[Numbers needed for treatment and their respective confidence intervals: useful tools to assess clinical significance and uncertainty associated with medical interventions].

This is a narrative review of a concept--number needed to treat (NNT). The proprieties of NNTs are described. Furthermore NNTs as important tools for medical decision are discussed. In particular, conditions that make calculus and the interpretation of NNTs somewhat more complex are commented. Namely, we consider situations where the results are not primarily binary or are expressed as "time to event" or come from a meta-analysis. The inconsistencies that might occur when NNTs are calculated from meta-analysis are discussed in some detail. Definitions for parameters derived from NNTs, as numbers needed to harm (NNH), unqualified success (NNTUS) and unmitigated failure (NNHUF) are given. Finally the usefulness of NNTs and absolute risk reductions to communicate risk to individual patients are considered. It is concluded that despite NNTs being powerful tools for medical decision, in the case of communicating risks to the individual patient absolute risks might be preferable.