EDITOR—The number needed to treat has become a popular summary statistic for the results of randomised controlled trials because it combines the treatment effect with the background level of risk in the population studied. Patients in a single trial are randomised for both of these factors, and a confidence interval can be calculated which estimates the statistical uncertainty of the number needed to treat in this particular population.1

Problems arise when comparisons are made between numbers needed to treat from different randomised trials, or when the numbers needed to treat from several trials are combined in a meta-analysis. Often the background level of risk varies between trials in a non-random fashion, depending on the entry criteria in each trial. If the relative benefit of the treatment is constant across these background levels …