Applying machine learning to identify unrecognized COVID-19 deaths recorded as other causes of death in the United States
- Mathew V. Kiang
- Zehang Richard Li
- Elizabeth Wrigley-Field
- Rafeya V. Raquib
- Dielle J. Lundberg
- Eugenio Paglino
- Benjamin Huynh
- Kirsten Bibbins-Domingo
- M. Maria Glymour
- Andrew C. Stokes
2026-03-18
The actual number of US deaths caused by severe acute respiratory syndrome coronavirus 2 infection has been investigated and debated since the start of the COVID-19 pandemic. Here, we use machine learning trained on US death certificates from March 2020 to December 2021 to predict 155,536 (95% uncertainty interval: 150,062 to 161,112) unrecognized COVID-19 deaths. This indicates that 19% more COVID-19 deaths occurred in the US than officially reported. Predicted unrecognized COVID-19 deaths occurred disproportionately among decedents with less than a high school education; decedents identified as Hispanic, American Indian, Alaska Native, Asian, and/or Black; counties with lower household incomes and worse preexisting health; and counties in the South. These findings suggest that the US death investigation system undercounted COVID-19 deaths unevenly, hiding the true extent of inequities.