Sequential analysis of latent variables using mixed-effect latent variable models: Impact of non-informative and...

Area of Science:

• Clinical Trials
• Biostatistics
• Psychometrics

Background:

• Sequential methods enable early trial stopping, reducing sample sizes while maintaining statistical error rates.
• Latent variables, such as quality of life or depression, require specialized models for analysis.
• Missing data in clinical trials can be non-ignorable, potentially biasing results.

Purpose of the Study:

• To evaluate the impact of informative and non-informative missing data on the double triangular test (DTT).
• To compare the performance of DTT with mixed-effects Rasch models (MRM) versus traditional score-based methods for latent variables.

Main Methods:

• The study utilized the double triangular test (DTT) combined with either the mixed-effects Rasch model (MRM) or a traditional score-based method (S).
• Simulations investigated type I and II errors, power, and sample size efficiency under various missing data scenarios (informative and non-informative).

Main Results:

• DTT with MRM demonstrated appropriate type I error rates, slightly increasing with informative missing data.
• DTT with MRM achieved high power (close to 0.95), while the S method was underpowered (approx. 23% reduction).
• DTT with MRM required fewer patients and was less affected by increasing missing data proportions compared to the S method.

Conclusions:

• Mixed-effects Rasch models (MRM) integrated with sequential analysis (DTT) offer a more powerful approach for latent variables than traditional methods.
• This combined approach remains robust even with non-informative or informative missing data, enhancing clinical trial efficiency.