Let’s talk numbers. I know what you’re thinking, I was promised no math. So no equations, just a couple of pretty pictures and a lot of words. I trust we’ve all seen the first chart – we know it, we love, we live it. 

Let’s be honest, it isn’t accurate, it’s idealized. 

The first thing I’ll note, is if the only thing that changes is that as time goes on more people become immune, the right side is not symmetrical with the left side, in fact the drop off is faster than the rise and that is because during the rise the rate of change in available infectees is less than during the descent because there are fewer available infectees. And yes, I created an Excel model and discovered this. 

The second thing, is that for this epidemic at least, you don’t start with protective measures (you know, the guidelines) in place, so the reality is there is this transition period as people react, and in the case of COVID in advance of any orders to, so there would be this period of lowering the transmission as behavior changed that would put you on a different slope.

The third thing is, people are people. And that means we have this tendency to adjust to risks so as we see the curve flattening (and thus the risk lowering), we would engage in riskier behavior, and as we see the curve steepening, we would engage in less risky behavior. Don’t believe me? Look at the history of automobile safety, where every safety advance is met with an increase in risky behavior such that deaths per mile decline less than what the safety feature should cause. 

So put that all together, and you have the actual reality of new cases just won’t ever look like those nice curves (oops, did I forget to mention random noise, so unsmooth the curves!) It will start steep, then flatten, but won’t ever go all the way down like you think it ought to. And there may well be multiple upticks depending on how long until either a vaccine or herd immunity kicks in.

I will just note at this point the area under the curves are not the same – the one that has lower transmission rates will have less area, i.e. fewer people will get infected. (Thanks again, Excel model). But since these are idealized, and the reality is messy and complicated, yeah, all bets are off on which one infects fewer people.

And that’s the real curve of actual new infections. Now let’s talk about the measured version. It isn’t the actual number, it’s a sample of the actual number.

So it’s going to have more noise (go to Worldometer, look at the new daily death histogram for the USA and explain to me why it varies 100% over a six day period – don’t worry, I’ll wait). 

And it’s going to depend on how many tests we run a day. More tests mean more new cases measured. If the actual number were flat, if you ran more tests you would see an increase in the number of cases. As I always said about software bugs, the sooner we stop testing the sooner we’ll stop finding problems. Normally we take care of that by normalizing, i.e. dividing the number that allows you to make meaningful comparisons like total number of people in a state to compare states or total number of tests (which is why I think Dr. Birx, whom I admire and respect, was always quoting the percent positive rates on tests as well as the totals). 

Now normalizing works well if you are consistent in your test criteria, e.g. people who have symptoms bad enough they are willing to have a swab shoved far, far up their noses (in the old days, till it came out the back). And test type. What happens when you starting with one test done by hand in state labs, and then keep adding different tests run on different equipment and you switch from shoving the swab 4 inches up someone’s nose to 2 inches? Do they all have the same sensitivity and error rates? But wait, that not the only changes you make, what happens when you keep increasing your testing to the point where you test everybody in a meat packing plant even though no one asked to have a swab shoved up their nose? IOW, you test both asymptomatic as well as symptomatic? You see another round of big jumps, like where Missouri found 373 workers were infected but asymptomatic. And at the same time you’re looking at new cases chart like the one I included, the State of Missouri is moving the date of the case from the test to when the person reports the first symptoms occurred.

So look at that chart and tell me are cases increasing, declining, or treading water. And while you’re at it, try to come up with a way to normalize across all those changes in testing. I don’t know about you, but when I try to look at that real chart and compare it to the idealized chart, I wonder how one can possibly relate to the other, and I’m vary comfortable with data, noise, uncertainty, and models vs. reality.

So I will say that Missouri, and I think every state at the request of the Federal government, provides a deeper dive into the data: 

https://health.mo.gov/…/novel-coro…/pdf/analytics-update.pdf

where they do cover many things, including COVID hospitalizations and percent positive test results (smoothed to a seven day average, which given that’s a work week seems right to me at the risk of loosing sight of short term trends) which shows a peak back on 3/21 at over 20% with a decline, a long plateau, and another decline to ~4% on May 2. I don’t know about you, but that makes me feel better, like the rate of infections really are declining.

So what does it all mean? I think we are getting better, but don’t expect the simple one wave if you do it right, two waves if you do it wrong concept, and don’t even get me started on why the second wave will by necessity be worse than the first. 

Now here’s another quick, back of the envelope calculation – chart shows about 200 new cases a day, say that represents worst case 10% of real infection rate so 2000 new cases a day, and given a population of 6,138,000 (I rounded up from wikipedia 2019 figure), it will take another 3,069 days before we all catch it here in Missouri, wait, subtract 30 days for what we’ve already gone through, so over eight years. I think we’ll have a vaccine by then. If you don’t like my calculations, get your own envelope.

And I think we should start calming down.

Tags: COVID19

I’ve come to the conclusion we are not thinking rationally about the COVID 19 pandemic. The problems are in large part driven by its novelty which leaves us vulnerable to our cognitive biases. First we have tunnel vision (AKA Cognitive Tunneling combined with Zero Risk Bias) and basically have focused on COVID 19 to the exclusion of all other considerations. This means we are looking at the results of the virus to consider every death from COVID 19 as a loss but ignore every other death, and infection from COVID 19 as a loss but ignore all other health and non-health problems. And loss aversion means we can’t take our focus off COVID, we only see the risks associated with COVID 19 infections and fatalities and are blind to all other risks. Essentially the enormous uncertainty and newness so terrified us that we pulled the covers over our head and refused to get out of bed. That’s OK for a day or two, but if we all do that how long can we keep it up?

So my suggestion is to lose the exclusive focus on COVID since it leads to a terrible local optimization problem where we are only optimizing for COVID and nothing else. Clearly COVID is a serious disease, but it isn’t the only serious disease we face. Instead, we should recognize the loss from COVID without action and count every decrement from that as gain – because we can think much more rationally about gains and in large part that’s how we think about all other disease and causes of death. 

So what does that mean? Let’s take that initial estimate of 2 million deaths in the US over the next two years (approximately) and use as our benchmark the 2.8 million non-COVID deaths (just to put an anchor out there, the case fatality rate for a year of life in the US is 0.86%) plus the 1 million COVID deaths per year and then strive to drive the number down from 3.8 million per year. And this is realistically the situation we confront. Since COVID would be the leading cause of death in this framing, clearly efforts should be significantly but not exclusively focused there as it would be only somewhat ahead of heart disease at over 647,000 and cancer at 599,000 deaths per year. This framing allows us to take a holistic system approach to the problem and we need our metrics to reflect that. The constant drumbeat from the media showing the daily death count from COVID alone doesn’t help us, it hurts us.

You may ask how is our current framing is hurting us. Keep in mind that the general rule of thumb (Pareto Principle) is that the first 80 percent of a solution is cheap and easy and the remaining 20 percent is increasingly expensive and difficult. In our zeal to drive one risk to 0, we are paying the full expense of getting to a complete solution for one problem (COVID) while starving all others of resources and attention – and in terms of fatalities those problems cause more deaths. If we don’t look at the full picture, we will be worse off and won’t make the right trade offs. Adult life is tradeoffs. We need to pull the covers down and get back to living.

Death statistics can be found at: https://www.cdc.gov/nchs/fastats/deaths.htm

Tags: COVID19