One of the simplest ways to get a handle on the clinical importance of a result is to consider the number needed to treat (NNT). This is the number of patients who must be treated in order for one, on average, to benefit.
If we return to our earlier hypothetical example, we found that 20% recovered on MiracleCureTM and 60% recovered on OldBoring. One of the ways we found of explaining this result was to say that an extra 40% recovered on OldBoring.This gives an NNT of 1/0.4 (or 100/40) = 2.5. One additional person recovers for every 2.5 people treated.
The idea of NNTs can be extended to other outcomes which are not binary in nature. The simplest way to do this for continuous outcomes is to establish a clinically relevant target value (eg a cholesterol of below 6) and then treat it as a binary outcome. For survival analysis, you need to choose a clinically relevant timepoint and estimate the proportions surviving from the Kaplan-Meier survival curves. The NNT may then be expressed as the number needed to treat for one additional person to survive up to, say, 5 years (a clinically relevant timepoint for many cancers).
Other means of expressing the benefit may be more useful in some circumstances. In survival analysis it is often most useful to consider the difference in median survival. A couple of weeks difference would be a lot less meaningful than 6 months or a year, and this consideration would be important when deciding the trade off with the severity and duration of side effects, which will be considered in the next section.
Click the "exercise" link below to test your knowledge...