The sample size of a clinical trial is the number of participants the trial aims to recruit. Sample size is a critical aspect of clinical trial design and has ethical and financial implications. The sample size depends on the target difference, the difference in outcome that the trial is powered to detect. This thesis aims to improve methods for specifying the target difference in randomised trials of osteoarthritis. I conducted a systematic review of sample size calculations in hip and knee osteoarthritis trials published in 2016. It found that most sample size calculations were poorly reported and could not be reproduced. The target difference in the sample size calculation was commonly justified by a published minimum clinically important difference (MCID). Several versions of the WOMAC (Western Ontario and McMaster Universities Osteoarthritis Index) were commonly used in hip and knee osteoarthritis trials. It was often unclear which version was used, hindering interpretation of trial results. I conducted a discrete choice experiment examining patient preferences when choosing between osteoarthritis medications. Duration of treatment effect was shown to be important to participants, viewed with similar importance to the amount of symptom relief provided and risks of the treatment. I analysed a cohort of people with osteoarthritis and showed that MCID estimates for the WOMAC varied across different follow-up time points. However, there was no visual trend in the change in MCID estimates over time. Longitudinal methods were feasible to calculate MCID estimates, but did not improve precision. A simulation study that I conducted found that the pattern of the treatment effect (its duration and consistency) affected the optimal statistical method of analysis for a randomised trial using the WOMAC as the primary outcome. Future research is needed to examine whether the findings are generalisable to different datasets, outcome measures and health conditions.
