Foundations of the minimal clinically important difference for imaging.

This article develops a generic conceptual framework for defining and validating the concept of minimal clinically important difference. We propose 3 approaches. The first uses statistical descriptions of the population ("distribution based"), the second relies on experts ("opinion based"), and a third is based on sequential hypothesis formation and testing ("predictive/data driven based"). The first 2 approaches serve as proxies for the third, which is an experimentally driven approach, asking such questions as "What carries the least penalty?" or "What imparts the greatest gain?" As an experimental approach, it has the expected drawbacks, including the need for greater resources, and the need to tolerate trial and error en route, compared to the other 2 models.