Forecasting the changes between endemic and epidemic phases of a contagious disease, with the example of COVID-19.

Background: Predicting the endemic/epidemic transition during the temporal evolution of a contagious disease.

Methods: Indicators for detecting the transition endemic/epidemic, with four scalars to be compared, are calculated from the daily reported news cases: coefficient of variation, skewness, kurtosis, and entropy. The indicators selected are related to the shape of the empirical distribution of the new cases observed over 14 days. This duration has been chosen to smooth out the effect of weekends when fewer new cases are registered. For finding a forecasting variable, we have used the principal component analysis (PCA), whose first principal component (a linear combination of the selected indicators) explains a large part of the observed variance and can then be used as a predictor of the phenomenon studied (here the occurrence of an epidemic wave).

Results: A score has been built from the four proposed indicators using the PCA, which allows an acceptable level of forecasting performance by giving a realistic retro-predicted date for the rupture of the stationary endemic model corresponding to the entrance in the epidemic exponential growth phase. This score is applied to the retro-prediction of the limits of the different phases of the COVID-19 outbreak in successive endemic/epidemic transitions for three countries, France, India, and Japan.

Conclusion: We provided a new forecasting method for predicting an epidemic wave occurring after an endemic phase for a contagious disease.