As you know, I am not a fan of models, such as the one that Tyler Cowen points to. I prefer an Ed Leamer “patterns and stories” approach.
One pattern, which Tyler agrees is significant, is that the virus is producing fatalities very unevenly across regions. For example, New York has 552 deaths per million, and Texas has 11. If you believe models, then you probably believe that Texas will catch up at some point. But I can imagine a story in which that does not happen.
Here are what I think are the causal factors for the differences.
1. Living conditions of the elderly. Consider Spain, which appears to have the highest per capita death rate of any major country (although Belgium might catch up). Wikipedia says,
Many nursing homes in Spain are understaffed because they are for-profit businesses and elderly Spaniards cannot necessarily afford sufficient care. In some nursing homes, elderly victims were found abandoned in their beds by Spanish soldiers mounting emergency response. Defense minister Margarita Robles said that anyone guilty of neglect will be prosecuted. By 23 March, a fifth of nursing homes in the Madrid area had reported cases of the virus. More than 65% of fatalities have occurred in those 80 or older, compared to 50% in Italy and only 15% in China. By 3 April, at least 3,500 Spaniards had died in nursing homes and another 6,500 contracted infections there. Thousands of elder care workers are also infected.
2. Concentrated poverty. Actually, it is not dire poverty. It is working conditions and living conditions that make it hard to socially distance.
Low income neighbourhoods in Barcelona have seven times the rate of infection of more affluent neighbourhoods. Part of the reason is that essential workers, who have kept going to work despite the epidemic, are likely to work in low-skill jobs such as supermarkets or elder care. Also, many care workers are immigrants, who lack access to unemployment benefits and live in some of the lowest category housing. Homeless people are also at risk and the charities that help them were forced to cease operations because of the disease.
3. Enclosed spaces where one might come into contact with someone with a high viral load.
I think that enclosure matters. I think some scientific papers suggest this, although I don’t have any links right now.
Actually, someone just sent me a link to Wendell Cox on exposure density which says
Exposure density, and thus infection, is likely to be less, all else equal, if common halls, elevators, crowded places and transportation facilities are avoided.
Cox says that exposure density is the likely explanation for why deaths have been much higher in New York City than elsewhere.
I am trying to come up with an explanation for why the virus has not killed thousands of homeless people in LA and SF. I think that the most plausible reason is that they live outdoors, and so do not inhale the large viral loads that you can inhale in a confined space. That is just a guess.
Note that if enclosure matters, then maybe health care workers would be safer treating patients in the parking lot than inside the hospital building.
So here are some hypotheses.
a) We will find that New York has some nursing homes that are run appallingly badly, and these will account for a significant share of the deaths there. Going forward, spikes in deaths in other locations will be tied to nursing homes that fail to maintain proper hygiene and to keep sick workers away.
b) The elderly and the very obese are vulnerable to fatal infection even from asymptomatic or mildly symptomatic spreaders. But the rest of us will have low infection fatality rates unless exposed to a highly symptomatic individual in a confined space.
c) Someone who is very symptomatic and gets into a crowded space, such as a bar or subway, can kill many people.
If these hypotheses are correct, then I believe that we can avoid a repeat of the New York fiasco with carefully targeted public health interventions and not require general lockdowns.
Keep in mind that I am not an expert and I have an aversion to lockdowns, so there are some obvious biases at work.
I came to a very similar conclusion, including a quick and easy to measure just how crowded and stagnant the air you’re breathing is… click through to my post.
I wish you were right about needing symptoms to be a super-spreader. But the choir practice in Skagit County Washington suggests otherwise. This is what makes this disease so problematic.
It is claimed that no one at the choir practice had symptoms. If the claim is correct, then this a very important counterpoint. But perhaps the claim is not correct.
Probably isn’t correct- who would admit to being the person knowingly responsible for going there with symptoms? I strongly suspect this is the explanation for many of the claims of ‘asymptomatic’ spread- “Not my fault, I didn’t know at the time I was sick!”
Well, the LA Times interviewed 8 people at the rehearsal, and they were unanimous in claiming that no one was coughing or sneezing or appearing ill. So, maybe somebody was able to fake wellness for 2 1/2 hours and they were actually at death’s door. Or possibly there is asymptomatic spreading of this thing.
Obviously no one knows the true answer at this point. But it seems to me that the best evidence we have points to asymptomatic spreading.
Lol! You don’t have to be at “Death’s Door” to have symptoms. You don’t even have to have a cough or sneeze all the time either.
If you want to show me the interviews with everyone who was there and their denials about being sick before the practice, that might be actual evidence, but interviews with 8 people about not hearing any coughs or sneezes is extremely unconvincing.
Re 3.: Japanese TV has had this guy, director (I think) of a 3rd-world/tropical infectious disease clinic in Tokyo, repeating advice to avoid places where the three 密 – 密閉 密集 密接 (tightly enclosed, people assembling close together, people in close contact) – occur together, for a month now.
“The elderly and the very obese are vulnerable to fatal infection even from asymptomatic or mildly symptomatic spreaders.”
Not sure if I can draw a firm conclusion, but looking at the NYC death statistics there https://www1.nyc.gov/site/doh/covid/covid-19-data.page
It looks to me like Age >65 also isn’t a mortality risk factor when there are no underlying conditions (not sure how common that is though )
pct of deaths with NO underlying conditions by age group as of April 14:
0 to 17 (0%)
18 to 44 (9.1%)
45 to 64 (4.2%)
64 to 74 (2.1%)
75 and over (1.1%)
This is a question I’ve had from the beginning and still can’t quite get a handle on – controlling for underlying health problems, what’s the remaining effect of age.
Perhaps the effect of age independent of health really is large. But it would’ve be surprising to find that an otherwise healthy 30 yo and an otherwise healthy 70 yo have roughly the same mortality. It’s just that the percentage of 70 yos who are otherwise healthy is so much smaller.
I wish someone would just publish the 2D table of death rates with age and health on the two axes.
FWIW- and I’ve looked at the data pretty extensively on the health care claims side – the age where age itself begins to predict one-year mortality is about 84. Meaning that before that age, actuarial risk is predominantly about comorbidities. After that the annual increase in mortality risk gets steep, because shocker: human beings are mortal. At 93, one-year risk of mortality is above 25% with age as the only predictive variable, and you can imagine it gets worse from there as you add cancer, heart disease, and everything else to the mix (to say nothing of an additional year on the tires.)
Interestingly, many medical studies show that age is not predictive of bad outcome or any decline in outcome when you look at interventions like chemotherapy or surgery. Doctors still undertreat the elderly by this metric, thinking the elderly will do worse. It’s frailty that predicts badness and I suspect that’s very true for COVID-19 also given the stories gathering about nursing homes and assisted living facilities.
The trouble is, only 137/6840 (2%) had “no underlying conditions”, but 1552 (11 times more – 23%), had “underlying conditions unknown”.
It’s impossible to do any meaningful analysis on the NUCs when those numbers are just a floor with a ceiling of adding in the UCUs that is literally an order of magnitude higher. My hunch, consistent with some small studies, is that most UCUs are indeed UCs, but we can’t tell just by looking at the NYC numbers.
The uncertainly in UCs is less, because UCU is a lot smaller, but even there, the potential error is up to 30%:
100% of those under 18 had UCs.
Between 79% and 92% of those from 18 to 44 had UCs.
Between 85% and 96% of those from 45 to 64 had UCs.
Between 76% and 98% of those from 65 to 74 had UCs.
Between 70% and 99% of those over 75 had UCs.
It looks like UCs are really important, though it would be nice to be able to compare to the prevalence of those conditions in the general population.
Compare Christopher M. Petrilli et al. (NYU), “Factors associated with hospitalization & critical illness among 4,103 patients with Covid-19 in NYC” — There are some surprises:
https://www.medrxiv.org/content/10.1101/2020.04.08.20057794v1.full.pdf
“Overall, we find that age and co-morbidities are powerful predictors of requiring hospitalization rather than outpatient care; however, degree of oxygen impairment and markers of inflammation are strongest predictors of poor outcomes during hospitalization. […]
In this report, we describe characteristics of 4,103 patients with laboratory-confirmed Covid-19 disease in New York City, of whom 1,999 required hospital admission and 650 required intensive care, mechanical ventilation, were discharged to hospice and/or died. We find particularly strong associations of older age, obesity, heart failure and chronic kidney disease with hospitalization risk, with much less influence of race, smoking status, chronic pulmonary disease and other forms of heart disease. Moreover, we also noted the importance of hypoxia despite supplemental oxygen and early elevations in inflammatory markers (especially d-dimer and c-reactive protein) in distinguishing among patients who go on to develop critical illness and those who do not. In the hospitalized population, measures of inflammation were much more important than demographic characteristics and comorbidities. […]
The risk factors we identified for hospitalization in Covid-19 are largely similar to those associated with any type of severe disease requiring hospitalization or ICU level care, though we were surprised that cancer and chronic pulmonary disease did not feature more prominently in the risk models.26 Moreover, the demographic distribution of hospitalized patients is also similar to other acute respiratory infections. For instance, while advanced age was by far the most important predictor of hospitalization and an important predictor of severe outcomes (as it is for most illnesses), 54% of hospitalized patients were younger than 65 years. This is typical of the hospitalization pattern in viral respiratory disease. Studies of influenza hospitalizations in the United States have found that people younger than 65 years account for 53-57% of influenza-related hospitalizations.27,28 While men made up a grossly disproportionate number of both hospitalizations and critical illness, this difference was attenuated by multivariable adjustment for co-morbidities such that gender was no longer one of the most prominent risk variables.
Surprisingly, though some have speculated that high rates of smoking in China explained some of the morbidity in those patients, we did not find smoking status to be associated with increased risk of hospitalization or critical illness. This is consistent with a handful of other studies that have previously shown a lack of association of smoking with pulmonary disease associated ARDS (i.e. from pneumonia), as compared with non-pulmonary sepsis-associated ARDS.
More striking were our findings about the importance of inflammatory markers in distinguishing future critical from non-critical illness. […]
[…] it is notable that the chronic condition with the strongest association with critical illness was obesity, with a substantially higher odds ratio than any cardiovascular or pulmonary disease. Obesity is well-recognized to be a pro-inflammatory condition.”
Perhaps an implication is that carefully targeted public-health interventions should attempt to shield or isolate individuals who are most vulnerable, per findings like these?
Wouldn’t this have been mostly obvious a month ago? Heck, wouldn’t this be common knowledge in 0 AD?
Population density/cities have always been associated with high infectious disease fatality and were generally population sinks. The poor living on top of each other usually got it worst.
Ironically, things everyone seems to hate (low density, driving a car, single family detached houses, absence of impoverished minorities, not shoving your parents in a home) proved the best ways to protect oneself.
Having semi-invalid grandparents and having had occasion to visit a Japanese old folks home, I can sympathize with both sides of the old folks facility debate. Having them live in with you can result in constant getting on one another’s nerves up to and including brawls and murderous intent (promoters of traditional living arrangements prefer not to mention this part), but avoids shoving them in an old folks home which is (perceived as) unkind/unfilial and is apt to get them killed as we see now. So one compromise is to let them live in separate dwellings/apartments close by, ideally within walking distance. On the other hand, if they can’t move around (too weak or too addled) then having no companionship at all and nothing to do is pretty hard, like solitary confinement except you’re not in prison, and that’s what the better kind of old folks homes sort of avoid. That’s one reason to keep in shape, so that this stage of one’s life is as short as possible or non-existent, and one reason I am not averse to building codes etc. mandating reasonable accomodation for wheelchairs and other low-mobility people.
My parents moved out of a nursing home (sort of, its for a bit more active then the invalid) just before this got going.
As of now they aren’t allowed out of their rooms and someone drops food in front of their tiny apartments three times a day. They are supposed to leave the used dishes outside the door. It may be necessary, but it sounds hellish.
3. Micro-droplet dispersion might account for both the correlation with enclosure and density. Nursing homes could do much worse than to occasionally open a window. The following short video on Japanese experiments. convinced me that I will be quarantining until a reliable vaccine is made available and I suspect it may convince other decrepit geezers to do the same: https://www.michaelsmithnews.com/2020/04/japanese-research-video-shows-how-long-micro-droplets-from-people-just-talking-linger-in-the-air.html
There was enormous regional and local variability in the morbidity and mortality of the 1918-19 flu pandemic. In many cases reasons for the variability are apparent, but in many cases not. Notwithstanding the many thousands of epidemiological studies of that pandemic that have been done, there is still no accepted general theory that fully accounts for why it was so much worse in some places than others. No reason yet to think we will do better this time around. For information on the 1918 pandemic, a good place to start is Crosby’s America’s Forgotten Pandemic: The Influenza of 1918.
I would guess that the prevalence and density of mass transit is also a significant factor. NY City has the subway, Texas does not. Also note that the incidence of infection in London and its suburbs is looking like it is high on a relative basis. London has The Tube.
Yeah, the subway seems like a very important factor. Apparently the NYC subway handles an order of magnitude more traffic than the next biggest US subway (Washington DC, I believe).
It’s even worse – the NYC Subway was especially bad for this in lots of ways, including:
1. The trains tend to travel quite slowly, and lots of people taking very long rides, both on the subway system, and on connecting systems (e.g., LIRR)
2. MTA responded to major reduction in traffic and fares by cutting service to the bone, which means efficient use of trains, and by efficient I mean “crowded”, with people stuck waiting together on platforms for long periods too.
3. A bunch of homeless folks hang out in the system all the time, and the social media take is that they are still there. Anyone who rides crowded trains all day is in a class beyond super-spreader, maybe ‘ultra-spreader’.
4. The culture and infrastructure for turnstile gatekeeping has decayed under the break-the-windows policy which ignores fare jumpers, which means it did not occur to people to have people standing there checking for temperature, requiring some kind of facemask.
5. I haven’t used it in about two years, but that was fairly recent, and plenty of locations were frankly filthy will bad odors and poor ventilation. Perhaps it’s a lot better now, but i remember smartphone service being choppy, which makes track and trace – already hard with mass transit – much more difficult.
6. There are no good alternatives to get around – even for obviously very sick people trying to get themselves to the hospital, and many people couldn’t switch to private vehicles, and even taxis or uber were of limited number and use, especially since some roads were closed.
So NYC is special, the city equivalent of one of those extra-social super-spreaders points in the social network. Just like our overall picture of what will happen to people in general is distorted by learning what happens to the first people to get sick – which are disproportionately super-spreader folks, our estimates of what will happen to the rest of the country are distorted by NYC getting his hardest, soonest, but in a way that won’t happen anywhere else.
Everybody wants information as early as possible. But if early people and places are special, then early data is still no good and misleading. You will be wrong to extrapolate from it, and decisions based on those extrapolations will deviate from the proper policies.
Unless you are doing big studies early, you would still have to wait until the early special cases are out of the way. If you are going to be passive and wait until actually good data becomes available, it will be too late to make any good use of it.
If NYC’s subway is the crucial factor, then we should expect NYC (and its commuter area) to have a *much* higher death rate than, say, Boston and its commuter area. (Ridership in the NYC subway is an order of magnitude greater than in the Boston T.)
Let’s check the relevant projected death rates for the first wave of the pandemic. IHME provides State-level data and projections, rather than metro-area level. IHME projects that the first wave will end three weeks later in MA than in NY. Take with a large grain of salt any projections. IHME models have earned a lot of criticism. With these caveats, here are the April 13 IHME projections for cumulative deaths in the first wave of the pandemic, March through May, for NY and MA:
https://covid19.healthdata.org/united-states-of-america/new-york
https://covid19.healthdata.org/united-states-of-america/massachusetts
Projection of cumulative Covid-19 deaths in NY: 14,542
Projection of cumulative Covid-19 deaths in MA: 8,219
Now, let’s divide each projection by roughly the relevant populations:
Populations of NYC + Westchester County + eastern Long Island = 12.4M
Population of MA = 6.9M
Death rate in NYC+W+LI = 14,452/12,400,000 = 0.12%
Death rate in MA = 8,219/6,900,000 = 0.12%
(Yes, each is 0.12%. Quite a coincidence.)
If IHME figures are in the ballpark, and if the subway is a crucial factor in NYC, then there must be other heavy equalizers in MA.
IHME was supposed to update its projections today (April 15), but didn’t. I will update these figures if IHME’s forthcoming projections change a lot.
An epidemiologist, although not one that creates forecast models, claims that IHME model is not a standard/good epidemiological model for forecasting the spread of the virus or deaths from the disease it causes.
IHME belatedly revised its models and updated its predictions today (April 17). The revised forecasts thru May are as follows, if assign NY deaths to NYC+Westchester+Eastern Long Island:
Death rate in NYC+W+LI = 21,812/12,400,000 = 0.18%
Death rate in MA = 3,236/6,900,000 = 0.05%
In other words, in four days, IHME changed its projections for the death rates per capita in these two regions from roughly equal to each other, to NY metro having a projected death rate 3.5 times MA’s projected death rate.
Arnold Kling told me to ignore computer models—I should have listened!
As Arnold Kling has noted in a later update, a careful study by Jeffrey Harris (MIT) provides systematic evidence for the hypothesis that, “The subways seeded the massive coronavirus epidemic in NYC.” I paste the article’s abstract here for the record (since I had raised questions here):
“New York City’s multitentacled subway system was a major disseminator – if not the principal transmission vehicle – of coronavirus infection during the initial takeoff of the massive epidemic that became evident throughout the city during March 2020. The near shutoff of subway ridership in Manhattan – down by over 90 percent at the end of March – correlates strongly with the substantial increase in the doubling time of new cases in this borough. Maps of subway station turnstile entries, superimposed upon zip code-level maps of reported coronavirus incidence, are strongly consistent with subway facilitated disease propagation. Local train lines appear to have a higher propensity to transmit infection than express lines. Reciprocal seeding of infection appears to be the best explanation for the emergence of a single hotspot in Midtown West in Manhattan. Bus hubs may have served as secondary transmission routes out to the periphery of the city.”
An interesting rebuttal to that study:
Automobiles Seeded the Massive Coronavirus Epidemic in New York City – Market Urbanism
Another rebuttal to the study by Harris:
Alon, “The Subway is Probably not Why New York is a Disaster Zone” (15 April 2020)
https://pedestrianobservations.com/2020/04/15/the-subway-is-probably-not-why-new-york-is-a-disaster-zone/
“1. Manhattan is the highest-income borough, with many people who can work from home. If they’re not getting infected, it could be from not commuting as much, but just as well from not getting the virus at work as much.
2. The Manhattan subway stops are often job centers, so the decline in ridership there reflects a citywide decline. A Manhattanite who stops taking the subway is seen as two fewer turnstile entries in Manhattan, whereas a New Yorker from the rest of the city who does the same is likely to be seen as one fewer Outer Borough entry and one fewer Manhattan entry.
3. Many Manhattanites left the city to shelter elsewhere, as seen in trash collection data.
4. Manhattan’s per capita subway usage is probably higher than that of the rest of the city counting discretionary trips, so 65% off the usual ridership in Manhattan may still be higher per capita than 56% off in Brooklyn or 47% in Queens. (But this is false on the level of commuting, where Manhattan, the Bronx, and Brooklyn all have 60% mode share.)”
Anecdote re 1. There have been 44 deaths and 100 tested positive for COVID at the Holyoke Soldiers Home in Holyoke, Massachusetts. There are vague stories that the place was poorly managed.
Regarding the homeless, in Boston one shelter last week tested everybody there, and found 36% inefected—all asymptomatic.
https://www.wbur.org/commonhealth/2020/04/14/coronavirus-boston-homeless-testing
I understand your objections to or criticisms of specific models but I just don’t understand why you describe yourself as “not a fan of models”.
You prefer “patterns and stories”.
I think I understand what you’re hinting at, and I’m sympathetic, but I’m still confused.
I’m confused because it sure seems like you have a model:
That sure seems like a model, i.e. if the model’s assumptions are true (and other not-explicitly stated assumptions are false), then the model’s conclusions/implications/output should be (roughly) true.
Tyler Cowen linked to this response by an epidemiologist to his (Cowen’s) posted complaints and questions about epidemiology and epidemiologists. The epidemiologist also posted a follow-up to their first post.
Some interesting quotes from the first post:
I define “model” in this context narrowly as a set of equations. If you define “model” to include any form analysis that tried to explain things, then everything is a model.
Fair enough – in my ‘circles’, a model is “any form [of] analysis that tried to explain things”.
I also don’t think that macroeconomic ‘equilibrium’ models are particularly good, but other models of the same basic structure, i.e. (approximate) solutions to some set of equations, seem perfectly valid in other (harder) sciences.
If your model (mathematical or otherwise) assumes that all places follow the same trajectory, then yes you’d assume Texas would catch up to New York. But, if you assume that actions (or locations) can alter the trajectory, then different places might follow different trajectories.
For one, contact density varies by location. Modelers (and the press) like to give Ro which by definition is a function of a virus’s infectivity, contact density and duration of infectivity. Since contact density varies by location, Ro for any virus is not a constant but varies.
Also, New York appears to have been further along the infection trajectory when we all learned about this thing. I assume awareness is the single biggest factory in changing the infection rate. New York got farther along the curve before anyone was aware and hence before anyone started to change their behavior. It is unlikely that we will return to the pre-awareness infection rates even after activity restrictions are officially lifted. (We all wash our hands ten times a day now.) The big question is will the post restriction level of infections be at a level our medical infrastructure can handle. That is a huge unknown. But, Auburn, AL (where I live) will probably still have a lower rate of infection than NYC as our proximity factors are greater: no crowded subways in Auburn.
Add weather to the factors? Deaths / 100K residents don’t seem as bad for most southern states. Hot and humid weather is known to slow down other coronaviruses, although it may not stop this one completely.
I think the lower class working populations are right. The numbers are much worse in Bronx and Brooklyn than Manhattan, even though Manhattan is denser.
Also, enclosed spaces could be important, but also homeless tend not to be obese.
+1
Latest info from NYC.
I think the co-morbidity info is important, to identify the Most At Risk folk. Also think that this part of the NYC data will be similar to other data, with differences more in measuring than in essence.
Be healthy – and not obese. Age is the biggest factor, tho perhaps because so many elderly have one or more other high risk Underlying Conditions.
https://www.medrxiv.org/content/10.1101/2020.04.08.20057794v1.full.pdf
Results: Among 4,103 Covid-19 patients, 1,999 (48.7%) were hospitalized, of whom 981/1,999 (49.1%) have been discharged, and 292/1,999 (14.6%) have died or been discharged to hospice. Of 445 patients requiring mechanical ventilation, 162/445 (36.4%) have died. Strongest hospitalization risks were age
≥
75 years (OR 66.8, 95% CI, 44.7-102.6), age 65-74 (OR 10.9, 95% CI, 8.35-14.34), BMI>40 (OR 6.2, 95% CI, 4.2-9.3), and heart failure (OR 4.3 95% CI, 1.911.2). Strongest critical illness risks were admission oxygen saturation 2500 (OR 6.9, 95% CI, 3.2-15.2), ferritin >2500 (OR 6.9, 95% CI, 3.2-15.2), and C-reactive protein (CRP) >200 (OR 5.78, 95% CI, 2.6-13.8). In the decision tree for admission, the most important features were age >65 and obesity; for critical illness, the most important was SpO20.5, troponin 64 and CRP>200.
Conclusions: Age and comorbidities are powerful predictors of hospitalization; however, admission oxygen impairment and markers of inflammation are most strongly associated with critical illness.