|Home | About | Journals | Submit | Contact Us | Français|
The dry period is now recognised as a critical time for the control of clinical and sub-clinical mastitis in dairy cattle. Infections that occur, or that are not cured, during the dry period often result in clinical mastitis or raised somatic cell counts in early lactation. There is known to be large variability between herds in the patterns of dry period intramammary infections (IMI) and yet, until recently, there has been no information on farm determinants of the risk of IMI, other than in relation to dry cow treatments. In this paper we consider new research on cow characteristics, farm facilities and herd management strategies during the dry period in relation to clinical mastitis and raised somatic cell counts (SCC) in early lactation. We then describe, within a Bayesian framework, the concept of synthesising existing knowledge with new data to facilitate decision-making on dry cow management for individual farms.
The importance of the dry period in the dynamics of intramammary infections in dairy cows is well established. Research in the UK over the last ten years has clearly demonstrated the importance of this period to mammary gland health in the context of UK dairy production and this has lead to practical recommendations, particularly in the field of preventive treatments, (Bradley and Green, 2000, Bradley and Green, 2001, Bradley et al., 2002, Berry and Hillerton, 2002, Bradley and Green, 2002, Huxley et al., 2002, Green et al., 2002a, Green et al., 2002b, Bradley and Green, 2005).
Infections present during the dry period are usefully split into those that persist from the previous lactation (existing infections), and those that enter between the time of drying off and calving (new infections), (Bradley and Green, 2004). It is helpful to consider these infection processes separately since in terms of prevention, different decisions are needed. Existing infections require decisions relating to effectiveness of a cure whilst new infections require decisions relating to the effectiveness of preventive strategies. The risk of a new dry period IMI is thought to be influenced by:-
Importantly, large differences have been identified in patterns of IMI during the dry period between farms, and within farms over time, (Cook et al., 2002, Dingwell et al., 2004, Green et al., 2005). However, reasons for these differences have not been identified or quantified. Differences in dry period IMI will lead to differences in the patterns of clinical mastitis (CM) and SCC in early lactation between farms (Green et al., 2002a). The large variability between herds in the rate of CM and the proportion of cows with an SCC ≥ 200,000 cell/mL in the first 30 days of lactation is illustrated in Figures Figures11 to to44.
Differences in dry period infection rates between farms are thought to arise, in part, from different management policies. This has been the subject of recent research in which cow characteristics, farm facilities and herd management strategies during the dry period have been investigated to assess their joint influence on the rate of CM and the proportion of SCC ≥ 200,000 cell/mL, in early lactation.
Fifty two commercial dairy farms located throughout England and Wales were included in the research (Green et al, 2007) and data were collected over a two year period, May 2003 to May 2005. To characterize farm facilities and management policies, data were obtained from farmer interviews and through on-farm observation, using structured, pre-tested questionnaires. Information was collected in May or June 2004 (for year one) and again in May or June 2005 (for year two) and covered details of dry cow facilities and dry cow management. Cow information was obtained from National Milk Records (Chippenham, England). Dates of clinical mastitis cases were obtained directly from farm records. Dry periods included in the analysis were from cows that were dried off after May 2003 and that calved before May 2005.
The outcomes of interest were the incidence of CM (first occurrence in lactation) and the magnitude of cow SCC, both within 30 days of calving. CM was also investigated as a time to the first occurrence during lactation (discrete time survival response). The data were highly correlated, consisting of a dry period for each cow (cow-dry period), within a 12 month time span on a farm (farm-year), within a farm and conventional multilevel (random effects) models (Goldstein, 1995) were specified so that these correlations were accounted for appropriately. Model building, fit and posterior predictive checks have been described previously (Green et al, 2007).
A total of 8,710 cow-dry periods (6,852 cows) were used in analysis for cows housed for the dry period and 9,964 cow-dry periods (7,576 cows) for cows at pasture. It was notable that management factors throughout the dry period, from the point of drying off until immediately post calving, were linked to CM and raised SCC in early lactation. A summary of significant factors that influenced clinical mastitis in the first 30 days of lactation are presented in Figure 5. Results of SCC analysis will be presented at the conference. Factors that had no impact on either clinical mastitis or SCC included the length of the dry period (the trend was for dry periods under 50 days to be associated with a higher risk of clinical mastitis and raised SCC in the following lactation) and specific therapeutic products (although a farm policy of appropriate product selection for different cows was protective).
The dry period in relation to udder health is an example of where, as clinicians, we make preventive medicine decisions based on information from a variety of sources. A quantitative analysis of multiple (nearly identical) studies is often termed a meta-analysis and is a useful technique for summarising information for use in ‘evidence-based’ medicine. This is particularly appropriate for multiple clinical trial analysis (although unfortunately these are rarely available in veterinary medicine) and techniques for this have been formalised (e.g see The Cochrane Collaboration, www.cochrane.org).
In some circumstances, we may need to move beyond a comparison of similar studies and combine the evidence of multiple studies that relate to different aspects of a particular disease process. We may also want to include current ‘expert’ clinical opinion within our models and possibly also assess the likely effect of new research on current clinical beliefs.
The essence of a Bayesian approach is to combine reasonable known evidence/information of a parameter (prior knowledge) with new research data, to determine a final probability for a parameter of interest (posterior distribution). The combination is carried out using Bayes Theorem and effectively weights the new data by the prior information. The method is flexible, is able to include diverse sources of evidence and is particularly useful for allowing predictions of future events. In this context, the results represent the probability of future events, given the prior and new data. The Bayesian approach provides an ideal framework to carry out such evidence synthesis and has been described in detail within a medical context (Parmigiani, 2002; Spiegelhalter et al., 2004).
In this case we consider the dry period and illustrate (Figure 6) a simplified pathway to describe the main processes of dry period intramammary infection. To understand and make decisions relating to this infection process, we need to combine knowledge of the following;
Each of the parameters used to describe the infection process during the dry period has some uncertainty associated with it, and we include this, as probability distributions, to assess the uncertainty of our model outputs. Using this approach we can make estimates, (for a given farm situation, for an ‘average’ farm or for a farm ‘at random’) of the impact of different decisions with an associated degree of uncertainty attached.
In a Bayesian context, we can extend this into a decision-based framework and ask questions such as, given specific management interventions, “what is the probability of reducing clinical mastitis within 30 days of calving by 20%” or “what is the probability of a return on investment from the intervention”. We can combine evidence from experts (if we wish) and literature as prior distributions, update this with new evidence, determine predictive distributions for future events and design outputs that facilitate decision-making.
An example of this is provided using a model of the infection process in Figure 6. ‘Expert’ clinical opinion for this model was obtained from a group of 9 experienced veterinary surgeons comprising the Nottingham Dairy Herd Health Group, using described techniques (O'Hagan et al 2006). After amalgamating the group opinions into probability distributions these were incorporated into a dry period mastitis model alongside new research evidence. The model can be used to estimate the consequences of new interventions (from research evidence) in the context of current clinical beliefs. Examples of model outputs are presented in Figures Figures77 and and88 and further results will be discussed at the conference.
The uncertainty in estimated parameters reflects uncertainty in the likely effect of different management interventions in different herd situations. Such variability is seen in practice when similar interventions result in different results on different farms. Estimating the expected variability in an outcome is important and useful since it helps decision-making in that an informed assessment can be based on resources required (time, money etc), the probability of improved health and therefore the probability of return on investment. Whether a probable return on investment is acceptable at (say) 50%, 75% or 90% is a farm based decision that depends in part, on the risk attitude of the farmer and vet involved. However, the aim of this modeling approach is to use a variety of sources of evidence to examine different ‘what if’ scenarios and thus to enlighten the decision-making process.
This research is funded by the Wellcome Trust (Grant Number WT076998 “Use of Bayesian statistical methods to investigate farm management strategies, cow traits and decision- making in the prevention of clinical and sub-clinical mastitis in dairy cows”): Martin Green is a Wellcome Trust Clinical Fellow. We would like to thank the Milk Development Council for funding earlier research and NMR for providing data data.