Quantitative assessment of changes in patients' constructs of quality of life: an application of multilevel models.

It has often been proposed that quality of life (QL) instruments should account for potentially changing conceptualisations of QL as patients adapt to disease (response-shift). Most instruments do not do this, and some that do are relatively complicated and burdensome for patients. The extent to which patients reconceptualise QL is unknown, and it is unclear whether this additional complication is necessary. This paper reviews existing methods for assessing response-shift and introduces an alternative approach using multilevel models. The method is described using data from a cancer clinical trial, and its performance is evaluated in simulations. The models reveal substantial response-shift in these cancer patients. Simulations under the null hypothesis of zero response-shift confirm that the method performs correctly in terms of its risk of type I error, and further simulations illustrate its statistical power to detect pre-defined levels of response-shift. The method is a relatively simple extension of familiar multiple regression models and yields parameters with a simple interpretation, representing the changes in importance of QL domains over time. It can be applied to existing datasets collected with other analysis strategies in mind and may have application in the investigation of response shifts and other manifestations of adaptation.