Fig 1.
Data flow of the DISPENSE tool.
The DISPENSE tool integrates data transfer, load and extraction into a one-stop-shop solution. In the back-end a database holds all relevant data for a visual data dashboard at the front-end that displays the requested information to the user. All modules run in a container-based infrastructure in components shown at the bottom of the scheme.
Fig 2.
Screenshot of the shiny front-end of the DISPENSE tool.
Welcome page displaying key data, performance indicators and plots to decision makers. The left-hand bar enables the user to dive deeper into a more detailed data analysis and prognosis.
Fig 3.
Compartment structure of the employed SEIR model.
The traditional model is extended by the normal ward and ICU compartments.
Table 1.
Overview of model parameters.
Table 2.
The second COVID-19 wave in Saxony in numbers.
Fig 4.
Age stratification of positively tested individuals in Saxony.
(a) Age distribution of positively tested persons in Saxony between October 1 and December 31, 2020. (b) Distribution of COVID-19 affected age groups per week as a function of time.
Fig 5.
Course of the second COVID-19 wave in Saxony.
Time series of 14-day sum of registered infected individuals. Data for (a) individual clusters and (b) all of Saxony shows dynamic effects caused by NPIs. The major difference between clusters is the amplitude; the overall picture is the same for each region and for entire Saxony.
Fig 6.
Bed demand in East Saxony during the second wave.
Time series (solid lines) of (a) number of normal ward beds and (b) number of ICU beds occupied by COVID-19 patients show effects caused by NCIs, changes in test strategy and the Christmas holidays. The predictions (dashed lines) provided an accurate seven-day forecast.
Fig 7.
Total bed demand for Saxony during the second wave.
Time series (solid lines) of (a) number of normal ward beds and (b) number of ICU beds occupied by COVID-19 patients are less noisy then single cluster time series and the prediction (dashed lines) more precise.
Fig 8.
Scatter plot of actual data and predicted values.
The plot shows a uniform corridor around the diagonal and displays no significant abnormalities.
Fig 9.
Time dependent distribution of relative errors among the three clusters.
The quality of the prediction increases in the course of the second wave.
Table 3.
Prediction quality of model.