How this instrument works
Number needed to treat answers a specific, practical question: on average, how many people like the ones in a study need to receive the treatment for one additional person to get a good outcome they wouldn't otherwise have had? It starts from the absolute risk reduction — the plain percentage-point gap between how often the bad outcome happened in the control group and how often it happened in the treated group — and takes its reciprocal. A smaller number means a more powerful, more concentrated benefit.
The measure was introduced in 1988 by Laupacis, Sackett and Roberts in the New England Journal of Medicine, specifically as a corrective to relative risk reduction, a figure that can make a modest benefit sound dramatic. Cutting a risk from 2% to 1% is a 50% relative reduction, a headline number, but it is also a 1-percentage-point absolute change — an NNT of 100, meaning a hundred people are treated for every one who benefits. Reporting both figures side by side keeps the arithmetic honest.
The result depends entirely on the two rates fed into it, which is exactly the point: the same treatment, applied to a higher-risk population where bad outcomes are common to begin with, produces a smaller, more compelling NNT than the identical treatment applied to a low-risk population, even when the underlying relative effect is unchanged. NNT is a property of a specific study population, not a universal constant attached to a drug or a procedure.
- Enter Control event rate (%), how often the outcome occurred without the treatment.
- Enter Experimental event rate (%), how often it occurred with the treatment.
- Read the absolute risk reduction — the percentage-point gap between the two rates.
- Read NNT — patients needing treatment, on average, for one additional good outcome.
Worked example — control 20%, treatment 15%
A trial reports a bad outcome in 20% of the control group and 15% of the treated group. The absolute risk reduction is the plain gap: 20 minus 15 leaves 5 percentage points. Dividing 100 by 5 gives an NNT of 20 — treat 20 patients like the ones enrolled to prevent one additional bad outcome.
Compare a second trial where the gap is wider: control at 30%, treatment at 10%. The absolute risk reduction is 20 points, and 100 divided by 20 gives an NNT of just 5, a far more concentrated effect than the first example despite both trials describing a treatment that works. The size of the gap, not the treatment's existence alone, drives how small the number gets.
Questions
Why not just report relative risk reduction instead?
Because relative risk reduction can make a tiny absolute benefit sound dramatic. A drop from 2% to 1% is a 50% relative reduction, an eye-catching figure, yet the absolute change is only 1 percentage point, an NNT of 100 — a hundred people treated for each one who benefits. NNT was introduced specifically to sit alongside relative figures and keep the actual scale of benefit visible.
Is a lower NNT always better?
Generally yes — a lower NNT means fewer patients need treatment for one additional benefit, a more concentrated effect. But NNT has to be read against the seriousness of the outcome being prevented and the treatment's own risks and costs; an NNT of 50 for preventing a fatal event can be a strong result, while an NNT of 5 for a minor, low-stakes outcome may matter less in practice.
Why does the same drug get a different NNT in different studies?
Because NNT depends on the baseline event rate of the population studied, not just the treatment's relative effect. The identical relative risk reduction produces a small, striking NNT in a high-risk population where bad outcomes are common, and a much larger NNT in a low-risk population where they are already rare — so an NNT reported in one trial should not be applied blindly to a different group of patients.
What does a negative or undefined NNT mean?
If the treated group actually does worse than the control group, the same arithmetic produces what is sometimes reported as number needed to harm rather than treat — a distinct figure describing how many people would need the treatment for one to be harmed instead of helped. An event-rate gap of exactly zero makes the reciprocal undefined, since dividing by zero has no meaningful result.
Can I compare NNT figures across two entirely different trials?
Only with real caution. Two NNT figures are directly comparable only when the trials studied similar populations, similar follow-up periods and the same outcome being measured. A short trial preventing a minor complication and a long trial preventing a fatal one can both report NNT of 10, yet describe wildly different clinical value — the number alone doesn't carry that context.
References
- Laupacis A, Sackett DL, Roberts RS. N Engl J Med, 1988 (PubMed)
- Centre for Evidence-Based Medicine, Oxford — Number Needed to Treat
Read this first: This instrument computes a screening figure from population formulas — it is not a diagnosis, and it cannot see the whole picture a clinician can. Use it to inform a conversation, not to replace one.