Genetic analyses of binary longitudinal health data in small low input dairy cattle herds using generalized linear mixed models

Genetic parameters were inferred for the health traits mastitis, metritis, retained placenta, ovarian cysts and acetonemia from 1247 Brown Swiss cows in first parity kept in 53 organic and low input farms in Switzerland. For genetic analyses, univariate animal and sire models, repeatability animal and sire models, and random regression sire models (RRM) in a “generalized linear mixed model (GLMM) context” were applied. The five health traits were defined as binary data, count data, and longitudinal binary data in the interval between −1 and 120 d in milk (DIM). Firstly, binary data were analyzed by applying linear animal and sire models, and threshold animal and sire models with a probit link function. Secondly, data of total number of disease cases recorded within the defined time span were analyzed by using GLMM animal and sire models with a log link function for Poisson distributed count data. Thirdly, for longitudinal health data, linear repeatability animal and sire models, linear sire RRM, threshold animal and sire repeatability models, and threshold sire RRM with a probit link function were applied. Disease incidences of the five health disorders in organic farms were on a generally low level, with a highest incidence of 5.78% for mastitis within the time span of 120 d. With regard to mastitis, moderate heritabilities with an average value of 0.15 were realized from univariate models and binary data, and from GLMM with the log link function and count data. Heritabilities for mastitis were smaller (<0.10) when using the longitudinal data structure in combination with repeatability models and RRM. Repeatabilities and heritabilities for longitudinal data as realized from repeatability models were on a quite similar level. Only for longitudinal ovarian cysts, heritabilities substantially differed from repeatabilities. Heritability was 0.02 from the animal model and 0.01 from the sire model, but repeatabilities were 0.14, which indicates a substantial permanent environmental effect. Daily heritabilities for all health traits from linear and threshold RRM at the beginning of lactation and at the end of the defined interval were three times higher than corresponding heritabilities in the middle of lactation. Bayesian information criterion (BIC) and heritabilities themselves favored threshold models over linear models. However, linear models converged more easily than threshold models, and genetic parameter estimates had smaller standard errors. Similar BIC values were found when comparing animal with sire models, although generally higher heritabilities were realized from sire models. For RRM applications, BIC was smaller and heritabilities were higher for linear sire compared to threshold sire models.