Speaker
Description
Background
The COVID-19 pandemic has highlighted the need for real-time infectious disease surveillance and forecasting systems to identify trends in transmission. In this study, we compare short-term forecasting models for COVID-19 hospital admissions that make predictions 1 to 4 weeks ahead based on retrospective electronic health record data from the Bern region of Switzerland.
Methods
We extract different variables, e.g., the number of visits to the emergency department per day, from an individual-level patient dataset that covers several hospitals located in the region of Bern, including Bern University hospital, during the period of February 2020 to June 2023. We apply last-observation carried forward (baseline), linear regression, XGBoost and neural networks (RNN and LSTM) to leave-future-out training time series with multiple cutting points and optimize hyperparameters. We systematically evaluate the performance of the different algorithms using the root-mean-square error between the predictions and the observations.
Results
Among the evaluated models XGBoost performs the best on average across all forecasting periods. We show that adding predictors like the number of patients admitted to hospital with fever and complementing the hospital data with measurements of the concentration of viral RNA in wastewater can improve the accuracy of the forecast. We propose to combine the best performing algorithm with the set of variables with the highest predictive power into a final prediction model.
Conclusions
This study presents a thorough and systematic comparison of methods applicable to routine hospital data for real-time epidemic forecasting. With the increasing availability and volume of electronic health records with decreasing delays, tools such as ours will contribute to more precise and timely information during epidemic waves of COVID-19 and other respiratory viruses, which may improve evidence-based public health decision-making.
| Expedited Notification | No thanks, I do not require Expedited Notification |
|---|
