Comparing algorithms performance for monitoring endemic disease: a simulation study based on the Danish PRRSV monitoring program

Surveillance systems are critical for accurate and timely monitoring and effective disease control. The use of statistical quality control methods for monitoring endemic diseases which are part of compulsory surveillance programs has not been previously explored. It is important to monitor changes of for instance disease prevalence, which might indicate disease spread. Thus allowing control efforts to be triggered immediately.

In this study, we investigated the performance of three univariate process monitoring control algorithms with the aim of building a monitoring system that can detect changes in the proportion of positive herds for endemic diseases in an accurate and timely way. Additionally, the effect of the sample size (or magnitude of the surveillance system) in the algorithm performance was also assessed. The Porcine Reproductive and Respiratory Syndrome Virus (PRRSV) monitoring program in Danish breeding herds was used as model to design this study.

Three algorithms commonly used for biosurveillance were compared: Shewart P chart (PSHEW), cumulative sum (CUSUM) and exponentially weighted moving average (EWMA). In order to simulate a baseline scenario, the weekly number of positive herds was obtained from a binomial distribution with a probability (p) of 0.1 and a sample size equal to the actual number of Danish breeding herds tested for PRRSV each week from 2007 to 2014. Increases of the number of positive herds were simulated for changes in the prevalence from p=0.1 to p= 0.15 and p= 0.20 during 4, 8, 24, 52 and 104 weeks. Thereafter, the performance of the algorithms was compared by examining their detection capability under the different scenarios.

The results showed that EWMA and PSHEW had higher cumulative sensitivity (CumSe) when compared with the CUSUM. Changes from 0.10 to 0.20 in sero-prevalence were easier detected (higher CumSe) when compared with changes from 0.10 to 0.15 for all three algorithms. EWMA and PSHEW detected changed showed similar results based on the median time to detection. CUSUM detected changes in the sero-prevalence later compared to EWMA and PSHEW for the different scenarios. Increasing the sample size 10 times resulted in half time to guarantee detection (CumSe=1), whereas 100 times sampled sized reduced the time to CumSe=1 by a factor of 6.

In summary, we showed that small changes in diseases sero-prevalence can be detected by using these algorithms. Increasing the sample size provides a faster detection for PRRS. However, the associated costs of increasing the number of herds tested and the disease should to be taken into account when making a decision.

Disclosure of Interest: None Declared.