Decision Support for Disease Management in Pig Production via Statistical Modelling

PhD Thesis by Michael Höhle

Dina Research Report No. 105, June 2003

Summary

Infectious diseases are an inherent component of modern pig production. An understanding of the mechanisms of infection are a necessity in order to be able to control them. Larger farms and efficient production systems have increased the production, but at the same time caused environmental and health related problems. Medicine consumption and aerial pollution are examples of current issues in the public debate. Figures from the Danish Veterinary and Food Administration show that in 2002, 34% of the pigs slaughtered at Danish abattoirs received a health annotation due to wounds, abscesses, joint damage, tail biting and especially scars induced by respiratory diseases. In 1979 the corresponding figure was just 13%. Even though the increased percentage is partly caused by improved registration methods and younger slaughter age, the numbers are still alarming. The incidence of pleurisy at slaughter is partly related to the number of pneumonia and influenza outbreak in the herds – infectious diseases that prosper under the above mentioned conditions.

The present thesis deals with management of infectious disease in pig production. Temporal and spatial progress of a disease through the pens and sections of a herd are modelled using statistical models. The aimof such amodelling is to gain both knowledge about the current disease conditions in a herd as well as being able to predict new outbreaks. A goal is thus the development of a decision support system to deal with e.g. respiratory diseases or diarrhea. A premise for such a system is the ability to calculate the consequences of decisions within a decision theoretic framework.

Tied together by this main thread, the thesis consists of five separate articles. The methods developed in the articles are generic and not bound to a specific disease or environment. Treatment of respiratory diseases with fatteners is partly used as example because the need here is especially apparent and data are available. For instance, on-farm data for registered treatments against pneunomia are analysed in the thesis. Simple visualizations constitute a first prototype for decision support for the workers of the farm. Attempts to use the data to perform predictions although show that the prediction based on treatment data is insufficient. A way to compensate for the uncertainty in treatment data is to conduct transmission experiments, where a specific disease is closely monitored in a well-defined population. Analyzing such data is done using classical mathematical models describing infectious diseases. However, an adaptation to the specific conditions of pig production, i.e. confinement units including pens and sections, production cycle, use of various diagnostic test, etc., is necessary before these models can be applied. The methodical adaptations are carried out in the thesis and motivated by the analysis of an experiment with classical swine fever in a test stable.

Even on the assumption of an acceptable predictive model, there is a long way to go before optimal on-farm treatment strategies against diseases can be computed. Therefore, two articles of the thesis deal with making sequential decisions in domains characterized by incomplete observations. A graphical and object-oriented notation is presented allowing a compact specification of the above mentioned decision problems. Even such a user-friendly specification language can, however, not hide that an exact solution of the specified decision problem quickly becomes computationally intractable. The last article of the thesis thus deals with methods to compute upper and lower bounds for the exact solution.

This thesis is a step towards providing decision support for management of infectious diseases in pig production – a topic of great current interest. The emergence of new diseases in swine, e.g. postweaning multisystemic wasting syndrome (PMWS), demands a constant flow of new initiatives. That PMWS in the media often (but also somewhat erroneously) is called swine-AIDS shows that mathematical modelling of infectious diseases is quite independent of the application domain. A conclusion is that in a modern society infectious diseases are an inherent part of the agenda – it does not matter whether it is foot-and-mouth disease, email virus, influenza or the severe acute respiratory syndrome (SARS). Fear can be observed in connection with these diseases, but with a better insight a better basis to act in rational ways is obtained. For pig production this could mean fewer sick pigs, lower medicine consumption, and simultaneously a better production economy.

Full thesis

The thesis is available as a PDF file.