Dr. Kent A. Weigel, Department of Dairy Science, University of Wisconsin
Dairy cow longevity is critically important, economically speaking, but adding this trait to genetic selection programs is anything but easy. Each step, including trait definition, data collection, data validation, and statistical analysis, has several pitfalls.
Challenges in trait definition are evident in the variety of names for this trait, such as "longevity", "survival", "productive life", and "herdlife". These may be prefaced by "true" or "functional", in an attempt to differentiate voluntary (good) culling and involuntary (bad) culling. What is the meaning of statements such as "the average culling rate is 38%”? One herd may have a 40% turnover rate, but 20% are sold as dairy replacements, and 20% are slaughtered due to low production. Another may have a 30% turnover rate, but 15% die in the barn, and 15% leave due to mastitis, lameness, or infertility. Economic consequences and animal welfare implications differ greatly.
Collection of longevity data is unlike that of any other trait. The trait is farmer-recorded but has no measurement error (the cow is either standing in the barn, or she's not). However, the trait is highly susceptible to bias, and a cow’s risk of culling is influenced by many factors beyond her control, such as availability of replacement heifers, plans for expansion, competence of the herdsman, and milk quota restrictions.
Data validation also poses challenges, many of which are limitations of the DHI milk recording system. For example, culling codes (i.e., sold for dairy, sold for beef, sold due to mastitis, sold due to infertility) lack specificity and flexibility. Furthermore, cows culled before first test in first lactation may not enter the system, and cows culled before first test in later lactations may have incorrect culling dates.
Statistical analysis is hampered by a skewed, non-normal distribution of survival times and a high percentage of censored records (from cows sold for dairy or cows still alive). Time-dependent explanatory factors are also a challenge – one must account for management changes within and between lactations. Factors such as disease exposure, modernization of facilities, and milk prices change frequently, and all affect a cow’s risk of culling at a given time. Proper analyses of culling data should feature two components. First, data should be analyzed using failure-time methodology that can account for non-normality, multiplicative relationships, censored records, and time-dependent covariates. In this way, the instantaneous risk of culling for a given cow can be calculated at any specified time for any set of explanatory variables. Second, inference should be based on the ratio of culling risk for high-producing cows, versus average cows, and the ratio of culling risk for low-producing cows, versus average cows. Optimal management schemes, and optimal sires for genetic selection, should yield a low risk of involuntary culling among high-producers and a high risk of voluntary culling among low-producers.
Lastly, genetic selection for the primary components of longevity, namely health and fertility, should augment (or even replace) direct selection based on culling data. Recording of health disorders and/or veterinary treatments is common on large commercial farms, and mechanisms for validating and storing such data (across herds) should be explored. Traits such as pregnancy rate, body condition score, lameness, mastitis, and ketosis can be measured early in life, and substantial genetic variation exists between sires in the health and fertility of their daughters.
