You've probably thought about this more than I have, but your "take down" of NNT doesn't work for me. Let's use your example. We treat 500 people and observe

You've probably thought about this more than I have, but your "take down" of NNT doesn't work for me. Let's use your example. We treat 500 people and observe another 500. 5 of the untreated die, and 2.5 of the treated die (I don't know how either). We observe that the NNT is 200. And we this is great! If a drug is inexpensive, has minor but not severe side effects, and saves 1/200 people from dying within a year you should absolutely be giving it to everybody. It's a home run. The problem isn't the NNT, it's the expectation of what constitutes a a high or low number. By contrast, consider another study. We treat 500,000,000 people, and observe another 500,000,000. We observe a 50% relative risk reduction among the treated. The drug is inexpensive, and has minor but not severe side effects. Should we create everyone? I have no idea. It has to depend on how much risk we started with. If the mortality rate is 1% per year among the untreated, then almost definitely yes. But if we observed 2 deaths in the untreated and only 1 from the treated, probably there is a better area to spend our efforts, even if we somehow knew the life we saved just noise. So yes, the reduction in risk matters, but only in concert with a concept of the baseline risk. By itself, it's not very helpful, but in concert with other data it's great. NNT is a way of expressing this in a single number. There is no apriori reason to believe that 200 is high or low. It may not be perfect, but it's better than just a bare relative risk percentage.