COVID-19 Pandemic

Early days

R₀ = how contagious

CFR = how deadly.

When the new COVID pathogen first emerged, there was a lack of information about the lethality or case fatality ratio (CFR). While individuals are most concerned about the CFR (because they don't want to die) the government systems and hospitals get stress tested harder if R₀ is large, even if CFR is comparatively lower, because of the sheer number of infected people needing care all at once.

A high R₀ can quickly overload a nation's healthcare system and subject the newly-infected to triage/prioritized/rationed care, which can then increase CFR. This is perhaps the most unappreciated concept by those who advocate for personal choice/100% freedom/no restrictions. Once a local healthcare system is overwhelmed (beds are full, half the doctors and nurses are sick, no more ventilators, etc) what should have been a low death rate virus becomes a much higher death rate virus. It may also be more difficult for the public to see the need to voluntarily adopt restrictions that slow the virus's spread if the CFR is low.

There will be virtually no test kits available, which means it is impossible to track the pathogen's movement in near-real time. Deaths reported in a community allow one to "look in the rear-view mirror" and know about how many people were infected about a month ago.

Redneck Epidemiology

Initially, the CFR will be unknown, but we will really want to know what it is. It may be possible to do some back-of-the-napkin or "redneck" epidemiology as demonstrated by Carter Mecham, retold by Michael Lewis in the book The Premonition pg. 173-174:

At midnight on January 26, the Chinese authorities announced 2,700 new cases and 80 new deaths. "I thought back to 2009 H1N1," wrote Carter at six the next morning, "and remembered how we used the 1918 Pandemic as the model in our heads (the movie) and a lesson I learned would be to be careful the next time to not cling to a single model (movie in my head) but use a range. I almost fell into the same trap this time by focusing on SARS. I did pull the 2009 H1N1 data but didn't look at that data as closely as I should have. I finally did last night." What he saw in that data was that, while the number of deaths looked a lot like the number in the early stage of the SARS outbreak, the speed at which the disease moved did not. It was moving much, much faster, and very much like the swine flu had moved. "The movie isn't SARS. And the case ascertainment isn't SARS- it is more like H1N1." It was as if he'd taken the virus to a clothing store and tried trousers on it until he found the best fit. H1N1 fit this new virus in the speed at which it was traveling. There was good news and bad news here. The good news was that it meant that a lot more people were surviving the virus than anyone knew. The bad news was that the virus would infect, and kill, vastly more people than the original SARS. Carter found a study that the CDC had made, after the fact, of the cases of swine flu that had gone undetected, or at least unrecorded, back in 2009. The numbers were incredible. For every case that had been recorded, somewhere between eighteen and forty cases had been missed.

He then asked: What if right now health authorities around the world were detecting only between one in eighteen and one in forty cases? "Yesterday we had 2700 cases and 80 deaths," wrote Carter. "Let's assume the real number of cases is 18-40 times greater, or 48,600-108,000." The 80 deaths were the result of some lower number of infections that had occurred roughly two weeks earlier; to figure out the fatality rate of the virus, you needed to know how many cases there had been. Carter did some rough math, using a reproductive rate of 2, on the low side, and 3, on the high side that is, each week, the number of cases was either doubling or tripling. "The case count 2 weeks ago would be 1/4 or 1/9 of 48,600-108,000 or 5,400-27,000," he wrote, and was likely doing the math in his head as he did. "So 80 deaths with a denominator of 5,400-27,000 projected cases 2 weeks ago, gives us a case fatality rate of 0.3%-1.5%. But these are very crude estimates."

To me, it is impressive that just a few weeks into the pandemic, Carter was able to accurately estimate the infection fatality ratio range based on a few news clips. The actual all-age IFR determined later that year was 0.68% (0.53–0.82%).

Typical Progression

COVID-19 emerged as the "alpha" variant, then mutated a year later and a less-lethal, more-virulent mutant "delta" took over. The lower CFR and higher R₀ also imparts immunity (after infection) to the alpha variant, so it took over. A year after that, Omicron emerged with an even lower CFR and a higher R0. While alpha started at 20-50x deadlier than the seasonal flu, the subsequent variants each had 25-50% of the original's CFR. Combined with the fact that an increasing proportion of the population had a prior infection, the pandemic would wind down after 2 yrs.

The Spanish Flu pandemic of 1919 killed 2% of infected people, but the virus that did the damage--H1N1--still circulates today. We have enough herd immunity that it no longer causes a pandemic. While some will advocate for letting the virus run completely unhindered through society to quickly achieve this desired herd immunity, in fact this must be done in a gradual fashion so as not to run out of hospital beds along the way. Once a hospital is saturated, they switch into "field clinic" mode of care: people are on gurneys in the hallway, doctors spend less time with each patient, equipment is rationed, and death rates climb higher than they would have under normal care conditions.

Case Ascertainment Rate

For each new case identified via PCR or rapid test, some larger number of actual cases exist because not every infected individual seeks testing or care; many just "stay home sick" until recovered. The case ascertainment rate (CAR) is expected to be low in the early days of a pandemic, unless extraordinary efforts are made. For COVID-19, South Korea undertook the efforts needed to procure tests and aggressively trace the infections in the early days of the pandemic. How can we tell? By looking at the case fatality rate (CFR) data.

Early in the pandemic, all nations are dealing with the same COVID variant. Total deaths are divided by total confirmed cases. But, if a nation is undertesting, there will be disproportionally more deaths per case. If we compare USA to S Korea, we see S Korea has a relatively constant CFR, or slightly declining as doctors got better at treating over time. Compared to S Korea, the USA shows up to a 3x higher CFR in the early months. This is not because the variant in the USA was 3x deadlier, it is because the USA was doing 3x less testing, per capita.

The other way to show that a nation has a very low CAR is to check the results of serosurveys. By selecting a random subset of a country and testing everyone's blood for antibodies, one can determine the viral prevalence over the past few months, compared to the number of positive tests during the same period. For example, Italy did not have enough tests early in the pandemic either; the ENE serosurvey would later show only 1 in 10 cases was positively identified. This leads to two confusions:

  • News agencies and laypeople looking at the data (dark blue only) often drew the conclusion that the second wave had more infections and we should "brace ourselves" for the pile of dead bodies. Except that pile fails to materialize in the second wave, leading to conspiracy theories or mutation discussions. Some invoked the "dry tinder" argument to explain that the first wave was particularly lethal because all fragile people quickly expired as the virus circulated.

  • Secondly, many news agencies never retrospectively update their case counts after serosurvey data is available. Progress defeating the pandemic looks slower than it actually is because 100,000 new cases in 2021 has a bigger % change impact on the total when the original total is undercounted.

Model projections

The CDC would gather model projections from universities and epidemiologically-inclined groups and average them together to produce 4-week look-ahead projections. I gathered these forecasts and retrospectively compared them to actual case counts. Usually, the model ensemble is no better than a linear extrapolation of the previous 4 weeks of cases. Also, no single modeling group was a "stand out" to look at and trust for next time.

Vaccine development

The mRNA "vaccine" was released in December 2020, just 13 months after the virus first appeared. I put quotes around "vaccine" because it did not perform the way most expected:

  • One Tetanus vaccine injection protects against infection for 10+ years

  • Four Polio vaccine injections protects against infection for life

But the mRNA vaccine required two injections and the protection waned after about a year. A third injection "booster" was recommended to restore protection, but that too waned after several months. The COVID vaccine behaved must more like a flu shot:

  • Must be dosed annually

  • Protection is variant-specific; if the wrong variant spreads, the flu shot offers limited protection.

  • Protection fades, must be administered yearly

Since the COVID-19 mRNA vaccine actually behaves like a flu shot, it should not have been called a vaccine in the first place. The fact that vaccinated people began contracting the virus after 9+ months led to vaccine hesitancy from the unvaccinated.

Kaplan-Meier Curve

Pfizer put the test results for their vaccine online. 21,669 people received the vaccine and 21,686 received a placebo and their infection rates were tracked for 112 days. The cumulative incidence curve, or Kaplan-Meier curve, shows the results experienced by the former in blue and the latter in red:

The vaccine becomes effective 14 days after the initial dose. This is a wonderfully simple way to display the effectiveness with no statistical permutations. It is so simple, Randall Munroe made a comic about its simplicity:

XKCD 2400: We reject the null hypothesis based on the 'hot damn, check out this chart' test.

Vaccine Hesitancy

Among the people I know, only about 2/3 elected to get the vaccine. The rest were hesitant and wanted to wait and see if there were any long term side-effects. To them, it seemed silly to be the first to get vaccinated, but in their defense there are a couple examples of novel medications causing damage and being withdrawn from the market (see: thalidomide or the Tuskegee study). Brand new safety devices like airbags, life jackets, vaccines, etc may save your life or harm you, and the wait-and-see approach is useful for a time. Perhaps someone thinks: "it takes 2 years (or some number) for the nefarious side-effect to show up" and so they think they should wait 2.5-3 years before getting vaccinated. This is operating under the fallacious notion that everyone who got the vaccine will simultaneously show the side effect at precisely the 2 year mark. But that's how it works out if you gave a single dose of a side-effect-causing vaccine to a single person. When a billion people get the vaccine, the side effect(s) emerge over a time range. Specifically, other side effects show up according to a gamma distribution function:

The tail of the distribution on the near-term side of the gamma distribution peak does not need to be very large to show a meaningful number of instances of the side effect occurring when a large number of doses are administered. We can expect the side effect to emerge:

  • Only months after hundreds of thousands of vaccines are administered

  • Only weeks after hundred of millions of vaccines are administered

This requires one to consider the likelihood that they will catch the disease and become severely ill or die after the vaccine is available but before they decide to get vaccinated. I cannot see this time frame justifiably being longer than ~6 months after there have been 10 million doses. If you're still holding out at that point, it's for a different reason.


Masks were mandated in public places by local governments to minimize the spread of the respiratory virus. Masks do relatively little to stop an inbound particle carrying the virus unless they are tight-fitting (most aren't). However, masks are reasonably effective at capturing outbound (coughed/sneezed/spittled) particles. That is, masks became a way to ensure more people covered their cough. Far too few people cough into their elbow and instead hold a closed fist an inch or two in front of their mouth (which is useless).

Support/opposition for masking emerged along party lines: republicans generally opposed, democrats generally in favor. Republicans generally felt "if you don't like the risk, stay home" and see protection is a self-responsibility. Democrats generally felt that we each had a responsibility to one another to do the right thing, even if that meant bearing a personal inconvenience.

It did not fall strictly across party lines. My observation is: there are "me people" and "we people" in this world and the "we people" would wear masks, while the "me people" would not. We People = our collective survival rates improve if we bear this burden together. Me People = my liberties come above all else, "The obedient always think of themselves as virtuous rather than cowardly." -Quote from Robert A Wilson


Schools, restaurants, bars, and public gatherings may be suspended to decrease the spread of the pathogen. These inflict a cost on small business owners and a social cost on those who crave personal interaction. They won't be popular and they can't be sustained for more than a few weeks. I observed cities and towns enacting these lock-downs when their ICUs and hospitals were effectively empty and not rising in occupancy. It seemed like the wrong thing to do. Yet there will be other cities in the news with overflowing ICUs and I read the testimonial of an exhausted doctor or nurse driving home after a 16 hr shift past a strip of open bars/restaurants/clubs full of unmasked people acting like absolutely nothing was going on. The frustration is real: when the medical staff is overwhelmed, we shouldn't be out in large groups partying and making things worse.

Therefore, I am only in favor of lock downs occurring when hospitals are full or filling quickly. Until then, use other social distancing measures, decreased restaurant capacity, increased ventilation, increased sanitizer use, etc.