Validating the European Health Literacy Survey Questionnaire in people with type 2 diabetes: Latent trait analyses applying multidimensional Rasch modelling and confirmatory factor analysis

5 years ago

Validating the European Health Literacy Survey Questionnaire in people with type 2 diabetes: Latent trait analyses applying multidimensional Rasch modelling and confirmatory factor analysis

Gun Nordström, Bodil Wilde-Larsson, Kjell Sverre Pettersen, Hanne Søberg Finbråten, Øystein Guttersrud, Anne Trollvik

Aim To validate the European Health Literacy Survey Questionnaire (HLS-EU-Q47) in people with type 2 diabetes mellitus. Background The HLS-EU-Q47 latent variable is outlined in a framework with four cognitive domains integrated in three health domains, implying 12 theoretically defined subscales. Valid and reliable health literacy measurers are crucial to effectively adapt health communication and education to individuals and groups of patients. Design Cross-sectional study applying confirmatory latent trait analyses. Methods Using a paper-and-pencil self-administered approach, 388 adults responded in March 2015. The data were analysed using the Rasch methodology and confirmatory factor analysis. Results Response violation (response dependency) and trait violation (multidimensionality) of local independence were identified. Fitting the “multidimensional random coefficients multinomial logit” model, 1-, 3- and 12-dimensional Rasch models were applied and compared. Poor model fit and differential item functioning were present in some items, and several subscales suffered from poor targeting and low reliability. Despite multidimensional data, we did not observe any unordered response categories. Conclusion Interpreting the domains as distinct but related latent dimensions, the data fit a 12-dimensional Rasch model and a 12-factor confirmatory factor model best. Therefore, the analyses did not support the estimation of one overall “health literacy score.” To support the plausibility of claims based on the HLS-EU score(s), we suggest: removing the health care aspect to reduce the magnitude of multidimensionality; rejecting redundant items to avoid response dependency; adding “harder” items and applying a six-point rating scale to improve subscale targeting and reliability; and revising items to improve model fit and avoid bias owing to person factors.

Publisher URL: http://onlinelibrary.wiley.com/resolve/doi

DOI: 10.1111/jan.13342