for all randomized controlled trials of breast cancer screening, the N.N.T. for recommending screening to prevent one death from breast cancer after 13 years of follow-up is 1,477. But further analyses show that the one woman would have probably died of other causes anyway. There may be no benefit at all with respect to preventing death from all causes.

    Screening with mammograms can cause harm, though. They lead to overdiagnosis, encouraging the provision of therapies that provide no benefits — but do carry risks, and therefore are considered harms.

    If we look at those same studies, for every 333 women who are assigned to have a screening mammogram, one extra will undergo a lumpectomy or mastectomy as a result. One in every 390 women assigned to have a screening mammogram will undergo an extra course of radiation therapy as a result. (In these randomized controlled trials, patients are either assigned to get screening mammograms or they are not. The study then usually looks at the outcome for all who were assigned to get the mammogram, whether they actually did or not.)

    In other words, for about every 1,500 women assigned to get screening for 10 years, one might be spared a death from breast cancer (though she’d most likely die of some other cause). But about five more women would undergo surgery and about four more would undergo radiation, both of which can have dangerous, even life-threatening, side effects.

    Thus, N.N.H., paired with N.N.T., can be very useful in discussing the relative potential benefits and harms of treatments. As another example, let’s consider antibiotics for ear infections in children. There are many reasons that parents and pediatricians might consider treatment. One commonly cited reason is that we want to prevent serious complication from untreated infections. Unfortunately, antibiotics don’t do that, and the N.N.T. is effectively infinite. Antibiotics also won’t reduce pain within 24 hours. Antibiotics have, however, been shown to reduce pain within two to seven days. Not all children will see that benefit, though. The N.N.T. is about 20 for that outcome.

    Antibiotics can cause side effects, however, including vomiting, diarrhea or a bad rash. The N.N.H. for side effects in this population is 14.

    This means that when a child is prescribed antibiotics for an ear infection, it’s more likely that he will develop vomiting, diarrhea or a rash than get a benefit. When patients are presented with treatment options in this manner, they are sometimes more likely to agree to watchful waiting to see if the ear infection resolves on its own. For most children with ear infections, observation with close follow-up is recommended by the American Academy of Pediatrics.

    A wealth of N.N.T. and N.N.H. data based on clinical trials is available on a website developed by David Newman, a director of clinical research at Icahn School of Medicine at Mount Sinai hospital, and Graham Walker, an assistant clinical professor at the University of California, San Francisco. But it’s important to understand that results from clinical trials do not always reflect what happens in the real world. As criteria for treatment become more permissive beyond those applied in trials, the N.N.T.s can go up. But importantly, N.N.H.s often do not. Healthier people are less likely to see a benefit from antibiotics or an aspirin. But they are not less likely to have a side effect or complication.

    This is because the harms associated with treatment usually have nothing to do with the underlying illness. They are caused by the therapy, regardless of the reason for use. Children will develop diarrhea, vomiting or rashes from antibiotics in the same relative amounts no matter why we are using them. Put another way, clinical trials are designed to target the class of patients that most likely benefits from treatment, but they are not targeted to those more or less likely to experience harm. When treatments are applied in real-world clinical settings, we generally don’t see changes in the proportion of patients harmed by them relative to trials.

    When we stray from recommendations for therapies, and broaden the population given studied treatments, the N.N.T.s often go up, but the N.N.H.s stay the same. Things are often even worse than the data in studies make them look. Fewer people benefit, but just as many are harmed.

    We hope that every therapy has a benefit. The N.N.T. shows us that benefits are often much less likely than many might think. The N.N.H. can show us how likely we are to have a harm compared with a benefit. Considering both, especially in light of how practice often differs from studies, can help us make better decisions about how to care for ourselves and those we love.

    Share
    Comments closed