Complementary nowcasts based on our fever curve

Dear Donors,

For the past year, our fever monitor has compared two distinct signals: The detections of anomalous vital signs and weekly incidences of COVID-19. Both signals have unique advantages and disadvantages when it comes to monitoring current trends of the pandemic:

  1. The incidence curve describes the overall state of the pandemic very well, but due to lags in reporting of numbers, is only really reliable up to a couple of days in the past.
  2. Our fever curve instead can provide a real-time monitoring (with no reporting-lags) of current physiological and physical stress, but also picks up on other signals not necessarily related to COVID-19, such as acute respiratory diseases, including influenza-like illnesses, or even changes in weather conditions.

A nowcast based on vital-signs

Today, we explore how to exploit the advantages that come with both sources of information for a short-time monitoring of the pandemic’s current state. Specifically, we encompass the history of the pandemic through the detection rate of anomalous vital signs and the (reliable) incidences from the previous week. Combining this historic information with the current detection rate thus potentially enables us to provide a so-called nowcast of COVID-19 incidences.

Below, you can see a first comparison of our proposed monitoring system with later confirmed incidences. For testing purposes we have restricted ourselves to all available data until the end of July. The gray dot indicates our nowcast for the last week of July for which reliable incidence numbers were not yet readily available by the time of testing our monitoring system.

Weekly nowcast based on the fever curve and comparison to officially reported average COVID-19 incidences per week.

You see that in hindsight our (still) comparatively simple monitoring system matches the reported incidences quite well. It even correctly picks up the turning points of the second and third wave!

Unique advantages

Even though the numbers do not match up exactly, our new approach has unique advantages that make it a promising complement to existing nowcasting methods:

  1. Thanks to the informative power of the vital data, we only require three single data points (detections and incidences from the last week and detection from the current week) to provide an estimate of current incidences. Therefore our methodology could in the future be used from the very start of a pandemic situation.
  2. The system is automatized and can in the future work in real-time such that it provides continuous updates once new information becomes available through our donors with no delays in reporting.
  3. It adapts to the current state of the pandemic and estimates how many cases we expect for each fever detection depending on whether we are in a phase with low or high incidences.

Outlook

We will now continue to evaluate and improve our vital sign-based nowcasting scheme which thanks to the help of you, the donors, provides an additional data-driven methodology to assess current trends in the pandemic. For this purpose we will next update our systems so that your donated data will be processed as fast as possible once it becomes available to us. We have already included the new nowcasting method into our fever monitor to give you a better impression of how your data is used to further improve the understanding of the current pandemic.

Marc Wiedermann
Alumnus

Researcher and Data Scientist with strong interests in time series and network analysis, predictive models and low-dimensional dynamical systems for the spread of human behavior.

Dirk Brockmann
Dirk Brockmann
Professor

Head of Research on Complex Systems Group