Publikationsdatenbank

Narrative review of infectious disease spread models developed in Poland during COVID19 pandemic (2023)

Art

Poster

Autoren

Jarynowski, Andrzej (WE 16)

Belik, Vitaly (WE 16)

Kongress

XLII Max Born Symposium

Wroclaw, 14.09.2023

Quelle

Sprache

Englisch

Verweise

URL (Volltext): https://fens2023uwr.files.wordpress.com/2023/08/abstract_book_v3.pdf

Kontakt

Institut für Veterinär-Epidemiologie und Biometrie

Königsweg 67
14163 Berlin
+49 30 838 56034
epi@vetmed.fu-berlin.de

Abstract / Zusammenfassung

Introduction:
Models are elementary tools for timely and targeted responses in the fight against pathogens. The use of Data-based modelling to generate forecast of the epidemic dynamics by harnessing vast, register based and open-source data have potential to be game changing and highly sustainable (i.e. in optimization of the resources [doi:10.15503/emet2020.100.122]). COVID-19 pandemic sped up infection disease spreading model’s development and applicability. However, models’ predictions for SARS-CoV-2 in Poland and in general in Eastern Europe were often disappointing [doi:10.15503/emet.2021.112.124]. On the other hand, due to the extreme amount of work to gather, analyse and understand data during the time of pandemic with lessons learned in Polish modelling community, for future epidemic, modelling could provide the scientific basis for public health decisions and intervention measures.

Methods and Data:
We have performed qualitative and quantitative benchmark of COVID-19 forecasting models for Poland till Omicron wave. We divided models into three categories: 1) system dynamics (i.e. differential equations); 2) agent-based models (i.e. microsimulation); machine learning (i.e. autoregressive models, neural networks). We proposed three categories of forecasting horizons: 1) short term prediction ~ 7 days; 2) medium ~ 4 weeks; 3) long ~ full infection season (from a few months up to one year). The selection of the models in qualitative part [doi:10.5604/01.3001.0015.0281] was based on availability principle (both scientific ‘white’ and unpublished ‘grey’ sources) and models were included if they provide forecast in any time horizon for at least one outcome for any kind of geographical unit of Poland (i.e. whole country, voivodeship, city): No. cases (any definition), No. hospitalisation, No. COVID-19 related deaths (any definition). For the quantitative part we took data from the European Covid-19 Forecast Hub and compared outcomes for short and medium term forecasts for registered COVID-19 cases.

Results:
The so-called “modelling zoo” allows us to compare methods repertoire from medicine, biology, physics, mathematics, economics, geography, computer science, etc. Only few models provide a scenario competition for more than one intervention [DOI: 10.14746/sr.2020.4.3.01]. The “dark figure” undiagnosed cases) and confusion matrix of tests is also rarely considered [DOI: 10.31373/ejtcm/147842]. The model which performed the best in predicting registered cases in short term is a simple mechanistic ARIMA with spatial component machine learning class model. Agent-based UW-ICM model performed the best in medium time prediction.

Conclusions:
Due to COVID-19 infectious disease modelling concept has “spread” over the whole Poland (both academic and business entities). As there is no model fit all, both forecasting hub exercise and narrative review of the models allow us to see pros and cons for each approach.