Nearly 156,000 more Americans may have died of COVID-19 in the first year of the pandemic than officially reported, with disparities by race, education, and other factors, a machine-learning study estimates.
For the study, published yesterday in Science Advances, a team led by a Stanford University investigator used machine learning (artificial intelligence) trained on US death certificates to predict unrecognized COVID-19 fatalities from March 2020 to December 2021.
They estimated unrecognized COVID-19 deaths using the predicted number of such deaths in out-of-hospital settings and an adjusted reporting ratio (ARR) that estimated underreporting of COVID-19 deaths in all settings. They did so by weighting the estimated degree of misclassification of COVID-19 fatalities in non-hospital settings by the proportion of deaths that occurred in non-hospital settings.
Most research on unrecognized US COVID-19 deaths has relied on excess-mortality models, which estimate deaths attributable to the pandemic by comparing observed all-cause deaths to those expected based on prepandemic trends, the researchers said. All-cause death data also include fatalities from external causes, such as injuries, unlikely to be caused by COVID-19 in the short term.
A 2021 study by Boston University researchers reached a similar conclusion, estimating that COVID-19 killed 20% more Americans—particularly racial minorities—than officially reported. The authors of that study said the findings, published in PLOS Medicine, underscore the need to adapt policies to address deepening racial and sociodemographic disparities.
Previous studies with different modeling specifications have estimated that excess deaths surpassed those of reported COVID-19 deaths in 2020 by 28%, 38%, and 14%.
“Accurate and timely mortality statistics are critical for health system responses during public health emergencies,” the authors of the new study wrote. “Throughout the COVID-19 pandemic in the United States (US), official COVID-19 mortality reporting was often delayed or incomplete.”
In-home deaths could be 160% higher than reported
From March 2020 to December 2021, the estimated 995,787 US COVID-19 fatalities (95% uncertainty interval (UI), 990,313 to 1,001,363) exceeded the 840,251 officially reported COVID-19 deaths (those with COVID-19 listed anywhere on the death certificate) by 19%. This equals 155,536 predicted unrecognized COVID-19 deaths in out-of-hospital settings during this time.
The number of COVID-19 deaths occurring in homes was 160% higher than officially reported, indicating 111,245 predicted unrecognized at-home COVID-19 deaths. In total, 17,346 misattributed fatalities occurred in hospices, 14,832 took place in outpatient settings and emergency departments, and 9,452 occurred in unclassified settings.
The number of COVID-19 deaths occurring in homes was 160% higher than officially reported.
COVID-19 deaths were most likely to go unrecognized in southern states. The estimated number of COVID-19 deaths was 31% higher than officially reported in the West South Central region, 26% higher in the Middle Atlantic, and 25% higher in the East South Central region.
The states where COVID-19 deaths were most likely to be mischaracterized were Alabama (ARR, 1.67), Oklahoma (ARR, 1.51), and South Carolina (ARR, 1.47). The states with the largest estimates of unrecognized COVID-19 deaths were Texas (24,024 deaths), New York (23,005), California (11,613), Alabama (11,501), Florida (7,718), and South Carolina (7,224).
Official and predicted unrecognized deaths correspond
COVID-19 fatalities were most likely to go unrecognized in the first pandemic wave (highest monthly ARR, 1.49), declining in likelihood in the second surge (highest monthly ARR, 1.19) and subsequent waves. The most predicted unrecognized COVID-19 deaths occurred in January 2021 (35,665 deaths) and April 2020 (32,110).
Peaks in such deaths corresponded to monthly peaks in officially reported COVID-19 deaths, with the largest number occurring during the same months. “This correspondence supports the finding that the unrecognized COVID-19 deaths we predicted were COVID-19 deaths,” the authors wrote.
The COVID-19 deaths of people aged 65 to 84 years were most likely to be miscategorized (ARR for 65 to 74 years, 1.21; ARR for 75 to 84 years, 1.22). Deaths in male patients were more likely to go unrecognized than in female patients (1.22 vs 1.14 ARR).
COVID-19 fatalities were also more likely to be mischaracterized for people with less than a high school education (ARR, 1.29; ARR for high school diploma or equivalent, 1.18; ARR for some college, 1.15).
COVID-19 deaths among Hispanic people were most likely to be unrecognized (ARR, 1.31). Relative to White people (ARR, 1.15), deaths were also more likely mischaracterized on certificates indicating American Indian and Alaska Native (AIAN; ARR, 1.24), Asian (ARR excluding Indian, 1.24), (ARR for Indian, 1.23) and Black (ARR, 1.19).
In absolute terms, 78,561 predicted unrecognized COVID-19 deaths occurred among White adults, 43,132 in Hispanic adults, 23,617 in Black adults, 5,772 deaths in Asian adults, and 2,318 in AIAN adults. The unrecognized fatalities were attributed to underlying causes such as Alzheimer’s disease and related dementia, cardiovascular disease, and diabetes.
Uneven effects on people of different races, backgrounds
Deaths in counties in the lowest quintile of median household income (ARR, 1.34) and the lowest quintile of residents with some or more college (ARR, 1.37) were more likely to be miscategorized. Similarly, deaths were more likely to be unrecognized in counties with the greatest proportion of people reporting poor or fair health (ARR, 1.30) and diabetes (ARR, 1.33).
The authors said COVID-19 fatalities may go unrecognized because of factors such as decedent and family attitudes and stigma, death investigators’ political roles and conflicts of interest, community testing rates, funding for death investigation offices, and racism, ableism, and classism.
The primary contribution of this study is our predictions showing that unrecognized COVID-19 deaths occurred unevenly, disproportionately affecting decedents,” of different races and backgrounds.
“Our national estimate of total predicted COVID-19 deaths is similar to the estimate of the US Centers for Disease Control and Prevention (CDC) of just under 1 million excess deaths for this period,” they wrote.
“The primary contribution of this study is our predictions showing that unrecognized COVID-19 deaths occurred unevenly, disproportionately affecting decedents,” of different races and backgrounds,” they added. “Together, this study indicates that the US death investigation system reported COVID-19 deaths in a systematically inequitable way that hid the true extent of pandemic mortality and inequities.”
The researchers cautioned that while machine learning approaches may help improve death estimates, larger investments in death investigation system reform, efforts to study and address bias in the system, and programs to build death investigation infrastructure globally are needed.
