DEVELOPMENT AND IMPLEMENTATION OF AUTOMATED METHODS TO DETECT CHANGES IN PATTERNS OF DIAGNOSTIC TEST RESULTS AT THE ISU VDL

<b>Project Methods</b><br> Study design. Statistical applications will be developed to scan the SDRS data on an automated basis, every time it is updated in the secure ISU server. In case of significant changes in number of submissions, number of negative tests, and/or number positive tests, the ISU VDL director, and the SDRS advisory board will be summoned for a web-meeting to discuss the relevance of detected changes. Relevant findings will be reported in the ISU VDL webpage, and communicated to the swine industry (e.g. via the American Association of Swine Veterinarians', and Swine Health Information Center's newsletters).Aim 1. Consists of developing statistical process control (SPC) and cyclic regression-based algorithms, to generate expected baseline levels for number of negative, and positive test results for PRRSv, TGEv, PEDv, and PDCoV. Moreover, the method will be used to detect significant changes from baseline of each test, identifying potential disease threats. For this phase of the study, it will be used historical ISU-VDL data. We will use previously established cyclic regression model algorithms, which consists of generating the expected levels of test results for seasonal diseases, with upper and lower confident limits [3-5]. The enteric coronaviruses dataset (PEDv, PDCoV, and TGEv) will be used to validate the model: PEDv was first detected in the ISU-VDL in the week # 11 of 2013. For the purpose of this analysis, that date will be considered as the reference for 'time to detect' signals on TGEV testing, and dates prior to February 2013 will be used to assess false alarm rate signals for TGEV. We expect that the baseline levels will be season-dependent. More specifically, there will be more cases in winter months than in non-winter months.Aim 2. The algorithm developed on aim 1 will be incorporated in SDRS live data. The code will be able to detect changes in test results for the previously specified diseases/tests, detecting potentially emerging / re-emerging diseases. An information technology programmer will incorporate the algorithm (written on the SAS, and/or R platform) into the existing SDRS dashboard. The code will be able to identify signals on an automated basis every time that the SDRS is uploaded with more data. We currently have developed this ability on a different project, where we monitor breeding herds for changes in selected productivity parameters using similar automated coding. Producers receive an automated email notification every time there is an unexpected increase in parameters such as abortions, or mortality. Our preliminary results with productivity data proved the concept that it is possible to monitor continuous variables on an ongoing and automated basis, notifying project coordinators in case of any trigger. Moreover, we will establish an Advisory Board including senior veterinarians and producers to discuss the clinical relevance of findings. This will confirm if there is need to warn the industry.Aim 3. Implement the system, and report for 6 months. We will implement the algorithms, and keep tracking changes in the previously listed pathogens for at least 6 months. The goal is to demonstrate the value and feasibility of keeping the project 'live' going forward. We expect that information from this project will be of high relevance to veterinarians and producers, who will have data to confirm or rule out perspectives on selected diseases, and be able to ask the right questions on changes of disease ecology under field conditions. In that case, additional funding would be requested at the end of aim 3 of this project. Target funding agencies include the USDA, the Foundation for Food and Agriculture Research (FFAR), NIH, and livestock commodity groups.Outcomes and statistical analysis. The project will deliver data on changes of trends of pathogen detection over time, geographical region, specimen, and/or age group. The cyclic regression algorithm will be developed using SAS (SAS Institute), and/or R (R-project.org), and incorporated into the existing Microsoft Power Business Intelligence (powerbi.com) dashboards available for the swine disease reporting system. A minimum of 156 observational points is needed to generate accurate 'time-to-detect', and 'false alarm rates' for the model. We have access to all cases submitted to the ISU VDL (at least 70,000 swine cases per year since 2007), which will give robust estimates for the parameters.