Since it’s already Tuesday and I’ll grab another week’s data on Thursday, I’ll make this quick.
There’s large amounts of day-to-day variation in the new cases by date of illness onset in the raw data and even in the 3-day windowed average, but there’s a lot less in the 5-day windowed average (as we’d expect given it’s an average over 5 days). The 5-day windowed average shows that cases stopped their rapid growth after July 8 and stabilized for a bout 10 days. Hopefully that’ll continue to be flat or even drop. Unfortunately, due to reporting lag we can’t trust the data since July 23 (that’s the point after which we can’t trust the data to be within 80% of it’s final value) and even before then the data could still increase by another 20% due to reporting lag. The cases graphs are shown in the gallery below (top row).
The percentage of tests showing positive COVID cases stopped climbing as well, but that could be an illusion. I haven’t collected or mined enough data from previous weeks yet to determine how long this takes to settle, but a quick subtraction showed that literally every single day from March 12 to July 27 (the latest date I had data for given the datasets in this update) had been adjusted a little, with the largest changes back in March and April. The graph at right is the positivity graph.
New hospitalizations by date of illness onset may have peaked, but there is so much variation in the hospitalization data that this is quite likely to change. In fact, the raw data has the local peak on July 18 while the 5-day windowed average has the peak on July 9. The 80% threshold I’ve assigned as “trustworthy data” starts on July 15, so the raw data peak is very likely to go up and may go up quite a bit. Even the July 9 data could go up as much as 10-20% still. So while new cases by date of illness onset may have stopped rising, new hospitalizations probably have not.
And new hospitalizations are a more meaningful metric of the transmission of COVID across Colorado than new cases. So it’s odd how people point to the drop in cases to prove that COVID is getting better when cases are falling yet point to stable or dropping hospitalizations as proof that the new cases can’t be trusted when cases are rising.
New deaths by date of illness onset continue to increase back in June and early July, exactly where we’d expect them to increase given long reporting lags and the mean delay between getting sick and dying of COVID (now 22 days, up from 19 in early July).
And new deaths by date of death continue to be under 10 and show large day-to-day variability. But new deaths by date of death are also increasing, with the largest local spike on July 28. This is well within the zone of “don’t trust this data yet,” however, so we should expect it to change pretty significantly before it settles down. Unfortunately, data this recent is almost always revised upward.
I’ve plotted the comparison of deaths by date of illness to deaths by date of death since July 1 at the right. As with last week when I introduced these graphs, you can see that the increase in deaths by date of death is showing up as an increase back in June and July in the deaths by date of illness onset. And I added the pink regions that show the period of time that the data is still changing so much that we can’t trust the results to be within 20% of the final value (and the final value is always higher).
Finally, I’d like to add my standard caveat here. I’m not an epidemiologist or infectious disease or other public health expert, just an engineer who has been tracking the progress of COVID-19 in Colorado using CDPHE data since April. If anything I say is contradicted by authentic experts, listen to them over me