Appendix 31 Epidemiological models for global surveillance of Foot-and-Mouth disease Andrés M. Perez1*, Mark C. Thurmond1, Tim E. Carpenter1, Thomas W. Bates2, Wesley O. Johnson3, Brett A. Melbourne1, Shagufta Aslam1, Rebecca B.Garabed1, Young-ku Choi3, Adam J. Branscum3, Miryam L. Gallego1, and Paul W. Grant1 1
The FMD Modeling and Surveillance Laboratory, Department of Medicine and Epidemiology, School of Veterinary Medicine, University of California, Davis, CA 2 Lawrence Livermore National Laboratory, Livermore, CA 3 Department of Statistics, University of California, Davis, CA
Abstract: This paper summarizes some of the epidemiological models we have developed to describe the global temporal-spatial distribution and risk of FMD and to identify factors that predict changes in FMD status or movement, as applicable to real-time global FMD surveillance. Regional models focused on the time, space (geographic location), and time-space evolution of FMD in specific countries. Countries being studied include Iran, Pakistan, Afghanistan, Turkey, Nepal, Mongolia, Colombia, Bolivia, and Argentina. Country-specific models were validated using methods of cross-validation and expert opinion, as solicited from veterinarians in each country. The results of risk models are projected to assess when and where FMD can be expected and to forecast globally changing risks of FMD. These models could be applied via an FMD web-portal currently under development for realtime global risk surveillance of FMD to characterize changes in time, place, and transmission of FMD, and to identify new anomalous and unexpected FMD cases or precursor events. Real-time global risk surveillance for FMD, utilizing prediction, forecasting, and anomaly detection models, can improve our awareness and assessment of FMD globally and provide fundamental information to enhance biosecurity in free areas and control strategies in infected areas. Introduction: It is becoming increasingly apparent that, if we are to control and perhaps eradicate FMD on a global scale, a comprehensive understanding of the global distributions and changes in risk and movement of the disease must be developed. Currently, we lack of information relating to where FMD can be expected to be found and the new or emerging risks of FMD world wide. Key goals of our laboratory are to disseminate information on FMD relating to outbreaks and isolates, make available modeled distributions of FMD in regions where the information is uncertain, and project or predict risk of infection in FMD-free areas of the world. In this paper we summarize some of the models and approaches explored by our group in the past year. Methods and Results: BioPortal system: The FMD BioPortal is a web-based system designed for the real-time capture and dissemination of data, diagnostic results, and FMD-related risk information to and from countries, agencies, and laboratories. The aim of the portal is to offer a secure and confidential mechanism for rapid transfer of data, analyses, and maps needed to assess changes in FMD risk for specified geographic locations and times. FMD-related data may be obtained automatically or by ‘hand’ upload from websites of the international organizations, such as OIE, FAO, or the WRLFMD, as well as from individual country or agency databases. The database outputs are reformatted into a standardized structure via a data adapter specific to a particular messaging system. The data can be subjected to a variety of epidemiological and statistical analytic and mapping methods, including spatio-temporal clustering, anomaly detection, and spread or movement prediction. Temporal models: The goal of temporal models is to provide an estimate, during some time i, of the expected number of FMD outbreaks in a given region of the world in time i+1. The mathematical process that best models the time series depends on the behavior of the infectious disease (trend, sporadic, seasonal, annual, and secular cycles). We are currently examining the potential application of different Bayesian models that alternatively use autoregressive, random mixed effects, stochastic processes, or the combination of these models to predict the number of FMD outbreaks in Iran. The number of outbreaks reported by Iran from January 1996 through December 2001 is used to select the model that best fit the distribution of the data, using estimates of the mean absolute prediction error and deviance information criterion. Then the selected model is used to predict the number of outbreaks in Iran for the next year, and the resulting prediction is compared with the actual observed number of outbreaks (as part of validation). 203