Reducing bias in risk indices for COVID-19

Abstract

Spatiotemporal modelling of infectious diseases such as coronavirus disease 2019 (COVID-19) involves using a variety of epidemiological metrics such as regional proportion of cases and/or regional positivity rates. Although observing changes of these indices over time is critical to estimate the regional disease burden, the dynamical properties of these measures, as well as crossrelationships, are usually not systematically given or explained. Here we provide a spatiotemporal framework composed of six commonly used and newly constructed epidemiological metrics and conduct a case study evaluation. We introduce a refined risk estimate that is biased neither by variation in population size nor by the spatial heterogeneity of testing. In particular, the proposed methodology would be useful for unbiased identification of time periods with elevated COVID-19 risk without sensitivity to spatial heterogeneity of neither population nor testing coverage.We offer a case study in Poland that shows improvement over the bias of currently used methods. Our results also provide insights regarding regional prioritisation of testing and the consequences of potential synchronisation of epidemics between regions. The approach should apply to other infectious diseases and other geographical areas.

Access to healthcare as an important moderating variable for understanding geography of COVID-19 outcomes – preliminary insights from Poland

“…Only triangulated pictures based on reported cases (with estimates of undiagnosed infections), deaths (due to COVID-19 and excess mortality), and hospitalizations (separately with and due to COVID-19), may allow for epidemiological interpretation. Thus, there are mathematical approaches of reducing the bias by combining multiple epidemiological indexes or constituting their derivatives [41][42]. For instance, crude fatality rates [43] seem to be less dependent on HCA than estimates of IFR in Poland.…”