时间：2024.10.11（周五），上午10:30-11:30（北京时间）

地点：数学楼2-3

报告人：何岱海, 香港理工大学

报告题目：The Effectiveness of the COVID-19 Vaccination Campaign in 2021: Inconsistency in Key Studies

报告摘要：In this work, we revisited the evaluation of the effectiveness of the COVID-19 vaccination campaign in 2021, as measured by the number of deaths averted. The published estimates differ a lot: from one widely referenced paper by Watson et al. (2022) estimating 0.5-0.6% of the USA population being saved, to average-level estimates of 0.15-0.2%, and to some estimates as low as 0.0022%. For other countries, Watson et al. gave much higher estimates than all other works too.

We reviewed 30 relevant papers, carried out an in-depth analysis of the model by Watson et al. and of several other studies, and provided our own regression-based analysis of the US county-level data.

The model by Watson et al. is very sophisticated and has many features; some of them that make it more realistic (age-structured epidemiology, “elderly first” vaccination, healthcare overload effects), but others that are likely inaccurate (substantial reinfection rates (i.e., immunity loss) for the Alpha and Delta variants, possible overfitting due to overly flexible time-dependent infection transmission rate) or questionable (45% increase in fatality rate for the Delta variant). Yet, the main argument is that Watson et al.’s model does not reproduce the trends observed in the county-level US data.

Eventually, we concluded that Watson et al.’s 0.5-0.6% is an overestimate, and 0.15-0.2% of the US population saved by vaccination-as estimated by regression studies on subnational-level data (e.g., Suthar et al. (2022) and by He et al. (2022))-is much more plausible value.

In our view, in order to be considered reliable, mathematical models should be tested on more detailed real data that was not used in model fitting. On the other hand, detailed data bring about new challenges in statistical modelling and uncertainties in data reliability.

报告人简介： 何岱海，香港理工大学应用数学系教授。西安交通大学工学博士(1999)和加拿大麦克马斯特大学数学博士(2006)。曾在北京师范大学、密歇根大学和以色列特拉维夫大学做博士后。主要研究兴趣是传染病建模和医疗数据统计分析。在PNAS、Science Advances、Annals of Internal Medicine、European Respiratory Journal、Journal of the Royal Society Interface等期刊发表论文140余篇。Google H-index 47. 连续三年入选斯坦福大学发布的全球年度2%顶尖科学家榜单(2021-2024)。

报告题目：The Effectiveness of the COVID-19 Vaccination Campaign in 2021: Inconsistency in Key Studies

报告摘要：In this work, we revisited the evaluation of the effectiveness of the COVID-19 vaccination campaign in 2021, as measured by the number of deaths averted. The published estimates differ a lot: from one widely referenced paper by Watson et al. (2022) estimating 0.5-0.6% of the USA population being saved, to average-level estimates of 0.15-0.2%, and to some estimates as low as 0.0022%. For other countries, Watson et al. gave much higher estimates than all other works too.

We reviewed 30 relevant papers, carried out an in-depth analysis of the model by Watson et al. and of several other studies, and provided our own regression-based analysis of the US county-level data.

The model by Watson et al. is very sophisticated and has many features; some of them that make it more realistic (age-structured epidemiology, “elderly first” vaccination, healthcare overload effects), but others that are likely inaccurate (substantial reinfection rates (i.e., immunity loss) for the Alpha and Delta variants, possible overfitting due to overly flexible time-dependent infection transmission rate) or questionable (45% increase in fatality rate for the Delta variant). Yet, the main argument is that Watson et al.’s model does not reproduce the trends observed in the county-level US data.

Eventually, we concluded that Watson et al.’s 0.5-0.6% is an overestimate, and 0.15-0.2% of the US population saved by vaccination-as estimated by regression studies on subnational-level data (e.g., Suthar et al. (2022) and by He et al. (2022))-is much more plausible value.

In our view, in order to be considered reliable, mathematical models should be tested on more detailed real data that was not used in model fitting. On the other hand, detailed data bring about new challenges in statistical modelling and uncertainties in data reliability.

报告人简介： 何岱海，香港理工大学应用数学系教授。西安交通大学工学博士(1999)和加拿大麦克马斯特大学数学博士(2006)。曾在北京师范大学、密歇根大学和以色列特拉维夫大学做博士后。主要研究兴趣是传染病建模和医疗数据统计分析。在PNAS、Science Advances、Annals of Internal Medicine、European Respiratory Journal、Journal of the Royal Society Interface等期刊发表论文140余篇。Google H-index 47. 连续三年入选斯坦福大学发布的全球年度2%顶尖科学家榜单(2021-2024)。