The vaccines are here, but not in enough numbers. The shortage has meant there has been a lot of debate on prioritization. After a CDC committee proposed putting essential workers ahead of the elderly or adults with medical conditions in November, I wrote an Insight piece arguing that using age as the criteria for prioritization made the most sense. 

This issue of Insight includes a substantive and thought-provoking Counter by Whitney R. Robinson, associate professor at the UNC Gillings School of Global Public Health. She studies racial/ethnic disparities in health outcomes and explains the complexities of such prioritizations: exposure, health inequities, slowing transmission. (As announced earlier, I’m inviting contributors to write Counters for Insight: providing space for arguments that engage and counter my own pieces and that deepen the discussion). 

Some of the charts she includes on racial inequities in COVID-19 outcomes in the United States are maddening, and her Venn diagram she drew in response to these discussions (using, she says “a black marker, my kids’ crayons and a large peanut butter jar”) is an excellent visualization of how complicated these topics can get. 

Dr. Robinson’s diagram is responding to an incident at Stanford Medicine. The large teaching hospital vaccinated just 7 of its 1,300 residents—who regularly treat COVID-19 patients—from its first 5,000 vaccines. Who got vaccinated instead? Some administrators, including those who never see patients. After a protest in the hospital, 

Stanford management blamed “the algorithm” (so on brand!). Reporters for Technology Review Eileen Guo and Karen Hao obtained this “algorithm”:

It’s basically a simple formula that pretty much ensures that those over 65 (largely administrators) and those under 25 get a lot of points (why under 25??). There were increasing points awarded for age—easily overriding the points for exposure. 

I was struck by the hospital leadership’s attempt, in the wake of the protest, to blame “the algorithm” for the debacle. As if it were some complicated, messy calculation or a machine- learning algorithm that had surprised everyone--including the management who came up with the algorithm that favored vaccinating themselves instead of frontline medical workers. It is a stark example of how systems can and will be gamed by those with power and privileges.

On December 20th, the same CDC committee met and finalized their proposal for the stage of vaccine rollout following health care workers and long-term care facility residents. They shifted their stance, now recommending prioritizing those 75 and older, as well as “Frontline essential workers.” I believe this is a great improvement over their earlier proposal, but still, as the Stanford example and Dr. Robinson’s post below demonstrate, the complexities and the problems remain. Who exactly is a frontline essential worker? (Right now, it is defined by industry, not exposure levels). What about age or medical risk among them? How will the process of allocation work? And so much more. I’ll be writing more myself on all this.

So, without further ado, here’s Dr. Robinson’s thought-provoking essay. See you in the comments.

Known Knowns, Known Unknowns, and Unknown Unknowns: the Complexities of COVID-19 Vaccine Prioritization

By Whitney R. Robinson, PhD, MSPH

I got very excited when I read about the Stanford Medicine vaccination controversy in mid-December.  Like other hospitals across the United States, Stanford’s hospital received priority doses of COVID-19 vaccines. Stanford being Stanford, the hospital leadership used an algorithm to decide on a distribution plan for the first round of 5,000 doses. 

That’s where the trouble started.  Stanford has 1,300 residents working in their hospital. Residents, who often have the most direct contact with ill patients, are the most junior and lowest-paid doctors in the hospital. Despite the enormous risks they face, just seven of the 1300 residents were selected for first doses of the vaccine.  

The residents were not pleased. In a letter to the Stanford leadership, residents protested,

“Many of us know senior faculty who have worked from home since the pandemic began in March 2020, with no in-person patient responsibilities, who were selected for vaccination. In the meantime, we residents and fellows strap on N95 masks for the tenth month of this pandemic without a transparent and clear plan for our protection in place.”

It is so rare that we as Americans broadly agree on anything.  So it felt cathartic to see that most of the reactions from my fellow Americans were angry on the residents’ behalfs.  It felt good to see us channel our outrage in unison.  

I was also intrigued because the situation perfectly illustrates a key point of almost all my academic research: patterns of exposure to disease risk can vary dramatically across the population and affect risk of illness and mortality.  That basic principle is surprisingly easy to forget.

The knowledge that we have about risk for COVID-19 death and illness reminds me of the famous quote from then-Defense Secretary Donald Rumsfeld.  He separated knowledge into three domains.  First are things we know (“known knowns”). Then, there are things that we don’t know, but at least we know that we don’t know them (“known unknowns”). And finally are the things that we don’t know that we don’t know (“unknown unknowns”). When it comes to COVID-19 risk, I believe that all three are present. Let’s walk through some of them.    

Known known #1: One of the most remarkable things about SARS-CoV-2 (the coronavirus that causes the COVID-19 disease) is the age gradient of the infection fatality rate (IFR).  

The older a person is, the more likely a SARS-CoV-2 infection is to be deadly.  An 80-year-old infected with the virus is ~100x more likely to die of it than an infected 40-year old.  That same 80-year-old is about 1000x more likely to die than an infected 5-year-old.  The steep age dependence of the IFR is crucial information for any COVID-19 prevention or vaccination strategy: if we want to minimize fatalities, we need to prioritize the well-being of older people. 

But the infection fatality rate is not the only factor that strongly influences the risk of dying from COVID-19.  The IFR is what’s called a “conditional” statistic.  It is the risk of death *if* one is exposed to the SARS-CoV-2 virus AND *if* one is infected with that virus.  And those are both important ifs. 

Known known #2:

Exposure risk varies tremendously across people and settings.  The huge variation in exposure risk is obvious on a global scale: in a country like New Zealand, the chance of being exposed to the coronavirus today is almost infinitely small. Because the average Kiwi’s risk is so small, the average American is probably thousands of times more likely to be exposed to SARS-CoV2. 

Known known #3:

But we’ve also learned that the risk of being infected with SARS-CoV-2 can vary dramatically even when there’s an exposure.  The best explanation I’ve seen on this is an Atlantic article by Zeynep Tufekci.  She wrote:

“There are COVID-19 incidents in which a single person likely infected 80 percent or more of the people in the room in just a few hours. But, at other times, COVID-19 can be surprisingly much less contagious. ...A growing number of studies estimate that a majority of infected people may not infect a single other person.. . .  Multiplestudies from the beginning have suggested that as few as 10 to 20 percent of infected people may be responsible for as much as 80 to 90 percent of transmission, and that many people barely transmit it.”

There are reports of a new, potentially more infectious strain of SARS-CoV-2 circulating.  Therefore, this “known known” will continue to be updated.  But, even given this new strain, a basic truth still seems to hold: some kinds of places and interactions foster more infection than others.  For instance, nursing homes, prisons and jails are all notably deadly in terms of COVID-19. These super-spreader locations combine high exposure risk (lots of staff and residents coming and going) with high infectiousness (shared living spaces, limited PPE, unavoidable close contacts), and a sizable percentage of older people (that IFR age gradient).  That combination makes the lethality of these settings impossible to overlook. 

What intrigued me about the Stanford Medicine vaccination brouhaha was how perfectly it tapped into people’s subconscious intuitions about the interconnectedness of exposure risk, infection risk, and IFR for shaping mortality risk (their “unknown known”?).

Using a black marker, my kids’ crayons and a large peanut butter jar, I drew three overlapping circles.  The one of the left was for those at high risk of exposure to SARS-CoV-2.  The one on the right was for settings characterized by high risk of SARS-CoV-2 infection if exposed.  And, finally, the bottom circle was for those with high infection fatality rates, the people most likely to die if exposed and infected with the novel coronavirus.  

 By putting a few examples in the diagram, I tried to make a few points:

1. Exposure risk can vary as much as IFR. On a given day in December 2020, a 30-year-old Stanford resident working in the Emergency Department has a coronavirus exposure risk of basically 100%.  In contrast, because of their financial resources and medical knowledge, the Stanford work-from-home senior faculty can have a risk approaching zero if they take all precautions available to them.  Let’s conservatively say that, on any given day, the exposure risk for the Stanford resident is about 100 times higher than for the senior faculty member. I’m assuming the senior faculty has a 1.0% risk of exposure every day, which I think is a big overestimate.  but the point is the scale: like the IFR, exposure risk can vary by orders of magnitude across the population.  In this case, the 100-fold difference in exposure risk is the same as the IFR between a 70-year-old and a 30-year-old.  There are circumstances where exposure differences are so large that they essentially cancel out the IFR: a heavily exposed 30-year-old can be as likely to die of COVID-19 as a 70-year-old with 1/100th the exposure risk.  

2. Infection risk matters. Exposure is not enough to cause infection.  With proper mitigation (like the PPE, protocols, and ventilation available to doctors at Stanford), infection risk can be dramatically reduced. See the purple area in the Venn diagram.

3. Most importantly, even among those who are high-risk because of older age, there is a subset that is even *higher* risk.  I denote these folks in dark green in the Venn diagram.  They are biologically susceptible (high IFR) because of their age, but they are also disproportionately likely to be exposed in high-infection circumstances.  These folks are the highest of the high risk.  Vaccinating these folks first would reduce mortality most dramatically. 

Known Known #4: Lightly regulated, first-come-first-serve systems for allocating scarce high-value commodities favor the wealthy and well-connected.  In other words, in states like Florida, the people in dark green in the Venn diagram almost certainly will not be vaccinated first.  

When pressed about their vaccine allocation plan, Stanford officials explained that their algorithm prioritized older people over younger people.  But they also said that a bug in their algorithm failed to count the exposure risk of the residents.  There appears to have been an easily fixable problem with the way the algorithm coded missing work location data for residents.  According to the residents, the problem was noted before the vaccinations began.  

When Florida’s governor issued an executive order last week adding all people aged 65 years and older to the first phase of vaccine administration, my heart sank.  Federal guidelines stratified by age more finely than this, putting people aged 75 and older ahead of 65-74 year olds.  States are already far short of the amount of vaccine it would take to vaccinate those 75 years and older.  There are 367,000 doses of vaccine arriving in Florida this week and about 2 million people over age 75.  By extending the first priority group to those 65-75 year old, the governor added another 2.5 million or so people to the top priority group.  Why add another *huge* group of people to the mix?  Why promise 367,000 doses of vaccine to over 4.5 million people?  Well, at the governor’s press conference, several local political allies received some of the first doses of vaccine delivered in a community setting.  There are many wealthy and well connected people between ages 65 and 74 in Florida.  Now, with a first-come, first-serve system, they will be able to push through those Silent Generation folks and get to the front of the line.  

The above is what I think that I know about COVID-19 risk: how it works and who is most at risk. Unfortunately, there’s still much we don’t know--we just know that we don’t know it.

Known unknown #1: When it comes to high likelihood of exposure or infection, I haven’t given you figures, like I have in that gorgeous IFR graphic above. That’s because we don’t have those numbers.  And without those exact numbers for exposure and infection risk, you can’t calculate all the data to know exactly where the different risks balance out, to see where the tipping point is between the 45-year-old grocery store clerk and the 72-year-old well-off retiree.  

We have the technology and expertise to have calculated these exposure and infection differentials.  With massive testing, sampling and genotyping of positive cases, and aggressive “backwards” contact tracing, we could know so much about who is exposed, who is infected and where the virus is most concentrated.  This kind of work was possible, but it would have put many business interests on the defensive.  Instead of funding research like this, many in Congress were focused on a Liability Shield for businesses.  I’ll let you read up on it on your own.  But I think it points to the strong incentives businesses have to avoid documentation of the exposure, infection, and mortality risks that their workers and their families face.  

One of the papers I’ve seen most referenced in support of purely age-based vaccine prioritization schemes is this preprint.  In a conversation with one of the authors on Twitter, I asked about incorporating data on differentials of exposure and infection.  She responded, “I completely agree that considering differences in exposure by occupation/setting is vital for assessing risk. Our model doesn’t include these differences, not bc of a lack of importance but a lack of data.” Those limitations mean that we can’t accurately calculate COVID-19 mortality risk across the population or know whether strictly age-based algorithms or some other vaccination scheme would prevent the most deaths in the next months.  

Known unknown #2:  I study racial/ethnic disparities in health outcomes for a living.  I know how stark the differences are for many diseases.  But this figure still makes me gasp and makes my stomach sink every time I see it. 

People who think that racial differences are all biological might say that all these non-White groups have suffered so much excess death because of that bottom circle, because of greater biological susceptibility.  Recent studies have evaluated this hypothesis and found that it’s not true.  Instead the answer is simpler: Black and Latino/a people in particular are dying of COVID-19 at such staggering rates because they are more likely to be exposed to the virus in infectious settings, particularly workplaces.  If a variable as crude as 6-category race/ethnicity reveals huge disparities in exposure and infection risk, what would we find if we measured it directly by occupation and neighborhood?  

Known unknown #3: People recently infected with SARS-CoV-2—but who have recovered—appear to be at low risk of immediate re-infection.  These individuals could be de-prioritized in the first rounds of vaccination if we had a good national registry of confirmed infections.  Alas, we do not have a good national registry of confirmed infections.  

Known unknown #4: I’ve focused here on individual mortality risk as the major factor to consider in a vaccination scheme. But Long Covid is also an important consideration.  We do not know enough.   

Known unknown #5: Another way to think about prioritizing vaccination is to slow transmission as quickly as possible. The evidence suggests that non-elderly adults are the main drivers of transmission risk, particularly the folks in light green in the Venn diagram.  But we don’t know the extent to which the Moderna and Pfizer/BioNTech vaccines reduce infection and transmission in those vaccinated.  If they do stop or slow transmission, there could be a strong argument for prioritizing the vaccination of those people and at those sites at the center of SARS-CoV-2 superspreader networks.  

Known unknown #6: When I hear people say, “Well, vaccinating by age tiers will be the simplest,” I realize that they don’t understand the unique challenge facing us.  Unlike a universal, single-payer health care system like the United Kingdom has, the US health care system is famously fractured and decentralized.  In the United Kingdom, the National Health Service has a roster of nearly everyone in the country and an existing point of contact through their assigned primary care providers.  So there’s the infrastructure there to identify everyone in a community of a certain age and contact them during their vaccination tier.  US health departments do not have such lists and infrastructure.  Our first rounds of vaccination were setting-based (hospitals, nursing homes).  I suspect that these will be the easiest.  Workplace settings have updated rosters of employees and residents and ways to track them.  Moving this effort to the community setting will be enormously complicated. 

The unknown unknowns: Of course, I could be wrong.  Perhaps the Florida free-for-all will be the best strategy because it gets everyone vaccinated the fastest.  Perhaps the declining reproduction numbers (Rt, a measure of how fast the virus is spreading) we were seeing in US states in mid-December will keep dropping, buying us time.  Then the exact details of vaccine prioritization will be less of a determinant of who lives and dies in the coming months.  Perhaps witnessing a mad uncoordinated rush to get vaccinated will increase the desirability of vaccination and counteract the vaccine hesitancy that was a major barrier to achieving herd immunity.  Perhaps the people who need this vaccine most, for whom it is literally life or death, will get it before others. I could be wrong about much that I’ve written above. I hope that I’m wrong.  

Note: I remain grateful to and humbled by the many scientists, doctors, journalists, and others who have elucidated and shared accurate information with the public. They are the reasons that we have “known knowns” and “known unknowns.” Today, I especially remember Dr. Li WenLiang, who died at age 34 of COVID-19, and Zhang Zhan, a citizen journalist in China recently sentenced to prison for her reporting on the early COVID-19 outbreak.