Identification of on-farm recorded data for the prediction of disease in dairy cattle

Abstract

Identification of cows at increased disease risk during the transition period is necessary to reduce the negative economic impact of disease and to improve animal welfare. Although timely identification of at-risk cows is a vital component of health management, it is challenging in modern dairy herds, where staff manage an increasing number of cattle. The consequent reduction in time available for individual animal observation has created a need for the development of decision support tools which facilitate individual cow monitoring. However, uncertainty exists as to which measurable traits best reflect cow health status, especially in the dry and transition periods where little monitoring of individual cows is performed. Therefore, the objectives of this project were 1) to quantify the effect of early lactation disease on productivity 2) to identify variables of routinely recorded herd data which could be used for disease prediction or as risk factors for disease and 3) to assess the feasibility of using such indicators in predictive disease modelling. Retrospective analyses were performed on 482 cow-lactations from the Langhill herd of Holstein cattle. Cow-lactations were assigned to 1 of 4 health groups based on disease incidence in the first 30 days of lactation. These groups were no clinical disease (NCD; n = 335, reproductive (REP; n = 77) (which included cases of retained placenta and metritis), subclinical mastitis (SCM; n = 53) (determined by somatic cell counts) and metabolic (MET; n = 17) (which included cases of displaced abomasum, ketosis, hypomagnesaemia and hypocalcaemia). The data were analysed using descriptive statistics, mixed models, and generalised linear mixed models, with a logit link, in SAS 9.3 and GenStat 16. There were significant differences in average milk yield between health groups throughout lactation. In the first 30 days of lactation, NCD cows had significantly higher (p<0.01) daily milk yield than either REP, SCM or MET cows. Days to first observed heat and first service were significantly higher in MET cows than all other groups (p<0.01) and was extended by 27 days compared to NCD cows. No difference existed between services per conception or calving interval across all groups however the 100 day in-calf rate was reduced amongst cows with disease compared to cows without disease. Preceding disease, milk yield at dry-off and the ratio of energy corrected milk to body energy content were found to be significantly different between health groups; both measures were significantly higher in SCM cows compared to REP and MET cows. Additionally, in the first 15 days of the dry period preceding disease diagnoses, REP cows had a significantly (p=0.02) greater rate of change in body energy content than NCD cows; -18.3±7.44 MJ per day vs. 0.6±5.11 MJ per day, respectively. Overall change in body energy content between dry off and calving was significantly greater (p<0.001) in REP cows than both NCD and SCM cows. The predictive ability of candidate indicators identified as being significantly different between health groups was assessed using further statistical analysis. The distribution of each candidate indicator was investigated before Pearson and Spearman correlation tests were used to quantify the relationships between indicators. Single candidate models, employing generalised linear mixed modelling with random effect for cow, were used to test the effect of each candidate indicator on each response measure (health group). Dry period length, change in live weight and body energy content across the dry period, condition score and body energy content at dry off and the rate of change in body energy content in the first 15 days of lactation were significant predictors (p<0.05) of reproductive disorders while the year of calving and live weight at calving were significant predictors (p<0.05) of subclinical mastitis when included in single candidate models. Multivariate models for each of the disease response measures (REP, SCM and MET) were developed using combinations of the candidate indicators as explanatory variables. Despite some highly significant relationships between the candidate indicator variables and response measures, the multivariate models developed do not currently have potential to predict risk of disease at an acceptable level of accuracy, as very few significant effects were found. This can be explained by the large individual cow variance components and a low incidence of disease in the current data set. Future research should focus on tracking candidate indicator data in individual cows with a view to establishing a baseline for each cow. This would allow each cow to be used as its own control, with deviations from the normal indicating potential disease challenge. This study has demonstrated that early lactation disease has both short- and long-term effects on productivity. Further, routine measures of herd data including body weight and body condition score, recorded in the dry period have been shown to be significantly different between cows of different disease status in the subsequent lactation. This study has shown that disease in early lactation has serious consequences for the productivity of dairy cattle and has shown the potential for predicting the risk of disease in the transition period in dairy cows. However, further work is needed with larger datasets and in different herds to develop greater accuracy in prediction.