The number needed to treat is how many must receive an intervention for one additional person to benefit, derived from the absolute difference between groups.
The number needed to treat is how many must receive an intervention for one additional person to benefit, derived from the absolute difference between groups. It expresses the same information as a risk difference in a form that is easier to reason about.
The same relative effect produces very different numbers at different baseline risks. A treatment halving a common risk has a small number; the same treatment halving a rare risk has a large one, and quoting the figure without the baseline is incomplete.
A number needed to treat over one year is not comparable with one over five. The duration belongs in the statement, and its absence makes figures from different studies look comparable when they are not.
The number needed to harm applies the same arithmetic to adverse outcomes. Presenting benefit in this form without the harm counterpart gives half the picture, and the two together are what a decision actually requires.
The figure is a property of a population under study conditions. It does not mean one person in that many will benefit in any identifiable way, and treating it as a personal probability misreads what was measured.
A relative reduction sounds impressive while a number needed to treat of several hundred sounds modest, and both can describe the same result. The second is harder to oversell, which is part of why it is less often quoted.
The figure inherits uncertainty from the underlying difference, and where that difference is not statistically distinguishable from zero the interval includes infinity. A point estimate alone conceals this entirely.
Ask for it whenever a relative figure is quoted, with its time horizon and its harm counterpart. Those three together make a result interpretable.
A number needed to treat of fifty to prevent a mild, temporary problem and one of fifty to prevent a permanent one are not equivalent, though the figures are identical. The measure strips out severity entirely, which is why it belongs beside a description of the outcome rather than alone.
The figure is one divided by the absolute risk difference between groups, both of which appear in any properly reported results table. A reader can compute it in seconds when a paper reports only relative measures, and the exercise is usually sobering.