A regime switch analysis on Covid-19 in Romania.

Area of Science:

• Epidemiology
• Computational Biology
• Data Science

Background:

• Pandemic prediction faces challenges due to unreliable reported case numbers and numerous influencing factors.
• The accuracy of COVID-19 data, particularly for infected and recovered individuals, is often questionable, while mortality data tends to be more reliable.

Purpose of the Study:

• To propose a robust three-stage analytical framework for modeling and predicting the evolution of COVID-19 in Romania.
• To address data limitations by refining epidemiological models and incorporating time-varying parameters.

Main Methods:

• A three-stage approach was employed, starting with a neural network-based SIR (Susceptible-Infected-Recovered) model for initial parameter estimation.
• The second stage refined the SIR model into a SIRD (Susceptible-Infected-Recovered-Deceased) model, separating deceased individuals and using grid search for daily parameter refinement.
• The third stage identified turning points (regimes) in model parameters to capture dynamic shifts in the pandemic's trajectory.

Main Results:

• The study generated daily parameter estimates for the epidemiological models.
• The refined SIRD model provided a more accurate representation by accounting for mortality separately.
• Identification of distinct pandemic phases (regimes) based on parameter fluctuations was achieved.

Conclusions:

• The proposed three-stage analysis, utilizing a refined SIRD model with time-varying parameters, offers a more reliable method for COVID-19 pandemic prediction.
• This framework provides a generalizable approach to pandemic forecasting, even with imperfect data.
• Understanding pandemic dynamics through parameter regime identification is crucial for effective public health strategies.