Regression methods when the eye is the unit of analysis.

Area of Science:

• Ophthalmology and Vision Science
• Biostatistics
• Statistical Modeling

Background:

• Vision research frequently uses the eye as the unit for outcomes and covariates.
• Measurements from a person's two eyes are correlated but not perfectly.
• Analytic methods for eye-specific outcomes using bilateral eye data have advanced.

Purpose of the Study:

• To illustrate regression approaches for analyzing correlated eye outcomes.
• To address eye- and subject-specific exposures and confounders.
• To compare different analytical strategies for vision research data.

Main Methods:

• Demonstrated regression techniques for paired eye data.
• Considered cross-sectional and longitudinal study designs.
• Examined discrete, continuous, and time-to-event outcomes.

Main Results:

• Ignoring eye correlation leads to underestimated P-values and confidence intervals.
• Analyzing eyes separately or persons reduces precision and data utilization.
• Regression models treating the eye as the unit maximize data, enhance interpretability, and use information efficiently.

Conclusions:

• Regression models with the eye as the unit of analysis are optimal for many vision research studies.
• These models appropriately account for the correlation between paired eye outcomes.
• This approach offers enhanced statistical power and more accurate covariate-outcome associations.