Lawson Andrew B, Xin Yao
Department of Public Health Sciences, College of Medicine, Medical University of South Carolina, Charleston, SC, United States.
School of Medicine, Usher Institute, University of Edinburgh, Edinburgh, United Kingdom.
Front Epidemiol. 2024 Nov 11;4:1403212. doi: 10.3389/fepid.2024.1403212. eCollection 2024.
During the COVID-19 pandemic, which spanned much of 2020-2023 and beyond, daily case and death counts were recorded globally. In this study, we examined available mortality counts and associated case counts, with a focus on the estimation missing information related to age distributions. In this paper, we explored a model-based paradigm for generating age distributions of mortality counts in a spatio-temporal context. We pursued this aim by employing Bayesian spatio-temporal lagged dependence models for weekly mortality at the county level. We compared three US states at the county level: South Carolina (SC), Ohio, and New Jersey (NJ). Models were developed for mortality counts using Bayesian spatio-temporal constructs, incorporating both dependence on current and cumulative case counts and lagged dependence on previous deaths. Age dependence was predicted based on total deaths in proportion to population estimates. This latent age field was generated as counterfactuals and then compared to observed deaths within age groups. The optimal retrospective space-time models for weekly mortality counts were those with lagged dependence and a function of caseload. Added random effects were found to vary across states: Ohio favored a spatially correlated model, while SC and NJ favored a simpler formulation. The generation of age-specific latent fields was performed for SC only and compared to a 15-month, 13-county data set of observed >65 age population. It is possible to model spatio-temporal variations in mortality at the county level with lagged dependencies, spatial effects, and case dependencies. In addition, it is also possible to generate latent age-specific fields based on estimates of death risk (using population proportions or more sophisticated modeling approaches). More detailed data will be needed to make more calibrated comparisons for future epidemic monitoring. The proposed discrepancy tool could serve as a useful resource for public health planners in tailoring interventions during epidemic situations.
在贯穿2020年至2023年及以后大部分时间的新冠疫情期间，全球记录了每日的病例数和死亡数。在本研究中，我们检查了现有的死亡计数及相关病例计数，重点是估计与年龄分布相关的缺失信息。在本文中，我们探索了一种基于模型的范式，用于在时空背景下生成死亡计数的年龄分布。我们通过采用县级每周死亡率的贝叶斯时空滞后依赖模型来实现这一目标。我们在县级比较了美国的三个州：南卡罗来纳州（SC）、俄亥俄州和新泽西州（NJ）。使用贝叶斯时空结构为死亡计数建立模型，纳入对当前和累计病例计数的依赖以及对先前死亡的滞后依赖。根据与人口估计数成比例的总死亡数预测年龄依赖性。这个潜在年龄字段作为反事实生成，然后与各年龄组内的观察到的死亡数进行比较。每周死亡计数的最佳回顾性时空模型是具有滞后依赖性和病例负荷函数的模型。发现添加的随机效应因州而异：俄亥俄州倾向于空间相关模型，而南卡罗来纳州和新泽西州倾向于更简单的形式。仅对南卡罗来纳州进行了特定年龄潜在字段的生成，并与一个15个月、13个县的65岁以上年龄人口观察数据集进行了比较。可以通过滞后依赖性、空间效应和病例依赖性对县级死亡率的时空变化进行建模。此外，还可以根据死亡风险估计（使用人口比例或更复杂的建模方法）生成特定年龄的潜在字段。需要更详细的数据才能在未来的疫情监测中进行更精确的比较。所提出的差异工具可以作为公共卫生规划者在疫情期间制定干预措施的有用资源。
