What is Number Needed to Treat (NNT)?

The number needed to treat (NNT) is how many people must take a treatment, over a stated time, for one of them to be spared an outcome they would otherwise have had. It comes from a trial that compares a treated group with an untreated one. In figures a medical teaching paper uses, 2.9 in 100 untreated 60-year-olds with high blood pressure had a stroke within five years, against 1.9 in 100 on a diuretic, a pill that lowers blood pressure. The gap is 1 in 100, and 100 ÷ 1 is 100, so the NNT is 100 over five years. Of those 100 on the pill, about 97 were never going to have a stroke, about 2 have one anyway, and one is spared. Nobody can say in advance which one. That is what a treatment working usually means: everyone takes it, and the benefit lands on the few whose outcome it changed. The NNT is that gap turned into people, and it only means something with its years attached.

The number needed to treat (NNT) is the count of people who have to take a treatment, for a set period, so that one of them avoids an outcome they would otherwise have had. It is worked out from a trial, where chance splits people into a treated group and a control group given a dummy, and the outcome is counted on each side. In figures the Canadian Medical Association Journal uses to teach the sum, 12.0 in 100 patients died within five weeks of a heart attack without streptokinase, a clot-dissolving drug, against 9.2 in 100 with it. The difference, 2.8 in 100, is the absolute risk reduction. Divide 100 by 2.8 and you get 36: treat 36 patients for five weeks to prevent one death. Those 36 fall into three groups. Roughly 32 would have lived anyway. Roughly 3 die despite the drug. One is alive because of it, and nobody can know in advance which one. The relative figure hides this. Deaths fell by about a quarter (2.8 ÷ 12.0), and a quarter sounds the same whatever the starting risk. The NNT does not. Cut a stroke risk of 3 in 100 by a third and the gap is 1 in 100, so the NNT is 100. Cut 30 in 100 by a third and the gap is 10 in 100, so the NNT is 10. Same drug, same third, ten times the benefit per person treated, because the higher-risk patient had more to lose. The period matters too: the gap between the groups usually widens as years pass, so the NNT falls, and an NNT with no time attached is incomplete. The same sum runs on side effects. If a bleed happens to 0.3 in 100 untreated and 0.9 in 100 treated, the increase is 0.6, and 100 ÷ 0.6 is about 167, the number needed to harm. Weigh the two numbers against each other and against how bad each outcome is: an NNT of 100 to prevent a death can be well worth it, and an NNT of 5 for a trivial gain may not be.

The number needed to treat is how many people must take a treatment for one of them to be spared an outcome they would otherwise have had. A shop testing a discount code measures the same thing. The shop emails a 20% code to 1,000 customers and holds back 1,000 similar customers who get nothing. Over the month, 50 of the code group buy and 30 of the held-back group buy. The code did not produce 50 sales, because 30 of those people were buying anyway. It produced 20 extra sales from 1,000 codes, so 50 codes went out for each sale the shop would not otherwise have made. Fifty is the shop's number needed to treat for that month. The customers sort the way a trial's patients do: 930 were never buying, 30 bought regardless, and 20 were turned by the code, with no way to pick them out beforehand. The match ends there: the shop can rerun the test next month on the same customers, and a trial cannot put the same patients through the same illness twice.