💾 Archived View for gemini.susa.net › tweeters › tweets_trvrb.gmi captured on 2020-10-31 at 01:15:51. Gemini links have been rewritten to link to archived content
⬅️ Previous capture (2020-09-24)
➡️ Next capture (2020-11-07)
-=-=-=-=-=-=-
Bio: Scientist @fredhutch, studying viruses, evolution and immunity. Collection of #COVID19 threads here: bedford.io/misc/twitter/
Location: Seattle, WA
I think this is the best explanation we have for what columns are in this file: nextstrain.github.io/ncov/data-prep
nextstrain.github.io/ncov/data-prep
Thanks Will! I'll try this out.
I could well imagine sustained daily case counts similar to July / now of over 50k per day for multiple months. I wouldn’t think sustained case counts of over 100k likely. This is given continued local responses to outbreaks to tamp down on transmission.
Hmm… Was July vs October surge in Utah in different parts of the state?
Interesting. To see if I’m following, would you detrend by fitting a linear regression and then taking residuals? Or would you do something like looking at increases / decreases in weekly case counts (ie look at deltas rather than values)?
Agreed. If you look at the network plot there are certainly a few states that don’t fall quite as neatly into these groups. Meant more as broad patterns.
I don’t think we know. There’s more work on these patterns in seasonal influenza. Two big takeaways there are:
1. Well connected metropolitan areas tend to have earlier flu season
2. Climate affects things with sharper epidemics in the north and shallower in the south
It’s unadjusted daily case counts. However, correlations are looking at relative peaks and troughs between timeseries. So it should be largely similar to cases per 100k (but I should confirm).
(Sorry to be absent these last couple months. I'll try to get back to more regular updates.)
Seasonal coronaviruses show seasonal circulation patterns. I expect that moving into fall and winter will make controlling COVID-19 more challenging, but I would generally expect continued circulation at case levels not dramatically different to those of the past months. 12/12
In group 3 are primarily states in the Midwest and the Mountain West that were not involved in either the first or second waves and that have seen steady increases in case counts since July. 11/12 pic.twitter.com/eJ8HI3nxIf
In group 2 are primarily states in the Sun Belt that were involved in the second wave, peaking in July and August. This was followed by sustained circulation but at lower levels than the summer peak. Many of these states have also been creeping up in the past ~6 weeks. 10/12 pic.twitter.com/UGz75MSydL
In group 1 are primarily states in the Northeast that had a first wave in March and April, followed by little summer circulation, but where case counts have been slowly creeping up in the past ~6 weeks. 9/12 pic.twitter.com/ns6Y5I9h6R
Network connections obviously differ based on the the threshold chosen. But I get very similar groupings at thresholds above or below the 0.60 correlation coefficient used here. 8/12
There appear to be three different "communities" in this network, or groups of states that resemble one another in their timeseries of confirmed cases. I've labelled these simply as groups 1, 2 and 3. 7/12
If we look at all pairs of states we can construct a network diagram where each state is a node and edges connect states that have a correlation coefficient of 0.6 of higher, ie states with similar epidemic timeseries. 6/12 pic.twitter.com/Y6OLbncFy0
We can look at correlations between states. For example, when case counts are high in New York they are also high in New Jersey (correlation coefficient of 0.95), while when case counts are high in New York they are low in Texas (correlation coefficient of -0.38). 5/12 pic.twitter.com/MNeEPwaJK0
Keep in mind that these numbers represent reported cases and that the proportion of infections reported as cases has increased over the course of the US epidemic as testing capacity has improved. 4/12
Using data from @COVID19Tracking, I plot daily confirmed cases for each state since March as a stacked chart. The three crests are obvious (though not clear how large the third will end up being). Different regions are contributing to each wave to different degrees. 3/12 pic.twitter.com/cTwk9LRgi8
I start with a simple coloring to group states in the West (in red), the Southwest (in orange), the Midwest (in green), the Southeast (in blue) and the Northeast (in purple). Color ramp borrowed from @andersonbrito_. 2/12 pic.twitter.com/Txc76DiWqY
Daily #COVID19 case counts are increasing in the US and we seem to hitting a third wave (or second surge if you'd prefer). Here I wanted to look at how case counts through time correlate across different states. 1/12
That’s a really great figure. Could I ask for it’s reference? Also, I strongly doubt that this effect came from viral evolution. Would explanation be that secondary infections of 15-35y cohort aren’t as deadly due to immunity after high attack rates in 1918-1919?
Our new Situation Report is now available:
nextstrain.org/narratives/nco…
8 months into the global transmission of #SARSCoV2 #COVID19, we look at the changing patterns of transmission we can see through viral genetics, including the initial fast spread & the effects of lockdown
1/3 pic.twitter.com/1h9GaQppOB
nextstrain.org/narratives/nco…
I agree with this.
This was my take as well. Tragic that we’re in this situation.
My impression was that false positives that occasionally occur are commonly from lab handing errors (contamination / swapped samples). The background rate of positivity will affect the FPR in this case.
I’m certainly not proposing this as a strategy. Just trying to understand why we see epidemics wax and wane. BTW Texas has had 528k confirmed cases in a population of 29M for a floor of 1.8% infected. We don’t catch every infection as a case probably only like 1 in 5 or so.
Effect of NPIs and effect of immunity should be multiplicative and not additive.
twitter.com/trvrb/status/1…
Interesting. I was assuming that 1 in 4 was something of a floor just due to spectrum of disease. Asymptomatic infections won’t seek out testing and many mild cases won’t either.
I hadn’t seen this. Thanks for sharing.
I believe a ratio of 1:6 is generally quite reasonable (based on recent CDC seroprevalence estimates).
Follow up #1: Some helpful further details to these estimates to consider from @youyanggu
twitter.com/youyanggu/stat…
Hi Youyang. Really good points here. Thank you. I’d obviously agree with the 4-8X reporting spread. And yes should take into account seroprevalence today vs when today’s confirmed cases were infected. And yes, there will be undereporting of deaths as well.
It looks like just accounting for lag in reporting of deaths gives consistent estimates for IFR: mobile.twitter.com/trvrb/status/1…. About 0.7% IFR assuming 10% infected and about half that with 20% infecting.
mobile.twitter.com/trvrb/status/1…
Seasonal flu has disappeared from circulation in Australia (during their regular flu season). Current societal experiment seems enough to drop flu’s normal R0 of ~1.5 to below 1. Curious if we’ll have a flu season in the US this winter.
I hadn’t seen this. Thanks for sharing. I’ll look into it.
Based on mentions, it was one of the most controversial things I’ve posted. mobile.twitter.com/trvrb/status/1…
mobile.twitter.com/trvrb/status/1…
I don’t have real data here, but my assumption has been that people are behaving based on perceived risk, with older individuals, on average, adhering more to social distancing.
I’ve commented on this previously: mobile.twitter.com/trvrb/status/1…. There has definitely been a shift in age of infections from March / April.
mobile.twitter.com/trvrb/status/1…
My best guess for fraction of Florida infected is about 12%, ie 525k cases ⨉ 5X ratio / 21.48M, though I could believe between 8% and 16% as reasonable. Lower than the 20% I threw out yesterday, but still high enough to impact epidemic spread if behavioral Rt is ~1.2. 16/16