Daily Status, Nov 16: It is getting bad; Please be careful and where a mask. Remember, some high-risk people need to go out and about

 Daily Status, Nov 16

Today is a full update

Reminder:  Any sections that are unchanged since yesterday are grayed out.  

 

I try to update every day between 11:00 and 12:00; however, because my updates are based on when the Virginia Department of Health (VDH) updates, it takes some time to analyze and collate the data. In addition, because I have my own health issues, there will be days when the update is late or is not published at all.  

 

The one thing to remember is that the day-to-day changes are minor. Any trend takes several days to identify and I will be looking at the data daily (that is easy) but may not update the blog.

 

Situational Awareness

 

Big picture: The fall surge is accelerating.  NOW IS THE TIME TO BE CAREFUL!

 

Local: The caseload in Vienna tripled since Sept. Be careful and wear your mask.

 

Yesterday, VA reported a record 2,677 new cases of COVID, though VDH reports that the number is anomalously high due to database issues over the weekend.   If, instead, we look at three day sum, we have a record setting three day stretch, with 5375 cases.  Looking at it individually, the data were reported by VDH on Nov 16, 2020 were nearly 4 standard deviation above the three-week bias-adjusted average of 1,252 cases.  In fact, each day in the last week were above the bias adjusted numbers.  We are experiencing rapid increase in case load for the last 8 weeks (since late Sept), averaging 10% per week, more than doubling our daily average of those 8 weeks.

 

Viewed over the last seven days, VA reported  a record 11,160 cases, or 1,594 cases per day, which works out to 19.2 cases/100K people. This is the highest verified weekly number during the pandemic, though it is likely the May numbers were significantly under-reported by a factor of four. The best way to demonstrate that is through hospitalizations which peaked at 1,600 in early May but are currently at 1080 (but increasing).  Since hospitalizations should not be impacted by testing levels that provides a strong indication that the caseload was probably higher in May, but underreported.

 

The trends over the last three weeks now are showing a minimal increase at 1.1% per day or 8% per week.  

 

By combining our current regional trends with the typical reporting for the day of the week, I expect about 1,546 cases tomorrow (Friday) with a 90% chance of the numbers falling between 1,200 to 1,993 cases

 

The testing numbers now show the percent positive to be above the 5% metric over the last week (7.65%) which is often used to indicate sufficient testing. This is concerning because, as the percent positive increases, it is possible that some cases are being missed as the number of positives is constrained by the testing availability.  VA is testing about 2.8% of the population every two weeks .  

 

When we look at the local ZIP code data, we see that the observed increases are almost universal across VA. I am comparing the current estimated % positive to that of one month ago. Note that almost all is a warmer color (further from blue and closer to yellow). This is an indication of the uniformity of the increase. As a practical matter, it means to be safe and careful, no matter where in VA you live, there is risk.


\
Current:


Oct 1:



 

Regions

In the spring, COVID-19 in VA was primarily a concern in the DC suburbs. Over three to four weeks, (from late May to early June), NOVA recovered and for about a month the disease was under control to the point that restrictions were eased. Unfortunately, in eastern VA/Hampton Roads, the easing of restrictions resulted in a surge in cases which peaked just before August 1st resulting in stricter restrictions in that area. Since then, with the exception of growth on college campuses, the disease has been stable, excluding the rural parts of the state where safeguards (social distancing and masks) are largely ignored. Starting in October our weekly case count has been increasing throughout the Commonwealth, except the SWVA, which may have peaked a few weeks ago.



 

Looking at the weekly case count, we see that the numbers are higher than at any other point in the pandemic.

 

Regional growth rates (in fraction per day) continue to show degradation over the past three weeks. Note: It is easier to show a decline when the prior numbers increased. The current growth rates for the different regions are shown below.

 

NOVA:                                            1.025--GMU

Central VA:                                     1.011--VCU

Hampton Roads/Eastern VA:         1.050--W&M, CNU & ODU

SW VA:                                           1.021--VA TECH & Radford

NW VA:                                            1.041 --JMU & UVA

 

The entire state is increasing with Rt=1.017

 

The following table shows the number per 100K for each region. Again, NOVA and Eastern VA are doing the best, and the mountainous regions in NW & SW continue having more cases. The concerning aspect is that in all regions the numbers for last week are significantly higher than the preceding three weeks. What is most concerning is that SW & NW VA are hitting significantly higher caseloads.

 

Daily Cases/100,000 

Region

Last month

Last week

NOVA

12.3

17.6

Eastern

10.1

15.1

Central

14.3

16.3

NW

12.7

32.7

SW

29.1

30.2

The following charts show all five regions of the Commonwealth over time.

The individual line charts show the unfiltered data per day, coupled with the trend lines. 

 

The trend lines show the different periods of growth.  

 

Early in the pandemic, the different parts of VA were functioning largely independently, with NOVA mimicking the northern states, and Hampton Roads mimicking the southern states. Since September 1, the regions have trended together. Starting in late September, NoVA and SWVA diverged from the rest of the state, a trend that continues today.

 

It is worth noting that all parts of the state are showing an increase in the growth rate over the last three to five weeks.

Note that the effects at both ends of the chart are probably artifacts of the (seven-day polynomial filtering I use for averaging); the filter is poorly constrained in the first and last few days of the time history.  For those technically inclined, the filter is called a Savitzky-Golay filter, basically a moving window polynomial filter. At the edge (first and last days of the time series), the filter will over-compensate for the trend as it is unconstrained. I recommend the Wikipedia article if anyone is interested in more information, or contact me.









 

Local/Northern VA:

 

After the early peak in May (~1,000 cases per day), NOVA saw a sharp drop in all COVID metrics, reaching a broad valley in mid-June (~200 cases/day), which lasted until around August 1st.  By Sept 1, NOVA increased to 300 but the caseload dropped to about 150 by late in the month. Since then we have had a steady increase averaging up to a current value of 440/day per day.

 

Fairfax Co.

 1.033

Arlington Co.

 1.033

City of Alexandria

 1.022

Prince William Co.

 1.017

Loudoun Co.

 1.035

The number above is Rt:  Rt is an exponential time constant, where the number of cases in a time segment is approximately, n=Ao Rt ^ t, where Ao is the number of cases at the start of the segment, Rt is the exponential growth rate, and t is the number of days since the start of the segment.  So, if Rt is greater than 1, it is growing exponentially, if it is less than one, it is decreasing each day.  

 

 

Another way to look at it, todays number are approximately the growth rate times yesterday's numbers.  This is the exponential time constant.  It is above 1 currently, meaning we are experiencing exponential growth. In addition, our case load is growing, as we are about 18/100,000; our estimated peak from may was around 30/100,000.  I expect to exceed that number by the end of the year.

 

Looking at the trends, the strong downward trend in daily case count we observed since around September 1st has ended.  We now see significant jump in cases in every jurisdiction. The cause of that is unknown but may relate to the cooler weather.  




The difference in the colors (contrast) in the NOVA map is increasing. In addition, the NOVA map is warming (as is happening throughout the Commonwealth). At this point, it seems likely that this is related to the fall surge others had predicted.

 



 

Most localities in NOVA have case counts near or above 10/100K/day.  In Vienna, for example, we were under five in late September but are now at 9.1/100K/day.

 

Region

Last month

Last week

Growth rate (%/day)

Fairfax County:

Vienna

10.8

12.0

 3.7

McLean

 7.7

12.5

 6.8

S. Alexandria

13.9

18.5

 3.1

Reston/Herndon

12.5

18.1

 3.9

Annandale/Fall Church

14.2

18.0

 3.7

Fairfax

 9.5

14.1

 6.1

Arlington/Alexandria:

 N. Arlington

12.0

18.0

 7.8

 S. Arlington

18.7

22.6

 6.6

 Alexandria

13.6

15.7

 3.3

 

Recently, the trends show two groups of rates for Fairfax County: McLean is doing the best.  All the other locations are grouped at a higher number -- almost every region is seeing small upticks in the last few weeks. 





 

In Vienna, the average daily case count has doubled; I do not why (I have not been out and about). While, the growth rate may have stabilized, the absolute numbers remain higher than they have been since May.  It is possible that the current 7 day average high is a random fluctuation, but the event is highly impropable.  More likely it is random fluctuations superimposed on a longer-termed tren2


Local Safety/Risk

I am attempting a different means of discussing risk. Risk is very personal-- people will react to COVID differently; however, we know how many people have died (and how many cases) for each age group in VA so we can get a case fatality rate for age groups. Also, based on serology studies, we know the case numbers are low by a factor of 2.4, so we can get an infection fatality rate (IFR).  We also know how co-morbidities play in. 

 

In the following tables, I have combined the current probability of a person being infected around Vienna to compute the probability that a person in a specific crowd is infected.  

 

Assuming one interacts with everyone in the crowd (big assumption), I assume if you interact with an infected person, you have a 50% chance of getting infected. I do not know what that rate is; it will be a function of how long you interact, how close, and if masks are used, etc. This is the big unknown.

 

T


hat gives me a probability of being infected based on the number of interactions. I then combine that with the IFR to estimate the risk of dying by age and number of interactions for people with and without co-morbidities.  

 

For comparison, the risk of dying in a car accident in per day is about one in three million. That is a baseline but none of us likely knows anyone who will not get into a car because of the risk of dying.

 

No Co-morbidities

#exposure        0-9      10-19      20-29      30-39      40-49      50-59      60-69      70-79        80+ 

         1   0.00e+00   4.73e-08   4.25e-07   1.89e-06   4.49e-06   1.26e-05   2.84e-05   4.58e-05   8.59e-05 

         5   0.00e+00   2.35e-07   2.12e-06   9.42e-06   2.24e-05   6.28e-05   1.41e-04   2.28e-04   4.28e-04 

        15   0.00e+00   6.99e-07   6.29e-06   2.80e-05   6.64e-05   1.87e-04   4.20e-04   6.78e-04   1.27e-03 

        50   0.00e+00   2.26e-06   2.03e-05   9.02e-05   2.14e-04   6.02e-04   1.35e-03   2.19e-03   4.10e-03 

       100   0.00e+00   4.30e-06   3.87e-05   1.72e-04   4.09e-04   1.15e-03   2.58e-03   4.17e-03   7.82e-03 

       500   0.00e+00   1.52e-05   1.37e-04   6.08e-04   1.44e-03   4.06e-03   9.12e-03   1.47e-02   2.76e-02 

      1000   0.00e+00   2.10e-05   1.89e-04   8.40e-04   2.00e-03   5.61e-03   1.26e-02   2.04e-02   3.82e-02 

      2000   0.00e+00   2.41e-05   2.17e-04   9.63e-04   2.29e-03   6.43e-03   1.45e-02   2.34e-02   4.38e-02 

Co-morbidities

 

 

#exposure        0-9      10-19      20-29      30-39      40-49      50-59      60-69      70-79        80+ 

         1   0.00e+00   4.73e-07   4.25e-06   1.89e-05   3.59e-05   6.31e-05   8.51e-05   9.17e-05   1.72e-04 

         5   0.00e+00   2.35e-06   2.12e-05   9.42e-05   1.79e-04   3.14e-04   4.24e-04   4.57e-04   8.55e-04 

        15   0.00e+00   6.99e-06   6.29e-05   2.80e-04   5.32e-04   9.34e-04   1.26e-03   1.36e-03   2.54e-03 

        50   0.00e+00   2.26e-05   2.03e-04   9.02e-04   1.71e-03   3.01e-03   4.06e-03   4.37e-03   8.19e-03 

       100   0.00e+00   4.30e-05   3.87e-04   1.72e-03   3.27e-03   5.75e-03   7.75e-03   8.35e-03   1.56e-02 

       500   0.00e+00   1.52e-04   1.37e-03   6.08e-03   1.16e-02   2.03e-02   2.74e-02   2.95e-02   5.52e-02 

      1000   0.00e+00   2.10e-04   1.89e-03   8.40e-03   1.60e-02   2.80e-02   3.78e-02   4.08e-02   7.64e-02 

      2000   0.00e+00   2.41e-04   2.17e-03   9.63e-03   1.83e-02   3.22e-02   4.34e-02   4.67e-02   8.75e-02

Age Distribution: 

I am not updating this section for the time being except for the charts.  I will leave it here as is for a while longer--at times it can be very interesting.  This is particularly so when specific age groups do not follow other groups. For example, teens and 20-somethings surged in early September while the other age groups did not due to the outbreaks at colleges.

 




Colleges

Overview: Things are static at most colleges. My data mining by ZIP code requires that I group them together when they share a ZIP code. 

 

My process combines the VDH h data and what is reported by the colleges. The assumption is that students feeling ill are going to seek health services; however, we have all heard anecdotal reports of people not doing that because they did not want to quarantine. The colleges usually update the dashboards after this post, or on Monday/Tuesdays. Except for Radford, all weekly updates are complete at this time and the numbers are encouraging (except for UVA).

 

As the process has evolved, it has become clear that I need to show active cases (over a ten-day period) in addition to cumulative cases.

 

I am assuming the colleges are promptly reporting their numbers to VDH. As it turns out, W&M is reporting the cases approximately two days after updating the dashboard. 

 

Note: I have been finetuning the following table to improve the estimated number of cases. Also, I am adjusting the pre-student caseload: to estimate the number of cases from students, I subtract out the average number of cases prior to the student arrivals from the reported cases. If the numbers are large (UVA, VA Tech, JMU & Radford), the impact is minimal. This makes it harder though to identify cases in more urban settings like ODU, VCU and CNU.

 


RED       Clear evidence for community spread

YELLOW There may be community spread; still ambiguous

GREEN   No evidence of community spread

 

College

% Positive

Cumulative

Last 10 days

 

Dashboard Cases (active)

 

% that had COVID

***

VDH Cases*

Est

Cases**

VDH Cases*

Est

Cases**

VA TECH

  5.6

   2146

 5369

    180

  189

67

 15.7 

GMU

  4.6

    176

  180

     62

   60

22

  0.5 

UVA

  1.0

    916

 1953

     41

   38

54

  8.2 

ODU

  8.2

    140

  283

     16

   22

33

  1.1 

JMU

  6.3

   1989

 8033

     74

   73

 22

 37.8 

CNU

  7.5

     53

  174

      1

    2

14

  3.5 

UMW

  6.9

      0

    7

      6

   11

3

  0.1 

RADFORD

 15.8

    776

 3738

     22

   75

19

 47.2 

VCU

  6.5

    292

  475

     63

   74

34

  1.5 

WL/VMI

  5.7

    223

  312

     28

   48

12/3

  8.1 

W&M

  2.6

     48

   50

     16

   16

<10

  0.6

   


 

College Communities:

When I started talking about communities the focus was on the safety for incoming students. Unfortunately, that concept has changed. Now we are seeing the colleges impacting the surrounding communities. If we look at the age distribution of cases in the communities of JMU, Radford and VA Tech (New River, and Central Shenandoah health districts), we see that, starting about four weeks ago, the number of cases for non-college age citizens has been increasing -- about two weeks behind the college-age curve. This suggests the disease is infecting the general population. 

So far it looks like about an extra 372 middle-aged and senior citizens have been infected; there have been 12 cases to 24 cases per day in those groups. Furthermore, in the five months from the beginning of the epidemic to early September, there was an average of 10 deaths per month; in the last seven weeks there have been 35 deaths. This is concerning as the return of students has appeared to result in 21 extra deaths in the Commonwealth today in just the New River and Central Shenandoah health districtsNone of the additional deaths were of student age. This trend has been observed elsewhere in the country where the college students infect the more vulnerable populations.





Attribution:

1) You can repost/ share this information in its entirety by forwarding the entire link, or,  2) If you want to share partial content, you must receive my permission. This is proprietary information and I need to make sure you understand what I am saying. If anyone sees that this work being used without attribution, please let me know as soon as possible. I am willing to have an informed discussion/debate on my approach, but I want to make sure the proper context is captured.

 

 

 

 

 

 

 

Comments

Popular posts from this blog

Daily Status, August 29

Daily Status, Nov 8: Every trend I was tracking 10 days ago remains. Which is not good.

October 24: Vienna is not doing well, Virginia is not doing well, the USA is not doing well (but I will not talk about the latter)