Tree-based item-response theory model for evaluating differential item functioning in patient-reported outcome...

Area of Science:

• Health Informatics
• Psychometrics
• Statistical Modeling

Background:

• Patient-reported outcome measures (PROMs) are crucial for assessing health status, but their validity can be compromised by differential item functioning (DIF).
• DIF arises from systematic differences in item responses among individuals with the same underlying health status, potentially leading to inaccurate conclusions and clinical decisions.
• Traditional methods for DIF detection may fail when covariates are unknown, necessitating robust alternative approaches.

Purpose of the Study:

• To introduce a user-friendly web application for implementing tree-based item response theory (IRT) models to detect differential item functioning (DIF) in patient-reported outcome measure (PROM) data.
• To provide a tool that accommodates potentially heterogeneous populations and unknown DIF-related covariates.
• To facilitate accurate interpretation of PROM scores by identifying sources of sample heterogeneity.

Main Methods:

• Development of an R Shiny web application for tree-based IRT modeling.
• The application supports flexible model specification, interactive data visualization, and customizable settings for various data types.
• Includes a tutorial for data preparation, model selection, and result interpretation.

Main Results:

• The web application allows interactive data upload (.CSV, .XLSX) and tests for DIF in dichotomous and polytomous items.
• Provides recommendations for parameter selection based on simulation studies.
• Outputs coefficients, item parameters, and plots to identify potential DIF sources.

Conclusions:

• The developed web application is an accessible and valuable tool for researchers and clinicians.
• It enhances the understanding of sample heterogeneity in PROM data caused by DIF.
• Facilitates more reliable inferences from PROM scores in clinical and research settings.