Till innehåll på sidan
Till KTH:s startsida Till KTH:s startsida

Helmut Küchenhoff and Felix Günther: Analysis of the COVID-19 pandemic in Bavaria: nowcasting and adjustment for misclassification

Tid: On 2020-10-28 kl 15.15

Plats: Zoom, email organisers

Medverkande: Helmut Küchenhoff and Felix Günther

Exportera till kalender

Daily numbers of newly reported COVID-19 cases are used at different regional levels to monitor the epidemic. There are three main problems for the interpretation of these numbers: first, the temporal assignment is problematic because the reporting day of a case does not coincide with the day of disease onset. Such reporting delays can lead to misjudgments with respect to the current state of the epidemic. Second, there may be errors in the examination procedure, which can also distort the assessment of the current situation. Third, not all infected persons are captured in surveillance data, because many infected persons have no or only mild symptoms, because persons refuse testing, or because the testing capacities are insufficient (case detetion ratio smaller 1).

In this talk, we present our approach to analyze COVID-19 surveillance data from Bavaria, Germany. First, we present a nowcasting strategy that can be used to adjust daily case reporting data for occurred- but-not-yet-reported-events. The basic idea of nowcasting is to estimate the reporting delay based on observations where both, the reporting date and the symptom onset date, are known. Considering such information, it is possible to estimate the epidemic curve, i.e., the daily number of new disease onsets, from the reporting data for days with incomplete information (near the current date). This solves the problem of temporal assignment of reported cases. The approach is based on a Bayesian hierarchical model [Guenther et al., 2020b] and considers changes in the reporting delay distribution over time and the day of the week. It builds upon a model developed by Höhle and an der Heiden [2014]. We introduce the model, discuss assumptions, and show results of a retrospective evaluation of the performance on the Bavarian data.

In the second part of the talk, we present a method to adjust aggregated case counts for miclassification in the person-specific disease examination [Guenther et al., 2020a]. We discuss plausible assumptions with respect to the sensitivity and specificity of the examination procedure and evaluate the distortion of the time-series of aggregated case counts based on different assumptions. A specificity smaller one leads to false-positive cases and is especially problematic, since the extent of distortion depends on the number of examined persons that changes over time and depending on testing strategy. The adjusted daily case counts can also serve as basis for nowcasting to obtain a misclassification adjusted estimation of the epidemic curve. In the Bavarian data, we find that the increase in case numbers starting in Mid-July might have been smaller than indicated by raw case counts. More false-positive cases due to a higher testing activity can, however, not fully explain the increase in cases during that period. During periods of higher case counts, the bias induced by false-positive examinations is less pronounced.

We close with a discussion and focus especially on potential consequences of a varying case-detection ratio over time.

Results of our analyses are available to the public and are updated on a daily basis ( corona.stat.uni-muenchen.de/nowcast/ ).

Preprint of the paper:

Place: Zoom. Please register by sending an email to Dmitry Otryakhin at otryakhin@math.su.se