Latent variable mixture models to address heterogeneity in patient-reported outcome data

Area of Science:

• Health Outcomes Research
• Psychometrics
• Biostatistics

Background:

• Patient-reported outcome (PRO) data often exhibit heterogeneity in sociodemographic and health characteristics.
• Unobserved heterogeneity can obscure true patterns within study populations.
• Latent variable mixture models (LVMMs) offer a method to identify homogeneous subgroups within heterogeneous PRO data.

Purpose of the Study:

• To review latent variable mixture models (LVMMs) and their application to patient-reported outcome (PRO) data.
• To demonstrate the use of mixture item response theory (IRT) models for PRO data analysis.
• To explore subgroup identification within heterogeneous PRO datasets.

Main Methods:

• Focus on mixture item response theory (IRT) models, combining latent class analysis with IRT.
• Application of LVMMs to the physical component of the SF-12 in 1391 total hip replacement patients.
• Assessment of model fit and class discrimination statistics to select the optimal number of latent classes.

Main Results:

• A three-class model provided the best fit for the SF-12 physical component data.
• Significant dissimilarities in item parameter estimates were observed across the identified classes.
• Class membership was associated with patient sex, arthritis, and back pain.

Conclusions:

• LVMMs are a valuable tool for exploring response patterns in heterogeneous PRO data.
• These models can uncover underlying subgroups, enhancing the interpretation of health and well-being measures.
• Further applications of LVMMs in PRO research are warranted.