Comparison between repeatability, reproductive stage stratified repeatability, and relative risk models for prediction of breeding values for functional survival in rotationally crossbred sows.

backgroundThe aim of this study was to compare different statistical models for predicting breeding values for sow survival with right-censored phenotypes from rotationally crossbred and commercial sows. We tested two hypotheses. First, we hypothesized that single-trait relative risk models predict more accurate breeding values than single-trait linear repeatability models. Second, we hypothesized that a reproductive stage stratified linear repeatability model predicts more accurate breeding values than the standard single-trait linear repeatability models. The single-trait models predict breeding values for survival between farrowings, while the reproductive stage stratified models predict breeding values for both survival between a farrowing and the next service, and survival between a service and the next farrowing. The validation criterion was the Pearson correlation between adjusted phenotypes for the lifetime number of litters produced and predicted breeding values for survival converted to lifetime number of litters produced. All validation criteria were compared to one another and against zero using appropriate statistical tests and correction for multiple tests. Each model was constructed with two different multi-breed relationship matrices to ensure that the results were not affected by the choice between them.

resultsThe values of the validation criteria for the single-trait models were significantly larger than zero and similar (0.02). The values of the validation criteria for the reproductive stage stratified linear repeatability models were both significantly larger than zero and significantly larger than those from the single-trait models (0.04 vs. 0.02).

conclusionsThe relative risk and linear repeatability single-trait models for survival between subsequent farrowings predicted equally accurate breeding values (0.02), while the linear repeatability two-trait models for survival from services to their subsequent farrowings and farrowings to the subsequent services predicted more accurate breeding values than the single-trait models (0.04 vs. 0.02). However, the accuracy of breeding values was small for all models because the survival phenotypes used for prediction were censored and the heritability of complete survival times was moderate (8-9%). Therefore, the comparison would benefit from reevaluation in other populations, and the models should be improved upon before implementation in practical breeding programs.