An improved score-type confidence interval for stratified risk differences involving rare events

Area of Science:

• Epidemiology
• Biostatistics
• Drug Development

Background:

• Stratified analysis is common in epidemiology, social sciences, and drug development for combining data across strata.
• Existing confidence intervals for risk differences in stratified analyses may perform poorly with rare events (incidence near zero), leading to coverage issues and computational failures.
• Algorithms can fail to produce valid confidence intervals when strata have zero events in both treatment arms.

Purpose of the Study:

• To evaluate the performance of common confidence interval methods for stratified risk differences when response probabilities are near zero or one.
• To propose an improved stratified Miettinen-Nurminen confidence interval.
• To address computational difficulties associated with rare events in stratified analyses.

Main Methods:

• Evaluation of existing confidence interval methods for stratified risk differences.
• Development and proposal of an improved stratified Miettinen-Nurminen confidence interval.
• Assessment of performance under boundary conditions (response probabilities near 0 or 1).

Main Results:

• Standard methods exhibit inadequate coverage probabilities for rare events.
• The proposed improved stratified Miettinen-Nurminen confidence interval demonstrates superior performance.
• The new method avoids computational issues encountered with rare events and boundary probabilities.

Conclusions:

• The proposed confidence interval method offers improved accuracy and reliability for stratified risk difference calculations.
• This method is particularly beneficial in scenarios involving rare events or when response probabilities approach boundary values.
• The improved method provides a robust solution for stratified analysis in epidemiological and clinical research.