PS40 Evaluating coherency of the Dermatology Life Quality Index scales using a nonparametric item response theory method: Mokken Scale Analysis

Abstract

Mokken Scale Analysis is a nonparametric approach within item response theory that evaluates whether a set of items forms a hierarchical, cumulative scale measuring a single latent trait. The method computes scalability coefficients and standard errors, where scalability refers to how well a set of items forms a unidimensional scale, meaning items measure the same underlying latent trait. The scalability, monotonicity, and invariant item ordering of the Dermatology Life Quality Index (DLQI) were evaluated using Mokken Scale Analysis of data from a European multicentre cross-sectional study of patients attending dermatological outpatient clinics in 13 countries. It is important to confirm the validity aspects of the most widely used questionnaire in dermatology. We calculated item-pair scalability (Hij), item scalability (Hi, measuring how strongly each DLQI item fits into the cumulative Mokken scale), Loevinger’s scale (H) coefficients, and Gplus/OPlus Guttman errors (person-level error counts showing how much an individual’s response pattern deviates from an ideal hierarchical scale). Additionally, we assessed the monotonicity and manifest invariant item ordering coefficient (MIIO) of items using HT-statistics. This is a practical method to check invariant item ordering – the difficulty consistent across all respondents, whatever the trait level. Reliability indices were determined, including alpha, omega and omega total. All DLQI items exhibited moderate-to-strong scalability (Hi ≈ 0.46–0.61), supporting monotone homogeneity. The overall H (≈ 0.52, 95% confidence interval 0.51–0.54) indicates a moderately strong cumulative structure. The MIIO HT-value (∼ 0.19) suggests weak invariant item ordering, which is consistent with observations that certain context-dependent items (e.g. work or study) can disturb strict ordering. A majority of respondents (97.3% < Tukey fence) had few or no violations, suggesting strong consistency and coherency with the ideal Guttman pattern and supporting the monotone homogeneity assumption. Reliability indices were high (Cronbach’s alpha ≈ 0.90, omega total ≈ 0.91), indicating high internal consistency. The DLQI forms a moderately scalable, unidimensional measure with strong monotonicity and coherent scales.