Background: The GRADE (Grading of Recommendations Assessment, Development and Evaluation) recommendations (for downgrading due to imprecision) encourage Cochrane authors to be 'confident' that a treatment is clinically unimportant if the mean difference and 95% confidence interval (95% CI) fall short of the clinical minimally important difference (CMID).
Objectives: To demonstrate that a mean difference alone cannot rule out important clinical benefit in a population.
Methods: A Cochrane Airways review will be used to present data to show that a mean difference that is clearly less than the CMID can lead to an important shift in the population of those who respond to treatment (also known as 'responders' who individually exceed the CMID). A responder analysis will be compared to the results of modelling the shift in the population mean.
Results: A statistically significant increase in the proportion of responders cannot be ruled out by a mean difference and 95% CI that are entirely below the CMID.
Conclusions: It is wrong to rule out a treatment as clinically unimportant on the basis of the mean difference alone. If the mean difference shows a statistically significant benefit (however small that might be), then a responder analyses is required to assess the size of the benefit in a population.