[Over- and Underestimation of Success Rates]

Abstract

The patient reported outcome in interventional studies is often measured with questionnaires at baseline and after the intervention. A person whose difference in outcome exceeds a critical threshold (the minimal important difference, MID) is classified as a responder, otherwise as a non-responder. The generally low reliability of differences causes misclassifications. False positives and false negatives usually do not cancel out: an MID above the average difference results in an overestimated proportion of responders, while an MID below the average difference results in an underestimated proportion of responders. Such misclassifications can be substantial. We introduce a new and simple method for estimating the true proportion of responders which is based on the assumptions of classical test theory. The consequences of the method are demonstrated with empirical data. It is recommended to report the estimates of true responders. This applies to settings with one study group as well as to settings with an additional control group.