Here's an interesting point - there have now been ~22,000 confirmed or probable COVID-19 deaths in NYC With an IFR of 0.9% (from serology), that implies that roughly 30% of the population has been infected
-
Show this thread
-
Replying to @GidMK
Here's the April serology (which is even higher now) in the areas hardest hit in NYC so you may need to rethinking your IFR assumptions for NYC:pic.twitter.com/Yh1YJe6KGH
1 reply 0 retweets 0 likes -
Replying to @12FreeBeer
Actually the 0.9% is based on the April serology. Depending on how you account for right-censoring, the serology from April implies an IFR of 0.77-1% in NYC
1 reply 1 retweet 1 like -
Replying to @GidMK
You have to weight sero based on areas where people died. Sero of 5% in highly populated rich neighborhoods were few people died skew your IFR est high. If 80% of deaths are in areas with 40+% sero then you need to weight that sero at 80%. That's NYC. Your IFR est is too high.
1 reply 0 retweets 0 likes -
Replying to @12FreeBeer @GidMK
A correct indicative serology sample to determine IFR is based exactly proportionally to the areas where deaths occurred. If 80% of the deaths are in 25% of your zip codes then your sero sample should be populated 80% from those zip codes not 25%. NYC sero sample didn't do that.
1 reply 0 retweets 0 likes -
Replying to @12FreeBeer @GidMK
To calculate IFR off of the broad based NYC serology is nonsense. That sample isn't appropriately weighted to where deaths occurred. All of these "back of the envelope" IFR calcs overestimate IFR & in NYC its a significant overestimate given how high sero is in heavy hit areas.
1 reply 0 retweets 0 likes -
Replying to @12FreeBeer
Epidemiologically, that's basically nonsense. You would expect both the seropositivity and death rates to vary by suburb, due to demographic and other factors, and so weighting by seropositive tests would unnecessarily bias your final number
1 reply 0 retweets 1 like -
Replying to @GidMK
It's math. If 100% of your deaths are in 10 neighborhoods representing 10% of your population, then sampling in the other 90% is useless. It's like including samples from Kansas when you're trying to figure out the IFR in Arizona. Your sero sample has to be where deaths occur.
3 replies 0 retweets 0 likes -
Replying to @12FreeBeer
Nope, that'd give you a totally useless estimate. The serosurveys should ideally be population representative and randomly sampled, otherwise all you're doing is picking a different bias for your sample
1 reply 0 retweets 3 likes -
Replying to @GidMK
Wrong. You take random broad demographic samples in the communities where people are dying & weight those samples based on community death proportion to figure out IFR & IFR by demographic. It's exactly how people who actually have to earn a living underwriting such risk do it.
1 reply 0 retweets 0 likes
Demographics absolutely, one of the weaknesses of current estimates is that they are not at least age-stratified. I think what you're describing is simply using smaller blocks for IFR - not entirely unjustified - not 'weighting' by suburb (which would be a bad idea)
-
-
Replying to @GidMK
Yes, IFR by borough, neighborhood & apartment complex as NYC is very heterogeneous. Have to make sure all appropriately represented (it’s not in NYC sero study, sample is also WAY too small). Some apartment complexes (& all care homes) with outbreaks not even included in sample.
1 reply 0 retweets 0 likes -
Replying to @12FreeBeer
Sample definitely isn't way too small, it's far more about design than specifically large sample sizes. Some of the more robust epidemiological surveys in the US have only 10k participants total
1 reply 0 retweets 0 likes - Show replies
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.