A comparison of breeding value predictors for longevity using a linear model and survival analysis

Area of Science:

• Animal Breeding and Genetics
• Quantitative Genetics
• Reproductive Biology

Background:

• Accurate prediction of breeding values for traits like longevity is crucial for genetic improvement in livestock.
• Traditional methods may not fully account for the complexities of survival data, including censored records.

Purpose of the Study:

• To compare sire breeding values for longevity estimated using different methodologies.
• To evaluate the impact of data censoring on the accuracy and correlation of these breeding values.

Main Methods:

• Breeding values were calculated using phenotypic averages, Best Linear Unbiased Prediction (BLUP), and survival analysis.
• Analyses were performed using both uncensored records and a combination of censored and uncensored records.
• Two datasets, small and large herds, were utilized to assess consistency.

Main Results:

• Sire rankings differed significantly across the prediction methods.
• Phenotypic averages showed weak correlations with other methods (≤ 0.46).
• REML BLUP and survival analysis (uncensored data) exhibited strong negative correlations (≤ -0.91), indicating agreement in ranking but opposite interpretation (longevity vs. culling risk).
• Inclusion of censored records in survival analysis weakened correlations with REML BLUP (≤ -0.60).

Conclusions:

• Methodological choices, particularly the handling of censored data, substantially influence breeding value estimation for longevity.
• Survival analysis and REML BLUP are more robust predictors than simple phenotypic averages.
• The choice of method and data inclusion criteria impacts sire selection outcomes.