Longitudinal random effects models for genetic analysis of binary data with application to mastitis in dairy cattle

Area of Science:

• Veterinary Epidemiology
• Animal Genetics
• Statistical Modeling

Background:

• Mastitis is a significant economic and welfare concern in dairy cattle.
• Longitudinal data analysis is crucial for understanding disease dynamics during lactation.
• Accurate modeling of mastitis risk requires accounting for genetic and environmental factors.

Purpose of the Study:

• To apply Bayesian analysis to longitudinal mastitis records in first-lactation Holstein cows.
• To compare different statistical models for predicting mastitis probability over lactation.
• To evaluate the impact of incorporating random genetic regressions and autoregressive processes.

Main Methods:

• Utilized 3341 test-day binary mastitis records from 329 first-lactation Holstein cows.
• Developed three Bayesian models for latent normal variables, including fixed effects, random genetic effects, and autoregressive processes.
• Calculated posterior means of heritability and Bayes factors to compare model performance.

Main Results:

• Heritability estimates for mastitis varied significantly across models and lactation stages, with Model 3 showing higher values (e.g., 0.57 at day 14).
• Bayes factors indicated that Model 2 and Model 3 provided better fit than Model 1.
• The probability of mastitis for an average cow, using Model 2, remained relatively low (0.05-0.07) across lactation days.

Conclusions:

• Incorporating random genetic regressions (Model 2) and autoregressive processes slightly improved mastitis prediction.
• The choice of statistical model significantly influences heritability estimates for mastitis.
• Bayesian analysis provides a flexible framework for modeling complex disease patterns in dairy cattle.