Multivariate dynamic linear models for estimating the effect of experimental interventions in an evolutionary operations setup in dairy herds.

Evolutionary operations is a method to exploit the association of often small changes in process variables, planned during systematic experimentation and occurring during the normal production flow, to production characteristics to find a way to alter the production process to be more efficient. The objective of this study was to construct a tool to assess the intervention effect on milk production in an evolutionary operations setup. The method used for this purpose was a dynamic linear model (DLM) with Kalman filtering. The DLM consisted of parameters describing milk yield in a herd, individual cows from a herd, and an intervention effect on a given day. The model was constructed to handle any number of cows, experimental interventions, different data sources, or presence of control groups. In this study, data from 2 commercial Danish herds were used. In herd 1, data on 98,046 and 12,133 milkings registered from an automatic milking system (AMS) were used for model building and testing, respectively. In herd 2, data on 3,689 milkings on test days were used for estimating the initial model parameters. For model testing, data from both bulk tank milk yield (85 observations) and test-day milkings (1,471) were used. In herd 1, the manager wanted to explore the possibility of reducing the amount of concentrate provided to the cows in an AMS. In herd 2, the manager wanted to know if the milk yield could be increased by elevating the energy level provided to the cows in a total mixed ration. The experiment conducted in herd 1 was designed with a treatment and a control group, whereas in herd 2 we used a pretest/posttest design. The constructed tool provided estimates (mean and confidence intervals) for each of 3 interventions carried out in both herds. In herd 1, we concluded that the reduction in concentrate amount provided in the AMS had no negative influence on milk yield. For herd 2, the increased level of energy had a significant positive effect on milk yield but only for the first intervention. In this herd, the effect of intervention was also evaluated for cows in the first lactation and without bulk tank records. The presented model proved to be a flexible and dynamic tool, and it was successfully applied for systematic experimentation in dairy herds. The model can serve as a decision support tool for on-farm process optimization exploiting planned changes in process variables and the response of production characteristics.

From biological models to economic optimization.

This article addresses the additional challenges being faced when biological models are used as a basis for decision support in livestock herds. The challenges include dealing with uncertain information, observation costs, herd dynamics and methodological issues in relation to the computational methods applied particularly in the dynamic case. The desired key property of information included in models is that it can be used as the basis for unbiased prediction of the future performance of the animals. Often there will be a tradeoff between uncertainty and costs in the sense that the level of uncertainty can be reduced (for instance through additional tests) at some cost. Thus, the decision about which (and how many) tests to perform can be seen as an optimization problem in itself. Another way of expressing the tradeoff is to talk about the value of information which can sometimes be assessed by modeling different approaches and levels of detail in data collection. Various optimization methods of relevance to herd health management are discussed with the main emphasis on decision graphs in the static case and Markov decision processes (dynamic programming) in a dynamic context.

Herd and sow-related risk factors for mortality in sows in group-housed systems.

Mortality of sows is a major problem for pig production worldwide. In this study, we used hierarchical multivariable logistic analyses to investigate different risk factors for mortality at the sow and herd level in herds with group-housed pregnant sows. Data included 3652 pregnant and 1266 lactating sows from 34 sow herds. A clinical examination for 16 clinical signs was carried out for each sow, and information about 16 herd related factors was obtained by interviews. Farm records were used to obtain information about whether or not sows died suddenly or were euthanized within 3 months after the clinical examination. Factors increasing the risk of sow mortality in the gestation unit were solid pen floors (OR=1.87), presence of vulva bites (OR=1.73) and unwillingness to stand when approached (OR=1.62). Factors increasing the risk of sow mortality in the lactation unit were pale vulva color (OR=12.69), body leanness (OR=4.11), and presence of shoulder ulcers (OR=2.89). The estimated between herd variation was small. Thus, the findings for the sow level variables may be generally applicable for sows in herds with group housed systems.

Modelling the economic impact of three lameness causing diseases using herd and cow level evidence.

Diseases to the cow's hoof, interdigital skin and legs are highly prevalent and of large economic impact in modern dairy farming. In order to support farmer's decisions on preventing and treating lameness and its underlying causes, decision support models can be used to predict the economic profitability of such actions. An existing approach of modelling lameness as one health disorder in a dynamic, stochastic and mechanistic simulation model has been improved in two ways. First of all, three underlying diseases causing lameness were modelled: digital dermatitis, interdigital hyperplasia and claw horn diseases. Secondly, the existing simulation model was set-up in way that it uses hyper-distributions describing diseases risk of the three lameness causing diseases. By combining information on herd level risk factors with prevalence of lameness or prevalence of underlying diseases among cows, marginal posterior probability distributions for disease prevalence in the specific herd are created in a Bayesian network. Random draws from these distributions are used by the simulation model to describe disease risk. Hereby field data on prevalence is used systematically and uncertainty around herd specific risk is represented. Besides the fact that estimated profitability of halving disease risk depended on the hyper-distributions used, the estimates differed for herds with different levels of diseases risk and reproductive efficiency.

Estimation of probability for the presence of claw and digital skin diseases by combining cow- and herd-level information using a Bayesian network.

Cross sectional data on the prevalence of claw and (inter) digital skin diseases on 4854 Holstein Friesian cows in 50 Danish dairy herds was used in a Bayesian network to create herd specific probability distributions for the presence of lameness causing diseases. Parity and lactation stage are identified as risk factors on cow level, for the prevalence of the three lameness causing diseases digital dermatitits, other infectious diseases and claw horn diseases. Four herd level risk factors have been identified; herd size, the use of footbaths, a grazing strategy and total mixed ration. Besides, the data has been used to estimate the random effect of herd on disease prevalence and to find conditional probabilities of cows being lame, given the presence of the three diseases. By considering the 50 herds representative for the Danish population, the estimates for risk factors, conditional probabilities and random herd effects are used to formulate cow-level probability distributions of disease presence in a specific Danish dairy herd. By step-wise inclusion of information on cow- and herd-level risk factors, lameness prevalence and clinical diagnosis of diseases on cows in the herd, the Bayesian network systematically adjusts the probability distributions for disease presence in the specific herd. Information on population-, herd- and cow-level is combined and the uncertainty in inference on disease probability is quantified.

An object-oriented Bayesian network modeling the causes of leg disorders in finisher herds.

The implementation of an effective control strategy against disease in a finisher herd requires knowledge regarding the disease level in the herd. A Bayesian network was constructed that can estimate risk indexes for three cause-categories of leg disorders in a finisher herd. The cause-categories of leg disorders were divided into infectious causes (arthritis caused by infectious pathogens), physical causes (e.g. fracture and claw lesions), and inherited causes (osteochondrosis). Information about the herd (e.g. the herd size, floor type and number of suppliers) and information about individual pigs (e.g. results from diagnostic tests) were used to estimate the most likely cause of leg disorders at herd level. As information to the model originated from two different levels, we used an object-oriented structure in order to ease the specification of the Bayesian network. Hence, a Herd class and a Pig class comprised the basic components of the object-oriented structure. The causal structure of the model was based on evidence from published literature. The conditional probabilities used in the model were elicited from experts within the field and from the published literature. To illustrate the behaviour of the model, we investigated the value of different levels of evidence in two fictitious herds with different herd characteristics related to the risk of leg disorders (e.g. purchase policy, production type and the stocking density in pens). The model enabled us to demonstrate the value of performing systematic collection of additional information (i.e. clinical, pathological and bacteriological examination) when identifying causes of leg disorders at herd level.