Identifying questions in the American Association of Swine Veterinarian's PRRS risk assessment survey that are important for retrospectively classifying swine herds according to whether they reported clinical PRRS outbreaks in the previous 3 years

The American Association of Swine Veterinarian's (AASV) Production Animal Disease Risk Assessment Program (PADRAP) is a web-based program that offers a set of risk assessment surveys being used by veterinarians who are members of the AASV. Members use PADRAP to help producers systematically assess risk factors that may be associated with clinical outcomes. As assessments are performed the completed surveys are added to the dataset maintained at the Iowa State University College of Veterinary Medicine. One of the surveys included in PADRAP is the porcine reproductive and respiratory syndrome (PRRS) Risk Assessment for the Breeding Herd. The aim of the study was to categorize questions in version 2 of the PRRS Risk Assessment for the Breeding Herd survey as important or unimportant for classifying herds according to whether they reported clinical PRRS outbreaks in the previous 3 years. The results elucidate the relative importance of risk factors and areas of risk factors for clinical outcomes and removing unimportant questions may reduce the time required to complete the survey without affecting the quality of information obtained. Surveys from 896 sow herd sites in the United States and Canada completed between March 2005 and March 2009 were included in the analysis. The survey contained a large number of questions with a complex correlation structure among the questions. Responses for several questions were dependent upon responses to others. To address these issues, an approach was developed using a series of statistical methods including random forest, principle component analysis, logistic regression and receiver operating characteristic (ROC) analysis to classify the herds using the questions in the survey as explanatory variables. Questions were ranked by importance and systematically excluded from least important to most important. The questions excluded, without significantly affecting the performance of the model for classifying herds were identified as unimportant. Thirty-eight of the 127 questions analyzed were identified as unimportant for classifying herds according to whether they reported clinical PRRS outbreaks in the previous 3 years. Sections of the survey where a large number of questions were identified as unimportant included (1) entry of semen into the breeding herd and (2) transportation of live animals. Sections with a high percentage of questions identified as unimportant included (1) characteristics of the site, (2) disposal of dead animals and waste management and (3) employee and visitors.