Presentation Information
[C07-01]Flaws in Simulations of COVID-19 Cases and Deaths Based on Transmission Models in Japan
*Hideki Kakeya1, Takashi Nakamura2, Yoshitaka Umeno3 (1. University of Tsukuba (Japan), 2. Tokyo University of Science (Japan), 3. The University of Tokyo (Japan))
Keywords:
SIR model,COVID-19,cross validation,sensitivity analysis,confidence interval
Kayano et al. (Scientific Reports, 2023) estimated the potential number of COVID-19 cases and deaths in Japan from February 17 to November 30, 2021, in a scenario where no vaccination was implemented. Their model predicted 63.3 million cases and 364,000 deaths, with a claimed 95% confidence interval of less than 1% of the estimated values. Kayano et al. (BMC Infectious Diseases, 2023) also projected that, in early 2022, the number of prevented cases in Tokyo would exceed the population of individuals under 49 years old, had no vaccination program been in place.
The objective of this study was to validate the transmission model used by Kayano et al. by simulating infection counts in a different time frame. We aimed to assess the impact of errors in the reproduction number within the counterfactual scenario and to compare infection curves under different settings.
To do this, we replicated Kayano et al.’s model to simulate infection dynamics in Japan during 2020. We then evaluated the model's performance by comparing simulated infection surges with actual data. We also analyzed how ±10% errors in the reproduction number of the Delta variant, corresponding to the reported 95% confidence interval in the original study, impacted the infection count estimates. Additionally, we compared infection curves produced under different simulation settings.
Our findings showed that while the model was able to reproduce the first infection surge in early 2020, it failed to accurately simulate the second and third surges in the summer and winter of the same year. Sensitivity analysis indicated that errors in the reproduction number caused infection count estimates to vary by -25% to +42%. Moreover, the infection curves generated by Kayano et al. under different methods aligned when scaling factors were adjusted.
These results suggest potential flaws in Kayano et al.’s counterfactual simulation, particularly regarding the accuracy of infection count estimates, the reported confidence intervals, and the authenticity of infection curves generated by different methods.
The objective of this study was to validate the transmission model used by Kayano et al. by simulating infection counts in a different time frame. We aimed to assess the impact of errors in the reproduction number within the counterfactual scenario and to compare infection curves under different settings.
To do this, we replicated Kayano et al.’s model to simulate infection dynamics in Japan during 2020. We then evaluated the model's performance by comparing simulated infection surges with actual data. We also analyzed how ±10% errors in the reproduction number of the Delta variant, corresponding to the reported 95% confidence interval in the original study, impacted the infection count estimates. Additionally, we compared infection curves produced under different simulation settings.
Our findings showed that while the model was able to reproduce the first infection surge in early 2020, it failed to accurately simulate the second and third surges in the summer and winter of the same year. Sensitivity analysis indicated that errors in the reproduction number caused infection count estimates to vary by -25% to +42%. Moreover, the infection curves generated by Kayano et al. under different methods aligned when scaling factors were adjusted.
These results suggest potential flaws in Kayano et al.’s counterfactual simulation, particularly regarding the accuracy of infection count estimates, the reported confidence intervals, and the authenticity of infection curves generated by different methods.