The story of COVID-19 death reporting in America
Are COVID-19 fatalities exaggerated?
From the time the SARS-COV-2 virus hit our shores in early February, most everyone I know has questioned the seriousness of the disease COVID-19. In the early days of the pandemic, cases of the disease accrued gradually, first one by one then steadily. By mid-March the cumulative number reached 10,000, then took off exponentially before our eyes to over 1.4 million cumulative confirmed cases in the US today. Even President Trump found the escalation unsettling, as the nation hung on to the words of his CDC team of experts during their televised reports to the nation. It wasn’t long before we learned that the exploding numbers were just the tip of a vast iceberg of unrecognized infections—people with mild, even undetectable, symptoms, impossible to count moving freely about the country spreading the disease.
By the end of February, we had our first death from COVID-19. The death occurred in Seattle Washington (postmortem testing later identified two earlier deaths in February in the San Francisco Bay Area). By mid-March, deaths confirmed as COVID-19 by medical authorities reached 500, then started to climb exponentially, just as COVID-19 infections had done but lagging behind a few weeks.
The nation trembled and in many places lockdown was imposed. For a moment we seemed to operate as one body. We are in this together. We need to hammer down the pandemic and “flatten the curve.”
At about this time, the concept of the case-fatality rate (the number of deaths divided by the number of reported cases) became a familiar term, showing up on news programs and suggesting a way to assess an individual’s risk of death from the pandemic.
Calculating the case-fatality rate (CFR) requires little more than a 5th grade education. There is a numerator (number of deaths) gleaned from the National Center for Health Statistics and a denominator (number of medically confirmed cases) reported by hospitals, laboratories, and medical clinics to the Centers for Disease Control (CDC). While counting deaths seemed straightforward, counting cases proved to be more difficult. The number of cases varies widely with testing capabilities and, as we shudder to recall, testing was sorely lacking in the early days of the pandemic, and is still not quite adequate. Thus the CFR has changed over time fluctuating between 2 to 6 percent in the U.S. But that is an average risk for an individual in the population. CFRs for people in specific age groups fluctuate even more—with negligible risk for the young, moderate risk in middle age, and a staggering risk from 4 to 15 percent, increasing with age, in the elderly population over age 65.
To get a better handle on the the CFR, studies are now underway in many places. The earliest such study, an antibody test performed by Stanford researchers (on which I’ve reported previously), attempted to estimate the the total number of people previously infected with the virus in Santa Clara County, California. The study was accomplished quickly but was criticized for possible selection bias and reduced specificity (too many false positive diagnoses). Nevertheless, the researchers concluded that the number of people who were previously infected with SARS-COV-2 (estimated as 2.8 percent of the population) was an order of magnitude larger than previously predicted—and the CFR was correspondingly orders of magnitude lower, estimated as 0.2 percent average risk of dying from the disease. Thus the COVID-19 epidemic was likely far less deadly than previously thought.
As people grew impatient with the lockdown, the question of how serious is this disease that causes death to some (sadly, mostly our elder and health-compromised population) but is so mild as to be undetectable to others (the young, the healthy) has persisted? Even with deaths reaching over 83,000 to date, the nation faces a dilemma. How do we balance the risk of illness and death to our most vulnerable populations incurred by the pandemic with the risk to millions of other citizens’ physical and psychological health, well-being, and livelihood incurred by the lockdown? Could some of the deaths attributed to COVID-19 instead be the usual, expected deaths due to such conditions as heart disease or emphysema that were spuriously concomitant with a COVID-19 infection? If true, this might suggest that the risk from COVID-19 is not as bad as once thought.
This concept was expressed by a Minnesota state senator and family physician, Dr. Scott Jensen, who appeared on Fox’s Ingraham Angle on April 9, 2020. The CDC had just released guidelines on how doctors should fill out death certificates for COVID-19 deaths. Laura Ingraham read aloud an excerpt from the 7-page guidelines:
In cases where a definite diagnosis of COVID–19 cannot be made, but it is suspected or likely (e.g., the circumstances are compelling within a reasonable degree of certainty), it is acceptable to report COVID–19 on a death certificate as “probable” or “presumed.” In these instances, certifiers should use their best clinical judgement in determining if a COVID–19 infection was likely. However, please note that testing for COVID–19 should be conducted whenever possible.
“So, what’s wrong with that?” Ingraham asked the senator-doctor. Jensen responded “ridiculous … the idea that we are going to allow people to massage and game the numbers is a big deal that undermines trust.” (I assume he implies here that the CDC has motives to manipulate the data.) He then claimed that there are additional, financial incentives to designate COVID-19 as a cause-of-death: hospitals receive specially guaranteed Medicare reimbursements for COVID-19 and are covered for treatments related to it. “Anytime healthcare intersects with dollars, it gets awkward,” he said. He then gave an example of a person getting hit by a bus and incurring a collapsed lung:
“They go to the emergency room and they’re there for 15–20 minutes and blood work comes back, COVID test comes back positive, and they died 20 minutes later because of their collapsed lung. We are going to put down as COVID-19?”
Although the doctor’s interpretation of the guidelines is fairly far askew (read further below), his inference that COVID-19 deaths may be exaggerated resonated with many viewers who were becoming increasingly uncomfortable and worried about the repercussions of the lockdown. According to Snopes, a fact-finding website, “Jensen’s commentary contributed to a narrative … that the ongoing COVID-19 coronavirus disease pandemic is not as deadly or serious as reported, and that social-distancing measures implemented to control its spread are unwarranted.”
Even the brilliant mind of businessman and thought-leader Elon Musk was influenced by the interview. In a May 7, 2020 podcast of “The Joe Rogan Experience,” Musk repeated both Jensen’s stories of the bus accident and the Medicare reimbursement incentive. Parodying a hospital administrator (at 1:31 hours into the program), Musk quips, “Well, the stimulus bill says we get all this money if it’s a COVID death … I’m like, okay, they coughed before they died.” Eleven million people, to date, have watched this podcast on Youtube.
What both Jensen and Musk imply about the previously mundane world of death certificate reporting (that it’s not very well conceived) and the new guidelines from the CDC (that there is an implicit bias) is simply untrue.
This is how death certificate reporting works:
Long before COVID-19, the CDC provided guidance to doctors on how to properly fill out a death certificate. There are rules for including qualifiers such as “probable” and “presumed” just as there are rules for every disease and condition. On page 2 of a 2004 summary document called Instructions for Completing the Cause-of-Death Section of the Death Certificate, the following caveat appears:
If the certifier is unable to determine the etiology of a process such as those shown above (list includes influenza), the process must be qualified as being of an unknown, undetermined, probable, presumed, or unspecified etiology so it is clear that a distinct etiology was not inadvertently or carelessly omitted.
There are codes for those qualifiers, as there are codes for hundreds of diseases, treatments, drugs, equipments, and more. The list is long and elaborate with instructions to assure standardization, accuracy, privacy, and more. An entire field of medical records and health information technology exists, employed by people who are “detail-oriented, proficient in computer systems and technology, and thoroughly familiar with general medical concepts and terminology.”
The attending doctor or certifier is the first and most important step in cause-of-death identification, followed by medical record analysts and coders who make sure nuanced and logically laid out sequences and conditions leading to the death are picked up, recorded, and retrievable later.
Elon Musk, later in the Joe Rogan podcast, pleaded for better data relative to COVID-19.
“Let’s clear up the data, clear up the data. So, like I said, somebody should be recorded as COVID only if somebody has been tested. Has received a positive, COVID test. Not if they simply have symptoms, one of, like, a hundred symptoms. And then if it is a COVID death, it must be separated. Was Covid a primary reason for death, or did they also have stage three cancer, or heart disease, emphysema, and got hit by bus and had COVID.” (Elon Musk from the May 7 podcast “The Joe Rogan Experience.”)
Mr. Musk is right to say “Let’s clear up the data.” Be assured, the data are coming. But, can he wait?
Everything related to COVID-19 is moving at an unprecedented pace. In ordinary situations, public health data are published only after they are reasonably complete and thoroughly cleaned, which takes time. Clarifying the data is the reason the CDC guidelines are specific in requesting that doctors and certifiers clearly denote when appropriate “probable” and “presumed” and include as many details as they wish to describe the causes of death. This information aids coders in doing their job and, later, insurance brokers, public health researchers, and government planners make sense of the data.
Could it be pneumonia or flu?
In the meantime, I decided to try to get to the core of the issue with existing provisional data. I’d been having a back-and-forth conversation with a friend about epidemiology and the nature of pandemics when she suggested that COVID-19 deaths might actually be normally occurring pneumonia or flu deaths. We went to the CDC’s flu page to find out.
We weren’t the only ones asking this question, of course. The CDC allocates a great deal of effort and resources to understand and prepare for the flu each year, and runs an ongoing, well-resourced Influenza Division. People there were undoubtedly trying to get at the same question.
With a bit of searching on the CDC website, I located a table that was designed to increase public understanding of flu, versus COVID-19, versus pneumonia. That table, Table 1, is here. Table 1 shows the number of deaths occurring during each week of the pandemic, by cause-of-death categories. I highly recommend that readers of this essay look carefully at the numbers and pay particular attention to the footnotes. They are important. You can also look up the International Classification of Disease codes recently added to this table at the header of each cause-of-death column for increased clarity. Note, importantly, that the table is provisional, which means data are not fully cleaned and tallied, especially in the final 2–3 weeks of the table (which are the most recent weeks of data).
Interpreting Table 1:
Prior to the week ending on 3/21/20, the number of COVID-19 deaths had not yet reached 100. Up until that week, entries on Table 1 looked normal. Pneumonia deaths were steady at about 3,600–3,700 per week and influenza deaths showed their usual seasonal trend, ranging from 500–600 deaths per week (these are unadjusted figures—see the note at the end of this article), peaking in early March. COVID-19 cases were at the time reaching 10,000 and lockdown had begun.
After the week ending on 3/21/20, the death period for COVID-19 begins to escalate. Below is an excerpt from Table 1. (I omit the last two weeks of the table because they are obviously incomplete and liable to cause confusion.) What do you see? Let’s go column by column.
Deaths due to COVID-19 (Col 2.) increase 5 times from week March 15–21 to week March 22–28, then 3 times during the next week, and 2 times through the following week. The COVID-19 mortality curve starts off steep then flattens a bit as lockdown ensues.
(Please keep in mind that this is a ballpark exercise. The more recent the week, the more the week’s tally is incomplete. Things can, and will, change later in this table.)
Next, note that all-cause mortality, the next column over (Col. 3), is climbing now week by week. If you go back and look at Table 1 on the CDC site, you will see that all-cause mortality was flat before (for weeks!) at about 57,000 deaths per week prior to the pandemic’s death period. In Col. 4 and 4.1, all-cause mortality for 2020 is compared to the expected mortality based on an average of previous years’ during the same calendar week (see footnote 2).
Columns 5-7 are the current year’s mortality for pneumonia, pneumonia concomitant with COVID-19, and influenza. The final column, Col.8, lumps all three causes together to account for any accidental miscoding of one condition for the other.
What do the trends in the columns tell you? Excess all-cause mortality (Col. 4.1) increases rapidly and tracks well with COVID-19 mortality, but is larger than can be explained by COVID-19 designations alone. Pneumonia deaths, which had previously leveled off, are also increasing, even as we know the data are incomplete. Flu (alone) deaths look somewhat normal as flu is seasonal and starts to drop off at this time.
To me, this table says that the cumulative number of COVID-19 deaths, as recorded on death certificates, is not exaggerated. On the contrary, the number is likely undercounted with many deaths missed where COVID-19 was the underlying cause. Remember, the entire table is prior to the CDC’s newly announced guidelines and does not reflect the clarifying instructions for “probable” and “presumed.”
There is now good evidence that undercounting of deaths due to COVID-19 is happening not only in the U.S. but in China, Italy, France, Germany and other places.
And, no, Dr. Jensen and Elon Musk were wrong when they said someone who gets hit by a bus and dies of a collapsed lung will get coded as COVID-19 if their blood sample comes back positive for the disease. This would be an incorrect application of long-established CDC rules, as spelled out clearly in old and new CDC guidelines. The three tiers of cause-of-death on the certificate (see example at the beginning of this article) must be sequentially linked in the etiology of the death, both logically and in time sequence, based on a designated underlying cause—in this case getting hit by a bus. Determining the causes of death in a logical, sequential order is standard operating procedure and the duty of physicians using their best judgement based on training, experience, and all that contributes to their profession.
Conclusion:
COVID-19 is a serious disease, probably much worse than most flus. Numbers of COVID-19 deaths are not overreported or exaggerated but are likely underreported, in particular during the early weeks of the pandemic. The CFR (case-fatality rate) is a measure of the odds of an individual dying once infected with the disease. The CFR is calculated by averaging a lot of people’s experiences. Because the CFR varies widely between different demographic groups and changes dramatically based on testing capabilities, the CFR is an unreliable statistic and might compromise the public’s understanding. Population mortality risk (deaths divided by the total population of a country) might be a better way to assess how the nation is doing as a whole. All-cause mortality can be investigated to shed light on the death burden during a pandemic. Unexplained increases in deaths from all causes can be investigated. Parsing of national vital statistics and health databases, as they become fully completed and cleaned, will in time resolve many of the issues plaguing our understanding today.
While it is our right as citizens to demand results from publically collected data, we must do so with humility and patience. Our public servants who work in the background to accomplish the necessary, detailed diagnosing, coding, tallying, publishing, and more, are stepping up to the plate in this hyper-focused time of need.
Note: On May 15, 2020, a Wall Street Journal article by Paul Overberg and Jon Kamp was publish which digs deeper into current death data analyzing by age, sex and race. I highly recommend.
Note: I mentioned above that Table 1 consists of provisional data. The data are provisional not just for COVID-19 but for influenza and pneumonia as well. You might come across other reports where official influenza numbers are higher than what you see in Table 1, not just because they are incomplete. The CDC routinely adjusts hospital-based data upward using a multiplier method to correct for under-reporting and have done so since the 2009 H1N1 pandemic.
