number needed to treat

Number needed to treat, or NNT, answers a wonderfully practical question: how many patients do we have to treat with a drug for one of them to gain the benefit we care about? If a medicine has an NNT of 20 for preventing a stroke, it means that, on average, 20 people must take it for one stroke to be prevented; the other 19 do not get that particular benefit. It turns a statistic into a tangible sense of how much a treatment really delivers.

The NNT is calculated from how much a treatment lowers the chance of a bad outcome compared with no treatment or a placebo. It is the reciprocal of that absolute reduction in risk, so a drug that cuts an outcome's rate from 2 percent to 1 percent has an absolute reduction of 1 percent and an NNT of 100. The smaller the NNT, the more powerful the treatment, because fewer people must be treated for each success.

Its great value is honesty about scale. A result can be reported as a dramatic relative reduction, say cutting risk by half, while the absolute benefit is tiny if the disease was rare to begin with; NNT exposes this by anchoring to real numbers of people. It is most meaningful when stated with the outcome and the time period, since treating for one year versus five years gives very different numbers.

NNT must be read alongside its mirror image, the number needed to harm, which says how many treated patients yield one extra case of a side effect. A drug with a low NNT for benefit but an even lower number needed to harm may do more damage than good. And because NNT comes from a particular trial in a particular population, it does not transfer cleanly to patients at very different baseline risk.