(8) COVID-19: Multifactorial Modelling to Predict Long-Term Trends

Publication Number: 682

Background: COVID-19, declared a pandemic in 2020, continues to persist. The true volume of infections cannot be measured directly with confirmed cases, though it is fundamental in understanding how / when surges will occur. Long-term modelling that considers the multifactorial complexity of the pandemic is essential to understand the transition to endemic COVID-19.

Objectives: Develop SEIRS long term forecast model using country specific data that predicts the waves of the COVID-19 pandemic in 45 countries.

Methods: Daily mortality data from Our World In Data were used to estimate the number of COVID-19 infections. Age-adjusted infection fatality rate (IFR) was derived from population based analyses in Iceland, where near total population testing occurred. The reproduction number (R0) came from fitting mortality curve and adjusted for: date, pre- and infectious periods, immunity duration by disease severity/vaccination status, vaccine efficacy/availability and uptake by age. The model incorporated historical data, scientific publications, and media reports. Adjusted daily mortality included fixed variables with R0 variation over time. Hospitalizations, ICU admissions and mortality were forecasted over one year for each country. Forecast accuracy was assessed quarterly once historical data were available.

Results: Before vaccines, R0 was affected by immunity duration, transmissibility and non-pharmaceutical interventions. In 2021, vaccination increased the model complexity, and was updated to estimate the susceptibility pool over time by country based on: vaccination/booster schedule by age, vaccine uptake, efficacy and immunity duration. These factors vary significantly and require country-specific modelling to predict COVID-19 mortality.
The forecast accurately modeled COVID-19 short-term and showed surges in the future. By July 2020, the model predicted 29.4 million (M) infections in the US, compared to the 17.1M - 35.1M infections estimated by the US National Institute of Health (NIH) seroprevalence study (based on 4.5M confirmed cases). By the end of 2021, the model showed 254M cumulative infections and 3.63M hospitalizations in the US, compared to 206M - 424M infections reported by NIH and 3.71M hospitalizations reported by Center for Disease Control (CDC). For 2022, the model projected 249M infections/4.2M hospitalizations in the US, and 2.8 billion infections/35.5M hospitalizations worldwide.

Conclusions: This SEIRS-based model that integrates country specific data accurately forecasts COVID-19 mortality and demonstrates an emerging persistent state. This represents the only model known to predict long-term scenarios under dynamic conditions for COVID-19 globally.