Excess mortality is often used to assess the health impact of the COVID-19 pandemic. It involves comparing the number of deaths observed during the pandemic with the number of deaths that would counterfactually have been expected in the absence of the pandemic. However, published data on excess mortality often vary even for the same country. The reason for these discrepancies is that the estimation of excess mortality involves a number of subjective methodological choices. The aim of this paper was to summarize these subjective choices. In several publications, excess mortality was overestimated because population aging was not adjusted for. Another important reason for different estimates of excess mortality is the choice of different pre-pandemic reference periods that are used to estimate the expected number of deaths (e.g., only 2019 or 2015–2019). Other reasons for divergent results include different choices of index periods (e.g., 2020 or 2020–2021), different modeling to determine expected mortality rates (e.g., averaging mortality rates from previous years or using linear trends), the issue of accounting for irregular risk factors such as heat waves and seasonal influenza, and differences in the quality of the data used. We suggest that future studies present the results not only for a single set of analytic choices, but also for sets with different analytic choices, so that the dependence of the results on these choices becomes explicit.

超额死亡率通常用于评估 COVID-19 大流行对健康的影响。它涉及将大流行期间观察到的死亡人数与在没有大流行的情况下反事实预期的死亡人数进行比较。然而，即使是同一个国家，已公布的超额死亡率数据也往往各不相同。造成这些差异的原因是超额死亡率的估计涉及许多主观的方法选择。本文的目的是总结这些主观选择。在一些出版物中，过高的死亡率被高估了，因为没有对人口老龄化进行调整。对超额死亡率进行不同估计的另一个重要原因是选择了不同的大流行前参考期来估计预期的死亡人数（例如，仅 2019 年或 2015-2019 年）。导致不同结果的其他原因包括指数期的不同选择（例如，2020 年或 2020-2021 年）、确定预期死亡率的不同模型（例如，前几年的平均死亡率或使用线性趋势）、不规则风险的核算问题热浪和季节性流感等因素，以及所用数据质量的差异。我们建议未来的研究不仅针对单个分析选择集呈现结果，而且针对具有不同分析选择的集合呈现结果，以便结果对这些选择的依赖性变得明确。前几年的平均死亡率或使用线性趋势），考虑热浪和季节性流感等不规则风险因素的问题，以及所用数据质量的差异。我们建议未来的研究不仅针对单个分析选择集呈现结果，而且针对具有不同分析选择的集合呈现结果，以便结果对这些选择的依赖性变得明确。前几年的平均死亡率或使用线性趋势），考虑热浪和季节性流感等不规则风险因素的问题，以及所用数据质量的差异。我们建议未来的研究不仅针对单个分析选择集呈现结果，而且针对具有不同分析选择的集合呈现结果，以便结果对这些选择的依赖性变得明确。