by Buzz Hollander MD
“Lies, damn lies, and statistics.” Most people would attribute that quotation to Mark Twain, and they would be wrong. Now, if I said, “Herd immunity requires 60% of the population to have covid-19 before transmission will start to slow down on its own,” most people would consider that to be true, and they would also be wrong. In fact, it is the most dangerous kind of statistic – not a mean-spirited lie, but a reasonable sounding assumption repeated so often that we all assume it to be true. Believing the oft-repeated statistics of herd immunity has made it hard to understand what we have been seeing with our own eyes. Only when we understand herd immunity can we begin to make sense of this pandemic.
Now, I do not want to imply that epidemiologists been deliberately lying to us since they became the stars of the intellectual universe four months ago. Heavyweights like Marc Lipsitch and Joel Miller have unveiled some dense and illuminating twitter threads on the matter of herd immunity. However, perhaps because of the incredible complexity of the subject, the implications never really caught on with the media version of herd immunity that made it to the public.
Most of us got the condensed version already, in which the herd immunity threshold is related to the reproductive rate, R0 (“R-naught”) by the following equation: 1-1/R0. If an infected person passes the virus to 2.5 other people on average, R0 is 2.5 and the threshold for herd immunity is 60%. That’s fine and good, albeit simplistic, when discussing immunization plans, the primary use of herd immunity concepts when they gained popularity in the 1970s. However, it is a terrible model for understanding the dynamics of a pandemic!
The dynamics of a pandemic are messy and complex. However, one critically important point needs to be made, and you are more likely to see a jackalope hop by your house than find this concept in a media article on herd immunity: the simplified (1-1/R0) definition of herd immunity implies “homogenous” spread of a disease through a population. Everyone is assumed to randomly bump into other people, and everyone they bump into is equally susceptible to the virus. Sound logical to you? I didn’t think so. This assumption is what we might call a model killer. Virus spread in the real world is actually “heterogenous.” Now we can start to talk about this like proper armchair epidemiologists.
While there are scores of factors which disrupt homogenous spread and can scupper a herd immunity model, let’s focus on four of particular importance: contacts, exposures, infections, and immunity.
1. Herds of People Are Not Like Herds of Sheep
We can start with an example: Think of your diabetic Aunt Ramona in New York City, who volunteers for ten organizations and loves her book clubs. Then there’s your retired Uncle David, who is fit as a fiddle and prefers long walks alone. Aunt Ramona probably got COVID-19 in the first wave, and possibly spread a good bit of it, too. She is one of the 20-25% of New Yorkers who tested positive for antibodies; David probably is not. Assuming some degree of lasting immunity (or death, but we are going to keep this example upbeat), Ramona is now out of the picture for any second wave consideration. The first wave infects Ramonas, not Davids! The 75% of un-protected New Yorkers are going to be comprised disproportionately of Davids, which drives down the effective transmission rate in future waves. This is Mother Nature’s way of “flattening the curve.” Now, it does very little good at the start of an outbreak — the Ramonas of a population are able to do a lot of damage. However, as time passes, if the Davids of the world don’t change their behaviors and become more gregarious or less hygienic, we can expect the real world, or effective reproduction number (Re) to drop – and, with it, the threshold for herd immunity.
“How much?” is a fair question to ask. In early May, two papers appeared as pre-prints that addressed this question, one a paper from a group led by Tom Britton, and the other by Gabriela Gomes. The Britton paper projected the herd immunity threshold for COVID-19 to be 43%; the Gomes paper modeled the threshold to be as low as 10-20%. The implications of this are huge, of course, but the past four months have taught us to take all models with a grain of salt.
One reason to suspect that SARS-CoV-2 is indeed spread in very heterogenous fashion has been its well-documented tendency towards case clusters and superspreader events. One model suggested that 10% of infected people might cause 80% of all cases. In terms of actual data, a study based on contact tracing in Hong Kong found that 20% of cases led to 80% of all infections. Yes, this implies that most people infected with SARS-CoV-2 did not spread the disease to a single soul. If the 10-30% of highly effective spreaders are largely out of the susceptible population early in the course of the pandemic, the disease loses its best soldiers, and the threshold to reach herd immunity drops.
2. All Exposures Are Not Created Equal
Not every social contact carries the same risks from exposure. If someone with an active case of covid-19 coughs in my face while I am swabbing them without any PPE, I am more likely to be infected than if I share a menu with a previous diner with covid-19. No one knows for sure just how many particles of SARS-CoV-2 are needed to cause someone to be infected rather than simply exposed, but there is general consensus that larger viral exposures lead to more likely infections, and the larger the exposure, quite possibly the more severe the infection, too. Not clear is whether severe cases of COVID-19 lead to higher viral loads and to larger future exposures. However, it is likely that the type of exposure can affect the likelihood of transmission — another way that our models might get confounded!
A great example is household exposure. In a helpful compilation of infection rates within households posted by Dr Muge Cevik, household transmission rates from studies in China were 13.2% of 157 cases; 11.2% of 391 cases; 19.3% of 349 cases; and 16.3% of 105 cases. This would imply the average COVID-19 patient would not infect a single person within a 4 or 5 member household. The home environment is a surprisingly unfriendly place for transmission of the novel coronavirus.
Contrast that to prisons, meat processing plants and cruise ships. There are a lot of problems with the type of exposure in these places. Density is high, so exposures tend to be closer range. The Davids of the world are not free to limit their interactions as they would in their home environments – everyone has a lot of social contacts. The number of daily interactions is extremely high. I will add one more possible critical factor. Whether it looks like this:
Editorial credit: Gianluca Piccin / Shutterstock.com
People take their meals in cafeterias: poorly ventilated spaces where folks linger in the company of each other; loud enough that they sometimes have to shout to be heard; and passing condiments around the table before touching their food and putting it in their mouths. Cafeterias are an ideal venue for a superspreader event, and create environments which can blow past a normal herd immunity threshold.
The risk is not just theoretical, of course. A cruise to Antarctica which embarked in mid-March (after the COVID-19 pandemic had been declared) screened the entire ship with PCR testing 20 days into their cruise: 59% were positive! Prisons are even worse. Now that some prisons have began testing their entire populations, we are seeing some astounding numbers: Ohio’s Marion Correctional Institute found that 80% of their over 2000 inmates were PCR-positive for covid-19; and Marion Correctional Facility (OH) had 73% test positive.
The takeaway is that exposure type is another factor which can reduce real-world herd immunity thresholds; people who live or work in high intensity exposure settings are more likely to be the Ramonas of the first wave, and therefore not be vulnerable to a second wave. The caveat is that if our symbolic David takes a cruise, starts working for a meat processing plant, or becomes incarcerated, all bets are off.
3. Not Every Infection is a Problem Infection
By now, everyone is aware that older, sicker individuals are more likely to become seriously ill if infected with SARS-CoV-2. However, we do not hear much about the people testing positive who get only mild illness (“paucisymptomatic”), or even no symptoms at all (“asymptomatic”) despite a positive covid-19 test. Some asymptomatic people might only be “pre-symptomatic” (i.e., about to get sick) but a well-designed study comes back to check on these people a few weeks after their positive test. We assume a lot of asymptomatic positives to be younger, healthier individuals. In the Mission District of San Francisco testing, for instance, people with medical comorbidities were exactly half as likely (38% vs 76%) to have asymptomatic covid-19 than those without.
However, the very high prevalence aforementioned “cafeteria” environments reveal an interesting phenomenon when examined more closely: a shockingly high asymptomatic rate, beyond what age alone would predict. Perhaps a cafeteria is a safer place for exposure than, say, a bar after an Atalanta-Valencia soccer match, or a choir practice in the Seattle area. The latter — granted, with an average age of 69 — had 87% of its attendees develop symptomatic illness. In a Washington state nursing home, 94% of positive tests were eventually symptomatic. Having worked in a nursing home prior to starting medical school, I can attest to the fact that there is a LOT of shouting in the ears of heard-of-hearing residents while moving them in and out of their beds.
In contrast, in large community studies, about 50% of positives in Iceland were asymptomatic, as were 45% in large scale community testing in Indiana, and 43% in the town of Vo in northern Italy. Moving up the spectrum, that Antarctic cruise that tested everyone on board found an 81% rate of asymptomatic positives despite Antarctic cruises having an average age well above a typical population. At the extreme end of the spectrum, a little over 90% of the positives at the St Joseph, MO, meat processing plant were asymptomatic. In correctional facilities, multiple reports detail the same trend: a four state review found an asymptomatic rate among 3277 covid-19 positive inmates to be 96%; and in two facilities that followed prisoners for 2 weeks after PCR testing, Montgomery County (PA) prison posted a 95% asymptomatic figure; and 99% of positive inmates at the Bledsoe County (TN) Correctional Complex lacked symptoms.
These are eye-opening numbers that largely escape discussion by the press – except for comments like “asymptomatic transmission has caused particular challenges.” I get it — it is hard to find and isolate asymptomatic carriers. However, I might add, “asymptomatic patients don’t take up hospital beds or die of covid-19.”
A recent, hotly-contested study suggested that aerosolized spread, not the widely assumed droplet spread, was the primary driver of covid-19 infections. This may or may not end up being borne out by future studies. The implication is that poorly-ventilated indoor gathering spaces could be spreading infection by circulating fine particles with small amounts of virus that evade the immune responses in our mucous membranes and head straight into our respiratory tract. This might just explain the low percent of symptomatic cases in this exposure setting versus others of the “direct hit,” large droplet variety — like nursing homes and bars after sporting events.
The hope is that these mild cases produce lasting antibodies just as well as serious cases. We are a long way from knowing this with confidence, due to a lack of systematic follow-up testing. However, a recent Lancet article detailed 9 PCR-positive patients from the Diamond Princess cruise. Unsurprisingly, 6 were completely asymptomatic, and all of them tested positive for IgG antibodies to SARS-CoV-2. This is a tiny but encouraging report. On a larger but less scientifically rigorous scale, two correctional facilities with a high proportion of asymptomatic covid-19 patients performed antibody testing on their entire populations: George W. Hill (DE) Correctional Facility found 43% of inmates to have antibodies, and Parnall Correctional Facility (MI) reported 92%! What does an exposure that produces antibodies while only rarely causing symptoms remind me of? A vaccine. Unfortunately, these numbers cast doubt on the WHO’s recent assertion that asymptomatic patients rarely spread the virus.
In any case, as economies re-open and people return to large-scale work environments, if the exposure types favor asymptomatic disease, we might expect a smoother passage to herd immunity than first-wave, lockdown data might predict.
4) Having Immunity
An individual’s immune system is the other essential factor which determines whether an exposure leads to an illness, and how severe that illness becomes. The assumption behind our basic (1-1/R0) calculation for herd immunity is that everyone entered 2020 equally susceptible to covid-19. The closing line of a depressing May 29 New York Times article on herd immunity was that there were “…328 million Americans who were susceptible to this when this started.” It’s worth looking at that line of thought a little more closely.
Our immune system is mind-blowingly complicated. We humans are exposed to viral particles in every social situation. As a first line of immediate defense, we have an “innate” immune response – things like skin, mucous, a cough reflex, and, more glamorously, immune system agents like cytokines and “Natural Killer” cells, that recognize antigens like viral particles that are “not self” and rapidly act to limit their ability to gain traction in our bodies. The “adaptive” or “acquired” immune response, on the other hand, involves recognition of antigens from prior exposure and the production of antibodies to neutralize antigens, and requires days to take full effect. These two types of responses are deeply interconnected, but those are the basics. The rubber hits the road when the immune system is exposed to viral particles that pass through our physical defenses.
Some people are going to be more vulnerable to an exposure to SARS-CoV-2. Old age is associated with weakening of the innate immune response. So, too, is chronic disease. Certain comorbidities dominate covid-19 cases: high blood pressure, obesity, diabetes, and heart disease. These seemingly disparate health problems all share one common ground: the metabolic syndrome. Put simply, metabolic syndrome is a problem of insulin resistance. Insulin resistance, we are beginning to discover, is a disease process that affects every aspect of our health, including our innate immune response. The pro-inflammatory state found in people with metabolic syndrome is likely related to the excess inflammation thought to play a role in the “cytokine storm” and other severe inflammatory reactions seen in the worst covid-19 cases.
Put simply, a healthy 40 year old is less likely to turn an exposure to SARS-CoV-2 into a significant illness than an obese or diabetic 40 year old. The same goes for a healthy 75 year old versus an unhealthy peer. What’s more, psychological stress was shown in a decades-old study to inhibit the immune response. The same goes for sleep deprivation. (The physician in me notes that healthier Americans would, in turn, be less vulnerable Americans in times of a pandemic.) Back to herd immunity, though — the most susceptible members of a population are most likely to turn an exposure to virus into an illness, and spread disease. These same people are more likely to have antibodies (or have died) from the first wave of a pandemic, and this phenomenon further lowers the herd immunity threshold.
Another factor worthy of mention is the possibility that significant “partial immunity” might exist to covid-19. The notion that the known cold-type coronaviruses, especially if acquired recently, might convey some immune protection was presented early in the pandemic. Watching much of the developing world resist major covid-19 outbreaks, some have posited — controversially — that the common tuberculosis vaccine, BCG, might confer some resistance; and, similarly, the oral polio vaccine. Whatever the source of pre-existing immunity, a pre-print paper released last month from the La Jolla Institute for Immunology found that 50% of their pre–2019 control blood samples had immune cells which reacted to SARS-CoV-2. Granted, the study was small (40 subjects); it had not been peer-reviewed; and all the action took place under a microscope and not in a human body; but, still, the notion is plausible. It is the old concept of “immune imprinting.” We don’t doubt that prior flu exposures to different strains from past years provide some protection against new strains of flu. The upshot of the La Jolla study, which was primarily designed to investigate vaccine targets, largely escaped media notice. That is, until May 27, when the Conservative Review ran the following tantalizing headline: “Horowitz: Bombshell study: Could half the uninfected population already be partially immune?”
The prison or cruise sub-populations which have reached 60-90% prevalence tell us that no, half our population does not appear to be immune to covid-19. However, any factor that contributes to a less-susceptible population will help lower the herd immunity threshold, although the durability of that effect is hard to predict.
5) Putting It All Together
The first question to ask would be: “Is what we are seeing happen in the real world in line with a 60% herd immunity threshold or more like 10-20%?”
To attempt to answer that, we in the Western world look to Sweden. While Sweden has taken a good many measures in the realm of social distancing, among them outlawing gatherings over 50 and closing secondary schools and universities, they have not approached the restrictive lockdowns seen elsewhere. As such, epidemiologists the world over warned that anything less than a total lockdown would doom Sweden to overwhelmed hospitals with tenfold ICU demand beyond capacity and a massive death toll.
What really happened was a more tempered outbreak. Here are screenshots of the actual ICU data, from the Swedish Intensive Care Registry; and below that, the actual total deaths per day, from Worldometer:
What began to push ICU and death rates down in late April over a month after restrictions were put in place? While the effective reproduction number was estimated to be 1.6 in Stockholm in early April modeling, why did it so quickly drop to <1, shrinking the outbreak, without any new restrictions?
If you answered, “summer backstop,” please try again. Not only do we not know why, or even if, a “summer backstop” exists, but the weather in Stockholm was equally dismal from late March onward; the high temperature in May did not even cross 60 degrees Fahrenheit until May 21! Something else was going on.
Recently released data suggests that only 5-10% in Stockholm would have had antibodies to SARS-CoV-2 back in mid-April. Low as this sounds, perhaps the the most likely explanation is that Sweden was already beginning to flirt with the 10-20% herd immunity threshold as described by Gomes. Otherwise, we are left with the rather flimsy theory that it took the Swedes 5 weeks to really get the hang of social distancing and hand washing.
This explanation sounds radical if you have only been reading media reports on herd immunity. That May 29 New York Times article on herd immunity reminds us no less than five times that herd immunity requires about 60% of the population to become infected, and concludes, “most places would have to see 10 or more times as many illnesses — and possibly, deaths — to reach the point where an outbreak would not be able to take off.” Is that really our future without a vaccine or strict social distancing?
Here in the US, most everyone in the medical community (myself included) predicted a sharp rise in covid-19 illnesses within weeks after states began the re-opening process in late April and early May. After all, if the effective reproduction number could only be controlled by social distancing restrictions, it stood to reason that relaxing those restrictions would lead to a substantial increase in disease burden. Using hospitalizations as our benchmark (due to the well-documented problems with relying on case numbers), what we have seen has told a different story. A screenshot of COVID-NET hospitalization data pooled from14 states shows a steadily slowing hospitalization rate:
US covid-19 death rates, again courtesy of Worldometer, echo this trend:
The news right now is filled with alarming reports about rising hospitalization rates in certain states. However, a June 6 ABCNews report revealed that only 8 states are seeing rising covid-19 hospitalizations, while 5 are about flat, and 37 are actually falling. We are seeing many more stories about the 8 states rising than the 37 states falling! If herd immunity is protecting the hardest hit states of the first wave, like New York, Michigan and Louisiana, we might expect the states struggling now to have largely escaped a broad first wave. Indeed, using deaths per capita due to covid-19 as a rough yardstick for disease prevalence, those 8 states — Arizona, Arkansas, Mississippi, North Carolina, South Carolina, Tennessee, Texas and Utah — rank 22, 42, 13, 33, 30, 40, 41, and 45, respectively. Of the only two in the top 30 of most deaths per capita, Arizona has been virtually flat in hospitalization levels since late March, and deaths have been flat-to-declining since mid-April; and Mississippi, which never started down from its first wave, has seen essentially stable ICU hospitalizations and deaths since mid-May.
Again, it’s hard to explain what we are seeing on the ground with a belief that only social distancing and mask-wearing can protect us from wide-spread outbreaks of covid-19. Once we allow for the possibility that herd immunity might already be limiting the transmission and severity of this virus, what we see starts to make sense.
To be clear, this is not an exercise in finger-pointing. Your perspective on lockdowns depends on how you value lives lost versus personal liberty and economic health. What’s more, a place like Sweden — or insert any lightly-affected region here — might stereotype as younger, healthier, more conforming, less dense, and less inter-connected than your home region. New York City serves as a warning – sprinkle in high density, multi-generational homes, tightly-packed public transit, and sub-populations with poor health, and you see a rise in prevalence to 20%, and double that in certain neighborhoods. Declaring the imminent arrival of herd immunity in New York City once prevalence hit 10% based on the above arguments would have been dangerously wrong.
The other X-factor is your optimism about a vaccine. Your openness to embrace the herd immunity threshold as a means to naturally check covid-19 transmission can vary greatly with your feelings about a safe, effective, available, and widely-accepted vaccine arriving in the months ahead. There are plenty of reasons for pessimism. However, if a public health policy maker has cause to believe their region could be largely vaccinated in January, then they might justify a policy of maintaining a low reproductive number via social distancing, combined with active testing/isolation/quarantine policies to minimize cases, hospitalizations and deaths, while waiting for a vaccine’s imminent arrival. On the other hand, if you don’t expect a vaccine for another year, that sort of aggressive program might seem socially untenable. Yes, letting go of social distancing and a “test and track” mentality will lead to more deaths from covid-19. In the context of influenza (some 50,000 deaths a year) and tobacco (closer to 500,000), however, our society seems to have been pretty comfortable with tolerating a substantial amount of suffering and death in the name of personal freedoms and economic growth.
Every region is unique in their likelihood to benefit from herd immunity at current low prevalence levels. I see four types of regions in this regard:
- Places like Hawaii, Alaska, New Zealand, Australia, and Japan, which have had very low prevalence to this point, possibly in the <1% range, and a minimal outbreak. They were able to harness island-type status to limit clusters and risk of superspreader events, so have almost no herd immunity protection at this point – but probably benefit from unique societal/geographic/climactic factors to help provide a ceiling to their transmission rates. They are probably more vulnerable to a Sweden-like experience than the nightmare of New York City.
- Some areas had explosive early outbreaks, suggesting a high reproduction number, that were rapidly tamped down by a strict lockdown, like Seattle and San Francisco. These places make me nervous if social distancing is abandoned.
- Other regions never seemed to heat up much, even with lackluster restrictions, like Georgia and south Florida; both states have seen covid-19 deaths stabilize or decline since mid-April highs, despite easing social distancing requirements, suggesting inherently low reproductive rates, and might already be able to benefit in terms of herd immunity dynamics from a moderate prevalence in the 5-10+% range.
- Then there are cities like New York, Detroit, and New Orleans, where per capita death rates many times that of flu season imply relatively high prevalence rates coupled with high reproduction numbers; on a broad scale, they are most likely to already be seeing substantial limiting of transmission from approaching their herd immunity threshold, but on the sub-population level, they carry high risk of explosive clusters in places largely spared by their first wave.
Policy should flow from these self-assessments, on a regional level. Public health decisions should not be cookie-cutter policies. A city, county or state’s decision making needs to incorporate their willingness to assume health risk relative to economic risk, their confidence in a vaccine, the vulnerability of their population, and the interplay of reproductive rate and herd immunity threshold in their region.
Regions that are high on the scale of trust in an impending vaccine and fears over an elderly or unhealthy population rationally could pursue a strategy of aggressive containment. They might allow only low-risk groups to gather — like school children and younger workers — and avoid large events, continue social distancing while stepping up protections for the elderly, and actively test/isolate/quarantine. Their hope would be that a vaccine comes along and swiftly takes them over the herd immunity threshold.
However, regions not bullish on an imminent vaccine, and with a less vulnerable population or early indications they are approaching a herd immunity threshold, might choose a more nuanced approach. Spread among the young and healthy might be deemed a positive, in that the sooner herd immunity dynamics kick in, the sooner there will be fewer cases of covid-19 to spread from the low-risk to the high-risk. After all, we have seen all too often how difficult it is to put a bubble around the elderly in places like nursing homes. This approach could tolerate low or moderate exposure settings among low risk populations: workplace cafeterias, movie theaters, concerts for the under 40 crowd, or even football games, for healthy non-seniors. The idea is to avoid pot-bellied, 75 year old Philadelphia Eagle fans screaming near each other, or 50 year old hypertensive Eagle fans yelling at each other in the bars afterward. Situations that congregate vulnerable populations in indoor settings for prolonged periods of time are the most concerning; for example, churches and casinos fall squarely in this category, and I pray for social distancing and good ventilation in places of worship and wagering, both. Nursing homes are one of the few places where I think at least weekly PCR testing should be required for visitors and staff.
Places that choose to embrace less restrictive approaches will require vigilance to minimize the risk of a New York-style outbreak. Periodic antibody testing of randomized samples of the local population will help reveal if spread is slowing in ways suggesting herd immunity. Close watch on hospitalization figures for signs of explosive growth of serious covid-19 cases could merit the re-establishment of practices likely to limit large transmission events, like banning large group gatherings and indoor dining.
If regions do things right, this country as a whole should avoid a panic-inducing second wave. Outbreaks will be unavoidable, and some of them probably will be severe. However, if we keep our eyes on the prize — herd immunity — we might find the road to a vaccine to be less rocky than predicted.
UPDATE 8-22-2020: This piece has aged fairly well, given that I started writing it almost 3 months ago! I think the experiences in states like AZ, FL, and TX – not to mention Sweden, which has made the strongest case for herd immunity threshold being as low as 10% in a favorable population – argue strongly that herd immunity dynamics probably kick in well under 20% even in these “hotspot” regions, unless you believe that more restrictive rules (i.e., mask mandates) and changes in public behavior were the primary drivers in each of these different states, and had a surprisingly rapid effect that coincidentally mirrored Sweden’s experience.
What would I change now? Well, clearly Georgia and south Florida had a much longer way to go to benefit from herd immunity than I expected in June! Their slow case growth in May/early June despite lax restrictions was more a combination of good fortune and a low R0 relative to, say, NYC or Milan, than any real herd immunity phenomenon. In all but the highest R0 regions, we are seeing how covid-19 can simmer for weeks or months before reaching exponential growth, a la Oahu right now. I also think the section on asymptomatic spread in congregate settings was grossly over-simplified, so I expanded on those thoughts and wrote a new (also long) piece.
All in all, the contrarian notion that herd immunity might explain the unexpectedly short burns of the worst hotspots is gaining traction, as evidenced by this New York Times article last week, with the headline, “What if ‘Herd Immunity’ Is Closer Than Scientists Thought?”