Probabilistic Modeling of COVID-19 Fatality Rates: A Method to Correct Reporting Bias
João Rodrigo Souza Leão*, Gilberto Cors and Gustavo Zampier dos Santos Lima
ABSTRACT
An accurate estimation of the COVID-19 case fatality rate (CFR) is crucial for understanding the severity of the disease, forecasting healthcare demands,and evaluating its impact on large populations. However, this metric is often distorted by underreporting and delayed outcomes. In this study, we present a probabilistic model that captures the temporal dynamics of the COVID-19 pandemic and provides corrected estimates of the case fatality rate. The model incorporates transition probabilities between disease states to simu- late the evolution of infections, recoveries, and deaths. It explicitly accounts for asymptomatic, mild/moderate, and severe cases, enabling the estimation of undiagnosed infections within the population. We validate the model by fitting it to official data from medium-sized cities, major metropolitan areas, and medium-sized countries, covering populations ranging from a few million to tens of millions.
Based on the inferred proportion of undiagnosed cases, we compute corrected case fatality rates, which range from 0.33% ± 0.02% to 1.14% ± 0.07%. Remarkably, these values exhibit a degree of universality, appearing largely independent of geographic, social, or demographic factors. Our results are consistent with independent seroprevalence studies and randomized testing, offering a refined understanding of COVID-19 fatality metrics. Additionally, the model provides estimates for the number of severe cases, ICU demand, and the true number of infections, making it a versatile tool for pandemic response planning. Beyond COVID-19, the proposed probabilistic framework is adaptable to future infectious disease outbreaks characterized by significant data incompleteness and reporting biases
