Swots' Corner:

NNTs and Confidence Intervals

Why Bother?

NNT values without confidence intervals may be better than nothing, but they do not tell you how likely the values, or 'point estimates', are to be true. A p value tells you that a result is seldom likely to occur by chance (less than 1 or 5% of the time). A confidence interval can tell you where the true value is most likely to be (more than 90 or 95% of the time). "You can be 95% certain that the truth is somewhere inside a 95% confidence interval". [1]

Calculating NNT

The NNT is the reciprocal of the absolute risk reduction.

Calculating Confidence Intervals

The pukka method is to "invert and exchange the limits of a 95% CI for the ARR" [2]. The calculation using the confidence interval derivation for proportions is:

When not to

If the odds ratio for the result is not significant (lower odds ratio confidence interval ≤ 1), then it is unwise to bother with the confidence intervals for the NNT - see Mulrow's excellent paper [4] featured in Bandolier 15 .

References

Sackett DL, Haynes RB, Guyatt GH, Tugwell P. Clinical Epidemiology: a basic science for clinical medicine. Boston: Little, Brown, 1991.
Cook RJ, Sackett DL. The number needed to treat: a clinically useful measure of treatment effect. British Medical Journal 1995; 310: 452-4.
Gardner MJ, Altman DG. Statistics with confidence. London: British Medical Journal, 1989.
CD Mulrow, JA Cornell, CR Herren, A Kadri, L Farnett, C Aguilar. Hypertension in the elderly. Implications and generalizability of randomized controlled trials. Journal of the American Medical Association 1994 272:1932-8.

previous or next story in this issue