How can we design trials to detect clinically important changes in disease severity?

1. Forty-eight British rheumatologists judged the change in disease activity in 50 sets of patient data drawn from life and presented as 'paper patients'. Each set comprised two values, recorded a year apart, for 10 commonly measured clinical variables. Doctors recorded the size of improvement or deterioration on a visual analogue scale (VAS) and whether the change was clinically important or not. 2. Clinical judgement policies were modelled using linear regression of the clinical variables on the VAS score. 3. Doctors showed little agreement over which patients had improved and which had not. Possible reasons could be discovered by inspecting their judgement policies. 4. The weights attributed to the clinical variables differed considerably between doctors. Furthermore weights the doctors believed they attached to the variables frequently differed from the weights in the regression models. 5. These models could be used to calculate the smallest change required in any clinical variable before it would be considered clinically important. However, the size of such changes was often outside the observed clinical range suggesting that the use of single outcome variables is unrealistic. 6. The modelling procedure described can be applied during the planning stage of the trial to participating physicians, patients, health economists or any other group having an interest in the results. The models themselves can then be used to reach a consensus policy for judging what is a successful outcome. This may be expressed as a linear combination of specific outcome measures. Its use may improve the power of clinical trials and the relevance of their results.