I recently came across this paper on “Estimating the number needed to treat from continuous outcomes in randomised controlled trials: methodological challenges and worked example using data from the UK Back Pain Exercise and Manipulation (BEAM) trial” by Robert Froud, Sandra Eldridge, Ranjit Lall, and Martin Underwood. The full text is available here:

 

http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=2702335

 

The title is misleading in that they did not estimate NNTs for outcomes with continuous measures: they used individual patient data and converted continuous outcome measures (disability scores) into dichotomous measures (“improved” (Yes/no) and “benefit” (Yes/no)). NNTs were then calculated for the dichotomous measures.

 

This made me wonder why I haven’t seen methods for estimating NNTs for continuous outcome measures (normally distributed) when you have the mean difference, standard deviation of the mean difference, and minimum clinically important difference? Have I missed them? Or, would it be too inaccurate?

 

Michael