Background and Purpose: Using numbers needed to treat (NNT) to report trial outcomes has been publicised as improving the interpretability of trial results. However, authors of back pain trials rarely report continuous outcomes using NNT, and if they do, rarely consider deteriorating patients. We performed a reanalysis of the UK BEAM dataset to explore the effects of reporting NNT on interpretation, and articulate difficulties with this approach.

Method and Results: 1,334 patients consulting for low back pain were randomised to receive either best GP care alone, or best GP care and exercise, manipulation or combined exercise and manipulation. The primary outcome was Roland Morris Disability Questionnaire (RMDQ) score at three and 12 months. Patient-response to a health transition question (TQ) was used as a secondary outcome. Using consensus thresholds for individual change on the RMDQ, and using the TQ, we calculated NNTs for improvement, and for benefit (incorporating one additional improvement gained or an additional deterioration prevented). In contrast to the small-to-moderate mean differences originally reported, NNTs were low, ranging from two to eight. Detailed results have been submitted for publication and will be presented at the meeting. Estimating NNT can be challenging due to difficulties in defining thresholds of individual change on continuous scales. However, using a TQ avoids some of these challenges.

Conclusion: In contrast to the small-to-moderate mean effects originally reported, corresponding numbers needed to treat for the interventions are attractive. Where possible, NNT should be considered for future use alongside more conventional reporting methods.