COVID round-up: Current Mortality, Bad Spreadsheets, Problem with COVID testing, Longer-lasting effects from COVID, and more
Deep breaths over (expected) bad practices
Here’s this week’s round-up of COVID-related stories, including one that popped up that I really didn’t expect, but once I read about it, found it depressingly familiar.
I will start with that story after a high-level review of current COVID mortality.
U.S. COVID mortality as of October 9, 2020
Using data from the CDC, the official (provisional) COVID death count is 200,499.
Out of 2,167,730 deaths. So fewer than 10% of deaths in the U.S. are currently attributed to COVID-19.
A visualization of the excess deaths so far (and the part attributed to COVID):
Even long after the “first wave” (which only occurred in an area centered on NYC), the only place with a similarly high mortality hit has been Arizona… but New York City and New Jersey still far outpace all the other locales for incidence. If you carved out the counties surrounding NYC (including the ones in NJ), you’d probably capture almost all of the huge COVID-19 mortality.
I direct your attention to the fourth column of the table, where they give you the ratio of 2020 deaths against “expected” deaths. NYC is showing a cumulative 72% excess mortality. That’s even though there has barely been any NYC COVID-19 deaths since May… at its peak in April, NYC had excess mortality of over 600%.
If you go to the CDC page, I want to note that some of the states have unnaturally low death counts due to their very large reporting lags to the CDC.
Two awful tastes that go horribly together: Excel and COVID
Yes, I’m talking about the spreadsheet software, and I really wish I weren’t.
A technical glitch that meant nearly 16,000 cases of coronavirus went unreported has delayed efforts to trace contacts of people who tested positive.
Public Health England said 15,841 cases between 25 September and 2 October were left out of the UK daily case figures.
Public Health England’s interim chief executive Michael Brodie said a “technical issue” was identified overnight on Friday, 2 October in the process that transfers Covid-19 positive lab results into reporting dashboards. He said the majority of the unreported cases had occurred in the “most recent days”.
It was caused by some data files reporting positive test results exceeding the maximum file size.
The government said the technical issue meant some cases during the week were not recorded at the time, so were included in Saturday’s data.
The daily total rose from 4,044 on Monday to a then-high of 7,143 on Tuesday. However, over the next four days the daily total remained stable at a time when continued increases might have been expected.
Then came the big leap in numbers – a far bigger day-on-day increase than at any time in the entire pandemic – which was announced on Saturday, five hours later than usual, and was accompanied by the government explanation.
In this case, the Guardian understands, one lab had sent its daily test report to PHE in the form of a CSV file – the simplest possible database format, just a list of values separated by commas. That report was then loaded into Microsoft Excel, and the new tests at the bottom were added to the main database.
But while CSV files can be any size, Microsoft Excel files can only be 1,048,576 rows long – or, in older versions which PHE may have still been using, a mere 65,536. When a CSV file longer than that is opened, the bottom rows get cut off and are no longer displayed. That means that, once the lab had performed more than a million tests, it was only a matter of time before its reports failed to be read by PHE.
I will get to that issue in a moment, but first, a different Excel issue that caused awful problems:
Errors from the spreadsheet software have even changed the very foundations of human genetics. The names of 27 genes have been changed over the past year by the Human Gene Nomenclature Committee, after Microsoft’s program continually misformatted them. The genes SEPT1 and MARCH1, for instance, have been changed to SEPTIN1 and MARCHF1 after they were repeatedly turned into dates, while symbols that were common words have been altered so that grammar tools didn’t autocorrect them: WARS is now WARS1, for instance.
I was so incensed by this issue on Monday, October 5, that I wrote a LinkedIn post on it:
However, while Excel is broadly known, and easy-to-use (and cause a lot of damage with), it is not a good tool for very important projects. It’s not secure. It has some controls, but you have to impose them on your files. It is extremely easy to mess up a single cell in your file and not detect it. And, in this case, you could go over file size limits and never notice it.
My main point is that Excel is a completely inappropriate tool to be used for something as crucial as updating official government statistics about COVID-19. Just as Excel (or any spreadsheet software) shouldn’t be used for financial reporting in multi-billion-dollar organizations.
That it sounds like the PHE will simply chop up the files into smaller bits to make them importable-into-Excel is showing me that they haven’t learned the lesson.
I didn’t hear one damn thing about governance/controls on making sure the data is appropriately imported into official reporting systems. This is a disgrace.
Yes, one can put controls on Excel, etc., but you should be using proper data management to begin with, dammit.
Go to the original STUMP post for all the coverage on this issue.
A few closing remarks on this:
Twitter has actually been a great resource throughout the pandemic. You do need to be able to detect the knowledgeable from the full-of-shit, but in general, the knowledgeable will give you links, detailed breakdowns, etc. Thanks to the twitter crowd! You guys are great! [Seriously, people need to learn to filter the junk. Twitter can be a fabulous resource for keeping current]
Too many people are half-assing it through a crisis.
The problem with COVID testing
A reader sent me an old NYT article, from 2007: Faith in Quick Test Leads to Epidemic That Wasn’t
Dr. Brooke Herndon, an internist at Dartmouth-Hitchcock Medical Center, could not stop coughing. For two weeks starting in mid-April last year, she coughed, seemingly nonstop, followed by another week when she coughed sporadically, annoying, she said, everyone who worked with her.
Before long, Dr. Kathryn Kirkland, an infectious disease specialist at Dartmouth, had a chilling thought: Could she be seeing the start of a whooping cough epidemic? By late April, other health care workers at the hospital were coughing, and severe, intractable coughing is a whooping cough hallmark. And if it was whooping cough, the epidemic had to be contained immediately because the disease could be deadly to babies in the hospital and could lead to pneumonia in the frail and vulnerable adult patients there.
It was the start of a bizarre episode at the medical center: the story of the epidemic that wasn’t.
For months, nearly everyone involved thought the medical center had had a huge whooping cough outbreak, with extensive ramifications. Nearly 1,000 health care workers at the hospital in Lebanon, N.H., were given a preliminary test and furloughed from work until their results were in; 142 people, including Dr. Herndon, were told they appeared to have the disease; and thousands were given antibiotics and a vaccine for protection. Hospital beds were taken out of commission, including some in intensive care.
[spoiler alert: none of them actually had whooping cough]
Now, as they look back on the episode, epidemiologists and infectious disease specialists say the problem was that they placed too much faith in a quick and highly sensitive molecular test that led them astray.
Many of the new molecular tests are quick but technically demanding, and each laboratory may do them in its own way. These tests, called “home brews,” are not commercially available, and there are no good estimates of their error rates. But their very sensitivity makes false positives likely, and when hundreds or thousands of people are tested, as occurred at Dartmouth, false positives can make it seem like there is an epidemic.
At Dartmouth the decision was to use a test, P.C.R., for polymerase chain reaction. It is a molecular test that, until recently, was confined to molecular biology laboratories.
This is what is used for COVID testing in general now. We will get back to that in a moment.
Dr. Katrina Kretsinger, a medical epidemiologist at the federal Centers for Disease Control and Prevention, who worked on the case along with her colleague Dr. Manisha Patel, does not fault the Dartmouth doctors.
“The issue was not that they overreacted or did anything inappropriate at all,” Dr. Kretsinger said. Instead, it is that there is often is no way to decide early on whether an epidemic is under way.
With pertussis, she said, “there are probably 100 different P.C.R. protocols and methods being used throughout the country,” and it is unclear how often any of them are accurate. “We have had a number of outbreaks where we believe that despite the presence of P.C.R.-positive results, the disease was not pertussis,” Dr. Kretsinger added.
At Dartmouth, when the first suspect pertussis cases emerged and the P.C.R. test showed pertussis, doctors believed it. The results seem completely consistent with the patients’ symptoms.
The Dartmouth doctors sent samples from 27 patients they thought had pertussis to the state health departments and the Centers for Disease Control. There, scientists tried to grow the bacteria, a process that can take weeks. Finally, they had their answer: There was no pertussis in any of the samples.
“The big message is that every lab is vulnerable to having false positives,” Dr. Petti said. “No single test result is absolute and that is even more important with a test result based on P.C.R.”
There are some big differences between pertussis (whooping cough) and COVID — for one, pertussis is caused by bacteria, and COVID comes from a virus. I don’t know how they test for viral samples, but I know such tests exist (though perhaps it’s just PCR in all cases).
So we have a situation where PCR testing has ramped up, and a lot of people testing positive may not actually have COVID. Both false negatives and false positives can cause trouble, too [I link to a GQ article below – there are people who almost definitely had COVID but never got a “positive” PCR result.]
If you’re interested in how false positives and false negatives can play out, here’s an interactive tool to play with from the bmj. I may do a video showing you the results of different combinations, explaining the results. That is not in this post, though.
GQ: Doctors Tell Me I Have COVID. Why Won’t the Tests? – on false negatives
Twitter thread: New Visual in Development: Testing, False Positives, and Infectiousness status of Daily testing numbers from CTP. – provides sources and methods.
Thread from twitter: IS TESTING WORTH THE PRICE?
Clinical Infectious Diseases: Correlation between 3790 qPCR positives samples and positive cell cultures including 1941 SARS-CoV-2 isolates
Longer-lasting effects on pensions from COVID?
Months ago, somebody asked me if the excess mortality from COVID would “help” pension plans, specifically, public pensions.
My very short, non-quantified answer was: not enough.
But now there is something semi-official out.
Issue Brief from the American Academy of Actuaries: Impact of COVID-19 on Pension Plan Actuarial Experience and Assumptions, Including Mortality
For many companies with calendar-year fiscal years, year-end accounting measurements typically do not reflect full plan participant status updates, due to timing considerations. Actuaries and plan sponsors are faced with determining whether excess mortality during 2020 is significant enough to warrant accelerating the participant census update or otherwise reflecting in year-end measurements. However, the level of excess mortality to date and the expected advanced ages at which most of the excess deaths occur would appear unlikely to have a significant impact on the liabilities for most plans. For example, few if any plans are likely to see a doubling of normal annual mortality rates due to the pandemic, and yet even a doubling of the one-year mortality rate is unlikely to reduce benefit obligations by much more than 1% for the typical plan.
The main issue is that it will be a one-time effect on mortality – what we saw this year. A one-year bump up in deaths will remove those portions of the liability, granted, but it is biased toward much older people and will not have a large effect on pension liabilities.
The issue is the long-term impact on mortality. If we had 10% excess mortality compared to 2019, say, for every year 2020 and after, then yes, liabilities would come down a bit. But we have no reason to believe that will be the effect. Indeed, improvements in hygienic practices in nursing homes, hospitals, and even the general public in reaction to the pandemic may increase longevity:
Along similar lines, some have suggested that habits developed during the current pandemic, such as more diligent hand-washing and mask-wearing, may serve to slow the spread of other diseases that currently exist as well as new diseases in the future, thereby lowering future rates of mortality. On the other hand, pandemics could become more common in the future. If so, it remains to be seen how the lessons learned during this pandemic will be applied to limit future spikes in mortality.
And this was something I was trying to get at on January 29, 2020. When I brought up the flu, it wasn’t to minimize Covid (though we didn’t know what its mortality effect would be, nor how infectious, at the time). I was trying to get people to take the flu seriously:
Now, I could argue (and I do argue) that people in the developed world don’t take regular flu seriously enough.
Tens of thousands of Americans each year die from the flu. Part of the reason it’s not taken overly seriously is that most of those who die from the flu are old.
BE CONCERNED ABOUT REGULAR SEASONAL FLU
There’s not much you can do about coronavirus right now, other than stay out of Hubei province in China.
More broadly, in the United States the CDC estimates, as of January 11, 2020, 13 million influenza illnesses, 120,000 hospitalizations, and 6600 flu-related deaths. The flu season starts around September, iirc.
It’s okay to be concerned about the Coronavirus 2019-nCoV, because I know I don’t trust the Chinese government to have good data (for a lot of reasons), and even if they did, they wouldn’t necessarily share it.
However, you really should take “ordinary” flu seriously, too.
And you should.
Related coverage: AI-CIO: Actuarial Projections: Virus Might Cause ‘Meaningful’ Increase in Death Rate for Elderly
Please take the flu seriously
Let’s look at the pediatric flu deaths for the 2019-2020 season:
Among the 188 reported pediatric flu deaths:
43% (81) occurred in children younger than 5 years old
12 occurred in children younger than 6 months and thus too young to get a flu vaccine
57% (107) deaths occurred in children 5-17 years old
So, there were 188 pediatric flu deaths, and as of October 9, 2020, there are 72 COVID deaths for kids under age 15.
It’s not directly comparable, but if you yell DO IT FOR THE CHILDREN, you really should do it for the flu, not just COVID.
Of course, far more elderly people die from the flu and have died from COVID. But “DO IT FOR GRANDMA” generally goes by the wayside in “normal” years, doesn’t it?
Americans do not take the flu seriously, and we should.
Let’s do some very rough estimates. In this piece – WHO: 10% of world’s people may have been infected with virus – more than 760 million people worldwide estimated to have had COVID already. [I don’t know how much to trust this].
If 1/10 of Americans have already had COVID-19, then that would be 33 million people. In a bad flu season, like 2017-2018, 45 million Americans had the flu. And that’s with a vaccine (yes, I know in bad flu year, the vaccine tends not to work so well). In a “normal” flu year, it’s about 30 million Americans; again, that’s with a vaccine.
I have no idea if 33 million or so Americans have had COVID already, or it’s a lot more or fewer. But I hope that people will take preventing infectious disease more seriously even after the current pandemic goes away. It would be nice if flu gets dampened as a cause of death in the U.S. because Americans pick up Japanese-level hygienic habits of wearing masks when they may be ill.
Now that I’ve got a bunch of nice-looking face masks, I know I’m just going to wear them in the future during cold/flu season. And don’t say I won’t, because I’ve been carrying around cotton hankies for years before now, even if it’s “weird”, and I refused to shake people’s hands when I was feeling ill as well as warn people to stay farther away from me.
By the way, yes, COVID is deadlier than the seasonal flu, even in a bad flu year. It’s not as deadly as the Spanish flu pandemic was, though. For the same number (maybe) of infections, fewer people die from the flu. To be sure, more children and young adults die from the flu than from COVID, looking at a rate basis. “Deadlier” is relative — COVID far deadlier for old folks than the flu; for young folks, it may be a draw.
I will make some comparisons in the future.
The dump of everything else:
MIT Tech Review: The CDC has finally acknowledged that the coronavirus can be airborne – meaning virus can hang around in the air.
IEEE Spectrum: What AI Can–and Can’t–Do in the Race for a Coronavirus Vaccine — AI can do certain things well
Nature: COVID has killed more than one million people. How many more will die? – this is worldwide.
The Atlantic: This Overlooked Variable Is the Key to the Pandemic
Burypensions: Counting NJ Deaths -September Update
The Economist (free to read): Is Pakistan really handling the pandemic better than India?
Illinois Policy Institute: ILLINOIS BECOMES ONLY STATE TO BAN HAUNTED HOUSES – yes, this is about COVID
WSJ: Wisconsin Struggles to Explain Sudden Covid-19 Spike – maybe it was Excel
Twitter thread: Mortality trends of multiple nations – interesting the variance in results of countries that are geographically near each other. They didn’t necessarily have different policies, but may reflect all sorts of things.
The last item is related to aging in general, not COVID, but I thought I’d throw in some good news here at the end:
Today’s research materials cover one of a number of studies to suggest that older people are becoming functionally younger over time, comparing the capabilities of age-matched cohorts of old people in past decades with old people of the same age today. Being 70 or 80 in 1990 was accompanied by greater loss of physical capabilities, such as walking speed or grip strength, than is the case at those ages today. This is what one would expect given the slow upward trend in life expectancy that has continued year after year for more than a century now, driven by a shifting combination of better lifestyle choices, greater control over medical issues throughout life, and slow improvements in treating age-related disease.
It is interesting to see just how much has been achieved without undertaking direct efforts to target the mechanisms of aging. While the reasons for a lesser burden of frailty and mortality in late life have changed over time, from a reduction in the burden of infectious disease across the 20th century to a lessening of cardiovascular disease over the last few decades, the theme remains an incidental reduction in the level of accumulated damage and dysfunction at a given age.
People really are staying well longer than before, even with increase of diabetes…and there are several reasons this may be happening.
I think the biggest difference is smoking: smoking ages people so much, and there has been a huge drop in how many people smoke from the peak. But it can also be related to dirty air, dirty water, leaded gas, and all sorts of toxic materials around people… and then there’s more effective drug interventions for all sorts of conditions which are treated more effectively now than 30 years ago.
If I look at pictures of how my mother looks now compared to how my grandmothers looked at the same age… yeah, I totally see the difference.