gms | German Medical Science

Introduction: Estimating COVID-19 mortality is impeded by uncertainties in cause of death coding. In contrast, excess all-cause mortality can be a robust indicator of how the COVID-19 pandemic impacts public health. However, correctly predicting baseline mortality is crucial for estimating excess mortality. Therefore, we modeled baseline mortality taking into account strong varying forces on a population's mortality experience. We compared excess mortality estimates to COVID-19 attributed fatalities to assess their agreement and evaluate possible indirect public health effects of the pandemic.

Methods: Using monthly data from 2010 to 2019 from Germany’s federal states, we built a baseline regression model using a Bayesian approach [1]. The model took into account age, sex, temperature, influenza activity, regional multiple deprivation, and cyclic patterns to predict the mortality rate. The model was validated by comparing observed and predicted mortality for 2019 after being trained on data from 2010 to 2018. Excess mortality from 01/2020 to 01/2022 was assessed as the difference between observed and predicted mortality. Prediction uncertainty bounds were calculated as the 2.5% and 97.5% quantiles of the posterior predictive distribution.

Results: Modelled excess mortality in 2020 and 2021 showed large heterogeneity with respect to regions and time periods, ranging from -16.4% to 47.5% relative to observed monthly mortality. Spatiotemporal excess mortality patterns were largely consistent with COVID-19 attributed mortality with a rank correlation of 0.50, and 83.8% of state-month combinations where COVID-19 attributed mortality was included in the 95% prediction uncertainty bounds for excess mortality. In Saxony and Saxony-Anhalt in 12/2020, COVID-19 mortality was high (37.0%, 23.9% of total mortality), but excess mortality even higher (47.5%, 28.9%). From 07/2021 to 12/2021, several federal states showed consistently higher excess mortality than COVID-19 attributed fatalities. For 2019, model validation showed a mean absolute percentage error between observed and predicted fatalities of 5.1% over months and federal states.

Discussion: The strong heterogeneity in excess mortality renders summary measures over large regions and time periods uninformative. The agreement of spatiotemporal patterns of excess mortality with those of COVID-19 attributed mortality until 06/2021 is consistent with the assumption that direct estimates of COVID-19 fatalities were without strong bias, and that only COVID-19 substantially contributed to excess mortality. However, as some excess mortality from 07/2021-12/2021 is not accounted for by COVID-19 attributed fatalities, further analysis using the registered causes of death are warranted.

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.

Earlier analysis of a subset of the data in [1].