Zhang Yuzi, Chang Howard H, Iuliano A Danielle, Reed Carrie
Department of Biostatistics and Bioinformatics, The Rollins School of Public Health of Emory University, 1518 Clifton Rd. N.E., Atlanta, GA 30322, USA.
Influenza Division, National Center for Immunization and Respiratory Diseases, Centers for Disease Control and Prevention, Atlanta, GA, USA.
Spat Stat. 2022 Aug;50:100584. doi: 10.1016/j.spasta.2021.100584. Epub 2022 Jan 4.
In the United States, COVID-19 has become a leading cause of death since 2020. However, the number of COVID-19 deaths reported from death certificates is likely to represent an underestimate of the total deaths related to SARS-CoV-2 infections. Estimating those deaths not captured through death certificates is important to understanding the full burden of COVID-19 on mortality. In this work, we explored enhancements to an existing approach by employing Bayesian hierarchical models to estimate unrecognized deaths attributed to COVID-19 using weekly state-level COVID-19 viral surveillance and mortality data in the United States from March 2020 to April 2021. We demonstrated our model using those aged years who died. First, we used a spatial-temporal binomial regression model to estimate the percent of positive SARS-CoV-2 test results. A spatial-temporal negative-binomial model was then used to estimate unrecognized COVID-19 deaths by exploiting the spatial-temporal association between SARS-CoV-2 percent positive and all-cause mortality counts using an excess mortality approach. Computationally efficient Bayesian inference was accomplished via the Polya-Gamma representation of the binomial and negative-binomial models. Among those aged years, we estimated 58,200 (95% CI: 51,300, 64,900) unrecognized COVID-19 deaths, which accounts for 26% (95% CI: 24%, 29%) of total COVID-19 deaths in this age group. Our modeling results suggest that COVID-19 mortality and the proportion of unrecognized deaths among deaths attributed to COVID-19 vary by time and across states.
自2020年以来，新冠病毒病（COVID-19）已成为美国主要的死亡原因。然而，死亡证明上报的COVID-19死亡人数可能低估了与严重急性呼吸综合征冠状病毒2（SARS-CoV-2）感染相关的总死亡人数。估算那些未通过死亡证明记录的死亡人数，对于了解COVID-19对死亡率的全面影响至关重要。在这项研究中，我们探索了对现有方法的改进，通过使用贝叶斯分层模型，利用2020年3月至2021年4月美国各州每周的COVID-19病毒监测和死亡率数据，估算因COVID-19导致的未被确认的死亡人数。我们使用那些死亡的[具体年龄]岁人群的数据来展示我们的模型。首先，我们使用时空二项回归模型来估算SARS-CoV-2检测阳性结果的百分比。然后，利用超额死亡率方法，通过SARS-CoV-2阳性百分比与全因死亡人数之间的时空关联，使用时空负二项模型来估算未被确认的COVID-19死亡人数。通过二项模型和负二项模型的波利亚 - 伽马表示法实现了计算效率高的贝叶斯推理。在[具体年龄]岁人群中，我们估算有58,200例（95%可信区间：51,300, 64,900）未被确认的COVID-19死亡，占该年龄组COVID-19总死亡人数的26%（95%可信区间：24%，29%）。我们的建模结果表明，COVID-19死亡率以及COVID-19所致死亡中未被确认死亡的比例随时间和各州而有所不同。
