Confidence intervals for exposure-adjusted rate differences in randomized trials

Area of Science:

• Biostatistics
• Clinical Trials Methodology

Background:

• Exposure-adjusted event rates are crucial for comparing interventions in clinical trials.
• Traditional methods using Poisson distribution assumptions can lead to undercoverage, particularly with rare events or over-dispersion.

Purpose of the Study:

• To review asymptotically robust methods for interval estimation of rate differences.
• To propose a novel interval estimator for rate differences that is robust to distributional assumptions.

Main Methods:

• Review of existing asymptotically robust methods for rate difference interval estimation.
• Development and evaluation of a modified interval estimator for rate differences.
• Assessment of finite sample properties, including performance with small samples, rare events, and over-dispersed data.
• Consideration of covariate adjustment capabilities.

Main Results:

• The proposed interval estimator demonstrates asymptotically nominal coverage for rate differences with arbitrary event count distributions.
• The new method exhibits good finite sample properties, avoiding substantial undercoverage in challenging scenarios.
• The approach is adaptable for covariate adjustment and implementable in standard statistical software.

Conclusions:

• The proposed method offers a robust and reliable approach for constructing confidence intervals for rate differences in clinical trials.
• This method addresses limitations of traditional Poisson-based intervals, improving accuracy for diverse event count data.
• The practical implementation and covariate adjustment features make it valuable for real-world clinical trial analysis.