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