My knowledge of epidemiology is mediocre at best, but health economist Robin Hanson is the smartest person I know. Even when he’s totally wrong, he improves our thinking. And even when he seems totally wrong, he is often right. In the face of great outrage, Robin here suggested that deliberately exposing some people to the coronavirus could save lives. Then he produced some simulations to confirm his intuition. In his latest post, he refines his simulation to show that under his assumptions, deliberately exposing the young has enormous health benefits:
I’ve changed my prior spreadsheet model so that the population is split into two groups who differ in their death rate, with the deliberately exposed taken only from one of those groups. Combining estimated COVID19 death rates by age with US population stats, I find that about half of the population is over 40, and has a COVID19 death rate about 23 times larger than the under 40 half.
So I compared four options regarding who is deliberately exposed: no one (baseline), random folks, folks 40 and over, or folks under 40. I assumed that just enough are exposed per day to fill a quarantine that holds 10M, and I varied the number of days of deliberate exposure within each option to min deaths in that option. (I also assume that higher death rates induce higher needed ICU days.)
Below I show graphs of contagious, deaths, and dead % of sick for the first 250 days of each scenario. In the baseline 14.3M die, while if random folks are exposed for 31 days, 11.2M die. If only the old are exposed for 15 days, 12.9M die, while if only the young are exposed for 54 days 8.1M die.
As the death of a young person is more of a tragedy due to their having more years left to live, we might want to adjust for this remaining-life-years-destroyed effect. Doing so, I find that when only the old are exposed for 15 days, there are 4.22M adjusted deaths, while when only the young are exposed for 54 days there are 2.78M adjusted deaths. That’s only 20% of the deaths in the baseline, and I have not yet searched for an optimal age cutoff, optimal cutoffs using gender and co-morbidities, an optimal schedule of exposures per day, or an optimal quarantine capacity!
Thus exposing the young seems better than exposing the old, which are both much better than exposing random people or no one. As I’ve said, this process rewards variance!
I urge people who know more than I do to ponder Robin’s thoughts closely and calmly.
READER COMMENTS
Christophe Biocca
Mar 16 2020 at 11:41am
I’m not expert but:
Is either an error or very misleading. His spreadsheet computes the adjusted deaths for the baseline and Expose All scenarios as 100% of the baseline deaths (so they’re effectively unadjusted). So he’s comparing adjusted to unadjusted in getting this 80% reduction (the F column even uses D$3 as the denominator).
So really it’s a combination of 43% reduction due to the strategy, and a 65.6% due to counting differently. He needs to compute QALY’s for the baseline scenario to get an apples-to-apples comparison.
Jonathan Monroe
Mar 16 2020 at 12:34pm
There are two more fundamental problems with Robin Hanson’s model:
The entire analysis assumes that the virus cannot be contained and that therefore the inevitable end state is infection of sufficiently many people to achieve herd immunity (and the number required to do this is a constant). So every intentional early infection means one fewer case later. Whether this assumption is true or not is the $64,000 question of COVID-19 policy – the various Asian countries that are successfully containing the virus through contact tracing and social distancing (China outside Hubei province, Taiwan, Hong Kong, Singapore, South Korea and Thailand as a minimum list) are making a trillion dollar bet that it isn’t, and they aren’t idiots. Hanson’s proposal reduces the death toll from 14 million to 11 million. Effective containment (if possible) reduces it to a few thousand.
The model assumes that the deliberate infection policy begins when the first case arrives in the US (not even when the first case is detected), and everyone who is going to be deliberately infected has been deliberately infected during the first 30 days, by which time there have only been 54 symptomatic non-deliberate cases. In fact, there is no way you can know a novel infection will be impossible to contain at the stage where there are only 54 symptomatic cases.
Jonathan Monroe
Mar 16 2020 at 12:48pm
Having tweaked Robin Hanson’s model it turns out that deliberate infection still works even if you don’t start until there are 100,000 symptomatic cases, at which point there is enough information to make decisions – and by which point China had already contained the epidemic and was well on the way to eliminating the disease (most COVID19 cases in China are now reimportations).
Robin Hanson
Mar 16 2020 at 4:10pm
Yes, it can help even if you start later. And it should be easy to factor in some estimated chance per day that a vaccine shows up and ends the model relevance.
Mark Bahner
Mar 16 2020 at 6:14pm
But it’s not only “a vaccine” that ends the model relevance, right? The model relevance can also be ended if contact tracing, and social distancing, and other preventative measures reduce the rate of infection from each newly infected person to less than one, with only a small fraction of the population infected. Correct?
South Korea…lots of testing limits spread
Mark Bahner
Mar 16 2020 at 5:55pm
Yes, either the three people who offered David Henderson bets that total deaths in the U.S. in 2020 from COVID-19 will be less than 100,000 people are way off, or Robin Hanson’s model assumption about the end infection rate seems way off.
Robin Hanson
Mar 16 2020 at 4:02pm
Adjustments are only needed for subgroups whose life years differ from the average. So no adjustments are needed in the Baseline or ExposeAll options.
Christophe Biocca
Mar 16 2020 at 6:53pm
Ok, that explanation makes sense, but the fact that you’re differentiating between two populations, where the people who will die more frequently count less, is doing a lot of the heavy lifting.
I made a copy of your spreadsheet and hard-set both “Deliberates per day” and “End Date Deliberates” to 0 in the “Expose Old” and “Expose Young” scenarios, and I’m still getting death reductions of 68% and 66% respectively. These are reductions entirely due to counting more accurately (since no one is being infected) I may be doing something wrong, but try it yourself and see?
Robin Hanson
Mar 17 2020 at 10:00am
I found an error in the spreadsheet. Try it again with the fixed version: http://hanson.gmu.edu/ExposeTheYoung.xlsx
Christophe Biocca
Mar 17 2020 at 12:28pm
This time I’m setting “Days Expose” and “Deliberates/Day” to zero for all scenarios on the summary page, and:
The total deaths are roughly the same for everyone (there’s a 9% reduction in total deaths for the Expose Old scenario, which doesn’t make much sense, maybe another bug? Or maybe because quarantine is still kicking in for those that naturally catch the disease).
But there’s still a 57% and 53% reduction in the “adjusted” numbers.
I assume issue 1. is some kind of implementation problem, but 2. seems inherent in the separation of the population into parts, because of the correlation between “likely to die” and “was going to die soon anyways”. If this were a disease that primarily killed the young, you’d see the opposite effect, where separating the population into slices would result in higher adjusted deaths than the baseline.
You can see the measurement effect (after zeroing the deliberate infections) by changing “Index” and watching the adjusted deaths change: in the absence of any deliberate infection there’s no behavior change driven by “Index”, so the only way it can change the numbers is by affecting the measurement.
Mark Bahner
Mar 16 2020 at 12:45pm
I don’t know much about this subject, but I do like cool ideas! 🙂
So…how about this one: Gather a group of volunteer, healthy 18-25 year-olds. I suggest university students. (But again, only volunteers.)
Expose them to COVID-19 virus. When they get well, draw blood to get their antibodies. See if the antibodies are useful in treating COVID-19:
Antibody therapy…bring back the 19th century
Mark Bahner
Mar 16 2020 at 1:10pm
Ooh! Ooh! An even better idea than volunteer university students (at least to start the program):
Volunteer (unmarried) members of the military.
Or volunteer (unmarried) students at the U.S. military academies.
Matthias Görgens
Mar 16 2020 at 11:38pm
That’s an interesting idea, but the coupling with the deliberate infection idea is not necessary: the infection produces plenty of recovered cases all on its own.
They are the vast majority of total cases in China these days.
You’ll find some volunteers to draw blood from amongst them.
Mark Bahner
Mar 17 2020 at 11:45am
Further thinking about the situation leads me to realize that, even if COVID-19 “only” killed one out of 1000 healthy young volunteers, that would still mean the program would be (probably properly) be viewed as actually causing the deaths of some healthy young people.
What I was trying to accomplish was to “kill two birds with one stone”…to implement Robin Hanson’s idea and to also gain the benefit of harvesting antibodies (from healthy young people).
But I’m having second thoughts about Robin Hanson’s idea, because I think the best way to reduce deaths, at least over the next year or so, is to identify COVID-19 cases, and make sure that on average each new case generates less than one additional case. I think that both China and South Korea have probably already accomplished the goal of having each new case generating less than one additional case.
Ray
Mar 16 2020 at 2:11pm
I suspect we’re going to get a natural version of this experiment anyways – older people will be mostly in isolation, younger people will be in the far less strict “social distancing” (at least, those younger people not living with their elders). If that can be made to hold, we can gradually expose younger groups to it at a rate that won’t overwhelm the system, assuming the system isn’t overwhelmed already.
Andre
Mar 16 2020 at 7:31pm
I joked last weekend that rather than close schools, that we should lock all kids in them for two weeks, with a few infected adults, a few medical personnel, blankets, lollipops, and games.
Generation inoculated, and the elderly much less vulnerable.
robc
Mar 17 2020 at 6:32pm
Joke?, sounds like a plan.
Andre
Mar 17 2020 at 10:36pm
I agree, but it is against the DNA of the left.
Christian Moon
Mar 17 2020 at 1:58am
In the UK the NHS puts a value on a quality adjusted life year (QALY) of somewhere up to £30,000 as a threshold for cost-effective medical treatment.
A GDP cost of 10% in lost output and wealth destruction as a result of COVID countermeasures would comfortably exceed the cost of providing 1% of the population with 7 years of QALYs.
For obvious reasons this is not being openly debated as a matter of intergenerational equity, but in view of the age profile of the mortality, inevitably that’s what it is.
Just how much should the young be sacrificing in order to extend the lives of the old and infirm? Just how much should we seniors want to take from our children and grand-children?
Comments are closed.