At this moment, the future evolution of COVID-19 is unclear. We propose the Network Inference-based Prediction Algorithm (NIPA) to forecast the future course of COVID-19. We apply NIPA to the current situation in The Netherlands on a province-based level using data from the RIVM: https://www.rivm.nl/coronavirus-covid-19/actueel. For more information on NIPA, please read our article on arXiv: https://arxiv.org/abs/2002.04482.
Predicting real epidemic outbreaks like COVID-19 is difficult due to underreporting. Please see our extended abstract on the fundamental limits of predicting epidemic outbreaks.
For the latest predictions on COVID-19 in the Netherlands, please visit: http://networkdatascience.ewi.tudelft.nl
Basic reproduction number R0
The basic reproduction number R0 is the average number of individuals that are infected by a typical infected individual. In Figure 2, we compute the reproduction number R0 from NIPA using formula (4) in "Reproduction numbers and sub-threshold endemic equilibria for compartmental models of disease transmission" by Van den Driessche and Watmough (2002). Before March 30, the reproduction number R0 is larger than 1, corresponding to an increasing number of newly infected cases. After April 14, the reproduction number R0 drops below 1, which indicates that the disease is slowly dying out.
Figure 2: The basic reproduction number is slowly decreasing over time. Before March 16, the number of
data points is small, leading to large fluctuations of the basic reproduction number R0.
Daily new cases
The number of daily reported COVID-19 cases per province is shown in Figure 3. Prediction of the future course of the epidemic is difficult due to the large variations in the data.
Figure 3: The number of daily reported COVID-19 cases per province. For clarity, only four out of 12 provinces are shown.
To evaluate the prediction accuracy of NIPA, we remove the data for a fixed number of days, say m, prior to April 2. The prediction is then computed based on observations from 29 February to 2-m April, 2020. Then, we predict the course of the disease up to April 2. The course of the disease is shown in Figure 4 for the removal of m=1,2,3,4 days. The short-term prediction is generally quite accurate, whereas the curves start to deviate (especially for Noord-Brabant) for 3 or 4 days ahead.
Figure 4: The prediction of the COVID-19 outbreak in The Netherlands by NIPA. For clarity, only five out of 12 provinces are shown. Each subfigure is obtained by omitting a number m=1,2,3,4 of days prior to April 2, and subsequently predicting the same number of days ahead in time. The omitted number of data points is equal to: (top-left) m=1 day, (top-right) m=2 days, (bottom-left) m=3 days, (bottom-right) m=4 days. The first prediction data point, for instance April 1 in the top-left figure, coincides with the last day that has been observed.