Unreliability in Simulations of COVID-19 Cases and Deaths Based on Transmission Models

Predictions on the number of COVID-19 infections by researchers have failed repeatedly in Japan. In this review we discuss the mathematical models that led to these failures in predictions. Specifically, we focus on two papers published in October 2023, both of which simulate counterfactual COVID-19 cases and deaths using transmission models. One paper estimates that the COVID-19 cases and deaths from February 17 to November 30, 2021 in Japan would have been as many as 63.3 million and 364 thousand respectively had the vaccination not been implemented, where the 95\% confidence interval is claimed to be less than 1\% of the estimated value. It also claims that the cases and deaths could have been reduced by 54\% and 48\% respectively had the vaccination been implemented 14 days earlier. The other paper estimates that the number of cases in early 2022, in Tokyo would have been larger than the population in the age group under 49 in the absence of the vaccination program. We reexamine the results given by these papers to find that the simulation results do not explain the real-world data in Japan including prefectures with early/late vaccination schedules. The cause of the discrepancy is identified as the low reliability of model parameters that immensely affect the simulation results of case and death counts. Finally, we review a series of failed predictions on the number of infections during the COVID-19 pandemic in Japan and discuss the implications for future public health.