The infection time line  
By mid April, the Covid  infection had rapidly spread
across the US  (40 states in only 13 days) and  wide  
variation in severity  (50x variation in total deaths, with  
the North East was the center of the pandemic.  A sort of
the total deaths, showed that there was a pattern of
neighboring states with declining levels of infection over
time, which consistent with community spread through hot
spots.


The only way the virus can get from New York to
Michigan is by flying as there is not a neighbor path.
Infections spread over time so I used the date of first
infection and then neighbors with declining infection
levels and later dates  to  create a map of the spread.



In the early stages, the infection appears to have started
in New York, and propagated through the industrial North
East.   I then colored the states based on the date when
5 deaths were first reported – each day has a different
shade, but roughly speaking Red is early to mid March,
Purple mid to late March, Blue is April.  
There is a pattern of spreading in infection. The first
infections occurred in a few early (red) hot spots, and
then spread to neighboring states with progressively later
(bluer) and smaller infections. WA was first to detect but
everyone was wary of the risks from China and so the
infection was minimized. NY was obviously worst hit, and
the progression of serious infection to neighbors can be
seen through the north east to NJ, CT, MA, MI, RI.
Within 4 days there were also  hot spots detected in LA,
MI, CO, GA but with smaller infections. These presumably
come from visitors flying in from NY with Mardi Gras
causing the biggest problem. Using the observation of
spreading in the north east as the model, each of these
new hot spots spread to their neighbors again with
smaller and later infections, the spread in the north
central region from MI to Il, IA, IN is the most sever. Texas
had a much later and smaller infection, so cancelling
SXSW was crucial. I show some possible spreading paths
created using the date of first infection for the
neighboring states and connecting them together
illustrated with the green arrows.










It looks like we can make sense of the path of the
infection, but why are there such large variations in
infection level. The next stage was to see if multiple
states could be modelled. I chose the state with the
highest infection level NY, the state that started earliest
but had one of the mid-levels of infection WA, and one of
the latest states that had one of the lowest levels of
infection TX.

The graph shows the 7 day average of daily deaths
along with model fit. In the model, the patient zero date
and the isolation date were adjusted for best fit. In NY the
death rate was nearly 100x higher than WA before they
isolated, as a result the infection took off. TX isolated
very early just like WA and had very low levels. Under
isolation, NY is declining fast, TX on the other hand has
started to reopen and the infections are rising again.  

My take away is that the model does a decent job of
tracking a wide range of infections, which further
suggests that the variation in infection level is simply a
function of timing and effectiveness of isolation. The
usual disclaimers apply, there is a lot of uncertainty
about the raw data in this fast moving problem.

The model parameter “fraction of deaths per infection
per day” was different between NY (1.1%), WA (0.6%)
and Texas (0.4%).  These differences are similar to
those reported for Case Fatality Rate (CFR) measured
by total deaths/total new cases. This will be the next
issue to focus on.



Patient Zero for TX was probably asymptomatic and
arrived in the first week in March. Based on changes in
mobility measured by cell phone activity, the TX
population were starting to modify behavior in the same
week. In early March the infections rose rapidly with 1
person infecting roughly 5 others.  By Mid March, the
State has issued “stay at home” orders. In late March,
the change in behavior was reflected in the Infections per
person (Ipp) starting to fall. By mid April, the infection had
stalled and just started to fall. The infection in TX was
relatively small, and in late March, it appears that people
has already started to relax their self-isolation, and in
early April infections started to rise again  transitioning    
to a modest level of Ipp of 1.3.






















Status Late May

The map for late May shows that for the average daily
cases per million are much more uniform across the
country compared to late April. The red color code shows
states with rising infections. The western states that had
low infection levels a month ago are now rising. TX in
particular is rising at around 5% a day. The poor rural
states such as Arkansas are also showing rising
infections. It appears that reopening has yet to show up
in cases levels.





The graph shows the trend for Texas with a steep rise
over the last 9 days. There is no way to know if this is a
long term trend. Unfortunately, this is what we should
expect to happen with re-opening. If this continues, we
reach worst case NY levels of infection by the end of
August. Another week will be telling.

Is increases testing to blame ?

Let’s put this testing BS to bed once and for all.
Hospitalization admissions are tracking cases for Austin
Texas, which conclusively shows that the case count is
real, not a testing artifact.  

If you get sick, the vast majority get counted based on
CDC criteria “A COVID-19 case includes confirmed and
probable cases and deaths.” A bunch of people with
some symptoms pass the screening questionnaire and
get tested and they are positive about 10% of the time.
These positives only increase the case number if they
were missed by the health case professionals OR DID
NOT GET SICK EVENTUALLY, in other words are
“asymptomatic” or “mildly symptomatic”.

The case number only gets inflated by the number of
asymptomatic’s who randomly happen to get tested.
Today, there are probably 1000 new individuals per day
per million who are asymptomatic (10x cases) and they
are infectious for around 10 days, so a total of 1% of the
population. The inflation of the case count every day
from the random chance that an asymptomatic gets
tested is about 1% of the current case count of 100
cases/d/M, or about 1 case. In other words, single digit
cases - NEGLIGIBLE.

Increased testing only increases the count from
individuals get sick and who are missed by the health
care professionals.

Increasing hospitalizations and hospitals running out of
beds are conclusive indicators that the infection is
growing.
Sorted list of the states by total deaths per million,
suggesting neighbor to neighbor community
infection.
Plausible map of the spread of Covid based on infection severity (pillar
height), date of first death (color code) for neighboring states.
The time line of deaths per day per million, for NY, WA and TX. The date of
patient zero and date of starts of isolation have a huge effect on the  severity of
the infection.
The time line for the infection in TX showing the probable date of arrival of first
asymptomatic individual, followed by the early change mobility, the impact of
isolation causing the start of the transition to lower infections per person. In
TX, there was an early move to reopen that caused a small increase in
infections per person.