The Autopsy / Case File · Chronic Disease Progression
Counted Once
The pandemic's death toll counts a specific kind of event: a person dying inside a window short enough that the certifier wrote the virus on the certificate. That is the count the country memorialized. It has almost nothing to say about the person who survived the hospitalization, went home, and died twenty-six months later of a heart that never worked right again.
The National Center for Health Statistics went looking for those deaths in the only place available at the time: the words typed on the death certificate itself. From January 1, 2020, through June 30, 2022, it found 3,544 of them — deaths coded to COVID-19 whose literal text also used a phrase like "long COVID" or "post-acute sequelae of COVID-19." That was three tenths of one percent of the 1,021,487 COVID-19 deaths in the period. Meanwhile a Department of Veterans Affairs cohort of 135,161 infected people, followed for three years against more than five million controls, measured 14.16 excess deaths per 1,000 hospitalized survivors in year two and 8.16 per 1,000 in year three.
Join the discussion →
A mortality statistic is not a measurement of how many people a disease killed. It is a measurement of how many deaths a system managed to attribute to that disease, using the rules and codes it had at the time. For an infection that kills within three weeks, the two converge, because the sequence is legible to whoever fills out the certificate: fever, hospital, ventilator, death. For an infection whose damage is still killing people two and three years out, they separate, and always in the direction of the smaller number. Nobody has to conceal anything. The form asks what caused the death, and by the time the death arrives, the honest answer looks like heart failure.
Three Thousand Five Hundred Forty-Four
In December 2022, NCHS published Vital Statistics Rapid Release Report No. 25, by Farida Ahmad, Robert Anderson, Jodi Cisewski, and Paul Sutton. Its subject was not long COVID mortality so much as the problem of finding it. The World Health Organization had approved ICD-10 code U09.9 for post-COVID conditions in September 2020, but the United States had not yet implemented it for mortality coding, so, as the authors put it, it was "not possible to assess long COVID mortality using ICD-10 codes." What was possible was reading the certificates. The team pulled every death coded to U07.1 — COVID-19, as underlying or contributing cause — and searched the literal text for nine key terms, from "chronic COVID" and "long hauler COVID" to "PASC" and "post COVID syndrome."
The search returned 3,544 deaths across thirty months. "Post COVID" did most of the work, appearing in 89.6 percent of them. The largest single month was February 2022, with 393. And the share of COVID-19 deaths carrying a long COVID mention peaked twice — 1.2 percent in June 2021 and 3.8 percent in April 2022 — both times, the authors note, during periods when the raw number of COVID-19 deaths was falling. That is not a coincidence and it is not noise. In a trough, the acute deaths recede first, and what is left standing in the data is disproportionately the slow kind.
What the Certificates Said Killed Them
The most revealing table in the NCHS report is not the count. It is the list of underlying causes. Across all COVID-19 deaths in the study period, COVID-19 itself was the underlying cause 88.2 percent of the time. Among the long COVID deaths, that figure dropped to 67.5 percent. Roughly a third of the deaths that a certifier explicitly labeled as long COVID were nonetheless coded to something else as the underlying cause: heart disease in 8.6 percent, cancer in 2.9, Alzheimer disease in 2.7, chronic lower respiratory diseases in 2.5, diabetes in 2.0, stroke in 1.8, hypertension in 0.7, Parkinson disease in 0.5, and a residual 10.1 percent spread across everything else.
That distribution is the mechanism in miniature. When a person dies eighteen months after an infection that scarred their lungs and stiffened their heart, the proximate cause is a cardiac or respiratory event, and the certificate says so. Long COVID appears, if at all, as a phrase further down the page — context a careful certifier supplied and a coding system was not yet built to capture. These 3,544 are, by construction, the best-documented cases in the country: someone at the bedside knew the history and wrote it down. Every death where that did not happen is invisible.
The demographic profile carries the same fingerprint. Long COVID deaths skewed older than COVID-19 deaths overall — 28.8 percent in the 75-to-84 bracket and 28.1 percent among people 85 and over. The age-adjusted rate for the twelve months ending June 2022 was 6.3 per million, running from 0.9 per million among people in their thirties to 117.1 per million at 85 and over, and it was highest among non-Hispanic American Indian or Alaska Native people, at 14.8. The sex split also moved: males were 56.0 percent of all COVID-19 deaths but only 51.5 percent of long COVID deaths, the acute disease's male lethality partly offset by a chronic sequela more common in women.
The Cohort That Kept Counting
To see the size of what the certificates miss, you need a study that does not depend on certificates at all. Miao Cai, Yan Xie, Eric Topol, and Ziyad Al-Aly published one in Nature Medicine in 2024: 135,161 people in the VA health system with SARS-CoV-2 infection between March 1 and December 31, 2020, of whom 114,864 were never hospitalized and 20,297 were, followed for three years against 5,206,835 uninfected controls.
For the non-hospitalized, the mortality story ends early. Their incidence rate ratio for death was 1.58 in the first year and then 0.97 in the second and 1.01 in the third — statistically indistinguishable from the controls. The excess deaths were 16.20 per 1,000 in year one and effectively zero thereafter. For people who had been hospitalized, the curve does not close. Their death rate ratio was 3.17 in year one, 1.44 in year two, and still 1.29 in year three, which is to say that thirty-six months after infection they were dying 29 percent faster than matched people who never caught it. In absolute terms: 58.85 excess deaths per 1,000 in year one, 14.16 in year two, 8.16 in year three.
Run that against the cohort itself. Years two and three together carried 22.32 excess deaths per 1,000 hospitalized survivors; across the 20,297 hospitalized veterans in this one study, that is roughly 453 deaths — more than an eighth of every long-COVID-mentioning death certificate filed nationwide over thirty months. One cohort, one health system, one infection year.
The certificate asks what caused the death. Two years out, the honest answer looks like heart failure — and heart failure is what gets written.
The same paper measured surviving burden in disability-adjusted life years, and the shape is identical. Cumulative DALYs over three years came to 91.2 per 1,000 non-hospitalized and 766.2 per 1,000 hospitalized, with years two and three accounting for 40.4 percent and 31.2 percent of those totals — the illness was still producing new damage in year three, at 9.6 DALYs per 1,000 for those never hospitalized and 90.0 for those who were. A disease still generating disability in year three is a disease still generating deaths in year four.
A Code That Arrived Late
The United States did eventually get the code. U09.9 became usable for American mortality data with 2023 deaths — more than two years after WHO approved it, and three years into the pandemic. Giuseppe Lippi and Fabian Sanchis-Gomar, working from the National Center for Health Statistics's own data, counted every U09.9 death recorded through the end of December 2024 and found 2,653 with post-COVID condition as the underlying cause, an age-adjusted rate of 0.089 per 100,000, higher in men (0.098) than in women (0.081), and higher in 2023 than in 2024 across every age group.
Two details deserve attention. The first is the size: 2,653 deaths over two years, in a country where 6.9 percent of adults told the 2022 National Health Interview Survey they had ever had long COVID and 3.4 percent said they currently did. The second is where the deaths happened — 33.0 percent at the decedent's home, 26.3 percent in a nursing home or long-term care facility, 24.1 percent as a hospital inpatient, 7.7 percent in hospice. A cause of death that kills people mostly at home and in nursing homes is one whose certification depends on a clinician who was not there, reconstructing from the chart a chain of events that began years earlier.
Why the Signal Stays Slow
Every structural feature of this problem pushes the recorded number down and none pushes it up. The exposure is nearly universal and undated, which is why the useful cohorts are the ones assembled early, from 2020 infections, before vaccines and reinfection blurred everything. The interval between cause and death is long enough that the two events are handled by different clinicians in different institutions. The proximate mechanisms — cardiovascular, pulmonary, thrombotic, neurologic — are high-frequency causes of death in older adults anyway, so a long COVID death does not look anomalous; it looks like an ordinary heart attack that happened to a person who once had COVID. And the group carrying the most residual risk, hospitalized survivors, skews old and comorbid — the exact population whose deaths are least likely to prompt anyone to go looking for a two-year-old precipitating event.
What that leaves is two numbers describing the same deaths and disagreeing by an order of magnitude. The certificate count is precise, official, and small. The cohort estimate is statistical, indirect, and much larger. The country built a memorial to the first one.
And Then You're Dead
The flags on the Mall counted an event that had a date on it. What the certificates cannot count is the person who was discharged, who went home with an oxygen concentrator and a follow-up appointment and died in the third year of something that will be filed under heart disease. The VA cohort says that person is 29 percent more likely to be dead than a neighbor who never caught it. The National Vital Statistics System says there were 3,544 of them nationwide in two and a half years. Both statements are accurate. Only one of them is a count.
That's it. That's the whole thing.
Sources
- The 3,544 deaths coded to U07.1 with long COVID literal text (January 1, 2020–June 30, 2022), 0.3% of 1,021,487 COVID-19 deaths; the nine search key terms; "post COVID" appearing in 89.6% of them; the February 2022 peak of 393 deaths; percentage peaks of 1.2% (June 2021) and 3.8% (April 2022) coinciding with declining COVID-19 death counts; COVID-19 as underlying cause for 88.2% of all COVID-19 deaths versus 67.5% of long COVID deaths; the underlying-cause table (heart disease 8.6%, cancer 2.9%, Alzheimer disease 2.7%, chronic lower respiratory diseases 2.5%, diabetes 2.0%, stroke 1.8%, hypertension 0.7%, Parkinson disease 0.5%, other 10.1%); age distribution (75–84: 28.8%, 85 and over: 28.1%, 65–74: 21.5%); sex distribution (51.5% male for long COVID deaths versus 56.0% for all COVID-19 deaths); the 6.3 per million age-adjusted rate for the 12 months ending June 2022, 7.3 for males and 5.5 for females, 0.9 at ages 30–39 and 117.1 at 85 and over; race and ethnicity rates (non-Hispanic AIAN 14.8, non-Hispanic White 6.7, non-Hispanic Asian 1.5); and the statement that U09.9 had not been implemented for U.S. mortality coding at the time of analysis — Ahmad FB, Anderson RN, Cisewski JA, Sutton PD, "Identification of Deaths With Post-acute Sequelae of COVID-19 From Death Certificate Literal Text: United States, January 1, 2020–June 30, 2022," Vital Statistics Rapid Release Report No. 25, National Center for Health Statistics, December 2022
- Cohort sizes (135,161 infected, 114,864 non-hospitalized, 20,297 hospitalized, 5,206,835 controls), infections between March 1 and December 31, 2020, three-year follow-up; death incidence rate ratios by year (non-hospitalized 1.58, 0.97, 1.01; hospitalized 3.17, 1.44, 1.29 with 95% CI 1.19–1.40); excess deaths per 1,000 (non-hospitalized 16.20, −0.91, 0.22; hospitalized 58.85, 14.16, 8.16); cumulative three-year DALYs per 1,000 (non-hospitalized 91.2, hospitalized 766.2); third-year DALYs per 1,000 (9.6 non-hospitalized, 90.0 hospitalized); and years two and three accounting for 40.4% and 31.2% of cumulative burden respectively — Cai M, Xie Y, Topol EJ, Al-Aly Z, "Three-year outcomes of post-acute sequelae of COVID-19," Nature Medicine, 2024
- The 2,653 U.S. deaths with ICD-10 code U09.9 as the underlying cause recorded through December 2024, the 0.089 per 100,000 age-adjusted rate (95% CI 0.085–0.092), the male rate of 0.098 versus 0.081 for females, higher rates in 2023 than 2024 across age groups, and the place-of-death distribution (home 33.0%, nursing home or long-term care 26.3%, medical facility inpatient 24.1%, hospice 7.7%) — Lippi G, Sanchis-Gomar F, "Mortality of Post-COVID-19 Condition: 2025 Update," COVID, volume 5, 2025, using U.S. National Center for Health Statistics mortality data
- The 6.9% of U.S. adults who had ever had long COVID and 3.4% who currently had it in 2022, from 27,651 respondents — "Long COVID in Adults: United States, 2022," NCHS Data Brief No. 480, National Center for Health Statistics, September 2023, using National Health Interview Survey data
- WHO approval of ICD-10 code U09.9 for post-COVID conditions in September 2020 — as reported in NCHS Vital Statistics Rapid Release Report No. 25, December 2022
- Photo: "COVID-19 Memorial Field of White Flags on the Mall — In America: Remember," installation by Suzanne Brennan Firstenberg, photographed September 26, 2021, by Amaury Laporte, CC BY 2.0, via Wikimedia Commons
Comments
Loading…