Providing probability distributions for the Gram-status of clinical mastitis cases in dairy cattle

Clinical mastitis (CM) can be caused by a wide variety of pathogens and a farmer has to start treatment before the actual causal pathogen is known. Knowing the Gram-status of CM cases would aid in the decision for the most appropriate treatment. By providing a probability distribution for the Gram-status, rather than only providing the most likely Gram-status, the involved uncertainty is visible for a farmer, thereby allowing a more informed treatment decision. The objective of this study was to examine the value of providing probability distributions for the Gram-status of CM cases to a farmer to take a more informed treatment decision. A naive Bayesian network (NBN) based on data from 274 Dutch dairy herds in which the occurrence of CM was recorded over an 18-month period was constructed. The dataset contained 3,534 CM cases, all classified accordingto their Gram-status. Two-third of the dataset was used for the construction and one-third was retained for validation. Information usually available at a dairy farm was included in the NBN under construction (parity, month in lactation, season of the year, quarter position, somatic cell count history and CM history, being sick or not, and color and texture of the milk). For getting insight in the quality of the constructed NBN, the accuracy was determined. The accuracy of classifying CM cases into Gram-positive or Gram-negative pathogens was 73%. Because only CM cases with a high probability for Gram-negative or Gram-positive pathogens will be considered for specific treatment, it was interesting to have a closer look at CM cases with probabilities > 0.90. We foundthat the accuracy of the classification increased with the calculated probability for Gram-negative or Gram-positive pathogens. The probability distributions for the Gram-status provide the farmer with considerable additional information about the most likely Gram-status of a CM case and the uncertainty involved.