Risks vs. Benefits
NNT is a conceptually simple measure of a drug’s effectiveness, but it provides little insight into the risks of a drug’s side effects. Both risks and benefits should be considered.
As an example, consider an earlier article by the New York Times’ health editor entitled “When Lowering the Odds of Cancer Isn’t Enough.” The article discusses the reluctance of some at-risk women to take tamoxifen, a drug designed to reduce incidences of breast cancer. The article notes that the drug does have potential side effects, but seems to suggest that the reluctance may be attributed to a phenomenon known as “omission bias.” “People tend to worry more about a low risk of harm [of taking a drug]…than about a higher risk of harm from doing nothing.”
But what are the risks and benefits in this case? Using the statistics from the article, based on a study involving women over 50 years old with family histories of breast cancer, let’s analyze the numbers.
Over a five-year period, without tamoxifen, the risk of developing breast cancer was 1.9%. With tamoxifen, the risk dropped to 1.0%. That’s 47% lower — a significant percentage improvement. But from an NNT perspective, it means that 9 women out of 1,000 would benefit. That’s an NNT of 111, i.e., one woman out of every 111 women taking the drug would benefit. Expressed this way, the benefits seem smaller, but are clearly positive (and would likely be higher still for higher-risk women seeking to prevent the reoccurrence of breast cancer).
When potential side effects are factored in, however, the analysis becomes more complex. According the article’s statistics, for every 1,000 women on the drug, you would expect an additional:
21 cases of endometrial cancer (a form of uterine cancer)
21 cases involving blood clots
31 cases of cataracts
Plus more numerous sexual and hormonal problems
To put this in an NNT context, a lesser used statistical measure used is NNH — the Number Needed to Harm. It is an indication of the number of patients treated with a drug that would result in one developing a specific side effect. In this example, the NNH for either endometrial cancer or blood clots would be 48 (i.e., 1000/21). The NNH for cataracts would be 32.
An alternative measure of risk, directly linked to a specific NNT, might be called Number Affected Adversely (NAA). It would measure the number of individual expected to suffer a side effect for every one person expected to benefit. Again using this example, the NAA of endometrial cancer (or blood clots) would be a little over two. In other words, for every one out of 111 patients likely to benefit from tamoxifin, two would likely develop another form of cancer (and a still higher number would suffer one or more of the other side effects).
Ideally, particularly if a potential side effect is as bad as the disease itself, you’d like to see an NNH > NNT or an NAA <1. If it is not, as in the example above, then the numbers at least provide a rational basis for a discussion of the potential benefits and risks,
The reluctance of some at-risk women to take tamoxifen — and we’re using tamoxifen only because of the availability of the statistics in the New York Times’ article itself — may have less to do with “omission bias,” and more to do with carefully balanced medical decisions reached by doctors and patients.