Statistical methods in ophthalmology: an adjustment for the intraclass correlation between eyes

Area of Science:

• Ophthalmology
• Biostatistics
• Medical Data Analysis

Background:

• Ophthalmologic studies often involve data from both eyes of an individual.
• The high correlation between eyes is frequently overlooked in statistical analyses.
• Standard methods may not adequately address correlated data structures.

Purpose of the Study:

• To present valid statistical methods for analyzing ophthalmologic data with correlated outcomes.
• To highlight the invalidity of treating eyes as independent variables when intraclass correlation exists.
• To provide accurate p-values for ophthalmologic research.

Main Methods:

• Development of statistical methods for correlated outcome variables (normal and binomial distributions).
• Comparison of proposed methods against the frequently-used independent analysis approach.
• Simulation of data with varying degrees of intraclass correlation between eyes.

Main Results:

• The independent analysis method yields true p-values 2-6 times larger than nominal p-values.
• Intraclass correlation significantly inflates Type I error rates when ignored.
• Accurate statistical methods are crucial for reliable ophthalmologic findings.

Conclusions:

• Current independent analysis methods are invalid for correlated bilateral ophthalmologic data.
• Correct statistical approaches are necessary to avoid misleading conclusions in eye research.
• Findings are potentially applicable to other medical fields with correlated replicate observations, such as otolaryngology.