Category: COVID Archives

Declines in WA, FL, NY, MA, GA, MI, PA, TX – increases in NC, CA, SC, VA

The big news today is that we had a record 321,000 tests reported today, and only 10.8% positive. Testing is now widespread, and we’re testing a lot of people who are not sick with COVID. Things are looking better for most of the states I’m tracking once again. Over the next 2 weeks, we’ll start to see if there is any correlation between various states restrictions and outcomes. Initial studies show no correlation, which doesn’t surprise me, since people will generally be cautious or not regardless of whether the government tells them to. Here is one from T.J. Rodgers, a person I know and respect: https://www.wsj.com/articles/do-lockdowns-save-many-lives-is-most-places-the-data-say-no-11587930911?mod=article_inline

As always, feel free to send me your questions about my assumptions, methodology, or modeling in general.

• Likely date of active case peak (Chalke modeling): April 10
• Likely date of peak deaths (IHME): April 15 (last revision on April 29)
• Short term projection for tomorrow: 1,125,000
• Total Test Results reported today: 320,628
• Total Pending tests reported today: 1,639 (a record low)
• National reported case Growth Rate today: 3.3% (very low)

Declines in FL, NY, MA, GA, MI, PA, TX – increases in NC, CA, VA

Things are looking better for most of the states I’m tracking. I was asked how I chose the states I’m tracking. I chose NY, WA, and CA as they were the first hot spots. I added MA, MI, and PA as they grew to statistical significance. I added GA, SC, and TX because they looked like they would be the first states to relax restrictions, and everyone is watching them to see if the propagation accelerates (I don’t think it will). And I track NC because I live there. VA is on my list because I got more than a few requests to add it.

As always, feel free to send me your questions about my assumptions, methodology, or modeling in general.

• Likely date of active case peak (Chalke modeling): April 10
• Likely date of peak deaths (IHME): April 15 (last revision on April 28)
• Short term projection for tomorrow: 1,088,000
• Total Test Results reported today: 205,012
• Total Pending tests reported today: 2,775 (a record low)
• National reported case Growth Rate today: 2.7% (very low)

Declines in WA, FL, NY, CA, SC – increases in VA, PA, NC

The great recovery continues. The numbers show that the worst is well behind us, and those infected are recovering at a faster rate than new people are contracting the disease.

As always, feel free to send me your questions about my assumptions, methodology, or modeling in general.

• Likely date of active case peak (Chalke modeling): April 10
• Likely date of peak deaths (IHME): April 15 (last revision on April 28)
• Short term projection for tomorrow: 1,059,000
• Total Test Results reported today: 230,442
• Total Pending tests reported today: 4,832 (very low)
• National reported case Growth Rate today: 2.7% (very low)

More on mortality rates…

Here is my first analysis based uniformly on COVID Tracking Project data. I was already using Tracking Project data for most of the charts and graphs, but I’ve redone the analysis going all the way back using only the Tracking Project.

I’ve already received quite a few questions about the mortality rates I quoted this morning for New York state, which were:

The common question is “what does it mean?”. These mortality rates represent the probability of death for those unlucky enough to be infected. The general population mortality rate (the probability of someone generally dying from COVID) is much, much lower. So these rates can be interpreted as “a person contracting COVID and age 59 and under has a 0.0009 chance of dying from the disease. Put another way, for every 1,111 people under age 60 contracting COVID, only 1 would be expected to die. For those over 60, the number is about 1 out of 40, a remarkable difference.

The second question I received in numbers was “is 0.09% low or high – relate it to something”. OK, so in the general population in the U.S., the average 50 year old has a 0.5% change of dying within a year (sorry for that news). This probability is roughly 5 ½ times higher than the probability of dying IF YOU CATCH COVID.

For the average 75 year old in the U.S., the probability of dying within one year is 3.6%. This is about 1 ½ times higher than the probability of dying IF YOU CATCH COVID.

How can this all make sense? Well, it seems Actuaries dwell on this the most, but each year about 3,000,000 people die of all causes in the U.S. When all is said and done, it looks like COVID deaths in the U.S. this season will make up about 2-3% of that number.

I’m going to pin this and this mornings mortality discussion to the bottom of the reports, since many new people join this list every day.

As always, feel free to send me your questions or comments.

• Likely date of active case peak (Chalke modeling): April 10
• Likely date of peak deaths (IHME): April 15 (last revision on April 27)
• Short term projection for tomorrow: 1,031,000
• Total Test Results reported today: 202,233
• Total Pending tests reported today: 4,206 (very low)
• National reported case Growth Rate today: 2.5% (very low)

1st cut at NY mortality rates

For well over a month now, I’ve been using data from Infection2020 for the daily growth rate, daily new cases, and cumulative reported cases. I’ve been using The COVID Tracking Project for state level data, testing data, and deaths. I’ve been reporting on active cases derived from both data sources. Lately, however, Infection2020 has been less reliable. They’ve been reporting only once a day (it used to be continuous). They note on the site that they are adding new sources of data and are under financial stress. You can donate on their site – I did. In any event, their data is diverging from other sources. As of late last night, they were showing over a million reported cases, while the COVID Tracking Project was showing 981K and JMU this morning is showing 988K. Also, at one point in the evening they were showing over 50,000 new reported cases, which is substantially out of line with other sources. As a result, I’m going to standardize my reporting and analysis on COVID Tracking Project data beginning later today. I will rerun all of the charts from the beginning with COVID Tracking Project data, and suspend use of Infection2020 data. Of course, I’ll still monitor a variety of sources to make sure I’m using reasonable source data.

Since The COVID Tracking Project uses a 4pm EDT cutoff, I will begin issuing my analysis in the early evening, instead of mid-day. So this will be the last daytime report, and you will receive another analysis around 6 or 7pm today, and each day thereafter at the same time.

Now, a thought or two about mortality. On April 23rd NY reported completing a somewhat random test of 3,000 New Yorkers for COVID antibodies. https://www.newsweek.com/test-shows-21-percent-new-york-city-residents-who-gave-samples-have-coronavirus-antibodies-1499878 Tests were done earlier that week for randomly selected individuals in grocery stores and shopping centers. Of the 3,000 tested, 15.9% were shown to have had the COVID infection. Since that time, New York has revised that number up to 16.9%. This is incredibly useful information, and provides us with a first meaningful clue about mortality. I first did a rough calculation of NY mortality rate by extrapolating the random test results to the NY population, then aligned deaths with the exposure period. From this I calculate a mortality rate of 0.53%. This is not unexpected, but it is much more interesting when broken down by age. When I look at cohorts of age 60 and over, and 59 and younger, and go through the same calculations, I get an estimate of mortality rate for these two age groups of:

Now there are lots of issues with my calculations. First, although the sample may have been random, there was some self-selection, as the sample pool was composed of those willing to leave their homes, which I bet skewed younger and less risk-averse. I also do not have a demographic breakdown of the sample. The only thing I know is that no one over age 75 was sampled. This is a major issue, as over ½ of all deaths in NY are over age 75. This leads me to believe that COVID prevalence is actually lower among the 60+ cohort, and higher among the 59- cohort. As a result, I believe that the mortality rate for the elderly is higher than that calculated above, and the mortality rate for those under 60 is lower than that calculated above. By the way, I would have preferred to use an age 65 cutoff, but unfortunately the NY Department of Health only reports deaths in deciles. Ideally, if I had the data, I could study mortality in 2 cohorts – those over 65 or with certain co-morbidities, and a second group of under 65 without certain co-morbidities. I believe you’d find a mortality rate in the first group well over the 2.50%, and the mortality rate in the second group much closer to nil.

So as we recover as a society, the methodology seems to settled that we’ll unlock one geographic area at a time. But we have 2 easily identifiable populations, once with relatively high risk, and one with a tiny fraction of the risk. It occurs to me that we could more safely unlock by doing it demographically, rather than geographically. We could extend shelter in place rules for seniors and those with definable risk factors, and eliminate them for the under 65 and healthy population. It would also be fairly easy to define perhaps one concentric circle around the vulnerable by maintaining shelter in place rules for certain caregivers and health care workers. So that’s my message for today: Unlock Demographically, Not Geographically.

OK that’s a lot for this morning. As always, if you have any questions about my assumptions, how I’m modeling, data sources, or want to know about Logistic modeling generally, drop me a line. If you’re bored with this daily report, let me know and I’ll remove you from the list.

• Likely date of active case peak (Chalke modeling): April 10
• Likely date of peak deaths (IHME): April 15 (last revision on April 27)
• Short term projection for tonight: 1,010,000
• Total Test Results reported yesterday: 152,416
• Total Pending tests reported yesterday: 4,077 (very low)
• National reported case Growth Rate yesterday: 3.4% (suspect – likely lower)

Daily deaths decline, new cases decline – growth rate falls to new low (again)

My headline has been the same for 3 days – Groundhog Day but that’s good news. COVID continues its steady decline in the United States, with the national growth rate for reported cases falling to 2.55%, a new low. This does not appear to be a Sunday reporting effect, as the tests reported yesterday were over a quarter million. Last Sunday 175,000 tests were reported. Yesterday we had about 5,000 fewer new cases than a week ago, but 75,000 more test reports. I like this.

We’re now 12 days past peak daily deaths (IHME), and 17 days past peak reported active cases (Chalke modeling). Tomorrow I’m going to report my first cut on mortality rates (my specialty). I’ll not preview this, but let me say it strikes me that we should be lifting restrictions demographically rather than geographically.

As always, if you have any questions about my assumptions, or how I’m modeling, data sources, or want to know about Logistic modeling generally, drop me a line. If you’re bored with this daily report, let me know and I’ll remove you from the list.

• Likely date of active case peak (Chalke modeling): April 10
• Likely date of peak deaths (IHME): April 15 (last revision on April 22)
• Short term projection for tonight: 999,000
• Total Test Results reported yesterday: 256,444
• Total Pending tests reported yesterday: 4,445 (very low)
• National reported case Growth Rate yesterday: 2.55% (a new low)

Daily deaths decline, new cases decline – growth rate falls to new low

The national growth rate for reported cases fell to a new low of 2.7%. This is not a Saturday effect, since we had near record tests of over 300,000 reported yesterday. I’m still challenged by the increasing anomalies and outright errors in the way some jurisdictions are reporting data, but I’m using it as it is, rather than trying to back out historical data reported as current or other data irregularities. These issues will show up for a day or two, then smooth out over time. The overall trend is the same for over 2 weeks now – the disease is in decline nationally. There are some states where it continues to grow, but most of the states I’m individually tracking are past peak and declining.

As always, if you have any questions about my assumptions, or how I’m modeling, or want to know about Logistic modeling generally, drop me a line. If you’re bored with this daily report, let me know and I’ll remove you from the list.

• Likely date of active case peak (Chalke modeling): April 10
• Likely date of peak deaths (IHME): April 15 (last revision on April 22)
• Short term projection for tonight: 974,000
• Total Test Results reported yesterday: 300,833
• Total Pending tests reported yesterday: 5,315 (very low)
• National reported case Growth Rate yesterday: 2.7% (a new low)

Daily deaths decline, new cases decline – growth rate falls to new low

The national growth rate for reported cases fell to a new low of 3.26% New deaths also fell for the 3rd day in a row. There are increasing anomalies and outright errors in the way some jurisdictions are reporting data. The floodgates for this opened when the CDC changed their guidance on reporting April 17th, allowing both “probable” cases and deaths. In addition, some states have been retroactively finding probably cases. I’ll comment more on this in the state by state view below.

If you’ve been following my analysis for a while, you know that the story of this disease is a story of diverse metropolitan areas. The vast majority of the progression is in cities, where the population density is high. I haven’t looked at this, but I suspect there is a strong correlation between population density and propagation speed. The point is, that each city is in a somewhat different phase of the curve. It’s readily apparent when you look at the state by state curves below. For those of you interested in the effect of relaxing restrictions in some of the early states unlocking now, I believe the earliest date we could see an effect is about 2 weeks following the lifting of restrictions, based on gestation period and testing lags.

As always, if you have any questions about my assumptions, or how I’m modeling, or want to know about Logistic modeling generally, drop me a line. If you’re bored with this daily report, let me know and I’ll remove you from the list.

• Likely date of active case peak (Chalke modeling): April 10
• Likely date of peak deaths (IHME): April 15 (last revision on April 22)
• Short term projection for tonight: 950,000
• Total Test Results reported yesterday: 223,552
• Total Pending tests reported yesterday: 4,396 (very low)
• National reported case Growth Rate yesterday: 3.26% (a new low)

Daily deaths decline, daily new cases rises

A few interesting things to talk about today. First, a warning – a slight bit of math talk to come. If you’re allergic to this, skip down to the daily stats below to continue reading.

To me, the biggest news yesterday was the random antibody testing in NY. A sample of about 3,000 tests were done throughout the state, and an amazing 13.9% tested positive. You can read about it here: https://www.newsweek.com/test-shows-21-percent-new-york-city-residents-who-gave-samples-have-coronavirus-antibodies-1499878

This gives us the first solid clue into mortality rates. NY state has a population of about 19.45 million. If we extrapolate the testing results across the state (more on this in a minute), we’d estimate that 2.7 million New Yorkers have had COVID. As I write this, NY has suffered about 21,000 deaths. Simple division suggests a mortality rate of about 0.8%. OK, so what I just did is not scientific. To do this properly you’d have to analyze the demographics and biases in the sampling, and more importantly, align the exposure period with the death count – BUT, this gives us an idea. Since deaths are concentrated among seniors, my gut tells me that the mortality rate is much higher in the 65 and over cohort, and much closer to zero for those under 65. I’m sure you’ll begin to see lots of analysis on this in the coming days. It’s critical to get a bead on mortality rates by demographics – I believe that this will prove to the be the most important factor driving public policy should we see another wave next winter.

Next piece of news: My Logistic model is not tracking the back side of the curve very well. I thought it was a little soon to conclude that, but Dr. Conyers (on this list) sent me a fascinating podcast interviewing the Director of the IHME (https://fivethirtyeight.com/?s=Podcast), whose model I track and respect. He talks about many of the same frustrations I’ve had with shifting data definitions and protocol, and concludes that the shape of this disease in the United States is not classically symmetrical. The IHME model has many similarities to mine – they are both non-linear functional forms fit to time series data, rather than assumption driven Epidemiological models. However, the classic functional form for communicable disease growth in a resource constrained system is the Logistic function, which has the property that the growth of the disease mirrors the later decline (symmetry). What we seem to be seeing is a classic growth pattern, followed by a lingering plateau, followed by a somewhat slower decline than the corresponding ascent. IHME have made recent adjustments in their model to reflect this pattern.

I theorize that this pattern is a result of the continual expansion of the definition of a COVID case, and the rapidly increased scope of testing. The data was stable so long as the cases captured were a stable percentage of actual cases. With widening testing, we are discovering more of the disease. This doesn’t mean it’s growing faster, it means that we know about more of it. I believe this explains the divergence between reported case count and reported deaths (case count increasing, death count dropping).

In any event, after today I’m retiring my Logistic model. It did a wonderful job of predicting the peak and the magnitude, even as far back as weeks ago. As a mathematician, I feel good about that. But you must abandon your tools when they are no longer useful. I will continue to model active reported cases, as my methodology has produced results which correlate well with hospitalizations, and in most cases, death statistics.

As always, if you have any questions about my assumptions, or how I’m modeling, or want to know about Logistic modeling generally, drop me a line. If you’re bored with this daily report, let me know and I’ll remove you from the list.

• Likely date of active case peak (Chalke modeling): April 10
• Likely date of peak deaths (IHME): April 15 (last revision on April 22)
• Short term projection for tonight: 920,000
• Total Test Results reported yesterday: 193,691
• Total Pending tests reported yesterday: 4,258 (very low)
• National reported case Growth Rate yesterday: 4.1%

Testing expands dramatically, cases do not – TX added

The big news is that over 300,000 test results were reported yesterday. Since the big unknown is the ratio of total cases to reported cases, I would have expected a spike in new reported cases, but that did not happen. This is good news, but I don’t know the nature of who is being tested and why.

The national growth rate yesterday rose slightly to 3.6%, but estimated active cases still fell slightly.

The IHME model was republished again yesterday. They raised the projected total death count to 67,641, up from about 60,000. They are staying with April 15th as the day we crested peak deaths. As I mentioned yesterday, they are now publishing a date for each state where they are estimating it is safe to relax restrictions, but I think their criteria is unrealistic. They’ve set the date for when a state falls below 1 active case per 1,000,000 of population. That would be 10 active cases in the state of NC for example. I suspect most states will relax restrictions far before this, if it in fact this metric is even attainable. And that’s not considering that no one is even counting active cases that I’m aware of (I model them based on certain assumptions).

I added Texas to the reporting today. Unless another state obviously flares up, I’ll stop there. The daily report is getting long as it is…

As always, if you have any questions about my assumptions, or how I’m modeling, or want to know about Logistic modeling generally, drop me a line.

• Likely date of active case peak (Chalke modeling): April 10
• Likely date of peak deaths (IHME): April 15 (last revision on April 22)
• Short term projection for tonight: 887,000
• Total Test Results reported yesterday: 311,381 (a new high)
• Total Pending tests reported yesterday: 4,191 (very low)
• National reported case Growth Rate yesterday: 3.6% (very low)