Determining a minimum clinically important difference between treatments for a patient-reported outcome

Patient-reported outcomes are important for assessing the effectiveness of treatments in many disease areas. For this reason, many new instruments that capture patient-reported outcomes have been developed over the past several decades. With the development of each new instrument, there is the ensuing question of what constitutes a minimum clinically important difference between treatments when using the new instrument. In this paper we describe a method for estimating a minimum clinically important difference between treatments for a patient-reported outcome through a desired difference in response rates for a definition of a responder. As well as being of interest in its own right, the use of a minimum clinically important difference on the patient-reported outcome scale is likely to lead to sample size advantages. We illustrate the method with data on neuropathic pain when responder is defined by requiring at least some improvement in the Patient Global Impression of Change and when responder is defined by existing responder definitions.