The number needed to treat: it is time to bow out gracefully

The number needed to treat (NNT) is a simple-to-understand absolute effect measure. However, it is only sensible when the risk difference is statistically significant. We highlight two important limitations of using NNT in the context of decision-making (developing a guideline, a policy decision, or a health technology assessment). The first limitation of NNT relates to difficulties in expressing and interpreting the confidence interval (CI) for the NNT when the CI of the risk difference includes the null (ie, the results are not statistically significant). This CI of NNT will be disjointed and will include implausible values. The second limitation of NNT relates to the increased complexity of trading off benefits and harms on the NNT scale. This proposal calls for abandoning the use of NNT from decision-making contexts. PLAIN LANGUAGE SUMMARY: The number needed to treat (NNT) has statistical and methodological limitations that make it unhelpful in the context of developing clinical practice guidelines and policy decisions.