Joint modelling of mixed outcome types using latent variables.

Area of Science:

• Statistics
• Biostatistics
• Statistical Modeling

Background:

• Correlated outcomes with mixed data types present analytical challenges.
• Latent variable models offer a framework to address outcome correlations.

Purpose of the Study:

• To quantify the consequences of using latent variables for correlated mixed-type outcomes.
• To contrast the effects of latent variables on marginal inference versus jointly normal outcomes.
• To demonstrate the practical benefits of joint models through simulation and real-world examples.

Main Methods:

• Theoretical and numerical calculations to quantify latent variable effects.
• Comparison of marginal inference in latent variable models versus jointly normal models.
• Simulation study to assess efficiency and bias reduction.
• Application to an osteoarthritis dataset.

Main Results:

• Latent variables significantly impact marginal inference, differing from jointly normal models.
• Joint models demonstrate substantial efficiency gains and bias reduction in simulations.
• The analysis highlights practical differences in osteoarthritis data interpretation.

Conclusions:

• Incorporating latent variables is crucial for accurate analysis of correlated mixed-type outcomes.
• Joint modeling provides superior statistical properties compared to traditional approaches.
• These methods offer practical advantages in fields like osteoarthritis research.