Advanced Spatial Statistical Modelling of COVID-19 Data

新冠肺炎大流行分几个阶段发生。2020年春季的第一波疫情的特点是重症监护病房(ICU)的占用压力很大，聚合酶链式反应检测的可用性有限，导致病例发现率较低。与第一波相比，第二波显示老年人口的高发病率，导致高死亡率和对ICU床位的需求增加。第一波和第二波都是以感染人数的强劲增长开始的。相比之下，2021年春季的第三次浪潮的特点是感染适度增加，死亡人数稳定，年轻患者导致ICU床位占有率和住院时间分布发生变化。这三波疫情的这些不同特征说明了大流行的动态方面，这些方面在数据方面反映在特定年龄的发病率、超额死亡率和ICU床位占用上。对于有效的大流行控制，需要对感染行为进行小规模的时空分析，并对遏制措施的有效性进行详细评估。然而，由于报告延迟、改变检测策略以及引入强制性快速检测等问题，对发病率数字的直接分析是有问题的。在对其他流行病指标的分析中也出现了类似的问题。此外，不同措施的同时实施以及数据受到测量误差的影响，使得遏制措施的有效性的统计量化变得困难。我们应用先进的统计模型来处理上述问题，以便展示新冠肺炎大流行的区域格局，并使用小区域二次数据来评估遏制措施对局部大流行情况的有效性。拟议的项目有两个目标：首先，我们研究社会经济、公共卫生和社交媒体数据与感染的相互作用，以探索和解释大流行的过程如何以及为什么在德国地区发展不同。这一观点旨在了解三波的不同模式，从而为未来的监测和早期热点检测提供更可靠的工具。其次，我们考察了住院人数和ICU占有率，并开发了一个模型来估计不同地区水平上的ICU入院率。这使我们能够反映当地的感染动态，并通过使用回归和变点模型来估计遏制措施在区域层面上的影响。此外，这些模型还可用于构建ICU占有率的短期预测模型。

The COVID-19 pandemic took place in several phases. The first wave in spring 2020 was characterized by a high pressure on the occupancy of intensive care units (ICU) and limited availability of PCR tests resulting in low case detection rates. The second wave showed high incidence rates in the elderly population resulting in high mortality rates and an increased demand of ICU beds as compared to the first wave. Both, first and second wave started with a strong increase of infections. In contrast, the third wave in spring 2021 was characterized by a moderate increase of infections, stable mortality numbers, and a change in the occupancy and length-of-stay distribution of ICU beds, caused by younger patients. These different characteristics of the three waves illustrate the dynamic aspects of the pandemic, which are data-wise reflected in age-specific incidences, excess mortality, and ICU bed occupancy. For an effective pandemic control, small-scale spatio-temporal analyses of infection behavior are needed, together with a detailed evaluation of the effectiveness of containment measures. However, a straightforward analysis of incidence figures is problematic due to issues such as reporting delays, changing testing strategies, and the introduction of obligatory rapid tests. Similar problems also occur in the analyses of other pandemic indicators. In addition, the simultaneous implementation of different measures and the fact that the data are subject to the measurement errors make the statistical quantification of the effectiveness of containment measures difficult. We apply advanced statistical modelling to cope with the above-mentioned problems in order to exhibit regional patterns of the COVID-19 pandemic and to assess the effectiveness of containment measures on the local pandemic situation using secondary small-area data. The proposed project has two objectives: First, we take a look at the interplay of socio-economic, public health and social media data and infections to explore and explain how and why the course of the pandemic developed differently in German districts. This view aims to understand the different patterns of the three waves and thus provides a more reliable tool for future surveillance and early hotspot detection. Second, we look at hospitalizations and ICU occupancy and develop a model for estimating ICU admissions on different regional levels. This allows us to mirror local infection dynamics as well as to estimate the impact of containment measures on a regional level by using regression and changepoint models. Furthermore, these models can be used to build a short-term forecasting model of ICU occupancy.