Methods to detect response shift in quality of life data: A convergent validity study

Abstract

When measuring changes in quality of life (QL) with a pretest-posttest design, response shift can affect results. We investigated the convergent validity of three approaches to detect response shift. (1) In the thentest approach, response shift is measured using a retrospective judgment of pretest QL-levels (thentest). (2) In the anchor–recalibration approach response shift is measured, assessing shifts in patients’ individual definitions of the scale-anchors (worst and best imaginable QL) over time. (3) In the Structural Equation Modeling (SEM) approach response shift is indicated by mathematically defined changes in factor solutions and variance–covariance matrices over time. Prior to and three months after invasive surgery, 170 cancer patients completed the SF-36, the Multidimensional Fatigue Inventory (as pre-, post-, and thentest), and the anchor–recalibration task (as pre-, and posttest). Results showed agreement between the thentest and SEM approach on the absence (6 scales) and presence (2 scales) of response shift in 8 of the 9 scales. For the ninth scale both methods detected response shift, but in opposite directions. Possible explanations for this discrepancy are discussed. The anchor–recalibration task agreed with the other approaches on only the absence of response shift in 4 of the 7 scales. The convergent results of thentest and SEM support their validity, especially because they use statistically independent operationalizations of response shift. In this study, recall bias did not invalidate thentest results.