A Rasch model analysis of two interpretations of ‘not relevant’ responses on the Dermatology Life Quality Index (DLQI)

A nyílt hozzáférést az EISZ és a kiadó között létrejött "Read and Publish" szerződés biztosította. Open access was provided "Read and Publish" contract between EIS and the publisher.

Abstract

Purpose Eight of the ten items of the Dermatology Life Quality Index (DLQI) have a ‘not relevant’ response (NRR) option. There are two possible ways to interpret NRRs: they may be considered ‘not at all’ or missing responses. We aim to compare the measurement performance of the DLQI in psoriasis patients when NRRs are scored as ‘0’ (hereafter zero-scoring) and ‘missing’ (hereafter missing-scoring) using Rasch model analysis. Methods Data of 425 patients with psoriasis from two earlier cross-sectional surveys were re-analysed. All patients completed the paper-based Hungarian version of the DLQI. A partial credit model was applied. The following model assumptions and measurement properties were tested: dimensionality, item ft, person reliability, order of response options and diferential item functioning (DIF). Results Principal component analysis of the residuals of the Rasch model confrmed the unidimensional structure of the DLQI. Person separation reliability indices were similar with zero-scoring (0.910) and missing-scoring (0.914) NRRs. With zero-scoring, items 6 (sport), 7 (working/studying) and 9 (sexual difculties) sufered from item misft and item-level disordering. With missing-scoring, no misft was observed and only item 7 was illogically ordered. Six and three items showed DIF for gender and age, respectively, that were reduced to four and three by missing-scoring. Conclusions Missing-scoring NRRs resulted in an improved measurement performance of the scale. DLQI scores of patients with at least one vs. no NRRs cannot be directly compared. Our fndings provide further empirical support to the DLQI-R scoring modifcation that treats NRRs as missing and replaces them with the average score of the relevant items.