Thread by @rubiconcapital_

Pointing out some interesting peculiarities in the recent Ontario #COVID19 case and mortality data. This brief thread is a bit technical.

These peculiarities are similar to those identified in the UK data by @ClareCraigPath and @RealJoelSmalley

https://twitter.com/RealJoelSmalley/status/1348303550875627527?s=20

The % Positivity and daily deaths by Date of Death trends meaningfully diverged for the first time during the first two weeks of November…

In the 1st wave, correlation between (i) deaths by date of death and (ii) fatal cases with an episode date 12 days prior was almost perfect, with R=0.983 and R^2=0.976…

The second chart shows the onset-to-fatality lag correlation signal…

In the 2nd wave, this correlation between episode date of a fatal case and deaths 12 days later weakens slightly (R=0.966, R^2=0.933)…

But the lag correlation signal is much weaker…

In the 2nd wave, starting in early November, the correlation impairs materially, and some “noise” appears to enter the dataset… (R=0.758, R^2=0.575)

Here are all three correlation/regression graphs shown again together… 1st wave, 2nd wave, 2nd wave (Nov 1 to Dec 15 only)….

Essentially what is implied by the data, is that the relationship between (i) symptom onset date and (ii) date of death *changed*, for the 1st time ever, beginning in November… consistent with divergence in %pos and # of deaths noted above. I don’t have an explanation for this.

We can see this visually a second way… a log chart showing the clear lag between (i) the episode date of fatal cases and (ii) deaths by date of death in the 1st wave, while the lag in the 2nd wave is a bit less pronounced, especially beginning early November...

One interesting (I think related) phenomenon occurring in Nov & Dec, however, is a material increase in the percentage of fatal cases (especially in an outbreak setting) that have a matching symptom onset date and test date… big change from first to 2nd wave and in Nov/Dec…

Same chart… perhaps it is that testing became more efficient in Nov/Dec (especially in outbreak settings) driving prevalence same day symptom onset and testing…

Either way, the basic question is: why are more and more fatal cases having matching symptom and test dates?

Putting the above out there for comment… @ClaireCraigPath / @RealJoelSmalley / @NickHudsonCT / @abirballan / @PanData19, any thoughts?

All data from the Ontario Open Data Catalogue... data.ontario.ca/en/dataset/con…