Early Detection of PRRSV Outbreaks in Breeding Herds by Monitoring Productivity and Electronic Sow Feed Data Using Univariate and Multivariate Statistical Process Control Methods

Porcine reproductive and respiratory syndrome virus (PRRSV) is one of the most impactful viruses in swine production worldwide. Early detection of PRRSV outbreaks is the first step in rapid disease response. A syndromic surveillance system can be implemented to detect early signs of PRRSV activity in breeding herds. This study aimed to integrate multiple data sources and test univariate and multivariate statistical process control charts to identify early indicators associated with PRRSV outbreaks in sow farms. From March 2022 until May 2023, 16 breed‐to‐wean swine farms were enrolled in the study. The following key clinical and productivity indicators associated with PRRSV outbreaks were investigated: number of abortions, number of dead sows, number of off‐feed events, preweaning mortality rate (PWM), and percentage of neonatal losses. The PRRSV status for the herd was determined by reverse transcriptase polymerase chain reaction testing using processing fluid samples, and it was considered as the reference to calculate the performance of the surveillance system. The exponentially weighted moving average (EWMA), cumulative sum (CUSUM), multivariate exponentially weighted moving average (MEWMA), and multivariate cumulative sum (MCUSUM) were the methods used to detect significant changes in the aforementioned parameters following the PRRSV outbreaks. Using the EWMA model, the indicators with the highest early detection rates were PWM followed by the abortions (71% and 64%, respectively), with the models raising alarms 4 weeks earlier on average than the processing fluids, respectively. For the CUSUM model, the weekly number of PWM, followed by abortions, were the indicators with the highest early detection rates (71% and 64%, respectively), with the models raising alarms 4 weeks earlier on average than the processing fluids for both indicators. Concerning the multivariate models, the MEWMA model with higher early detection used the PWM and neonatal losses (86%), with the models raising alarms 4 weeks earlier on average than the processing fluids, with the models raising alarms 3.5 weeks earlier on average than the processing fluids. For the MCUSUM, the model with higher early detection used PWM and neonatal losses (86%), with the models raising alarms 4.3 weeks earlier on average than the processing fluids. The models with the earliest time to detect signs associated with a PRRSV outbreak and with the lowest false negative and false positive were the multivariate models, MEWMA and the MCUSUM, using the combination of PWM and neonatal losses. Thus, monitoring multiple indicators outperformed the univariate models. With that, using multivariate models is the best option for disease surveillance using indicators, and it allows the decision‐makers to investigate potential outbreaks earlier.