Physicians make different choices about treatments according to how the information is presented [1]. They make more conservative choices with NNT than with relative risk reduction. The actual numbers, of course, remain the same, but they are framed. Any danger of interpretation lies in us, not in the numbers. A 50% reduction or increase sounds wonderful, but there is a huge difference between a 50% reduction or increase in a common event, and a 50% reduction or increase in a rare event. The NNTs make the point about how rare or common an event is because the absolute risk reduction is part of the calculation. As a result, we get the whole picture.

An extreme example of the danger of Dr. Jain's exclusive use of relative risk is the way in which information about risk for thromboembolism with oral contraceptive pills was presented in the United Kingdom [2, 3]. The official statement, that "combined oral contraceptives containing desogestrel and gestodene are associated with around a two-fold increase in the risk of thromboembolism," caused enormous concern for women and their physicians. Women stopped taking oral contraceptives, resulting in unwanted pregnancies and increased abortions [3].

For one woman to have a thromboembolic event while receiving the combined oral contraceptive compared with older preparations, the relative risk is 2.0 (95% CI, 1.1 to 3.7) and the NNT is 6700 (CI, 3600 to 54 000). Telling a woman that she has a risk for thromboembolism that has doubled because of the type of pill she takes compares poorly with telling her that, yes, there is an increased chance, that we think the size of that chance is 1 in about 7000, but that the uncertainty ranges from 1 in 3600 to 1 in 54 000. For comparison, in the United Kingdom, the chance of being killed in an automobile accident in any one year is 1 in 16 000. Any decision the woman makes may well change depending on the way the information is framed, just as it did for physicians [1].

We think that NNTs are useful, and in our Annals paper we discussed ways in which changes in baseline risks for individual patients might be handled. The power of systematic reviews combined with NNTs is that they allow us to see more clearly the choices between several treatments applicable to patients with similar or identical conditions. An example is the use of NNTs for the relative efficacy of oral analgesics for postoperative pain relief [4].