Repeatability of a Rasch model of the AQ20 over five assessments

Abstract

Rasch analysis is now used widely to assess the measurement properties of health status questionnaires. This study tested the stability of the AQ20 - a dichotomous-response measure of health status in asthma, using parameters estimated by a Rasch model. One hundred forty-four asthma patients completed the AQ20 on five occasions over 3 months. At visit 1, two items showed significant misfit and were deleted. At each visit, the overall mean item-person and item-trait interaction statistics from the remaining 18 items (AQ18) were very similar. The repeatability of the item calibrations was excellent (intraclass correlation coefficient 0.95), despite the patients' health having changed (repeated-measures ANOVA: FEV(1) and AQ18 score p<0.0001). Tests of differential item functioning (DIF) over time showed that one item increased in severity. This item refers to ability to garden, and changes in response patterns could be related to seasonal changes over the study period. We conclude that this study has highlighted the usefulness of multiple repeat assessments which allow items to be tested for DIF over time. This is important as inclusion of 'time-dependent' items in a questionnaire may reduce the reliability of the instrument.