Minimal change is sensitive, less specific to recovery: a diagnostic testing approach to interpretability

The pursuit of interpretability of longitudinal measures of patient outcome has led to several methods for defining minimal amounts of change or final states that are important. Little is known about the best method. The purpose of this study was to directly compare methods using diagnostic utility to evaluate their usefulness.

Secondary analysis of longitudinal cohort data of persons attending physiotherapy for shoulder pain. Disability of the arm, shoulder, and hand outcome fielded at baseline and 3 months. Published methods were used to define positive response in scores: minimal change, final state, and combined change and final state. Proportions described as improved were compared (Kappa) and diagnostic testing techniques used to evaluate the strengths of each.

Only moderate agreement was found between methods (Kappa=0.47). Minimal clinically important differences were most sensitive but not specific. Final states were less sensitive, more specific, and most accurate. Combinations were slightly less specific.

A new approach allowed us to evaluate the relative merits and risks of different approaches to interpreting longitudinal patient outcomes. Our study points to a combination of change greater than error and/or a final score within general population norms as being the most clinically sensible with strong diagnostic accuracy.