Abstract #401

# 401
Associations between metabolic diseases and reproductive performance of dairy cows using survival analysis and machine learning models.
O. Bogado Pascottini1, M. Probo2, S. Leblanc1, G. Opsomer3, M. Hostens*3,4, 1Population Medicine, Ontario Veterinary College, University of Guelph, Guelph, ON, Canada, 2Department of Veterinary Medicine, Veterinary Teaching Hospital, University of Milan, Lodi, Italy, 3Department of Reproduction, Obstetrics and Herd Health, Faculty of Veterinary Medicine, Ghent University, Merelbeke, Belgium, 4Department of Farm Animal Health, Faculty of Veterinary Medicine, University of Utrecht | Utrecht, the Netherlands,.

This study aimed to evaluate the associations between individual transition cow conditions and metabolic diseases (MD) and multiple MD (MD+) and the probability of pregnancy in Holstein cows. Kaplan-Meier, Cox proportional hazard, and machine learning models were used to analyze the associations of MD and MD+ with the pregnancy risk from a 1-year cohort with 1,946 calvings. The recorded MD were milk fever, retained placenta, metritis, ketosis, displaced abomasum, and clinical mastitis. Twinning was included as additional event due to its association with multiple MD. The cumulative 60-d milk yield variable (M60) was also included as co-variate for all the models. Survival models were stratified to account for differences by parity. For the machine learning modeling parity was included as a variable of interest. Using Kaplan-Meier models, in primiparous cows, the 120 DIM pregnancy risk was 62% for healthy animals. This was not significantly different for MD (58%) but was reduced for MD+ (45%). Among healthy primiparous cows, 80% were pregnant by 210 DIM, but pregnancy risk at that time was reduced for primiparous cows with MD (72%) and MD+ (62%). In healthy multiparous cows, the 120 DIM pregnancy risk was 53%, which was reduced for MD (36%) and MD+ (30%). The 210 DIM pregnancy risk for healthy multiparous cows was 70%, being significantly higher than the 210 DIM pregnancy risk for multiparous cows with MD (47%) or MD+ (46%). Survival analysis showed that the pregnancy risk was reduced when cows experienced more than one MD, irrespective of parity. Multivariable survival models were not able to accommodate complex MD interactions, which dropped the distinct number of cows per category. Decision tree and random forest models showed that parity was the most influential variable affecting fertility. Decision tree analysis selected ketosis and metritis while random forest also identified mastitis and M60 as important variables interfering with fertility. Machine learning methods helped exploring the complex interactions between the parity and each MD to study their hierarchical effect on the pregnancy risk in dairy cows.

Key Words: transition period, reproduction, decision tree analysis

Speaker Bio
Miel Hostens focused on optimization of productive and reproductive performances in small and large herds with an emphasis on nutrition during his PhD and post doc position at the Department of Reproduction, Obstetrics and Herd Health of Ghent University (Belgium). He was workpackage leader for 3 work packages with a focus on data management in the large EU FP7 project GplusE until 2018. He currently works as assistant professor at the ?Department of Farm Animal Health of Utrecht University (the Netherlands). where he is involved in the education of master and bachelor students in Veterinary Medicine, statistical training of PhD students in data science in the area of dairy cows and post academic and extension services in the area of herd health management in dairy cows.