Use of epidemiologic information in targeted surveillance for population inference Scott J. Wells a, *, Eric D. Ebel b , Michael S. Williams b , Aaron E. Scott c , Bruce A. Wagner d , Katherine L. Marshall d a Department of Veterinary Population Medicine, University of Minnesota, St. Paul, MN 55108, USA b Risk Analysis Division, Food Safety Inspection Service, USDA, Ft. Collins, CO 80526, USA c National Surveillance Unit, Centers for Epidemiology and Animal Health, USDA-APHIS-Veterinary Services, Ft. Collins, CO 80526, USA d National Animal Health Monitoring System, Centers for Epidemiology and Animal Health, USDA-APHIS-Veterinary Services, Ft. Collins, CO 80526, USA 1. Introduction Targeted sampling is emerging as an important concept in animal health surveillance. Classically, one type of targeted surveillance has been visual observation of animals within herds with specific clinical signs of disease. This form of targeted surveillance has been used histori- cally in veterinary medicine for case finding and has proven critical for detection and eradication of important diseases including foot and mouth disease, contagious pleuropneumonia, and others. The past 5 years, however, have seen a substantial effort by many countries to incorporate targeted sampling concepts into eradication programs for endemic diseases and for demonstrating freedom from disease (e.g., Sta ¨rk et al., 2006). The US Bovine Spongiform Encephalopathy (BSE) Surveillance Program has used targeted surveillance (BSurvE) to enhance the likelihood of detection of BSE in cattle with clinical signs of disease or other abnormal presentations (USDA-APHIS-VS, 2006). In this surveillance system, a complex model involving age and categories for exiting the population was used to derive point values for each Preventive Veterinary Medicine 89 (2009) 43–50 ARTICLE INFO Article history: Received 14 January 2008 Received in revised form 14 January 2009 Accepted 20 January 2009 Keywords: Targeted surveillance Prevalence estimation Risk ratio ABSTRACT Epidemiologic information, including animal characteristics (e.g., observable risk factors or clinical signs) predisposing to animal disease, is frequently used for design of targeted surveillance systems, but this information is infrequently used for population inference. In this study, we report the evaluation of use of epidemiologic information for population inference in targeted surveillance in three animal disease scenarios. We adapted sampling theory using Monte Carlo methods to determine target population sample size to detect disease with 95% confidence, using information from the epidemiologic parameters risk ratio and fraction of the population with the characteristic. These parameters and their uncertainties were derived from a reference population. The next step was to use a second (sampled) population to evaluate effects of sampling the targeted population. The focus of the study was on estimation of prevalence. Our results showed that if one is less certain of the epidemiologic parameters, a rational decision is to model the input parameter distributions reflecting this uncertainty, thereby increasing the sample size above the minimum needed for the detection of the disease with a known confidence. Targeted surveillance is appropriate for prevalence estimation when one has representative and justifiable estimates of key epidemiologic parameters. ß 2009 Published by Elsevier Ltd. * Corresponding author at: 136 ABLMS, 1354 Eckles Avenue, University of Minnesota, St. Paul, MN 55108, USA. Tel.: +1 612 625 8166; fax: +1 612 624 4906. E-mail address: wells023@umn.edu (S.J. Wells). Contents lists available at ScienceDirect Preventive Veterinary Medicine journal homepage: www.elsevier.com/locate/prevetmed 0167-5877/$ – see front matter ß 2009 Published by Elsevier Ltd. doi:10.1016/j.prevetmed.2009.01.007