So many folks are doubtful of specificity of the NYS test. The test probably has a specificity north of 97%
How do we know?
*Thread*
How do we know?
*Thread*
Let's start here.
Outside of NYC and suburbs, there were 32.8% of N=3000 tests done.
This means 984 tests.
Outside of NYC and suburbs, there were 32.8% of N=3000 tests done.
This means 984 tests.
3.6% of those 984 tests were positive.
This means 35 tests were positive.
Let's take essentially the worst case scenario that all 35 were false positives.
This means 35 tests were positive.
Let's take essentially the worst case scenario that all 35 were false positives.
That means 984-35=949 were true negatives.
The specificity is TN/(TN+FP).
This means specificity = 949/(949+35) = 96.4%
The specificity is TN/(TN+FP).
This means specificity = 949/(949+35) = 96.4%
We can calculate the 95% confidence interval here:
sample-size.net/confidence-int…
But to save you time it is 95.1% - 97.5%
And that assumes nobody in nearly 1000 people had the antibodies.
sample-size.net/confidence-int…
But to save you time it is 95.1% - 97.5%
And that assumes nobody in nearly 1000 people had the antibodies.
And if even half of the cases were true positives, then the specificity will rise accordingly.
Now keep in mind we know nothing about sensitivity, which would drive up the prevalence in NYC if it is <100%. How much? About 1% for every 5% decrease in sensitivity
There is also a latency to develop seropositivity. The exact amount will depend on the assay thresholds.
But it is very likely another 1-2% will seroconvert because of this. More if the samples were collected a while back.
But it is very likely another 1-2% will seroconvert because of this. More if the samples were collected a while back.
Keep in mind there are tons of big name virologists who are not liking this result for various results.
There is only one avenue wherein it could be massively wrong.
That is if they sampling somehow strongly enriched for people who were symptomatic.
There is only one avenue wherein it could be massively wrong.
That is if they sampling somehow strongly enriched for people who were symptomatic.
That would mean people were more likely to go for the test if they thought they had COVID19 either because of sx or friends/contacts with COVID pts.
So, let's think about that for a minute.
In some high prevalence areas, the people who got a CLINICAL test had ~50% rate of having COVID19. So half of these high risk people with symptoms had other causes (viruses, allergies, etc.)
In some high prevalence areas, the people who got a CLINICAL test had ~50% rate of having COVID19. So half of these high risk people with symptoms had other causes (viruses, allergies, etc.)
This implies a VERY strong sampling bias would have to have occurred to shift the 20% down to low single digits like some people are arguing.
Seems quite implausible to me.
Seems quite implausible to me.
The only other explanation is if the tests conducted outside of NYS + suburbs were someone not as good as the ones in NYC.
That could happen, but seems implausible.
That could happen, but seems implausible.
@threadreaderapp unroll
