Inference of nested variance components in a longitudinal myopia intervention trial

Area of Science:

• Ophthalmology
• Biostatistics
• Clinical Trials

Background:

• Myopia intervention trials generate complex nested repeated measurements.
• Correlations can arise from longitudinal observations and bilateral eye data.
• Accurate statistical modeling is essential for valid treatment effect estimation.

Purpose of the Study:

• To compare statistical models for analyzing nested repeated measurements in myopia intervention trials.
• To investigate the sources of correlation in bilateral eye data and longitudinal observations.
• To determine the best model for illustrating myopia intervention trial data.

Main Methods:

• Proposed three statistical models with different covariance structures for nested repeated measurements.
• Utilized commercial statistical software for model estimation and implemented Schwarz criterion for model selection.
• Conducted simulation studies to evaluate model performance and reanalyzed myopia intervention trial data.

Main Results:

• Atropine demonstrated effectiveness in reducing myopia progression rates.
• Myopia progression rates were found to be homogeneous across subjects.
• The statistical model with independent random effects for each eye provided the best fit.

Conclusions:

• Model selection is a critical step prior to making inferences from myopia intervention trial data.
• Incorrect model selection can lead to misattribution of correlation mechanisms.
• The findings emphasize the importance of accounting for all variance components accurately.