Measures of Clinical Meaningfulness and Important Differences

Abstract

Measures of clinical significance have been in use for several decades as a means of interpreting clinical findings and patient-reported outcomes. The most common of these measures is the minimal clinically important difference. With the rise in popularity of measurements of clinical significance, several common misconceptions have arisen that may impact their interpretation and application to clinical practice. The purpose of this article is to present a schema for understanding measurement of clinical significance and use this to highlight the reasons why misuse and misinterpretation have occurred. A new measure of clinical significance is then defined that is intended to be resistant to these issues. Clinical significance has long been a topic of importance to researchers looking to make their findings interpretable and has been quantified in diverse ways.1 Recently, there has been rapidly increasing interest in and use of an assortment of minimal (clinically) important difference measures. The range of their use is illustrated by the publications of reviews and meta-analyses in pain relief,2 cognitive interventions for dementia,3 and CT densitometry for patients with chronic obstructive pulmonary disease.4 Consensus has not been reached for how clinical significance should be defined. Despite this, current methods fall into 2 distinct approaches. The first estimates measurement error levels, and the other quantifies the ability of the instrument to predict clinical outcomes of interest. The conceptual differences between the 2 approaches have not been clearly delineated in the literature. Further, additional conceptual and practical issues exist for measures using the second approach because it has not previously been framed as a clinical prediction problem. It is the aim of this paper to develop a framework to guide researchers in the use of clinical importance measures and to introduce a new methodology for predicting clinically meaningful change. We first propose 2 types of clinical significance measures relating to what we call the Detection and Clinical Prediction Problems. Next, we discuss weaknesses of existing measures of clinical prediction within this unifying framework. Finally, we define a new measure of clinical significance using predictive values and demonstrate its use with simulated data.