Usage of Wastewater Data as an Early Indicator for Hospitalization Forecasting in Pandemic Situations

The COVID-19 pandemic has not only had a major impact on the everyday lives of individuals, but also posed major challenges for politics, the economy and the education system. To be better prepared for such situations and to be able to take appropriate measures, it is crucial to assess and predict the course of a pandemic as accurately as possible when deciding on measures. In this study, we investigate to what extent the addition of wastewater surveillance data can improve an otherwise autoregressive COVID-19 forecasting model for Germany and its federal states. We examine possible correlations between the viral load measured in wastewater and the COVID-19 hospitalization incidence as well as case numbers of GP diagnoses of various respiratory diseases. We compare the performances of autoregressive implementations of Random Forest regressors, XGBoost regressors, ARIMA models and linear regression models with their respective performances when including both time series as predictors. For the tree-based models we also investigate the performance of fully cross-modal models which solely rely on the viral load measurements to predict the COVID-19 hospitalization incidence.
We found that the wastewater data can be used as an early indicator for the detection of upcoming trends in hospitalization incidence on a national scale due to its high correlation to the hospitalization incidence. The wastewater data seems to precede the hospitalization figures by about six days. The inclusion of wastewater data into the prediction models does not significantly improve the forecast of COVID-19 hospitalization. The best-performing model was the ARIMA model without the inclusion of wastewater data with a Mean Average Percentage Error of 4.61%.