An Item Response Theory-Informed Strategy to Model Total Score Data from Composite Scales

Area of Science:

• Psychometrics
• Statistical modeling
• Health outcomes research

Background:

• Composite scale data, common in therapeutics, is discrete and bounded, often summarized into a total score (TS).
• Item Response Theory (IRT) models are the gold standard for analyzing composite scales but require item-level data, which is not always available.
• Existing methods for TS analysis often use continuous variable (CV) models, which may not fully capture the discrete, bounded nature of TS data.

Purpose of the Study:

• To develop and investigate IRT-informed functions for analyzing total score (TS) data when item-level data is unavailable.
• To enhance the fit and accuracy of both continuous variable (CV) and bounded integer (BI) models for TS analysis.
• To provide a formal framework for linking IRT models with TS models and comparing their relative information content.

Main Methods:

• Proposed a novel method using IRT-informed functions for expected values and standard deviation in TS analyses.
• Investigated the application of these functions within both CV and BI models.
• Evaluated the method using simulated data and real-world clinical data.

Main Results:

• The proposed IRT-informed functions significantly improved the model fit for both CV and BI analyses.
• IRT-informed disease progression enabled precise estimation of latent variable parameters.
• IRT-informed standard deviation (SD) allowed for modeling deviations from homoscedasticity, improving accuracy.
• The methodology facilitated joint analyses of item-level and TS data.

Conclusions:

• IRT-informed functions offer a robust approach to analyze composite scale total scores, even without item-level data.
• This methodology enhances the precision and accuracy of TS analyses by incorporating information from underlying IRT models.
• The approach provides a quantitative basis for comparing different TS analysis methods and enables integrated analysis strategies.