Asymptotic confidence interval construction for proportion difference in medical studies with bilateral data.

Area of Science:

• Medical Statistics
• Biostatistics
• Clinical Trial Design

Background:

• Bilateral dichotomous data are prevalent in comparative medical research involving paired organs.
• Accurate statistical analysis is crucial for interpreting results from studies comparing treatments for conditions affecting paired organs like eyes or ears.

Purpose of the Study:

• To evaluate confidence interval estimators for proportion differences in bilateral binary data.
• To assess the impact of ignoring data dependence on statistical accuracy.
• To recommend reliable methods for analyzing paired organ data in medical studies.

Main Methods:

• Simulation studies comparing various confidence interval estimators (Wald-type, Fieller, likelihood ratio, score, bootstrap).
• Analysis under both dependence and independence models for bilateral binary data.
• Performance evaluation based on coverage probability and expected width.

Main Results:

• Ignoring the dependence in bilateral data significantly distorts coverage probabilities.
• Confidence intervals derived from dependence models (Wald-type, score-type, bootstrap) demonstrate satisfactory performance across sample sizes.
• Empirical coverage probabilities closely align with nominal confidence levels when using dependence models.

Conclusions:

• Accounting for data dependence is essential for accurate confidence interval estimation in bilateral studies.
• Recommended methods include Wald-type, score-type, and bootstrap confidence intervals utilizing dependence models.
• These methods ensure reliable statistical inference for comparative medical research with paired data.