Evaluation of Approaches to Analyzing Continuous Correlated Eye Data When Sample Size Is Small

Area of Science:

• Ophthalmology
• Biostatistics
• Statistical analysis

Background:

• Correlated eye data analysis is crucial in ophthalmology research.
• Small sample sizes pose challenges for standard statistical methods.
• Choosing the correct statistical approach is vital for reliable results.

Purpose of the Study:

• To evaluate the performance of statistical methods for continuous correlated eye data with small sample sizes.
• To compare type I error rates and statistical power across different analytical approaches.

Main Methods:

• Simulated correlated continuous eye data under various conditions (sample size, inter-eye correlation, effect size).
• Analyzed data using paired t-test, two sample t-test, generalized estimating equations (GEE) Wald and score tests, and linear mixed effects models (LMM).
• Compared performance based on type I error rates and statistical power, validated with real datasets.

Main Results:

• In design 1 (eyes in different groups), paired t-test and LMM showed nominal type I error and higher power.
• In design 2 (eyes in same group), two sample t-test offered better type I error control but lower power.
• GEE Wald tests inflated type I error, while GEE score tests had lower power.

Conclusions:

• Common statistical methods may underperform with small sample sizes for correlated eye data.
• Paired t-test and LMM are recommended when eyes are in different groups.
• Averaging eye data with t-test is suitable when eyes are in the same group; study design is key.