Objective: Using serial N-of-1 trials and subsequent analysis with Bayesian methods may allow study of therapies using small numbers of subjects. Our research questions were: (1) Can serial N-of-1 trials analyzed with Bayesian statistical techniques be used to estimate the population effect of a therapeutic intervention? (2) Compared to placebo, how likely is it that low-dose amitriptyline therapy in children aged 10-18 years with active polyarticular-course juvenile idiopathic arthritis (JIA) results in a significant improvement in pain?
Methods: Six children (age 10.3-16.3 yrs, 4 girls) were enrolled. There were 3 pairs of randomized, double-blinded treatments (amitriptyline 25 mg or placebo) per participant. Each treatment lasted 2 weeks, with a 1 week washout. The primary outcome was pain, measured by 10 cm visual analog scale. Assessments were at the beginning and end of each treatment. A Bayesian statistical model was used to determine the treatment effect. Values < 0 indicated superiority of amitriptyline.
Results: Bayesian techniques were used successfully to obtain estimates of population effect, despite the small number of participants. The mean treatment effect for pain was 0.67 (SD 0.89, 95% credible interval -0.99, 2.55). The probability that the treatment effect was < 0 was only 16%.
Conclusion: These methods can be used successfully to estimate population effects when sample sizes are small. It is unlikely that amitriptyline reduced pain by a clinically significant amount in these children with polyarticular JIA. These methods may be particularly suited to pilot studies and the study of rare illnesses.