Salmonella serosurveillance: different statistical methods to categorise pig herds based on serological data

Area of Science:

• Veterinary Immunology
• Statistical Modeling
• Food Safety

Background:

• Salmonella seroreactors in pig herds pose a significant food safety concern.
• Accurate classification of herds based on serological data is crucial for effective control.

Purpose of the Study:

• To compare three statistical methods for categorizing pig herds into high and low Salmonella seroreactor groups.
• To evaluate the agreement between these methods and a regulatory approach.

Main Methods:

• Comparison of semi-parametric quantile regression, a naive mean-based method, and an activity region finder with random forest regression.
• Analysis of Salmonella-specific antibody S/P ratios from 13,649 pigs across 314 Belgian herds.
• Calculation of corrected and pair-wise agreement among methods, including a comparison with the FASFC-adopted method.

Main Results:

• Nearly 14% of herds were classified as high seroreactors by at least one method.
• The statistical methods demonstrated a high level of agreement, exceeding 92% corrected agreement.
• All methods showed advantages and disadvantages, influencing their suitability for specific policy objectives.

Conclusions:

• The evaluated statistical methods provide reliable tools for classifying pig herds based on Salmonella serological data.
• High agreement among methods supports their utility in developing targeted Salmonella control policies.
• The study offers recommendations for method selection based on specific classification objectives and policy requirements.