Estimating the true burden of SARS-CoV-2 infection and mortality proved challenging during the COVID-19 pandemic because case detection varied with testing availability. The authors evaluated how aggregation by cause of death and by spatial unit can mask underlying trends in pandemic-related mortality. Using mortality data from Michigan, they compared aggregated and disaggregated approaches to determine how cause-specific and local patterns—particularly for acute respiratory infections (ARI)—were reflected in excess mortality measures.
The analysis used mortality records from the state of Michigan spanning January 2015 through November 2019 to define baseline expectations. Expected mortality for December 2019 through November 2022 was estimated with negative binomial regression. Observed deaths in that later period were compared to expected counts to quantify excess mortality. The pandemic interval was divided into seven distinct periods based on dominant viral strains and on pharmaceutical and non-pharmaceutical interventions implemented during the pandemic.
Deaths were aggregated by cause and at two spatial scales. Cause was reduced into two coarse categories: acute respiratory infections (ARI) and all other causes (non-ARI). Spatial aggregation was performed at the county level and, in parallel analyses, at the finer census tract level. The study also evaluated spatial heterogeneity of excess mortality using the Theil index to separate variation driven by differences between counties from variation within counties.
The authors report a pronounced rise in ARI excess mortality early in the pandemic. The peak ARI excess mortality was described as an observed-to-expected ratio of 10.7; the confidence bounds as reported in the source were unclear. By comparison, all-cause excess mortality rose to a much smaller extent, reaching a ratio of 1.20 observed-to-expected deaths (95% CrI: 1.17–1.22). Across the evaluation period, non-ARI excess mortality remained essentially unchanged.
These contrasting patterns indicate that aggregating causes into a single all-cause metric can obscure large, cause-specific increases in mortality driven by ARI-related deaths during acute pandemic phases.
Comparing excess mortality estimates at the county and census tract levels revealed substantial and inconsistent differences for ARI. County-level ARI excess mortality estimates ranged from 0.42 to 1.2 times the corresponding census-tract estimates, demonstrating that aggregation to the county level can either understate or overstate local ARI burdens. In contrast, county-level estimates for non-ARI and for all-cause excess mortality were consistently closer to tract-level estimates, suggesting less sensitivity of these aggregated metrics to spatial scale.
The study also found that within-county variation in ARI mortality decreased during the pandemic’s acute phases and then slowly returned toward reference (pre-pandemic) levels. This pattern suggests that as ARI rates rose statewide, spatial patterning across census tracts became more homogeneous, reducing relative inequality within counties during peaks.
At the census tract level, greater social vulnerability was associated with higher ARI excess mortality during the first pandemic phase. The excess mortality rate ratio (EMRR) for social vulnerability was reported as 1.1 (95% CrI: 1.08–1.13). However, this association attenuated when results were aggregated into all-cause mortality or when analyses were performed at the county level. Thus, coarse aggregation by cause or place reduced the ability to detect important social gradients in pandemic-related ARI mortality.
The findings indicate that reliance on county-level all-cause excess mortality as a proxy for ARI or COVID-19-specific mortality may obscure meaningful signals. Using only all-cause, county-aggregated metrics limits the power to identify large shifts in cause-specific mortality and to detect local associations with social factors.
The authors suggest that even modest disaggregation—separating ARI versus non-ARI deaths—and use of finer spatial units such as census tracts can improve detection of local trends and of social determinants associated with elevated ARI mortality. These refinements can better inform targeted surveillance, healthcare allocation, and public health responses during infectious disease outbreaks.
The report used Michigan mortality data and statistical models as described; some reported uncertainty bounds in the source were unclear. The authors declared that the University of Michigan IRB provided ethical approval and that relevant reporting guidelines were followed. Data can be requested from the Michigan Department of Health and Human Services. The authors declared no competing interests.