The COVID-19 Bayesian style for epidemic detection (BSTI) has two key features that set it apart from other styles: Professor Gordon Pipa, head of the neuroinformatics studies organization on the AI campus of the University of Osnabruck, states: “On the one hand, the new approach provides a forecast horizon, which allows to evaluate the reliability of predictions. In addition, the style takes into account the effect of infection rates on neighboring districts. It also allows us to evaluate the propagation dynamics. “

One of the many demanding situations when breaking down forecasts in individual districts is the low number of cases. “A single forecast trajectory can be misleading because you can’t assess the reliability of the forecast,” explains Professor Pipa. The style we use not only calculates one of the most likely paths, but rather takes into account many imaginable paths that meet the data. This allows you to calculate forecast horizons as a measure of the probability distribution. This approach assesses the situation, adding statistical uncertainties, which can provide useful data even when the number of instances is low. “

In addition, the BSTI style calculates the effect of neighboring regions. An interaction core describes the extent to which a high or low number of cases in a neighboring region has an effect on the infection rate in a district. Research Group and the Robert Koch Institute effectively used the interaction core in 2019 to describe the progression of rotavirus infections, as well as Lyme disease and Campylobacter bacteria.

Experts at the J-lich Supercomputing Center (JSC) helped adapt the approach to COVID-19 knowledge and adjust the code for research in J-lich supercomputers. “Determining the forecast horizon is incredibly computational, as we implement many other models Therefore, statistical modeling requires much more computational time than strategies that do not use a forecast horizon. To do the daily analyses, without delay after the publication of RKI knowledge, we use the resources of the J-lich Supercomputing Center: desktop computers in general would be absolutely outpervised by the work,” explains Jens Henrik Gobbert of JSC.

The analyses, updated daily, and the option of spatial or temporal visual comparisons are freely accessible and presented in as understandable a format as possible. “We sought to make the effects temporarily and in an understandable format, so a giant organization of others can simply communicate about the content without delay without getting bogged down in technology,” Gobbert explains.

For example, visitors to the interactive site can freely choose districts to view their five-day forecasts, or they can compare the most recent knowledge reported through the Robert Koch Institute with estimated actual new infections. Due to delays in knowledge transfer, the reported figures differ. and infrequently significantly, based on the actual number of new cases. Therefore, the goal of the “immediate forecast” is to provide an initial assessment of the existing figures using statistical analysis. A forecast, in turn, provides an estimate of progress over the next five days.