Spatial disparities in the mortality burden of the covid-19 pandemic across 569 European regions (2020–2021)

In 2023, deaths due to COVID-19 fell markedly compared to 2020–2022, prompting the WHO to end the global health emergency on 6 May 2023. It is therefore timely to evaluate the overall mortality burden of the pandemic, focusing on 2020 and 2021 when its impact peaked. While early work relied on reported COVID-19 deaths, excess mortality—defined as the difference between observed deaths and those expected in the absence of the pandemic—has become the gold standard due to inconsistencies in surveillance data across countries. Most excess mortality studies are national-level, and regional analyses typically cover single countries using heterogeneous methods and indicators (e.g., life expectancy, death counts), hindering cross-country regional comparisons. Moreover, metrics like life expectancy and age-standardised death rates are not additive across years, limiting their ability to summarise the total burden over 2020 and 2021. This paper addresses these gaps by estimating ASYLL for 569 comparable regions across 25 European countries, forecasting expected (counterfactual) mortality with a CP-spline approach that accounts for regional diversity and yields confidence intervals. We quantify and map spatial disparities in excess mortality for each year and cumulatively, and analyse how geographic patterns shifted between 2020 and 2021.

The study situates itself within a literature that increasingly uses excess mortality to assess the pandemic’s impact, given limitations of reported COVID-19 death data. Prior work largely focuses on national estimates, with some single-country regional studies using differing baselines (pre-pandemic levels vs. forecasts) and indicators (life expectancy, death tolls), complicating cross-regional comparisons. Only two peer-reviewed studies enable simultaneous comparison of regional excess mortality across multiple European countries for 2020, and one study covers 200 NUTS-2 regions for 2020–2022 but omits Germany, the UK, Ireland, and Sweden. The UK ONS has provided weekly regional excess mortality for Europe online. These heterogeneous methods and indicators underscore the need for a harmonised, comparable regional assessment across countries and for both 2020 and 2021.

Data preparation: Subnational death and population counts by age and sex were compiled for 25 European countries from Eurostat, the Human Mortality Database, and national statistical offices. Spatial units followed the NUTS hierarchy to ensure comparability: NUTS-3 for Czechia, Denmark, France, Italy, Luxembourg, Poland, Slovakia, Spain, and Sweden; NUTS-2 for Austria, Belgium, Estonia, Finland, Hungary, Iceland, Latvia, Lithuania, the Netherlands, Norway, Portugal, Slovenia, Switzerland, England and Wales; NUTS-1 for Ireland, Northern Ireland, and Scotland. Germany used national "Raumordnungsregionen." Minor territorial adjustments were applied for consistency. Regional aggregates were validated against national HMD totals, showing negligible differences. Heterogeneous age groupings were harmonised to single years of age up to 95+, with ungrouping applied where needed. The final dataset comprised 569 regions with populations from ~40,000 (Bornholm, Denmark) to ~6.75 million (Madrid, Spain).

Baseline mortality forecasting: To estimate expected mortality absent the pandemic for 2020 and 2021, the study used CP-splines, combining two-dimensional P-splines with demographic priors. This non-parametric framework flexibly captures diverse age profiles and temporal trends across many small populations and yields smooth, plausible surfaces. For each region, time windows were optimised via a rolling starting year (up to 2010). Models forecasted 2019 and the Poisson deviance between observed and forecasted 2019 mortality was used to select the best starting year; the chosen window informed forecasts for 2020 and 2021. This approach accommodates regional heterogeneity and provides confidence intervals for excess mortality.

Excess mortality metric (ASYLL): Age-Standardised Years of Life Lost captures the cumulative years of life lost due to excess deaths, allowing additivity over time and comparability across populations. Steps: (1) estimate age-specific excess mortality (observed minus forecasted rates); (2) convert excess deaths to years of life lost using remaining life expectancy at each age; (3) sum across ages; (4) age-standardise using the 2013 European Standard Population (ESP) to remove the influence of population size and age structure. ASYLL is expressed as years of life lost per 1000 standard population; for example, ASYLL=20 indicates 20 years of life lost per 1000.

Software and reproducibility: Analyses were performed in R 4.3.1; ArcGIS 10.8.1 was used to process shapefiles and produce maps. Source data and code for reproducing French NUTS-3 estimates are provided in the supplementary materials and OSF repository.

• 2020: Notable ASYLL losses occurred in 362 of 569 regions; only 7 regions experienced gains (negative ASYLL). Eight regions exceeded 20 years of life lost per 1000 population.
• 2021: Notable losses rose to 440 regions; only 4 regions experienced gains. Seventy-five regions exceeded 20 years of life lost per 1000 population.
• Spatial patterns: In 2020, highest losses concentrated in early outbreak areas—northern Italy, southern Switzerland, central Spain, and Poland; many regions in France, Germany, southern UK, Finland, Iceland, Northern Ireland, Estonia, Latvia, and Hungary had modest losses; gains were found in parts of western/southwestern France and in Denmark, Norway, and Sweden. In 2021, an East–West gradient emerged: Eastern Europe showed the highest losses, especially among men, with Slovakia, Hungary, and Latvia exceeding 35 years per 1000; Western Europe mostly saw moderate losses or gains.
• Borders and gradients: German–Polish and German–Czech borders did not clearly demarcate high-loss zones; adjacent German regions showed intermediate losses relative to neighbouring Czechia and Poland.
• Combined 2020–2021: For both sexes combined, only 2 regions had a significant gain; 458 regions had notable losses, with 136 exceeding 20 years per 1000. High excess mortality was more frequent among men: 151 regions (men) vs 73 (women) with high ASYLL.
• Country contrasts: In Italy (2020), no excess mortality in Caltanissetta, Trapani, and Potenza (both sexes combined) versus >38 per 1000 in Bergamo and Cremona. In Germany (2021), near-significant gain among males in East Schleswig-Holstein versus >16.5 per 1000 in South Saxony and North Thuringia. In Poland (2020–2021), Poznań region (West) had ~32 per 1000 (both sexes combined) versus at least 60 per 1000 in Puławski region (East).
• Top values: Among men, very high ASYLL in Cremona (57.1 per 1000; 95% CI 49–65.2) and Bergamo (51.7; 95% CI 46.7–56.7). Over two years, many of the most-affected regions were in Poland and Slovakia; Estonia had the lowest male excess mortality among the Baltic states.
• Association with baseline mortality: Pooled analysis shows a negative association between pre-pandemic life expectancy (2015–2019) and ASYLL, weak in 2020 and strong in 2021. However, stratification by Central and Eastern Europe (CEE) vs Western Europe reveals no consistent within-stratum relationship (an illustration of Simpson’s paradox).

Estimating excess mortality at a fine regional scale reveals substantial heterogeneity that national aggregates obscure. Using ASYLL enabled additive assessment across 2020 and 2021, showing that while early hotspots in 2020 were concentrated in northern Italy, southern Switzerland, central Spain, and parts of Poland, the pandemic’s burden shifted in 2021 to an East–West gradient with Eastern Europe bearing the highest losses, especially among men. The ecological analysis suggests that the apparent negative association between baseline life expectancy and excess mortality in pooled data disappears when stratifying by CEE versus Western Europe, indicating that structural regional differences drive observed patterns. Potential contributors to the East–West gap include later epidemic onset in the East, greater vulnerability due to pre-existing health burdens and socioeconomic factors, selective migration, lower compliance with interventions and vaccination, and lower trust in authorities—factors with roots in historical and systemic contexts. Methodologically, the study leverages high-quality vital statistics, validates regional aggregates against HMD, and uses a robust, flexible CP-spline forecasting framework with uncertainty quantification. The focus on 2020–2021 isolates the period when excess mortality was most directly attributable to COVID-19, avoiding confounding from later dynamics (e.g., vaccination uptake, influenza waves in late 2022). Findings emphasise the need for regionally tailored public health responses and rapid measures targeting highly connected transit hubs to slow spread from initial outbreak areas.

This study provides a harmonised, cross-country regional assessment of the COVID-19 mortality burden across 569 European regions for 2020 and 2021 using ASYLL and a robust CP-spline forecasting framework. It documents stark spatial disparities, a pronounced shift from localized early hotspots in 2020 to an East–West gradient in 2021, and greater male excess mortality in many regions. The results demonstrate the value of regional analysis for understanding pandemic impacts and informing targeted policies, particularly rapid interventions limiting connectivity at major transit hubs to prevent widespread diffusion. Future research directions include: (1) linking regional excess mortality to contextual factors and policy measures (e.g., social distancing, international isolation, vaccination) through ecological and epidemiological analyses to identify drivers of spatial disparities; (2) comparing COVID-19 excess mortality with other historical mortality crises to contextualize its severity and guide preparedness; and (3) extending models to incorporate spatial dependence and cause-of-death information to capture harvesting effects and refine estimates.

• Spatial dependence not modeled: Regions were modeled independently without explicit spatial autocorrelation, which could reduce uncertainty if incorporated.
• Harvesting effects: The approach does not adjust for potential mortality displacement following initial waves, potentially influencing 2021 estimates.
• Temporal scope: Analyses are limited to 2020–2021 to maintain attribution to COVID-19; inclusion of 2022 would confound interpretation due to vaccination effects and severe influenza waves.
• Baseline modeling choices: Although CP-splines with optimized time windows are robust, alternative baseline specifications may yield different estimates; sensitivity analyses from other studies suggest potential variability.
• Data structure: Annual (not weekly) data were used; some prior work with weekly data has modeled spatial dependence but operates in a different framework.