Nowcasting with leading indicators applied to COVID-19 fatalities in Sweden (preprint)

The real-time analysis of infectious disease surveillance data, e.g., in the form of a time-series of reported cases or fatalities, is essential in obtaining situational awareness about the current dynamics of an adverse health event such as the COVID-19 pandemic. This real-time analysis is complicated by reporting delays that lead to underreporting of the number of events for the most recent time points (e.g., days or weeks). This can lead to misconceptions by the interpreter, e.g., the media or the public, as was the case with the time-series of reported fatalities during the COVID-19 pandemic in Sweden. Nowcasting methods provide real-time estimates of the complete number of events using the incomplete time-series of currently reported events by using information about the reporting delays from the past. Here, we consider nowcasting the number of COVID-19-related fatalities in Sweden. We propose a flexible Bayesian approach, extending existing nowcasting methods by incorporating regression components to accommodate additional information provided by leading indicators such as time-series of the number of reported cases and ICU admissions. By a retrospective evaluation, we show that the inclusion of ICU admissions as a leading signal improved the nowcasting performance of case fatalities for COVID-19 in Sweden compared to existing methods.

On assessing excess mortality in Germany during the COVID-19 pandemic (preprint)

Coronavirus disease 2019 (COVID-19) is associated with a very high number of casualties in the general population. Assessing the exact magnitude of this number is a non-trivial problem, as relying only on officially reported COVID-19 associated fatalities runs the risk of incurring in several kinds of biases. One of the ways to approach the issue is to compare overall mortality during the pandemic with expected mortality computed using the observed mortality figures of previous years. In this paper, we build on existing methodology and propose two ways to compute expected as well as excess mortality, namely at the weekly and at the yearly level. Particular focus is put on the role of age, which plays a central part in both COVID-19-associated and overall mortality. We illustrate our methods by making use of age-stratified mortality data from the years 2016 to 2020 in Germany to compute age group-specific excess mortality during the COVID-19 pandemic in 2020.

Epitweetr: Early warning of public health threats using Twitter data (preprint)

Summary Background ECDC performs epidemic intelligence activities to systematically collate information from a variety of sources, including Twitter, to rapidly detect public health events. The lack of a freely available, customisable and automated early warning tool using Twitter data, prompted ECDC to develop epitweetr. The specific objectives are to assess the performance of the geolocation and signal detection algorithms used by epitweetr and to assess the performance of epitweetr in comparison with the manual monitoring of Twitter for early detection of public health threats. Methods Epitweetr collects, geolocates and aggregates tweets to generate signals and email alerts. Firstly, we evaluated manually the tweet geolocation characteristics of 1,200 tweets, and assessed its accuracy in extracting the correct location and its performance in detecting tweets with available information on the tweet geolocation. Secondly, we evaluated signals generated by epitweetr between 19 October and 30 November 2020 and we calculated the positive predictive value (PPV). Then, we evaluated the sensitivity, specificity and timeliness of epitweetr in comparison with Twitter manual monitoring. Findings The epitweetr geolocation algorithm had an accuracy of 30.1% and 25.9% at national and subnational levels, respectively. General and specific PPV of the signal detection algorithm was 3.0% and 74.6%, respectively. Epitweetr and/or manual monitoring detected 570 signals and 454 events. Epitweetr had a sensitivity of 78.6% [75.2% - 82.0%] and PPV of 74.6% [70.5% - 78.6%]; and the manual monitoring had a sensitivity of 47.9% [43.8% - 52.0%] and PPV of 97.9% [95.8% - 99.9%]. The median validation time difference between sixteen common events detected by epitweetr and manual monitoring was −48.6 hours [(−102.8) - (−23.7) hours]. Interpretation Epitweetr has shown to have sufficient performance as an early warning tool for public health threats using Twitter data. Having developed epitweetr as a free, open-source tool with several configurable settings and a strong automated component, it is expected to increase its usability and usefulness to public health experts. Funding Not applicable Research in context Evidence before this study Previous reviews have shown how social media, including Twitter, have been used for public health purposes. Most recent studies, in relation to the COVID-19 pandemic, have shown the added value of early warning tools based on Twitter and other social media platforms. They also noted the lack of an open-source tool for real-time monitoring and surveillance. Added value of this study Epitweetr is a free, open-source and R-based early warning tool for automatic Twitter data monitoring that will support public health experts in rapidly detecting public health threats. The evaluation of epitweetr presented in this study shows the strengths of the tool which include having good performance, high degree of automation, being a near-real-time tool and being publicly available with various customisable settings. Furthermore, it shows which are the areas of improvement for the next versions of epitweetr. Implications of all the available evidence This tool can be further developed to include more automation and machine learning components to increase usability and information processing time by users.

Analysis of the early Covid-19 epidemic curve in Germany by regression models with change points (preprint)

We analyze the Covid-19 epidemic curve from March to end of April 2020 in Germany. We use statistical models to estimate the number of cases with disease onset on a given day and use back-projection techniques to obtain the number of new infections per day. The respective time series are analyzed by a trend regression model with change points. The change points are estimated directly from the data. We carry out the analysis for the whole of Germany and the federal state of Bavaria, where we have more detailed data. Both analyses show a major change between March 9th and 13th for the time series of infections: from a strong increase to a decrease. Another change was found between March 25th and March 29th, where the decline intensified. Furthermore, we perform an analysis stratified by age. A main result is a delayed course of the epidemic for the age group 80+ resulting in a turning point at the end of March. Our results differ from those by other authors as we take into account the reporting delay, which turned out to be time dependent and therefore changes the structure of the epidemic curve compared to the curve of newly reported cases.

Analysis of the COVID-19 pandemic in Bavaria: adjusting for misclassification (preprint)

We present a method for adjusting the observed epidemic curve of daily new COVID-19 onsets for possible misclassification in infection diagnostics. We discuss different assumptions for specificity and sensitivity of the person-specific COVID-19 diagnostics based on PCR-tests, which are the basis for the daily reported case counts. A specificity of less than one implies occurrence of false positive cases, which becomes particularly relevant with an increased number of tests. The recent increase in cases in Bavaria could therefore be smaller than reported. However, an increase in case counts can still be seen from Mid-July until September 2020. The additional consideration of a sensitivity less than one, i.e., the occurrence of false negative tests, results in an epidemic curve in which the daily case counts are increased by a constant factor, but the structure of the curve does not change considerably.

Real-time monitoring of COVID-19 dynamics using automated trend fitting and anomaly detection (preprint)

As several countries gradually release social distancing measures, rapid detection of new localised COVID-19 hotspots and subsequent intervention will be key to avoiding large-scale resurgence of transmission. We introduce ASMODEE (Automatic Selection of Models and Outlier Detection for Epidemics), a new tool for detecting sudden changes in COVID-19 incidence. Our approach relies on automatically selecting the best (fitting or predicting) model from a range of user-defined time series models, excluding the most recent data points, to characterise the main trend in an incidence. We then derive prediction intervals and classify data points outside this interval as outliers, which provides an objective criterion for identifying departures from previous trends. We also provide a method for selecting the optimal breakpoints, used to define how many recent data points are to be excluded from the trend fitting procedure. The analysis of simulated COVID-19 outbreaks suggest ASMODEE compares favourably with a state-of-art outbreak-detection algorithm while being simpler and more flexible. We illustrate our method using publicly available data of NHS Pathways reporting potential COVID-19 cases in England at a fine spatial scale, for which we provide a template automated analysis pipeline. ASMODEE is implemented in the free R package trendbreaker.