Confidence intervals for the number needed to treat
In a recent article  Douglas Altman describes the number needed to
treat as a useful way of reporting the results of randomised controlled
trials, and proceeds to demonstrate how confidence intervals for this
measure are calculated. However, as Altman shows, a confidence interval
for an absolute risk reduction (ARR) from, say, -5% to +25% inverts to a
confidence interval that goes from a number needed to treat to benefit
(NNTB) of 4, through infinity to a number needed to treat to harm (NNTH)
My impression from discussing such intervals with clinicians is
that they find them very difficult to grasp. Altman correctly argues on
general grounds against presenting confidence intervals only for
significant effects. Moreover the potential application in presentations
such as forest plots that put together the results of several studies
rules out the argument that the NNT should only be estimated at all when
it is significant.
In view of this, I believe that the NNT has as much potential to confuse
as to enlighten. The ARR is a more basic quantity with much less
potential to be misunderstood, and should be regarded as the primary
measure of effect size. The estimated ARR and its confidence interval are
most readily grasped when presented in percentages, as in Altman’s paper.
The NNT and its confidence limits are better regarded as secondary,
whether in numerical presentation of results or as an additional scale on
a diagram, a useful informal alternative way of interpreting an ARR when
it is well away from zero.
1 Altman DG. Confidence intervals for the number needed to treat.
BMJ 1998; 317: 1309-12.
Competing interests: No competing interests