The COVID situation here in Florida continues to improve based on new case and hospital admission data. Based on the fatalities curve — well, that is hard to say. That was the subject of my diary from 9/22/2021.
In case you want a closer view of the title picture, here’s an embiggen link
I ended that diary thinking by this point the expected fatalities plateau implied from the new cases data would become more prominent — well, maybe it has, but not as strong as I had thought. For the current “emerging plateau” graph I lowered the steps between data reports to 10 fatalities per day rather than 12; I also lowered the expected CFR for my Bedford curve from 1.82% to 1.74% (1/55 to 1/57). Still the emerging plateau seems to be getting harder to see…
A suggestion from the comments in my last diary was for me to look at the hospital admission data. I was able to find it on the CDC site, but had to manually transcribe it. To my surprise it seems that the admissions and new cases curves in this state are nearly coincident when scaled at 1:10...odd, but it seems to support my contention that most COVID tests here in Florida are being done once people are already quite ill.
I was tempted to write earlier in the week, but held off and instead tried again to model the data fill in process. To do this I took the ratio between the reported 7 day moving average on the same days between data reports, and then averaged the differences by “day” (not date)…I got this series of curves:
I then took the 9/27 data and iteratively applied the percentages up to approximately mid October. I found that the final estimate was a bit shy of the expected Bedford curve (which was when I lowered the expected CFR — the curve looked better, but there was still a gap)
When I looked at my estimate for today to the reported numbers, I was impressed — although the first 2 days were off significantly more than the rest of the curve.
But it should be expected that the further back in time, the better this method gets; the the error correction factor approaches 1 fairly fast. It was as I was creating the graphics for this diary that I saw how out of place the odd ratios from the initial data pull I grabbed on 9/10 were. I was prompted to recalculate the average fill in factor:
...And saw the estimate for today get really close to the reported numbers
...And the new estimated fatalities curve for mid month was much closer to the Bedford estimate after performing the next 4 iterations
In light of the little discrepancy between the new cases and admissions for the 2 weeks between about 8/24 and 9/8, maybe the Bedford curve would be better calculated off the latter? Yeah, you just know I can’t say no to doing that:
In another 10 days or so I am expecting 3 more reports to be published. Barring the very unexpected I hope to capture these and compare my estimates to the reports. As for the new cases and admissions —
First, I will continue to try and find a source where I can easily convert the available page source data to spreadsheet data for admissions. Barring that it should not be too hard to keep up with the new data by hand, as long as it is not revised much retrospectively.
Second — the new cases and hospital admissions curves seem to be flattening...so I’ll be continuing to look for signs that new cases begin to increase again. Knowing this state it’s to be expected.
Thanks for reading — stay safe!