I have been looking at graphs showing new cases and deaths for Santa Clara Country, where I live, and the contrast between the two patterns is striking. Judged by new cases, things started to get worse around the beginning of June, and rose pretty steadily thereafter. The peak was June 23rd, at a level about ten times what it had been a month before.
The pattern of deaths is entirely different. It peaked in mid-April at six deaths a day, fell by May 8th to one death a day, and has been averaging a little less than one death a day ever since. There is no sign that the increase in cases led to an increase in deaths.
One would expect deaths to lag detected cases by two or three weeks, but the increase started in early June and, as of July 5th, a month later, there is no visible increase in the death rate.
Two possible explanations occur to me. One is that increased testing has resulted in detecting many more of the mild cases, in which case the apparent increase in cases is an illusion. The other is that, at this point, the most vulnerable people either are being very careful, as I am — my family has been self-quarantining since mid-March — or have already gotten the disease and either died or recovered. The people who are getting it now are younger, healthier, and much less likely to die of it.
Looking at death rates for the U.S. as a whole, I see a similar pattern, minus the initial high, presumably because it was smeared out by happening at different times in different states. Death rates peaked on April 21, have been roughly constant ever since. The pattern of cases is again similar, but not identical, to that in Santa Clara Country, rising until the beginning of April, gradually declining until early June, then rising almost three-fold by the end of the month.
U.S. Cases
U.S. Deaths
The other striking thing, of course, is that this pattern doesn't seem to be mentioned in the media. There is a lot of talk about increasing number of cases, not accompanied, at least in what I have seen, by any mention of the fact that the number of deaths is not increasing.
Part of the reason may be that the pattern is less striking in the case of the U.S. as a whole, where effects are smeared out by different timing in different states. Allowing for about a three week lag between detection of a case and a resulting death, the effect of the increase in U.S. cases should be only starting to show up in the death rate. It's much clearer in the data from Santa Clara County.
The other part of the reason, I suspect, is bias. There is a lot of pressure to persuade people that Covid is a serious problem in order to get them to wear masks and engage in social distancing, so evidence that it is less of a problem than it seems, that the surge in cases is not resulting in a surge in deaths, is likely to be downplayed in most of the media. In addition, making Covid look bad makes Trump look bad, and most of the media are hostile to Trump.
P.S. A commenter points out that the topic has not been entirely missed by the media. The New York Times had an article on July 3d noting that the increase in cases had not yet been reflected in death rates and discussing possible explanations.
24 comments:
Different framing of your second explanation of the numbers: Vulnerable people are being equally careful, but younger people are being less careful. Tired of being cooped up.
Another explanation for media coverage: Persuading people that covid is a serious problem is *the reason* that vulnerable people are doing a good job staying isolated.
Overarching explanation: It's a blend of many explanations.
How likely do you think a contributing factor to this effect is the mutation of COVID-19? Like any virus, whose evolutionary interest is to learn to not kill the host, it's adapting, making it better at infecting human cells but be less fatal.
It has been mentioned in "the media". Here, for instance:
https://www.nytimes.com/2020/07/03/health/coronavirus-mortality-testing.html
1. some less-noticeable (less harmful per infection) strains are successfully spreading
2. young healthy people letting it rip and enjoying the outdoors together. some contagion but not much death.
3. lower viral load per infection - you get it but don't die because you were exposed at a lower level due to avoidance of superexposure events (church singing groups indoors). distance, masks, outdoors.
4. increased testing (true deaths per infection rate was always low).
5. kill-the-old atrocities like placing covid positives in nursing homes have run their course
6. gradually improving treatment protocols
I think a lot of those are contributing. I think better treatment is contributing, and increased hospital capacity.
I also think death rates are about to spike again.
-Bobboccio
In my country, there is an interesting phenomenon. The number of covid-19 positives is increasing at an alarming rate in the cities nearby where the major biomedical institute is located. They have locked down those cities now, but even with the lockdown, the rate of positives keeps increasing as more and more testing is being concentrated there. I live in the second biggest city, where no testing is being performed, and alas we have no lockdown and no new positive cases have been reported for several weeks now. Everybody is baffled by this.
The increase in cases can result from both (i) a faster spread of the disease, and (ii) a greater percentage of actual cases being discovered as there is increased testing. It is hard to know the relative magnitudes of these.
But if the effect of (i) is (as you suggest) mostly that more young people are getting milder cases, then the number of daily deaths should continue to go down or not increase significantly. This appears to be what was happening as of 2-3 weeks ago because daily deaths have continued to decline.
We will have to see if these effects continue to predominate.
Third possibility: Covid-19 is now less deadly. Admittedly, this is less likely than your two explanations, but possible. For instance, warm weather might make it less virulent or damaging; it might have mutated to be more infective but less severe; we could be using better treatments now to avoid deaths; and so on.
It is disappointing that we are not sufficiently immersed in the data to compare hypotheses. This seems to be another dimension a government failure.
Farrs Law of Epidemiology. None of this is new
I don't think it's true that there is no increase in deaths nationally. If you look at the number of deaths each day vs. 7 days prior, there had been a decrease for a while, but it had been getting smaller.
The holiday weekend complicates things, but as of today, it looks like we are back to increasing numbers. That holds up whether you look at just Wednesday or Saturday-Wednesday. Unfortunately, I expect that trend to continue.
There are definitely parts of the country where deaths are increasing. In Arizona, for example, since June 1 the number of deaths per day has roughly doubled, and current hospitalizations has roughly tripled.
From Business Insider another option: https://www.businessinsider.com/why-us-coronavirus-deaths-arent-rising-younger-cases-lag-time-2020-7
"But increased testing could also create a longer lag between when a case is confirmed and when a patient dies, since cases are being detected earlier in the course of an illness."
A "surge" in testing has accompanied the so called "surge" in cases. Some have argued that the increase in positives was caused by an increase in tests. Interestingly, the ratio of positives to tests has also increased. So, one needs to explain two things. 1) What caused the "surge" in testing and 2) what caused the ratio of positives to tests increase. I think much of the "surge" in testing results from two things. Last month about 4 million people returned to work. Many employers have insisted on people getting tested before allowing them to return. In addition people who already have the virus are tested repeatedly (every 3 to 7 days) to determine if they have recovered. This double counting can increase the number of tests quickly. The ratio increase can also be partially explained by this repeated testing. Most of those tests will be positive. Furthermore, the improvement of treatments and early detection because of increased testing results in more repeated tests, because the patient is in the process of recovery for a longer period of time.
The last few days has seen a very slight increase in the 7 day death reproduction rate from 1.027 to 1.032. I know, 3 decimal points, what accuracy. However that difference means a difference in 4 weeks for deaths to double. So, there is some evidence that deaths are rising faster than they have been. That rise in deaths, because its a lagging indicator points to a rise in "real" cases 14 to 21 days ago.
First of all, I dispute your reading of the graph. Cases rise a little bit in early June, hardly worth talking about. In mid June they rise substantially, which predicts deaths to be recorded about now.
If you think the issue is increased testing, you should look at the testing. Percent positive fell over the course of May, suggesting that the increased positive tests were due to increased testing. But percent positive rose starting at mid June, suggesting a real outbreak, but not that much bigger than in early June. Taking the percent positive as the measure of the size of outbreak, it is perhaps now 2x as big as in early June. The mid-June outbreak was not even that big.
Intermediate between tests and deaths are hospital beds. That starts going up about 21 June, a week delay from the increase in percent positive. So far, it has doubled.
In itself, doubling is not much. But if it comes from exponential growth, it is important to notice.
I think @DWAnderson hits one aspect. Increased cases are partly due to increased testing. This is twofold.
First, testing is more available. Someone who only had mild symptoms in late March was told to stay home and only try to get a test if symptoms got severe, because testing was restricted. Now, it's easy for someone who is only mildly symptomatic or worried to get a test. So you'll catch more mild cases.
Second, we're actually doing contact tracing. People who are known to have been exposed to a person with a positive result are getting tested. That means that we might even catch the asymptomatic or presymptomatic cases that we weren't catching before.
That said, I think the increased case rates ALSO indicate increased spread. It's not just due to testing.
@Rob has another good point. If you look at the slope of the decline in deaths, it's been consistently flattening. It went almost flat before the holiday weekend.
That might indicate a situation where you had states where things were getting worse balanced against states that were still declining, but the relative differences hadn't equalized or reversed.
That said, the last three days basically tell you exactly what has been going on. TX/FL/AZ have had growing daily death rates for the last week, which fit the lag time profile of the sharp increases in cases that occurred in mid-June. But their death rate increases weren't in "breakout" territory yet. California is similar, and it's only started to hit "breakout" numbers over the last three days.
Now the bad states are going exponential, and starting to dominate the national numbers. And we've now seen a reversal of the downward death trend to an increasing trend.
We're now seeing that the rise in cases CANNOT be solely due to a rise in testing, because if it was, the death rate wouldn't have reversed. The death rate HAS reversed, ergo the increases in cases is due to a surge in virus spread.
Consider this. That the bulk of the testing is non determinative in nature. This has been an issue ignored from the onset.
Why what we see is not aligned with what we read. (One substantial reason, anyway)
GIGO.
Your US covid death data is wrong. Your chart shows peak death rates in April of ~100 per day. In fact, according to the CDC, at the peak in mid-April it was ~2500 - so off by 25x. The CDC data can be fetched here: here.
I think we can safely ignore the so called cases. These covid deaths are hard to fake, although certainly there is some error (some folks actually died of other causes but also some died of covid that weren't tested). But I expect the errors are pretty small.
Since deaths are going to be proportional to time lagged infection rates among the vulnerable, it's clear those infections have been falling since mid-April. As of a couple of weeks ago they were still falling even though the "cases" were increasing. So "cases" isn't telling us anything important even if it's true.
Another thing to keep in mind is, when infections peak the population has reached herd immunity. So we reached that point nationally in early April a week or two before the peak in deaths. Most states have also, certainly the big ones where there have been many deaths (the CDC has deaths by state in the download). Once a population is at herd immunity, the disease must die off - in fact that' exactly what it means, the time rate of change in infections becomes negative. In practice, since there'll be a lot of infected people when the disease is at it's peak, the population rushes past herd immunity and infections soon start decreasing exponentially.
It looks like weekly deaths are about halving. They'll continue to. The crisis is well over. At this point more folks die from auto accidents than from covid by about a factor of 3.
The chart showing ~100 deaths per day in April is for California, not the entire US.
The Swiss immunologist Beda Stadler points out that the test itself will be positive if detects the viral debris that is a result of the immune system having defeated the virus.
"Even if the infectious viri are long dead, a corona test can come back positive, because the PCR method multiplies even a tiny fraction of the viral genetic material enough [to be detected]."
At the same time, we now know that the medical inflates not just the death numbers but also the case numbers. For example, if someone calls for a testing appointment and doesn't show up they are logged as positive, on the basis that they must have had the symptons if they called in the first place.
So part of the "massive" increase in cases is actually just population immunity, and part of it is fake news.
You should give death data 60 days especially from states and jurisdictions with overloaded public health departments. There's no centralized Federal reporting of deaths and it takes a while to reconcile.
The other big change is that we are far better at dealing with vulnerable populations in assisted living and skilled nursing facilities. That disproportionately contributed to the early mortality numbers. I still think it's prudent for those facilities to have strict precautions as well as the elderly and the obviously vulnerable like respiratory or have mechanical issues with breathing. The rest of the population, there's not a clear direction and is something worth debating.
I wish we could separate out institutionalized living from the community mortality for that discussion.
Even the Feds use this solution (not a scam), this is the Electronic Verification of Vital Events to deal with death data capturing.
https://www.naphsis.org/evve
I've been doing similar analysis for AZ.
One of the biggest problems is discrepancies in how the data is aggregated/displayed. WorldofMeters (can't link, it's an https addresss), which is pretty great, shows cases, testing, and deaths on the date they are reported. Putting it bluntly, this is crap, because there is significant lag time for all of the metrics.
AZ Dept of Health COVID dashboard's summary tells you what was reported that day, but when you go to the individual pages, you get confirmed cases by the date tested, number of tests performed each day, and deaths by date of death. This data gives you a much better idea of what's actually happening, but there is always a lag in each of these measures. One of the best is hospitalizations, again, number of cases hospitalized by the date the patient entered the hospital.
So looking at all this data, testing and positive cases are strongly correlated, as you would expect. The dates with the highest numbers of tests performed are also the dates with the highest number of positive cases. We've had days now with over 5,000 positive cases, all back in the last week of June. Looking at the deaths graph linked above, the greatest number of deaths we've had on one day is 42 again in the last week of June, but death reports lag deaths probably by a week, so the historical data in this graph is constantly being updated. Hospitalizations have dropped significantly after peaking in mid June. There is not note on the hospitalizations data about lag, but I haven't seen the historical data here changing very much.
Blah blah blah: Cases are way up, but the dashboard shows the vast majority of new cases are in younger people. Deaths are up somewhat, but overall, the case fatality rate is down to something like 0.018, which is a lot less than the flu. New hospitalizations are falling off, although if the "catching cases earlier" hypothesis means a longer lag time between diagnosis and required hospitalization, will we see hospitalizations start going up again? I doubt this, because again, most the of the new cases are among younger, healthier people.
IMO what's happening now is good: many many more cases with only a slight uptick in deaths shows the fatality rate is not nearly what we thought it was. Yay! Can we get back to our lives please?
Assume you have all seen this site? https://www.cdc.gov/nchs/nvss/vsrr/covid19/excess_deaths.htm
First of all, I am one of those better safe than sorry people when it comes to Covid and am lucky my job allows me to be. Fully understand that not all or even that many people can say that.
Secondly, my daughter posted a comment that people who play D&D know that you never want to roll the die for *any* saving throw, regardless how high your constitution score. Or as another gamer said, "Let's say it's a 2% roll. I can make it 0% (or close to it) by *not* meeting you in person."
Anyway, I am finding this thread interesting and appreciate the open & mostly non-political discussion.
My take:
1) Case number stats have too much noise to provide any real insights … except that more people are being tested. Which is good so they can a) get early treatment to improve their recovery chances and b) infect fewer people along the way. Also, younger people might be more willing to roll the above die so as not to miss a beach party, date or …
2) Hospitalization is a bit cleaner, though admission is unclean since the filters differ based on far too many variables. Better, but still not good.
3)Deaths, unfortunately, are the cleanest. The cause of death is somewhat unclean, since there might be a number of contributing factors and Covid/not Covid is overly binary. So…
4) I agree with the OP on measuring by 'excess' deaths, as grisly as that is. The excess deaths seems to be dropping, which is good, but the last few days appear to be ticking upwards, which is not.
Unknown pointed out that the death data is actually for California and not the whole US - I think I might be excused as the image is titled US Deaths and is juxtaposed with US Cases.
Even so, the graph is at least misleading. It gives the impression that California deaths haven't fallen much at all but weekly deaths have fallen by half from their peak in May. CDC data as of 7/10 is that 539 died in the week of 5/9 but 187 in week of 6/20. Both numbers will no doubt be revised up but I doubt the ratio will change that much.
Given all the lying by "public health experts," one can be forgiven for wondering if it's not intentionally misleading; I would not put it past the "health authorities" to report the data in such a way that there's a high daily peak in each week even though weekly deaths are decreasing.
I recommend the CDC's Pandemic Planning Scenarios. Not just for the scenarios, but more for other parameters, like the delay from symptoms to hospitalization and the delay from symptoms to death. Especially the 1 week delay from death to recording of death (probably not specific to covid). Most data sources are write-only and count deaths on the day that they are received, but Santa Clara counts deaths on the day that they occurred. This means that they are constantly updating numbers. They are more optimistic than CDC and think that it only takes 5 days to learn about a death. (Indeed, it is probably the county that tells the CDC and thus knows first. But does it get it on the web first?) Thus the last 5 days are marked in gray to indicate that they are provisional.
Most cases are asymptomatic
Post a Comment