Notes on interval estimation of the attributable risk in cross-sectional sampling

Area of Science:

• Epidemiology
• Biostatistics
• Public Health

Background:

• Attributable risk (AR) is a key measure for assessing the public health impact of risk factors.
• Accurate interval estimation of AR is crucial for reliable public health assessments in cross-sectional studies.

Purpose of the Study:

• To compare the finite-sample performance of five asymptotic interval estimators for attributable risk (AR) in cross-sectional studies.
• To evaluate coverage probability and average length of AR interval estimators across various scenarios.

Main Methods:

• Simulation study comparing five asymptotic interval estimators for attributable risk.
• Analysis focused on coverage probability and average interval length.
• Scenarios included varying risk ratios, exposure probabilities, and sample sizes.

Main Results:

• Leung and Kupper's estimators showed poor coverage when the risk ratio (RR) was 1.
• Fleiss’s estimator demonstrated good coverage for large sample sizes (>=100) and moderate exposure probabilities (>=0.20), but with potential efficiency loss.
• Fieller's theorem-based estimator outperformed Wald's for RR >= 2.
• Leung and Kupper's estimator was preferable for large RR (>=4) and non-small exposure probability (>=0.05).

Conclusions:

• The choice of interval estimator for attributable risk depends on study parameters like sample size, exposure probability, and risk ratio.
• Fleiss’s estimator is a robust choice for general use with adequate sample sizes.
• Specific estimators show advantages under particular conditions (e.g., Fieller's for moderate RR, Leung and Kupper's for high RR).