Time-series modelling for the quantification of seasonality and forecasting antibiotic-resistant episodes: application to carbapenemase-producing Enterobacteriaceae episodes in France over 2010–20

Abstract

Abstract Background Carbapenemase-producing Enterobacteriaceae (CPE) cause resistant healthcare-associated infections that jeopardize healthcare systems and patient safety worldwide. The number of CPE episodes has been increasing in France since 2009, but the dynamics are still poorly understood. Objectives To use time-series modelling to describe the dynamics of CPE episodes from August 2010 to December 2016 and to forecast the evolution of CPE episodes for the 2017–20 period. Methods We used time series to analyse CPE episodes from August 2010 to November 2016 reported to the French national surveillance system. The impact of seasonality was quantified using seasonal-to-irregular ratios. Seven time-series models and three ensemble stacking models (average, convex and linear stacking) were assessed and compared with forecast CPE episodes during 2017–20. Results During 2010–16, 3559 CPE episodes were observed in France. Compared with the average yearly trend, we observed a 30% increase in the number of CPE episodes in the autumn. We noticed a 1 month lagged seasonality of non-imported episodes compared with imported episodes. Average stacking gave the best forecasts and predicted an increase during 2017–20 with a peak up to 345 CPE episodes (95% prediction interval = 124–1158, 80% prediction interval = 171–742) in September 2020. Conclusions The observed seasonality of CPE episodes sheds light on potential factors associated with the increased frequency of episodes, which need further investigation. Our model predicts that the number of CPE episodes will continue to rise in the coming years in France, mainly due to local dissemination, associated with bacterial carriage by patients in the community, which is becoming an immediate challenge with regard to outbreak control.