Sample size calculation

A previous clinical trial of TCAs in individuals with chronic low back pain reported the mean (standard deviation (SD)) pain scores measured on the VAS to reduce to 5.70 (2.43) in the placebo group (n = 48) [34]. With n = 56 in each arm of the study, we have 90 % power to detect a difference of 1.5 (i.e. VAS to reduce to 4.20 (2.43) in the treatment group), which reflects the minimum clinically significant reduction in pain intensity (α = 0.05, two-sided significance).

In a recent study of individuals with low back pain, the control group (n = 22) had a pain score that reduced to a mean (SD) = 6.2 (2.8) on the Descriptor Differential Scale following treatment, while the pain score of the treatment group (TCA) (n = 19) reduced to a mean (SD) score = 4.5 (2.6) [15]. With an α = 0.05 and n = 60 in each arm of the study, we have 90 % power to detect a difference of this size (1.7), which reflects clinically significant reductions in pain intensity from ‘moderate’ to ‘mild’ or from ‘strong’ to ‘moderate’, which is equivalent to a decrease of 2 units (17 %) on the Descriptor Differential Scale (0–12 per descriptor) [35].

With 60 in each arm of the study, we will have 90 % power to detect a clinically relevant difference in disability (improvement of 13–16 % in disability or 3–4 points on the 24-point Roland Morris Disability Questionnaire after 26 weeks [36]) (α = 0.05, two-sided significance), which corresponds to a significant improvement in key functional activities, including walking and dressing.

More generally, with 60 per arm we have 90 % power to detect a difference of 0.60 SDs. With our primarily analyses involving adjustment for the baseline value of the outcome, we will have greater than 90 % power according to the size of the baseline-follow-up correlation. Given our previous experience in such studies we expect a maximum dropout rate of 20 % so we will recruit a total of 150 (75 in each arm of the study) [37].