Joint Models to Predict Dairy Cow Survival from Sensor Data Recorded during the First Lactation.

Area of Science:

• Animal Science
• Veterinary Medicine
• Data Science

Background:

• Early prediction of dairy cow survival is crucial for effective farm management and breeding strategies.
• Joint models for longitudinal and survival data offer a novel approach to analyzing complex biological and agricultural datasets.

Purpose of the Study:

• To explore the application of joint models for predicting the survival probability of dairy cows to subsequent lactations.
• To assess the performance of these models using sensor-derived longitudinal data from the first lactation.

Main Methods:

• Developed a joint model integrating daily first-lactation sensor data (milk yield, body weight, rumination time) with survival data (time to culling).
• Utilized data from 6 Holstein dairy farms, predicting survival to the second and third calving based on early lactation observations (60, 150, 240 days in milk).
• Employed 3-time-repeated 3-fold cross-validation to evaluate model performance using Area Under the Curve (AUC) and expected error of prediction.

Main Results:

• Model performance varied, with Area Under the Curve (AUC) ranging from 45% to 76% across scenarios and farms.
• The expected error of prediction ranged from 3.5% to 26%, indicating that predicted survival probabilities aligned with observed culling events.
• Model performance demonstrated stability across different farms, suggesting generalizability.

Conclusions:

• Joint models are a viable tool for predicting dairy cattle survival, providing valuable insights for management decisions.
• The accuracy and stability of the model warrant further investigation for practical implementation in dairy farming.
• This approach can enhance breeding programs and improve overall herd management by identifying at-risk animals early.