Clinical Scorecard: Accounting for Eye Correlation Improves Statistical Accuracy

At a Glance

Key Highlights

• Treating both eyes as independent leads to inflated false-positive rates.
• Mixed effects models and GEEs are preferred for eye-level predictors.
• Averaging measurements may suffice for subject-level predictors.
• Single-eye analysis shows lower power, especially with low interocular correlation.
• Incorrect modeling can mislead statistical power assessments.

Guideline-Based Recommendations

Diagnosis

• Avoid treating measurements from both eyes as independent observations.

Management

• Utilize mixed effects models or GEEs for eye-level predictors.

Monitoring & Follow-up

• Regularly assess model performance based on predictor type.

Risks

• Inflated Type-I error rates due to incorrect model specification.

Patient & Prescribing Data

Not applicable; focused on statistical methods in research.

Statistical modeling strategies impact research outcomes.

Clinical Best Practices

• Select statistical models that account for interocular correlation.
• Use mixed effects models for eye-level predictors.
• Consider averaging for subject-level predictors.

References

• Optometry and Vision Science