Comparison of methods for estimation of individual-level prevalence based on pooled samples

Area of Science:

• Veterinary epidemiology
• Biostatistics
• Disease surveillance

Background:

• Estimating animal disease prevalence from pooled samples is crucial for effective surveillance.
• Simple random sampling of pooled samples presents unique statistical challenges.
• Accurate estimation requires careful consideration of test characteristics (sensitivity and specificity).

Purpose of the Study:

• To review and compare frequentist and Bayesian statistical approaches for estimating animal-level disease prevalence using pooled samples.
• To identify the preferred estimation method based on prevalence scenarios and knowledge of test performance.
• To provide guidance on selecting appropriate methods when dealing with imperfect test results.

Main Methods:

• Review of frequentist (large-sample theory, exact estimates) and Bayesian methodologies.
• Analysis of pooled egg samples (n=20 per pool) for Salmonella enteritidis contamination.
• Comparison of estimation approaches under varying assumptions of sensitivity and specificity.

Main Results:

• When sensitivity and specificity are known or perfect, both large-sample and exact estimates are viable.
• When sensitivity and specificity are unknown, Bayesian methods or large-sample estimates are used, with exact methods required if frequentist lower confidence limits are negative.
• The study applied these methods to Salmonella enteritidis prevalence estimation in pooled egg samples.

Conclusions:

• The choice between frequentist and Bayesian methods depends on the prevalence level and knowledge of test accuracy.
• Bayesian approaches offer exact estimates and are particularly valuable when test characteristics are uncertain or frequentist methods fail.
• Accurate disease prevalence estimation is vital for targeted control strategies in animal populations.