Number needed to treat is how many people must take a treatment, for a defined period, for one of them to get the benefit a study measured. If the number needed to treat is 50 over five years, then 50 people take the treatment for five years so that one of them avoids the tracked event, while the other 49 do not. The figure is simply one divided by the absolute risk reduction, and that one step, turning a percentage into a head count, is what makes it one of the clearest ways to state what a treatment actually does.
Key points#
- Number needed to treat is the count of people who must be treated, over a stated period, for one additional person to avoid a defined bad outcome. It equals one divided by the absolute risk reduction.
- It cannot be calculated without the baseline event rate, so it carries that baseline inside it and cannot hide how common the event was, which a relative reduction can.
- The identical drug with the identical relative effect can produce very different counts. When the outcome is ten times rarer, a number needed to treat of 50 grows to 500.
- To read it fairly, ask what exact outcome was prevented, over how long, and who was studied. A figure from a high-risk group overstates the benefit for a lower-risk person.
The definition, in one line#
Number needed to treat is how many patients you would treat, over a stated period, so that one extra patient avoids a defined unwanted outcome. Two words in that sentence do the heavy lifting.
"Extra" (or additional) matters because the count includes only benefits that would not have happened anyway, not every good outcome among treated people. "Defined" matters because the figure means nothing until it names the exact event and the time window. A 25 to prevent one heart attack over ten years is a different creature from a 25 to prevent one headache over a week. A bare "25" with no event and no clock attached is not yet a fact you can use.
The formula and where it comes from#
Start with the absolute risk reduction, the plain difference in event rates between the treated group and the comparison group. Invert it, and you have the number needed to treat:
number needed to treat = 1 / absolute risk reduction
That inversion is the whole trick. A large absolute risk reduction inverts to a small count (few people treated per benefit), and a tiny absolute risk reduction inverts to a large count (many people treated per benefit). Because you cannot even reach the formula without knowing how often the event happened in the comparison group, the baseline risk is baked into the answer.
A worked example you can run in your head#
Round, invented numbers keep any real product out of it. Picture 2,000 people split into two equal groups of 1,000, followed for three years for one unwanted event.
In the comparison group, 80 people have the event, an absolute risk of 8 percent. In the treated group, 60 have it, an absolute risk of 6 percent. The absolute risk reduction is 2 percentage points, or 0.02 as a fraction.
Now invert. One divided by 0.02 is 50, so the number needed to treat is 50. Over three years, you treat 50 people like these for one of them to avoid the event. Stated in relative terms, dropping from 80 events to 60 is a 25 percent reduction. Twenty-five percent and a head count of 50 describe the very same result. One phrasing sounds impressive; the other tells you the size of the job.
Now change only the baseline and watch the count move. Suppose the event were ten times rarer, so the comparison group had 8 events and the treated group 6. The relative reduction is still 25 percent, but the absolute risk reduction is now 0.2 percentage points, and the number needed to treat jumps to 500. Same drug, same relative effect, ten times as many people treated for each one helped. That sensitivity is the figure exposing what a relative number leaves out.
Why a count is harder to spin than a percentage#
Most ways of describing a benefit flatter it a little. A relative reduction reads the same whether the underlying risk was high or nearly nonexistent, because it says nothing about how common the event was in the first place. The number needed to treat cannot be produced without that baseline, so it refuses to detach the effect from how often the outcome actually occurs.
It is also candid about the people who do not benefit. A number of 50 tells you, in the same breath, that 49 people carried the cost, the routine, and any side effects without the measured payoff. Plenty of worthwhile treatments have large numbers needed to treat, because we accept treating many people to spare a few from something serious. The figure just puts that trade where everyone can see it.
The number needed to harm#
Harm has a mirror-image measure. The number needed to harm is how many people you would treat, over a comparable window, for one additional person to suffer a specific side effect. It is one divided by the absolute change in that harm.
Set the two side by side. A treatment with a number needed to treat of 50 and a number needed to harm of 200 helps roughly four people for each one it harms in that particular way. Placing benefit and cost in the same unit, people treated, tells you more than any pair of loose percentages ever could.
Reading a number needed to treat without being misled#
Three questions keep the figure honest.
First, what exact outcome was prevented? A small count for a surrogate marker, a lab value that shifts, means far less than a slightly larger count for something a patient would actually feel, such as a stroke avoided. Pull the endpoint into view before you are impressed.
Second, over how long? A number needed to treat of 30 over one year and a number needed to treat of 30 over ten years are not the same offer. The clock is part of the fact.
Third, who was in the trial? The figure is anchored to the baseline risk of the people studied. A number from a high-risk population overstates the benefit for a lower-risk person, because the same relative effect spread over a smaller baseline yields a larger real-world count. Treat any single value as an estimate with a range around it, and give more weight to a benefit whose reported range stays comfortably finite than to one whose upper edge drifts toward the thousands for a single gain.
The count is not a verdict. A small one does not order you to treat, and a large one does not forbid it, because how much a given outcome is worth is for you and your clinician to weigh together. What the figure does is keep the many people who carry the effort without the reward in the same frame as the one who benefits. When a treatment is offered, ask for that count, the event it prevents, and the years it covers. A claim that can answer all three deserves your attention. One that cannot is not finished.
Sources and further reading
Questions and answers
Is a lower number needed to treat always better?
Usually, but not on its own. A low count means fewer people treated per benefit, which is generally good, yet it only matters once you know what outcome was prevented and over what period. A low number for a minor, short-term endpoint can be less meaningful than a higher number for a serious event prevented over years.
How does number needed to treat relate to relative risk reduction?
They can describe the exact same trial and still feel very different. Relative risk reduction is a percentage that ignores baseline risk, so it looks identical whether the event was common or rare. Number needed to treat is built from the absolute risk reduction, so it always reflects how often the outcome actually happened.
What counts as a good number needed to treat?
There is no universal threshold. For a mild, reversible problem, only a small count may feel worthwhile. For a devastating outcome such as a stroke or a hip fracture, many people reasonably accept treating hundreds to prevent one. The number is an input to that judgment, not a substitute for it.