Underestimated Toll: AI Reveals Hidden COVID-19 Deaths, Disparities

New research indicates the official U.S. COVID-19 death toll significantly underestimated the true number of lives lost, particularly among Hispanic communities and people of color, and in several Southern and Southwestern states. A study published in Science Advances utilized artificial intelligence to identify approximately 155,000 additional deaths likely attributable to COVID-19 that were not initially recorded as such.

The Hidden Deaths: A Deeper Look

While official reports from the Centers for Disease Control and Prevention (CDC) document over 1.2 million COVID-19 deaths in the U.S. Since early 2020, the new study suggests the actual number could be considerably higher. Researchers analyzed death certificates, comparing patterns from hospitalized patients with confirmed COVID-19 diagnoses to those who died outside of hospitals. This analysis revealed a significant number of deaths attributed to conditions like pneumonia or diabetes may have been, in fact, due to undiagnosed COVID-19.

Disparities in Uncounted Deaths

The study pinpointed specific demographic groups and geographic areas where the undercounting was most pronounced. Undiagnosed COVID-19 deaths were more common among:

• Hispanic individuals and other people of color: These communities experienced a disproportionately higher rate of unrecognized COVID-19 deaths.
• Individuals in the early months of the pandemic: Limited access to testing and evolving understanding of the virus contributed to underdiagnosis during the initial surge.
• Residents of Southern and Southwestern states: States like Alabama, Oklahoma, and South Carolina showed a higher prevalence of uncounted deaths.

Challenges in Accurate Death Reporting

Several factors contributed to the underestimation of COVID-19 deaths. These include:

• Limited Access to Testing: In the early stages of the pandemic, at-home testing was not widely available, and many individuals who died outside of hospitals were not tested for the virus.
• Variations in Death Investigation Systems: The quality of death investigations varies across the country. Some areas rely on elected coroners who may lack the specialized training of medical examiners.
• Potential for Political Influence: Research suggests that partisan opinions may have influenced testing decisions and post-mortem investigations. Some coroners reported pressure from families to exclude COVID-19 as a cause of death.

The Role of Artificial Intelligence

Researchers led by Andrew Stokes of Boston University employed machine learning to analyze death certificate data. By identifying patterns in confirmed COVID-19 deaths among hospitalized patients, they were able to estimate the number of likely undiagnosed deaths outside of hospitals. Steven Woolf, a researcher at Virginia Commonwealth University, described the team’s use of machine learning as “intriguing.”

COVID-19 and Excess Deaths

It’s important to note that the pandemic similarly led to deaths from other causes. For example, individuals with pre-existing conditions may have died given that they could not access routine medical care due to hospitals being overwhelmed with COVID-19 patients. Increases in drug overdose deaths, linked to social isolation and disrupted treatment access, also contributed to overall mortality during the pandemic. However, Stokes and his colleagues focused specifically on deaths directly attributable to coronavirus infection.

Looking Ahead

The findings underscore the need for improvements in the nation’s death investigation system and highlight ongoing health disparities. As Steven Woolf noted, individuals on the margins continue to experience disproportionately high mortality rates due to limited access to healthcare. Addressing these systemic issues is crucial to ensuring a more accurate and equitable response to future public health crises.