Improving the specification of the target difference in the sample size calculation of a randomised trial of treatments for osteoarthritis