Number needed to treat (revision 8)
Old revision·18:27, 3 Jan 2025·SurpassSunniva
| Number needed to treat | |
|---|---|
| Abbreviation | NNT |
| Definition | Reciprocal of the absolute risk reduction |
| Requires | A stated time horizon |
| Counterpart | Number needed to harm |
| Topic infobox · conventions | |
The number needed to treat (NNT) is the number of people who must receive a treatment, rather than the comparator, for one additional person to experience the benefit over a defined period. It is the reciprocal of the absolute risk reduction.[1]
Its value is that it expresses effect on an absolute scale. A relative risk reduction of 20% describes a large benefit where the baseline risk is high and a negligible one where it is low; the number needed to treat carries the baseline risk within it.[1]
It is meaningless without a time horizon. A treatment with a number needed to treat of 50 over one year and one of 50 over five years are very different propositions, and quoting the figure without the period is the commonest error made with it.[1]
Calculation
[edit]\text{NNT} = 1 / (\text{risk}\text{control} - \text{risk}\text{treated})
For SELECT, the primary composite occurred in 8.0% of the placebo group and 6.5% of the semaglutide group over a mean 39.8 months. The absolute risk reduction is 1.5 percentage points, so the number needed to treat is 1/0.015 ≈ 67 over roughly 3.3 years.[2]
The corresponding relative figure — a hazard ratio of 0.80, or a 20% relative reduction — is the same result expressed differently and sounds considerably larger. Neither is wrong; they are the same arithmetic.[1]
Confidence intervals transfer awkwardly. The interval for a number needed to treat is the reciprocal of the interval for the absolute risk reduction, and it behaves badly when that interval spans zero, which is one reason the measure is best used for statistically significant results.[3]
Number needed to harm
[edit]The same arithmetic applied to an adverse outcome gives the number needed to harm, and the two are only comparable when computed over the same period in the same population.[1]
In SELECT, discontinuation for adverse events occurred in 16.6% on semaglutide against 8.2% on placebo — an absolute difference of 8.4 percentage points, giving a number needed to harm of about 12 for that outcome over the same period.[2]
References
- ^ a b c d e Laupacis A, Sackett DL, Roberts RS. "An assessment of clinically useful measures of the consequences of treatment." New England Journal of Medicine 318(26):1728–1733 (1988). PMID 3374545.
- ^ a b Lincoff AM, Brown-Frandsen K, Colhoun HM, et al. "Semaglutide and cardiovascular outcomes in obesity without diabetes." New England Journal of Medicine 389(24):2221–2232 (2023). PMID 37952131.
- ^ Altman DG. "Confidence intervals for the number needed to treat." BMJ 317(7168):1309–1312 (1998). PMID 9804726.