When does a difference make a difference? Interpretation of number needed to treat, number needed to harm, and...

Although great effort is made in clinical trials to demonstrate statistical superiority of one intervention vs. another, insufficient attention is paid regarding the clinical relevance or clinical significance of the observed outcomes. Effect sizes are not always reported. Available absolute effect size measures include Cohen's d, area under the curve, success rate difference, attributable risk and number needed to treat (NNT). Of all of these measures, NNT is arguably the most clinically intuitive and helps relate effect size difference back to real-world concerns of clinical practice. This commentary reviews the formula for NNT, and proposes acceptable values for NNT and its analogue, number needed to harm (NNH), using examples from the medical literature. The concept of likelihood to be helped or harmed (LHH), calculated as the ratio of NNH to NNT, is used to illustrate trade-offs between benefits and harms. Additional considerations in interpreting NNT are discussed, including the importance of defining acceptable response, adverse outcomes of interest, the effect of time, and the importance of individual baseline characteristics.