The Hidden Death Toll
A new study published this week estimates that approximately 155,000 COVID-19 deaths that occurred outside of hospitals went uncounted during the first two years of the pandemic, meaning the official U.S. death toll for 2020 and 2021 may have been understated by roughly 16 percent. The research, which used machine learning methods to analyze patterns in excess mortality data, adds to a growing body of evidence suggesting that the pandemic's human cost was even larger than the staggering official figures suggested.
About 840,000 COVID-19 deaths were recorded on death certificates in 2020 and 2021, making it the third leading cause of death in the United States during that period. But a team of researchers found that when they examined all-cause mortality data — the total number of deaths from any cause, compared against historical trends — there was a substantial gap between expected and observed deaths that was not explained by the recorded COVID toll or by other known factors.
Why Deaths Go Uncounted
Death certificate reporting is imperfect under any circumstances, and the early months of the pandemic created conditions that amplified those imperfections. Hospitals and medical examiners were overwhelmed. Testing was severely limited, meaning many patients who died of COVID-like illnesses never received a confirmed diagnosis. In the absence of a positive test, clinicians had to make judgment calls about whether COVID-19 was the underlying cause of death — and those calls were inconsistently applied across jurisdictions.
Outside of hospitals, the attribution problem was more severe. People who died at home, in nursing facilities, or in rural settings with limited healthcare access were less likely to receive post-mortem testing or detailed medical review. If a 75-year-old with underlying conditions died at home in April 2020 without receiving medical care, their death might be recorded as natural causes or attributed to heart disease — particularly early in the pandemic, when COVID's symptom profile was not yet well-understood by clinicians.
How the Study Was Conducted
The research team used a form of artificial intelligence — specifically supervised machine learning models trained on pre-pandemic mortality patterns — to forecast expected deaths in different demographic and geographic groups during the pandemic period. By comparing predicted mortality to observed mortality, the researchers could identify excess deaths: deaths that occurred above what historical trends would have predicted in the absence of COVID.
They then applied statistical methods to separate excess deaths attributable to COVID from those attributable to other pandemic-era disruptions, such as delayed medical care for non-COVID conditions, increases in drug overdoses and accidents, and changes in healthcare utilization. What remained after accounting for these factors was attributed to unrecognized COVID-19 deaths.
The 155,000 figure carries uncertainty — the study's authors provide confidence intervals rather than a single precise estimate — but the central finding is consistent with earlier work using different methodologies. Multiple prior studies using excess mortality frameworks have reached similar conclusions: the pandemic killed substantially more Americans than the official COVID death count captures.
Geographic and Demographic Disparities
The study spotlights dramatic disparities in where and among whom undercounting occurred. Rural communities had higher rates of unrecognized COVID deaths than urban areas, reflecting both the more limited healthcare infrastructure and the diagnostic challenges in communities with less access to testing. Racial and ethnic minority populations — who experienced disproportionate COVID mortality throughout the pandemic — also had higher rates of undercounted deaths, compounding documented disparities in the official toll.
Southern and rural states, where testing was more limited in early 2020 and where some jurisdictions were slower to update death reporting protocols, appear in the analysis as sites of concentrated undercounting. This geographic pattern is consistent with what public health researchers observed in real time during the pandemic.
Implications for Public Health
The finding matters beyond the historical record. Accurate mortality data is foundational to pandemic preparedness planning. If death counts are systematically undercounted during a pandemic, the data inputs that guide modeling, resource allocation, and policy decisions are compromised from the start.
The study also raises questions about the adequacy of the U.S. vital statistics system for capturing mortality during crises. While reporting infrastructure has improved since 2020, the structural factors that contributed to undercounting — inconsistent jurisdiction reporting, limited rural diagnostic capacity, inadequate post-mortem testing — have not all been resolved.
Public health officials have called for investment in death investigation infrastructure, real-time mortality surveillance systems, and standardized COVID attribution protocols as preparedness measures for future pandemics. The new study provides additional quantitative evidence for why those investments matter.
This article is based on reporting by STAT News. Read the original article.
