- Machine learning algorithms estimated 995,787 deaths from COVID-19 in the United States from March 2020 to December 2021, 19% more than officially reported.
- Deaths from coronavirus among older adults are likely to be 21% to 22% higher than officially reported, and deaths among Hispanic individuals are likely to be 31% higher.
- Many of the uncounted coronavirus deaths were attributed to Alzheimer’s disease, cardiovascular disease, and diabetes.
The official U.S. coronavirus death toll may have missed more than 150,000 unconfirmed deaths during the first two years of the pandemic, a machine learning modeling study suggests.
New estimates show that 995,787 people died from coronavirus between March 2020 and December 2021, nearly 20% higher than the official U.S. tally of 840,251 deaths recorded on death certificates from coronavirus over the same period (adjusted reporting rate). [ARR] 1.19, 95% uncertainty interval [UI] According to Dr. Andrew Stokes of Boston University et al., 1.18-1.19).
These unrecognized coronavirus deaths are more likely to occur among older people, nonwhites, low-income people, and residents of certain regions of the country, the researchers reported. scientific progressfound that “the U.S. death surveillance system systematically reported COVID-19 deaths in inequitable ways, concealing the true extent of the pandemic’s mortality and inequities.”
“If we improve the quality of our death investigation systems and national death surveillance data, we will be far better prepared to detect the next pandemic earlier and more completely,” Stokes said. today’s med page. “These machine learning techniques can also be applied to understand other causes of death that have long been undervalued, such as suicide mortality, Alzheimer’s disease and dementia mortality, diabetes, and drug overdose.”
Previous studies have relied on excess mortality models to estimate the number of deaths from unidentified coronaviruses. Studies using these models have compared observed all-cause deaths with expected pre-pandemic trends and estimated that in 2020 alone, excess mortality was 14% to 38% higher than reported coronavirus deaths.
According to one calculation, the number of excess deaths worldwide in 2020-2021 reached a staggering 18.2 million, compared to the 5.9 million officially attributed to COVID-19.
However, Stokes said these models cannot reveal which deaths are uncounted coronavirus deaths rather than indirect pandemic effects such as interruptions or delays in care. Excess mortality models use highly aggregated data, which also makes detailed subgroup analysis difficult.
To make the estimates accurate, he explained, the machine learning algorithm used the “silver standard” of correctly classified in-hospital COVID-19 deaths (where SARS-CoV-2 testing was near universal during the early stages of the pandemic) to predict whether out-of-hospital deaths (where such testing was less common) were likely to be related to COVID-19.
Based on this approach, Stokes noted that many of the predicted out-of-hospital, unidentified coronavirus deaths would be due to causes such as Alzheimer’s disease, cardiovascular disease, and diabetes.
The researchers found that unrecognized coronavirus deaths were most likely to occur during the early waves of the pandemic from March to May 2020 (ARR 1.49, 95% UI 1.48 to 1.51) and declined in subsequent waves. January 2021 and April 2020 had the highest number of predicted deaths due to unidentified coronavirus infections, at 35,665 and 32,110, respectively.
States most likely to experience estimated unconfirmed coronavirus deaths included Alabama (ARR 1.67), Oklahoma (ARR 1.51), and South Carolina (ARR 1.47). The states with the highest absolute numbers of estimated deaths were Texas (24,024), New York (23,005), California (11,613), Alabama (11,501), and Florida (7,718).
Hispanic ethnicity (ARR 1.31, 95% UI 1.30-1.32) was associated with more unrecognized deaths, as was male gender (ARR 1.22, 95% UI 1.21-1.23) and older age: 65-74 years (ARR 1.21, 95% UI) 1.21-1.22), age 75-84 years (ARR 1.22, 95% UI 1.21-1.23).
Other sociodemographic factors associated with undercounting COVID-19 deaths include:
- West South Central region residence (ARR 1.31, 95% CI 1.29-1.33)
- Mid-Atlantic region resident (ARR 1.26, 95% UI 1.24-1.27)
- Less than high school education (ARR 1.29, 95% UI 1.28-1.31)
- Counties in the bottom quintile of median household income (ARR 1.34, 95% UI 1.31-1.36)
- Counties with more residents reporting poor or good health (ARR 1.30, 95% UI 1.28-1.33)
“Communities affected by undercounting of COVID-19 deaths can be interpreted as patterns of structural racism, classism, and ableism in death surveillance systems that require further research and policy attention,” Stokes and colleagues concluded.
Limitations of the study include the assumption that the training dataset for the machine learning model on deaths among hospitalized patients assigned to COVID-19 is correct.
