The PACASurvE laboratory network for real-time infection surveillance and alert at a regional scale

Abstract : irregular ratios. Seven models issued from time series analysis and three ensemble stacking models (average, convex and linear stacking) were used to describe and forecast CPE episodes. The model with the best forecasting's quality was then trained on all available data (2010-2016) and used to predict CPE episodes over 2017-2020. Results: Over 2010-2016, 3,559 CPE episodes were observed in France. Compared to the average yearly trend, we observed a 30% increase in the number of CPE episodes in September and October. On the opposite, a decrease of 20% was noticed in February compared to other months. We also noticed a 1-month lagged seasonality of non-imported episodes compared to imported ones. The number of non-imported episodes appeared to grow faster than imported ones starting from 2014. Average stacking gave the best forecasts and predicted an increase over 2017-2020 with a peak up to 345 CPE episodes (95% PI [124-1,158], 80% PI [171-742]) in September 2020. Conclusions: The number of CPE episodes is predicted to rise in the next years in France because of non-imported episodes. These results could help public health authorities in the definition and evaluation of new containment strategies. Key messages: Time series modeling predicts an increase in the number of CPE episodes in France in the next few years with a quicker rise of non-imported episodes. An increase of 30% in the number of CPE episodes was observed in September and October with a 1-month lagged seasonality impact of non-imported episodes compared to imported one.