• Ante Lojić Kapetanović, Marina Lukezić, Ajka Pribisalić, Dragan Poljak, and Ozren Polašek.
• Ozren Polašek, Department of Public Health, University of Split School of Medicine, Šoltanska 2, 21000 Split, Croatia, ozren.polasek@mefst.hr.
• Croat. Med. J. 2022 Jun 22; 63 (3): 295-298.

AimTo facilitate the development of a COVID-19 predictive model in Croatia by analyzing three different methodological approaches.MethodWe used the historical data to explore the fit of the extended SEIRD compartmental model, the Heidler function, an exponential approximation in analyzing electromagnetic phenomena related to lightning strikes, and the Holt-Winters smoothing (HWS) for short-term epidemic predictions. We also compared various methods for the estimation of R0.ResultsThe R0 estimates for Croatia varied from 2.09 (95% CI 1.77-2.40) obtained by using an empirical post-hoc method to 2.28 (95% CI 2.27-2.28) when we assumed an exponential outbreak at the very beginning of the COVID-19 epidemic in Croatia. Although the SEIRD model provided a good fit for the early epidemic stages, it was outperformed by the Heidler function fit. HWS achieved accurate short-term predictions and depended the least on model entry parameters. Neither model performed well across the entire observed period, which was characterized by multiple wave-form events, influenced by the re-opening for the tourist season during the summer, mandatory masks use in closed spaces, and numerous measures introduced in retail stores and public places. However, an extension of the Heidler function achieved the best overall fit.ConclusionsPredicting future epidemic events remains difficult because modeling relies on the accuracy of the information on population structure and micro-environmental exposures, constant changes of the input parameters, varying societal adherence to anti-epidemic measures, and changes in the biological interactions of the virus and hosts.

8 keywords...

hide…