Potential of ARIMA models of piglet birth and mortality for syndromic surveillance

Abstract

The feasibility of ARIMA models for syndromic surveillance was evaluated to determine anomalies in births and mortality in swine offspring from a typical commercial farm, and to predict alarm thresholds. For this purpose, data series of these variables aggregated by month over an eight-year period (2010-2017) were analyzed. Models were selected based on the verification of the admissible, parsimonious and stable assumptions, as well as the Akaike criterion for predicting values and their 90 % confidence interval. The best fitting model for both births and mortality was an ARIMA (1,1,0). The lower limit of prediction, set as an alarm level for births, was 109 offspring. In the case of mortality, the alarm level corresponding to the upper limit of the prediction interval was 12.67 %. The ARIMA models established were feasible for syndromic surveillance strategies based on birth and mortality data series, with prediction of alarm thresholds (births <109 offspring and mortality > 10.6 %) that enable early warning and the development of timely responses to correct deviations of productive parameters and their implications.