Key takeaways (warning, this is a long post):
1) Be cautious about scary numbers that don’t pass the smell test.
2) Reflexivity will greatly lower the ultimate impact.
3) The biggest risk is hospital capacity.
4) If you are not a senior and don’t have other health risks, don’t panic if you get sick. Worry more about not passing it on to high risk groups.
I am not here to say “don’t worry about corona. It’s just the flu“. I am also not here to say you should worry about it excessively and treat it like it’s smallpox.
My message simply is there is a lot of misleading information out there as, because we are human, we tend to focus most on the scariest news. Therefore, my agenda here is to present some facts you may not be aware of so you can have a better sense of what the worst case, best case, and in between outcomes might be.
A Possible Worst Case
So let’s start with the really scary stuff first. The true doomsday estimate about what happens if we do nothing are speculated at 100m infections. If we take the 3-4% mortality rate, then as many as 4m Americans would die. That’s pretty dire!
There’s one likely problem with that approach though. We know there are lots of unreported cases. What we don’t know is how many. For reasons I’ll explain later, it’s likely the actual mortality rate is well under 1%. For now, let’s use 1% and say a worst case is 1m deaths if we leave the disease unchecked.
An Inherent Logical Conflict
Before proceeding to what more favorable outcomes might look like, there’s a fundamental mathematical flaw that needs to be addressed. You can’t assume worst case mortality AND worst case incidence. Let me explain.
China has 80,000 reported cases. How can we have the potential for 100M when they only have 80K? Something is amiss. I know, they’re underreporting cases! OK, but by a factor of 100? or 1000? Unlikely.
So, it seems hard based on other country’s experience to suggest we will have 100M cases or anywhere close to that. Let’s say for argument’s sake China is underreporting by 5x and actually has 400k cases.
Note, this doesn’t change the fact that 3,000 Chinese died. That is a constant. What does change is the mortality rate. Instead of the scary 3-4%, it becomes 0.6-0.8%. It is wrong to say that a) there will be 100M cases AND b) there will still be the same 3-4% mortality rate that we find in smaller sample sizes that are not representative of the actual infected population due to undercounting.
A Possible Best Case
Sometimes the best approach is to simplify. Let’s try this exercise. Instead of trying to guesstimate incidence rates and R0 and mortality by age, let’s just consider actual deaths by countries who have been effected so far.
The three main areas impacted so far are Hubei province in China, Italy, and South Korea. One interesting commonality is the population of each is 50-60M. The highest mortality is in China at close to 3,000 (some Chinese deaths were outside Hubei). Italy is half that for now but likely trending up towards that 3,000. South Korea is at under 100 and it’s cases likely have peaked.
If we normalize for the US having 5X the population of Hubei and Italy, this would suggest we will have 15k deaths. Perhaps this sounds like a lot but the annual flu season takes close to 50k lives each year.
There are reasons to think we will have better results per capita than other countries. For example, we know smoking is a major risk factor for those who get coronavirus. Smoking rates in China are twice the US and Italy is 50% greater. Also, Hubei’s health system was initially overwhelmed and Italy’s remains so.
If South Korea proves the better comparison, we would have easily under 1000 deaths. We are currently having 5-10 deaths/day which is similar to the South Korea experience. By this point in the Italian experience, there were already over 100 dying each day.
More On Estimating Incidence
You may have read about the New York man who has infected 50 others leading to the New Rochelle lockdown. He is a good example of the potency of the virus unchecked. I’m sure there will be other cases similar to him. However, they are likely to be the exception rather than the norm.
From the WHO report on China, patients who were meticulously tracked infected only 1-5% of the people they came in contact with. Now, to be fair, the person who had a 1% transmission rate was connected to 25,000 people! This means this person infected 250 others, but it’s still 1%. If you only have contact with 50 people, you will not infect 50 people.
Another way to look at this is the Diamond Princess (the Japanese cruise ship). This is about as close to a controlled experiment as we can have. There were no people coming in or out of the system and safeguards were relatively poor. Yet, under 20% of the passengers got the virus.
Unfortunately, more people hear stories like the one that there are supposedly 100,000 likely cases in Ohio. We know this is false. Based on the tracking from Wuhan, it took 90 days to reach 100K cases there.
That would suggest the virus first reached Ohio in December which is obviously an error. It doesn’t help curb panic when supposedly authoritative sources like the Ohio Department of Health and the Governor spread bad math.
Reflexivity of Incidence
I’m sure many of you are familiar with this concept of R0 by now. You’ve probably heard how it’s twice as bad for corona as the flu. However, as I wrote about last week, this assumes we do nothing. We can dramatically influence R0 with behavior change. Basic hygiene and distancing will cut spread materially. Shutting down the economy will cut it even further.
There was an interesting paper published recently looking at how changes in R0 impact the spread of the virus. If we practice enough social distancing to reduce transmission 50%, then the # of cases will be reduced by 80% a month from now. If we effectively reduce beginning cases 75%, the # of cases will be 93% lower in a month.
Note, this research ignores impacts from other actions such as closing schools, travel bans, etc. So, do you think we’re 50% more cautious about getting close to others over the last week? I would suggest that’s a layup. The 75% reduction sounds reasonable to me. Now, throw on the other measures to keep people from assembling at all like event cancellations, school closures, etc.
All of a sudden a 95% decline in potential incidence sounds feasible. That 100m worst case we started with might only be 5m. At a 1% mortality, that would result in 50k deaths, aka a normal flu season.
By the way, if you want a really crude check on incidence. Famous people represent 1 in 2000 to 1 in 10,000 Americans based on rough estimates. It is normal that one famous person would have it by now. If next week there are 100 famous people diagnosed, that would suggest we have something like 500K cases even if the reported numbers don’t show that yet.
More on Estimating Severity
Back to the WHO report, 80% of cases are mild and 20% are severe (17% with recovery, 3% fatal). This 80% has been widely cited in the media. However, this assumes incidence is reported properly.
Let’s take my guesstimate from earlier that the real cases in China were 5x what was reported. How does that change the proportion with mild symptoms?
Cases | Mild | Severe | Fatal |
100 | 80 (80%) | 17 (17%) | 3 (3%) |
500 | 480 (96%) | 17 (3.4%) | 3 (0.6%) |
Wow, that’s a quite a difference isn’t it? 96% of people will have mild symptoms once we get more accurate reporting while the mortality rates falls to 0.6%. Things just got a lot less scary.
If this proves out to be the case, it suggests a much different prevention framework than the broad brush of scaring everyone. Instead, we should focus our effort on those who should be fearful: seniors and those with pre-existing co-morbidities.
If you are on a college campus where you are around young healthy people, it would be better if you could stay there. Professors who are in a high risk group should teach their class remotely, but everyone else is better off on campus than going home where they might infect high risk people in their family or at a store or restaurant.
By scaring everyone, we are actually in some cases going to increase risk for high risk people. There needs to be a more nuanced response where those who avoid high risk populations should largely go about their normal activities (with enhance diligence about hygiene). If they get a mild form that only circulates among similar low risk people, they can get immunity and reduce the risk to the larger community in the future.
More on Estimating Mortality
Let’s start by digging deeper into why there is so much discrepancy about mortality rates. There are two main sources of variability: measurement error and different characteristics of each country’s population.
Measurement Error
I’ve already touched on the underreporting of cases suppressing the denominator and thus elevating the reported mortality rate. An interesting representation of this is the German data.
Germany only has a 0.2% mortality rate. Norway and Sweden have similar low rates. Yet, Germany has the 6th most reported cases in the world. What should we conclude from that? Germany has done a far better job counting cases than other countries.
This means Germany is reporting something far closer to an actual mortality rate than a guesstimate mortality rate. There are certainly caveats that may lead their rate higher in the coming weeks (given they are earlier in the process serious cases would still be in the severe category but will eventually die) but it does show that the high mortality rates elsewhere are largely a function of undercounting cases.
Recall the case of the Diamond Princess. This is an easy to measure group because of the isolation. The mortality rate was 1%, even though the population was older and there were limited treatment options. Of the seven deaths, at least five were 75 or older (I can’t find an age on the other two).
And for one more example, the reported mortality rate in China is 4%, but that is heavily influenced by the early experience in Wuhan. From the WHO report, the early mortality rate was 17%! However, patients who developed symptoms in February only had a 0.7% fatality rate. Additionally, Chinese cases outside Wuhan had a similar 0.7% mortality rate.
The overall 4% is strongly influenced by the inadequate early healthcare response. Italy, in addition to its demographic issue (see below), is going through a similar healthcare challenge.
Population Differences
Compare Germany to Italy. The mortality rate is a horrific 7%! Hopefully, much of that is undercounting cases. However, there are conditions specific to Italy that suggest it should have a higher fatality rate. As mentioned above, it has 50% more smokers. It is also the oldest population in Europe with 23% of its residents over 65. A study of 105 deaths showed an average age of 81.
Corona incidence appears to be higher for older populations. The data out of China suggests over half of infections are people 50 or older. This compares to a little over a third of flu patients who are above 50.
In an analysis of ~1000 deaths in China, 81% were over age 60 and 94% were over age 50. People 80 and older had a 15% fatality rate!
Age | Mortality Rate |
<39 | 0.2% |
40-49 | 0.4% |
50-59 | 1.3% |
60-69 | 3.6% |
70-79 | 8.0% |
80+ | 14.8% |
There is an important caveat to this data though! The overall mortality rate in the sample was 2.3%. If my estimate of undercounting incidence by 5X is correct, then each of these #s should be lower (though not by the same amount). Given the undercounting is likely greatest at younger ages, then that 0.2% could possibly be 0.02% while the 80+ is likely still double digit.
The other area of difference by country is the impact from pre-existing conditions. I mentioned smoking differences earlier. Additionally, there are health conditions like heart or respiratory disease. There is obviously a correlation to old age among this population which is unfortunately not separated in the data.
Back to the Chinese data, those with no pre-existing conditions had a fatality rate of 0.9%. This compares to a rate of >5% for those with other underlying health conditions.
Is This Like The Flu?
Yes and no. The mortality rate is far worse. In terms of who is impacted, this is strikingly similar to the flu.
In a typical year, 30M Americans will get the flu and half of those will seek medical treatment. The mortality rate is 0.15% so somewhere between 1/3 and 1/25 of corona.
Roughly 80% of flu deaths happen to seniors and the flu mortality rate for seniors is close to 1%. So you can see the proportion of deaths hitting seniors are similar with the flu and corona. What is different is the mortality rate.
The Hospital Capacity Variable
There has been a lot of discussion about flattening the curve to spread out the impact of the virus. Notice I didn’t say to lower the number of infections. While that may be a byproduct, the real reason officials are so focused on flattening the curve is because of fear of a shortage of hospital beds if the peak is too high.
There is good reason for this. If you look at the poor results in Italy and initially in Wuhan, they were largely related to unpreparedness and a lack of medical capacity. Governments that have been able to promptly treat people have had much lower fatality rates.
The fear is if we did nothing and had that worst case scenario of 100M infections, we would not have enough hospital beds to treat everyone. The most commonly cited numbers are we could handle 300K hospitalizations and only have 100K respirators.
This is probably a little misleading as under a true state of emergency we could mobilize military facilities and take over hotel rooms, but let’s ignore that for now. One thing to remember is all 300K people would need to be sick at once. In a typical flu season, 500K are hospitalized and we have had years above 800K. Because it was spread out, it wasn’t a problem. Thus, “flatten the curve”.
So, we would need to have a call it one week period of 300K severe cases. By my earlier math, it would take 9M cases (at a 3.4% severity rate) to produce 300K hospitalizations. Note, that’s not 9M cases in total over the entire episode, but a peak of 9M simultaneous cases.
Could that happen? Yes, it’s not impossible. Is it likely based on everything I described about the compounding of actions to reduce transmission? No, it’s not likely. That said, if you’re a public health official, making sure we keep the peak below the maximum hospital capacity is obviously your biggest priority.
Random Other Thoughts
Human beings are funny animals. We are famously irrational. Your risk of dying from operating a vehicle (35K annually) is similar to or higher than that of dying from corona. Yet, we fear coronavirus and don’t think twice about getting behind the wheel. Even funnier, we fear self driving cars that could reduce those 35K death rates meaningfully.
What’s the difference? We have control over driving a car (though not over the other driver). We don’t have control over the virus (although we can by washing our hands). We’d apparently rather die controlling our fate rather than powerlessly, even if the odds are the same.
We don’t think enough about secondary impacts. I can all but guarantee we’ll see a rise in suicides if we have a recession. Those deaths count too. Financial stress also raises the risk of heart attacks, immune diseases, and other leading causes of death. They count too.
If you work at the CDC, you measure victory only by reducing the virus deaths. You unfortunately don’t think to count the others. I’d also suggest the political motivations are askew here. If you scare everyone that millions will die and only thousands do, you are a hero! You stopped a catastrophe. There is an incentive to promote the worst case even if it has dire economic consequences, particularly for people at the low end of the ladder.
If you do go out to shop or eat, or even if you order delivery, try to support the people who are barely hanging on financially. Give a little extra tip to the delivery driver, order from the family owned restaurant not the chain, go to a movie (it will be empty enough you can sit six feet away from anyone else). If someone you know gets the virus, send them flowers from the local florist. Small things can make a difference between someone keeping their business and losing it.
Prediction Time
I have a history here of making predictions so I’m not going to be cowardly and refuse to do one now. I’m going to go out on a limb and say US deaths will be under 10,000, perhaps significantly less.
I’ll also suggest the mortality rate will be less than 0.5%. This implies as much as 2M cases (actual, not reported). Remember, 95% of these will be mild. This would require fewer than 100K hospitalizations. We have more than enough capacity to handle that.
If I’m wrong and we have 100K deaths or more, I’m sure some of you will remind me. I’m not saying it can’t happen. I think the actions to essentially barricade the economy make it pretty unlikely though.
Oh, and if anyone has other data sources they think I should incorporate, please feel free to send them along.
Wash your hands regularly, practice safe spacing, but also try to find ways to support your small businesses. They are in crisis too.
Thanks for sharing these thoughts Ian.
Would love to have a post on update to GDP – which is where an event risk (Coronavrirus) spills into a Cyclical downturn because of store closures, hit to incomes & spending, etc.
Yeah, first step was quantifying risk, next is evaluating whether cost to mitigate risk is worth it and what can we do to mitigate more efficiently than blanket measures.
Ian, excellent analysis. thanks for the hard work.