Construction and evaluation of epidemiologic simulation models for the within- and among-unit spread and control of infectious diseases of livestock and poultry

Abstract

Epidemiologic modeling is an increasingly common method of estimating the potential impact of outbreaks of highly contagious diseases, such as foot-and-mouth disease (FMD) and highly pathogenic avian influenza (HPAI), in populations of domesticated animals. Disease models are also used to inform policy decisions regarding disease control methods and outbreak response plans, to estimate the possible magnitude of an outbreak, and to estimate the resources needed for outbreak response. Although disease models are computationally sophisticated, the quality of the results of modeling studies depends on the quality and accuracy of the data on which they are based, and on the conceptual soundness and validity of the models themselves. For such models to be credibly applied, they should realistically represent the systems they are intended to reflect, should be based to as great an extent as possible on valid data, and should be subjected to careful and ongoing scrutiny. Two key steps in the evaluation of epidemiologic models are model verification and model validation. Verification is the demonstration that a computer-driven model is operating correctly, and conforms to its intended design. Validation refers to the process of determining how well a model corresponds to the system that it intended to represent. For a veterinary epidemiologic model, validation would address issues such as how well the model represents the dynamics of the disease in question in a population to which the model is applied, and how well the model represents the application of different measures for disease control. Among the steps that can be taken by epidemiologic modelers to facilitate the processes of model verification and validation are to clearly state the purpose, assumptions, and limitations of a model; to provide a detailed description of the conceptual model for use by everyone who might be tasked with evaluation of a model; document steps already taken to test the model; and thoroughly describe the data sources and the process used to produce model input parameters from data. The realistic representation of the dynamics of spread of disease within individual herds or flocks can have important implications for disease detection and surveillance, as well as for disease transmission between herds or flocks. We have developed a simulation model of within-unit (within-herd or within-flock) disease spread that operates at the level of the individual animal, and fully incorporates sources of individual-level variation such as variability in the durations of incubating and infectious periods, the stochastic nature of disease spread among individuals, and the effects of vaccination. We describe this stochastic model, along with the processes employed for verification and validation. The incorporation of this approach to modeling of within-unit disease dynamics into models of between-unit disease spread should improve the utility of these models for emergency preparedness and response planning by making it possible to assess the value of different approaches to disease detection and surveillance, in populations with or without some existing level of vaccine immunity. Models rely not only on realistic representations of the systems of interest, but also on valid and realistic information. For spatially explicit models of the spread and control of disease in populations of livestock and poultry, this means a heavy reliance upon valid spatial representations of the populations of interest, including such characteristics as the geographic locations of farms and their proximity to others in the population. In the United States, limited information regarding the locations of actual farm premises is available, and modeling work often makes use of artificially generated population datasets. In order to evaluate the accuracy and validity of the use of such artificially generated datasets, we compared the outcomes of mechanistic epidemiologic simulation models that were run using an empirical population dataset to those of models that made use of several different synthetic population datasets. Although we found generally good qualitative agreement among models run using various population datasets, the quantitative differences in model outcomes could be substantial. When quantitative outcomes from epidemiologic models are desired or required, care should be taken to adequately capture or describe the uncertainty in model-based outcomes due to the use of synthetic population datasets.