Here’s the current state of the primary analytics that I maintain on a daily basis. The data for these comes from the Johns Hopkins University and from the AZ DHS. Occasionally the data diverges between these two organizations, but not wildly.
I thought it might be interesting to share what I look for in these tables and graphs, especially now that the case rates are slowing down in Arizona and across much of the US.
The above table is familiar to anyone who has read one or two of these pages. It sorts each state by their overall case growth rate. I call this the Instantaneous Rate of Change (IROC) because what I’m measuring is the current slope of the COVID Case curve. The states can be compared because the COVID cases are normalized by each state’s population. This table can help one to quickly identify where the hotspots in the US are. The Confirmed Case IROC tells us how many Cases per 1000 persons per day each state is experiencing. I sometimes refer to this as the case growth rate. The dIROC_Confirmed column tells us how fast the case growth rate itself is increasing. Above, we see that Arizona has the largest case growth rate in the country but that this growth rate is decreasing by 0.0275 cases per 1000 per day — every day. When a state’s dIROC_Confirmed starts slowing, it is an indicator that the case growth rate peak is near. When the dIROC_Confirmed goes negative, it is a sign that the peak is in the past and the case growth rate is now decreasing. I captured this table all the way down to California, who about 2 weeks ago had the highest case growth rate in the country by a pretty good margin. As you can see their case growth rate continues to decrease every day. It looks like Arizona will be seeing similar numbers in about a week since it’s peak was about a week and a half ago.
This table also tells us interesting things about cumulative numbers. You can see that Arizona, Rhode Island, and Utah all have very high cumulative cases per 1000 population. This probably reflects the overall populations of these states being smaller or more confined to larger population areas. Much of Rhode Island’s population is near Providence, Utah’s population is primarily in the Salt Lake area, and a majority of Arizona’s population is in two metro areas, Phoenix and Tucson. Also, most of these regions have seen recent case surges while states like NY, MA, and NJ have had a long lull since their primary surge back in March. One other metric on this chart that’s useful to understand the state’s situation is Deaths per 1000. The Northeastern states still have the largest number of deaths per 1000 persons, but other states like Louisiana, Mississippi, and Arizona are catching up.
Arizona Case, Death, and Hospitalization Data
Now I’m switching primarily to Arizona numbers. These may or may not be similar to what is seen in other states, but I have pretty good access to Arizona data as well as the motivation (being an Arizona resident) to tracking them more closely.
The above chart is the visualization of the Case per 1000 curves for five different age demographics in Arizona. During the initial summer case surge, it was clear that all the age groups between 20 and 54 tracked together and had the same case rates when normalized by their population numbers in the state. The 65+ group has tended to have lower cases per 1000 and the under 20 group has far lower cases per 1000. This chart tells us that this trend continued until about mid-November, when the 55-64 group really separated from the herd (this is the green line above). It also appears to the naked eye that the 65+ line (maroon) also had a slightly increased slope over the under 55 groups starting around mid-November. As I’ve mentioned in other postings, this likely represents contribution to cases from winter visitors to Arizona, none of whom are represented in the Arizona population for their age group. Therefore, they’re contributing to the numerator (cases) without contributing to the denominator. This may just be a curiosity, or maybe it has a hand in Arizona’s particular COVID case and hospitalization challenges since the beginning of October. It appears to me too that while the age groups under 54 (red, yellow, and blue) have started significantly flattening out — meaning that their growth rates are slowing) — the over 55 groups seem to be slowing less.
The above is a chart that I look at a lot and like because it compares the case curves for Maricopa County (Phoenix) and Pima County (Tucson). This is interesting as a natural experiment, because it seems provable that Pima County has imposed much more structure around COVID protocols than Maricopa County and has done so pretty consistently since the beginning. The Blue line is the polynomial equation fit to the Maricopa actual case data (light grey) and the Red line is the equation for Pima actual data (dark grey). The change in the current surge rates is revealed where the real data diverges from the polynomial. This started around the first week of January, but I could see signs of it beginning at the start of 2021. The other interesting signal in this chart is the moving average of tests per day in the state. I’m only showing the numbers of tests, not the positivity of these tests, in this chart because my theory is that the number of tests is a leading indicator for case growth (this likely measures how sick people think they are… few people get COVID tests for fun, I’ve observed). Looking above, you can see that the number of tests moving average hit a low around 9/08 when the case rates were increasing only a slight amount and appeared to be doing so linearly. The case rates started curving upward somewhere starting early October and the test numbers increased along with it. The test number peaked somewhere around the holidays and then gradually slowed. This may have anticipated the case rates starting to slow somewhere in early 2021. This may or may not really be a leading indicator, but the data I have makes it seem interesting at least.
Comparing Deaths over 65 with under 65
The over 65 population in Arizona makes up about 13% of everyone in the state, but has experienced most of the deaths. At some point, I started tracking the Over65 to Under65 ratio as an indicator. You can see the above as the green line. This is the 5 day moving average of this ratio. The light blue bars are raw deaths for the under 65 group (87% of the population) and the red is the raw deaths for the over65 group. The right way to show this data would be normalizing it by each group’s population, but if I did that, the blue wouldn’t be visible unless I plotted it on a logarithmic scale (very hard to explain to non-math majors). The green line is maybe the most interesting element of this because of the comparison of the ratio during the smaller summer outbreak (June/July) where the ratio hovered around 2.5 with the ratio during the current surge, where the ratio jumped quickly and has been ranging between 3.0 and 4.0. This is hard to analyze, but it appears like the virus is more deadly now for the older demographics than it was during the summer. Perhaps the cooler weather compounds the virus’ effect or perhaps there are less susceptible persons under 65 now than there were during the summer? If you’re curious about the spikiness of the ratio during the lull from August until October, I’m pretty sure it’s just due to the small numbers of deaths during that time frame. Small data’s statistics can behave oddly.
Above is what I call my “experimental” metric on a ratio of current hospitalizations to cases from one week earlier. As you can see above, I do this ratio for each age demographic. The basis for this analytic is that I was curious about the rate that people get hospitalized one week after they are confirmed with COVID. Perhaps 2 weeks might have been a better gap, but I chose to use one week because I felt that might be a more common time frame between a case and a hospital visit (any kind, ER, ICU, Outpatient, etc.). Each datapoint represents the daily ratio and since the trends are hard to see with the naked eye, I fit a regression line to each age group so I could see the trends. It doesn’t show it here, but though all the trendlines on this ratio show a decreas now, but back around November, the over 65 group’s trendline was increasing pretty steeply. This changed sometime around the holidays and all the age groups have been trending down ever since. This indicates to me (but doesn’t prove, of course) that hospitalization surged in the over65 group pretty strongly early in this outbreak. It was likely this over65 group that filled the hospitals (see below) very rapidly starting November 1. You can see above that the maroon and green (over 65 and 55-64) dots around that timeframe frequently ranged between 50 and 75%. This means that the number of people hospitalized during that timeframe over 55 was 50-75% of the Confirmed Case counts for those demographics one week earlier. Now the numbers for over65 are generally under 25% and the 55-64 are generally under 10%. This is not a perfect metric, but it does seem to be illustrating a trend and was possibly a good leading indicator for the COVID hospitalization easing that can be seen below.
Finally, I’ll re-show the latest from the AZ DHS on the breakdown of COVID, non-COVID, and empty ICU beds in the state. This is a stacked bar chart and the three states above add up to 100%. So for instance, at the hospitalization peak around Jan 5th we could see just over 60% of the ICU beds in the state occupied by a COVID patient, about 30% of the ICU beds occupied by a non-COVID patient, and about 10% of the ICU beds empty (my guess is that these were the broken ones??). Today, we see about 50% COVID, 40% non-COVID, and still 10% unoccupied. If this follows the trend from the summer, we’ll soon see the COVID hospitalization numbers start dropping even faster.