Mathematics finds a number of uses in dairy herd management, from calculating the proper portions for a TMR to balancing revenue and costs. But imagine if one could actually predict the future, assigning the probability of an outcome down to decimal places. What potential does this power hold? A recent study published in the Journal of Dairy Science explores this prospect.
A model to predict the risk for postpartum problems was developed by a team of veterinary researchers and graduate students at the University of Wisconsin – Madison and Cornell, both with a common interest in the diagnosis and management of ketosis.
Together, they compiled and analyzed data collected in 2010 from more than 2,500 fresh cows from four commercial dairy herds in New York and Wisconsin.
Noting the incidence of postpartum diseases such as milk fever, ketosis, retained placenta, metritis, lameness, pneumonia and displaced abomasum, these individuals attempted to identify factors that could predict which cows were at greatest risk for one or more treatments or removal from the herd within the first 30 days in milk (TXR30).
“We came into this as more of an open book and said, ‘Let’s see what we can do,’” reflects Dr. Gary Oetzel, corresponding author and professor of food animal production medicine at UW – Madison School of Veterinary Medicine.
“We did find other literature, which we did cite in the paper, that talks about things related to culling risk, but what we did was quite a bit more comprehensive than anything else that had been done. So we were mostly figuring out how to do this as we conducted the statistical analysis.”
For multiparous cows, important factors that increased the odds of TXR30 were birthing abnormalities, including twins, stillbirths and dystocia, as well as a greater number of lactations. Oetzel says that previously, high milk production was actually protective, as it was an indicator of well-being in earlier lactations that is likely to be repeated.
Lameness, which increases the pain and stress an animal experiences, was also a predictor of postpartum problems, although not as much as researchers initially predicted. This may be a reflection of the superior quality of management on the four farms, which resulted in few animals with poor locomotion scores.
For primiparous cows, the lack of data from previous lactations limited the researchers’ ability to predict postpartum problems. However, birthing difficulties and gestation length were identified as significant predictors.
“We expected older cows to have different risk factors than younger cows. One of the nice things in this study was it forced us to really wrestle with that and wind up with a totally different model for first-calf heifers,” Oetzel says.
In both groups, interactions between some factors were also found to either exacerbate problems or help mitigate negative effects. For example, multiparous cows with calving difficulties and longer previous gestation lengths were at very high risk for TXR30.
These researchers hope that someday producers will be able to use predictive models to identify each cow’s probability of encountering problems and then individualize their monitoring and husbandry for highest-risk cows.
“It is always better to prevent diseases than to treat them,” says Dr. Jessica McArt, a previous member of the department of clinical sciences at Colorado State University and current assistant professor at Cornell University.
“Knowing ahead of time which animals are at risk of disease and/or culling and what factors increase that risk can be quite helpful from a management standpoint. It can assist with better preventative management, improved and focused observation and faster intervention.”
Oetzel and McArt caution that predictive models have limits. Results are associative, and extrapolating data from these few recorded herds should be done prudently. Further, statistical odds of TXR30 based on a predictive model cannot identify every cow that is inclined toward trouble.
“While these results can be used to help focus preventative management changes on the animals with the highest post-calving risk, we should not forget that we still need to closely monitor all cows during the transition period,” McArt explains.
Oetzel hopes that additional research studies with larger numbers of animals can validate the utility of predictive modeling in fresh cow management. Future studies could include a control group which is treated the same as the rest of the herd and a treatment group of cows that are assigned to different risk groups with adjusted management practices, such as closer observation, increased preventative treatments or extra space.
One recent study from Cornell submitted to the Journal of the American Veterinary Medical Association tested such hypotheses in relation to lameness; it found that milking severely lame cows only twice daily rather than three times actually increased daily milk yield by 3.5 pounds.
Oetzel also believes models can be useful in predicting other high-probability events with easily measured variables, such as a cow’s likelihood of becoming pregnant after her first service.
“We are not telling people to go out just yet and use these models, but we believe it is important to get this idea of predictive modeling out there,” he says. “We really want it to get into people’s thinking and maybe to encourage others to look at things this way.”
This transition from subjective observations to objective statistics is important in increasing the efficiency of dairy operations. No matter how closely a producer may take note of his or her cows’ activities, the absolute future of their health is impossible to conceive. However, predictive modeling helps the industry clear the haze of the unknown, taking steps to prevent undesirable outcomes before they even occur. PD
Holly Drankhan is a senior at Michigan State University with plans to attend vet school. She is a 2014 Progressive Dairyman editorial intern.
