Applying linear mixed models to estimate reliability in clinical trial data with repeated measurements

Area of Science:

• Psychiatric Health Sciences
• Biostatistics
• Psychopharmacology

Background:

• Reliability is crucial in psychiatric health sciences, often assessed using repeated measures.
• Traditional methods may not fully capture the complexity of reliability in longitudinal studies.

Purpose of the Study:

• To demonstrate the application of linear mixed models for deriving test-retest reliability with continuous or quasi-continuous data.
• To explore the advantages of mixed models in handling complex variance structures and covariate effects.

Main Methods:

• Utilizing linear mixed models to analyze repeated measures data.
• Implementing complex variance structures, including random intercepts, random slopes, and serial correlation.
• Applying the methodology to data from clinical trials on schizophrenia treatments.

Main Results:

• Linear mixed models provide a flexible framework for reliability analysis, accommodating various covariance structures.
• Complex variance structures yield time-dependent reliability functions.
• Classical methods are a special case when the variance structure simplifies to a random intercept.

Conclusions:

• Linear mixed models offer a powerful and versatile approach to assessing test-retest reliability in psychiatric research.
• This methodology allows for nuanced reliability estimation, accounting for time-dependent effects and covariates.
• The findings are applicable to evaluating treatment efficacy in conditions like chronic schizophrenia.