Editverse Free Calculator
NNT & NNH Calculator
Enter the control and experimental event rates over the same follow-up to get the absolute risk reduction (ARR = CER − EER) and the point estimate of the number needed to treat, with the R command. When the experimental rate is higher, ARR is negative and the number shown is the number needed to harm (NNH). Read a full interpretation guide below. Validated against R 4.3.3.
In Simple Words
The number needed to treat (NNT) answers a practical question: how many patients must receive a treatment for one extra person to benefit? It is the reciprocal of the absolute risk reduction (ARR) — the difference between the event rate in the control group and in the treated group. An NNT of 10 means treating 10 patients prevents one event.
Smaller NNTs mean more effective treatments. The mirror-image measure, the number needed to harm (NNH), counts how many patients must be treated for one extra person to experience a harm. Both numbers turn abstract percentages into something a clinician and patient can weigh together.
When to Use NNT and NNH
Use NNT to communicate the benefit of an intervention from a randomised trial or cohort study, and NNH to communicate its risks, so the two can be compared on the same scale. NNT is far more interpretable at the bedside than a relative risk or odds ratio, because relative measures hide how common the outcome actually is.
Always report NNT with its time horizon and its 95% confidence interval — an NNT of 20 over one year is very different from 20 over a decade, and the interval shows the precision. When the treatment effect is not statistically significant, the confidence interval for the NNT is unusual (it spans from a finite NNT through infinity to a finite NNH); report it carefully rather than as a single number. This calculator returns the point estimate only; for the 95% interval by the Altman method, enter event counts and group sizes in the advanced NNT calculator, which labels the result as NNT benefit or NNT harm. Not sure which effect measure to present? Our interactive Test Selector can help.
cer <- {cer}; eer <- {eer}
arr <- cer - eer # > 0 benefit, < 0 harm
if (arr > 0) c(ARR = arr, NNT_benefit = ceiling(round(1/arr, 6))) else c(ARR = arr, NNH_harm = ceiling(round(1/(eer - cer), 6)))User guide
When to use. Express treatment benefit as the number of patients needed to treat to prevent one event.
Assumptions. Event rates are from comparable groups over a defined time horizon.
Common mistakes. Always report the time horizon; NNT is meaningless without it.
Worked example
Click Load example data above, then Run Analysis to see a fully worked result with figure, formula and interpretation for this test.
References
- Laupacis A, Sackett DL, Roberts RS. An assessment of clinically useful measures. N Engl J Med. 1988;318:1728-1733.
Worked Examples
Example 1 — A prevention trial
If 30% of control patients have an event versus 20% of treated patients, the absolute risk reduction is 0.30 − 0.20 = 0.10, so the NNT is 1 / 0.10 = 10. Treating 10 patients prevents one event over the trial’s follow-up period. Always state that period alongside the number.
Example 2 — Number needed to harm
If a side effect occurs in 8% of treated patients versus 3% of controls, the absolute risk increase is 0.05 and the NNH is 1 / 0.05 = 20. Presenting NNT = 10 alongside NNH = 20 lets a clinician weigh one prevented event against the rate of harm on the same scale.
Assumptions to Check
Comparable groups and a fixed time horizon. The two event rates must come from comparable groups over the same defined follow-up. NNT changes with the period, so it is meaningless without one.
Constant baseline risk. A single NNT assumes patients share a similar baseline risk. For populations with very different risk, the NNT differs and should be reported by risk stratum.
Event rates, not rates of a continuous outcome. NNT applies to binary outcomes. For continuous endpoints, report the mean difference and an effect size instead.
Common Mistakes
Omitting the time horizon
An NNT of 25 over 5 years is very different from 25 over 5 weeks. Always attach the follow-up period; without it the number cannot be interpreted.
Reporting NNT without a confidence interval
A point NNT hides uncertainty. Report the 95% confidence interval, and note that a non-significant effect produces an interval spanning NNT to NNH through infinity.
Deriving NNT from a relative measure alone
A relative risk reduction cannot give an NNT without the baseline (control) event rate. The same relative reduction yields very different NNTs at high versus low baseline risk.
Rounding down
NNT is conventionally rounded up to the next whole patient, since you cannot treat a fraction of a person to prevent a whole event.
Frequently Asked Questions
What is a good NNT?
It depends on the outcome and the treatment’s cost and harms. A low NNT to prevent a serious event is excellent; a higher NNT may still be worthwhile for a cheap, safe intervention.
How is NNT related to absolute risk reduction?
NNT is simply 1 divided by the absolute risk reduction. An ARR of 0.05 gives an NNT of 20.
Can I calculate NNT from a hazard ratio?
Yes, for time-to-event data NNT can be derived from the baseline risk and the hazard ratio at a specified time point, but it then applies only to that time point.
Does the result match SPSS, R, and Stata?
Yes. The calculator uses the standard ARR and reciprocal formulas and shows the exact R command so you can reproduce it.
Related Calculators
References
- Cook RJ, Sackett DL. The number needed to treat: a clinically useful measure of treatment effect. BMJ. 1995;310(6977):452–454. PMID 7873954.
- Chesnaye NC, Ortiz A, van Diepen M, et al. How to interpret the number needed to treat for clinicians. Nephrol Dial Transplant. 2026. PMID 40839106.
- Laupacis A, Sackett DL, Roberts RS. An assessment of clinically useful measures of the consequences of treatment. N Engl J Med. 1988;318(26):1728–1733.
Need This Analysis Done for Your Manuscript?
Our PhD biostatisticians compute your NNT, NNH, and absolute risk measures with confidence intervals, and deliver publication-ready tables with reproducible R output.