I’m going to post my thoughts regarding the NYC antibody tests in full. I’d love for people to point out the faults in my logic.
The antibody tests used detect IgG antibodies, which develop on average after 14 days — and 95% of which have developed after 21 days. The specificity is appraised at between 93% and 99%, meaning somewhere between 1-7% will show as a false positive. The sensitivity assumption is lower, and the pool of true negatives vs. true positives is skewed highly towards false positive prevalence over false negative prevalence.
The audience considered for testing are grocery-store shoppers, which is likely to bias the infection rate higher than the city-wide average, as daily shoppers are more likely to be represented and are more likely to have been infected. Cautious people who rarely leave their homes are less likely to be represented and less likely to be infected.
Average time to death is appraised at a 5 day average to symptom onset, upon which an 18.8 day average to death. This means that roughly half of deaths will have occurred 23.8 days after infection, or that antibody results on average develop 9.8 days slower than death total counts, ranging from 9.8-2.8 days behind. This means that the death toll from antibody prevalence is not fully realized in the statistics for up to a week after testing.
With these caveats in mind, let us look at the results.
New York City deaths:
Positive swab-year deaths (high prevalence at hospitals): 12,287 (April 26th)
Clinically diagnosed deaths (high prevalence at hospitals): 5,228 (April 27th)
Total: 17,515 deaths
Excess mortality: 20,900 (NY Times)
For the sake of this, we will assume that the excess mortality is not comprised solely of SARS-COV-2 deaths, but it stands as an important marker in realizing the death toll of this virus. We can assume that wrongly clinically diagnosed deaths can replaced from the excess mortality source instead, meaning this number stays constant. The majority of deaths are recorded at hospitals, meaning techniques such as Lung CT scans for diagnosis have instead been used in lieu of swab-tests.
New York Population 2019: 8,330,000. This means that 0.21% of the city has died of the novel coronavirus according to the official death count. This leads us to the first antibody study they concluded just a week ago.
21.2% antibody prevalence in New York City. Keeping the false positive ratio in mind, this gives us anywhere from a 14.2%-21.2% infection rate. This study is interesting due to the fact that the average time to death vs antibodies is now reflected in the statistics for this test.
21.2% of 8,330,000 = 1,765,960 individuals.
17,515 / 1,765,960 = 0.98% of all infected have died. This is our absolute lower-end estimate.
—
The recent antibody results from yesterday indicate a 24.9% infection rate, meaning anywhere from 17.9% - 24.9% have been infected. This will be fully realized in the statistics next week as the average time to antibodies/average time to death is matched.
24.9% of 8,330,000 is 2,074,170 individuals.
17,515 / 2,074,170 = 0,84%. This is the lowest bound our IFR can be moving forward.
Now, there are many factors regarding the death total that must be adjusted for in the search of the IFR. I will name them below but we shall not adjust for this.
Firstly, the relatively young population in New York City will skew the IFR lower. One in eight residents of NYC are 65+, comparable to around one in five in most European nations. Considering the lethality of this infection rises considerably with age, this population distribution likely effects the IFR negatively when comparing to Europe.
The health of New York City residents is remarkably poorer than that of European nations, with a higher obesity and diabetes rate. However, obesity is not remarkably over-represented as a risk factor, with old age remaining a deadlier risk factor than either obesity or diabetes. This will lead to the lowering of the IFR in comparison, but when adjusting for the population distribution differences, the IFR will still skew higher.
Unresolved deaths/the state of NYC hospitals. Currently around 780 patients are in intensive care in New York City, a marked decline from their peak. Mortality rates are around 90%, meaning that roughly 700 of these ICU patients will die. This will skew the IFR higher. New York City’s hospitals have not collapsed like those seen in Italy, although standard of care has likely diminished due to stress. This will skew the IFR higher than natural — but not by much, as everybody who requires care receives it.
The backlog. 3,000 excess mortality deaths are noted and the backlog likely contains a percentage of these deaths. When this is accounted for, the IFR will skew higher.
Missed deaths. People living alone at home may not necessarily be reported as dead immediately. There is a small crack here that allows for deaths to slip in between as even clinically diagnosed deaths are majority hospital-reported. This will skew the IFR higher.
Failure to form antibodies. I remember reading a South Korean study that stated 3% of those infected failed to produce measurable antibodies after infection. Comparing this to another study that claims 100% of infected produce antibodies, we can preliminarily assume 0-3% of infections will not be accounted for, skewing the IFR down.
Conclusions:
As 0.21% of New York City has died due to the novel coronavirus, it is clear that this pandemic should not be underestimated and that previous massive iceberg assumptions are false. This is a pyramid, reflecting upon the situation in the city 24 days ago.
Our absolute lowest bound estimate is a 0.84% IFR from these findings.
Prediction: 0.98% - 1.2% IFR in New York City, likely higher for European countries with larger share of elderly population.
Range: 0.84% to 1.2% IFR.
0.84% IFR assumes that no ICU patients will die, no further people will die as average time to death vs average time to antibodies is matched — no excess death backlog is reported, no missed deaths reported. 1.2% assumes 700 ICU deaths, backlog reporting, missed deaths and no failure to form antibodies.
Final notes:
These are the results based upon a no-false-positive appraisal of the antibody tests using the official death counts from New York City. Using the excess mortality results we can estimate a:
First round antibody testing: 1.1% IFR
Second round antibody testing: 1% IFR
Assuming any ratio of false positives in these results will skew the IFR higher considerably. For example, assuming 7% false-positive IFR using official death counts:
First round antibody testing (14.2%, 1,182,860 individuals infected): 1.4% IFR
Second round antibody testing: (17.9%, 1,491,070 individuals infected): 1.1% IFR
But it’s highly unlikely this is the prevalence of false positives accounted for in this testing, these calculations are simply theoretical to show the false-positive skew, or the base high-bound IFR.
The antibody tests used detect IgG antibodies, which develop on average after 14 days — and 95% of which have developed after 21 days. The specificity is appraised at between 93% and 99%, meaning somewhere between 1-7% will show as a false positive. The sensitivity assumption is lower, and the pool of true negatives vs. true positives is skewed highly towards false positive prevalence over false negative prevalence.
The audience considered for testing are grocery-store shoppers, which is likely to bias the infection rate higher than the city-wide average, as daily shoppers are more likely to be represented and are more likely to have been infected. Cautious people who rarely leave their homes are less likely to be represented and less likely to be infected.
Average time to death is appraised at a 5 day average to symptom onset, upon which an 18.8 day average to death. This means that roughly half of deaths will have occurred 23.8 days after infection, or that antibody results on average develop 9.8 days slower than death total counts, ranging from 9.8-2.8 days behind. This means that the death toll from antibody prevalence is not fully realized in the statistics for up to a week after testing.
With these caveats in mind, let us look at the results.
New York City deaths:
Positive swab-year deaths (high prevalence at hospitals): 12,287 (April 26th)
Clinically diagnosed deaths (high prevalence at hospitals): 5,228 (April 27th)
Total: 17,515 deaths
Excess mortality: 20,900 (NY Times)
For the sake of this, we will assume that the excess mortality is not comprised solely of SARS-COV-2 deaths, but it stands as an important marker in realizing the death toll of this virus. We can assume that wrongly clinically diagnosed deaths can replaced from the excess mortality source instead, meaning this number stays constant. The majority of deaths are recorded at hospitals, meaning techniques such as Lung CT scans for diagnosis have instead been used in lieu of swab-tests.
New York Population 2019: 8,330,000. This means that 0.21% of the city has died of the novel coronavirus according to the official death count. This leads us to the first antibody study they concluded just a week ago.
21.2% antibody prevalence in New York City. Keeping the false positive ratio in mind, this gives us anywhere from a 14.2%-21.2% infection rate. This study is interesting due to the fact that the average time to death vs antibodies is now reflected in the statistics for this test.
21.2% of 8,330,000 = 1,765,960 individuals.
17,515 / 1,765,960 = 0.98% of all infected have died. This is our absolute lower-end estimate.
—
The recent antibody results from yesterday indicate a 24.9% infection rate, meaning anywhere from 17.9% - 24.9% have been infected. This will be fully realized in the statistics next week as the average time to antibodies/average time to death is matched.
24.9% of 8,330,000 is 2,074,170 individuals.
17,515 / 2,074,170 = 0,84%. This is the lowest bound our IFR can be moving forward.
Now, there are many factors regarding the death total that must be adjusted for in the search of the IFR. I will name them below but we shall not adjust for this.
Firstly, the relatively young population in New York City will skew the IFR lower. One in eight residents of NYC are 65+, comparable to around one in five in most European nations. Considering the lethality of this infection rises considerably with age, this population distribution likely effects the IFR negatively when comparing to Europe.
The health of New York City residents is remarkably poorer than that of European nations, with a higher obesity and diabetes rate. However, obesity is not remarkably over-represented as a risk factor, with old age remaining a deadlier risk factor than either obesity or diabetes. This will lead to the lowering of the IFR in comparison, but when adjusting for the population distribution differences, the IFR will still skew higher.
Unresolved deaths/the state of NYC hospitals. Currently around 780 patients are in intensive care in New York City, a marked decline from their peak. Mortality rates are around 90%, meaning that roughly 700 of these ICU patients will die. This will skew the IFR higher. New York City’s hospitals have not collapsed like those seen in Italy, although standard of care has likely diminished due to stress. This will skew the IFR higher than natural — but not by much, as everybody who requires care receives it.
The backlog. 3,000 excess mortality deaths are noted and the backlog likely contains a percentage of these deaths. When this is accounted for, the IFR will skew higher.
Missed deaths. People living alone at home may not necessarily be reported as dead immediately. There is a small crack here that allows for deaths to slip in between as even clinically diagnosed deaths are majority hospital-reported. This will skew the IFR higher.
Failure to form antibodies. I remember reading a South Korean study that stated 3% of those infected failed to produce measurable antibodies after infection. Comparing this to another study that claims 100% of infected produce antibodies, we can preliminarily assume 0-3% of infections will not be accounted for, skewing the IFR down.
Conclusions:
As 0.21% of New York City has died due to the novel coronavirus, it is clear that this pandemic should not be underestimated and that previous massive iceberg assumptions are false. This is a pyramid, reflecting upon the situation in the city 24 days ago.
Our absolute lowest bound estimate is a 0.84% IFR from these findings.
Prediction: 0.98% - 1.2% IFR in New York City, likely higher for European countries with larger share of elderly population.
Range: 0.84% to 1.2% IFR.
0.84% IFR assumes that no ICU patients will die, no further people will die as average time to death vs average time to antibodies is matched — no excess death backlog is reported, no missed deaths reported. 1.2% assumes 700 ICU deaths, backlog reporting, missed deaths and no failure to form antibodies.
Final notes:
These are the results based upon a no-false-positive appraisal of the antibody tests using the official death counts from New York City. Using the excess mortality results we can estimate a:
First round antibody testing: 1.1% IFR
Second round antibody testing: 1% IFR
Assuming any ratio of false positives in these results will skew the IFR higher considerably. For example, assuming 7% false-positive IFR using official death counts:
First round antibody testing (14.2%, 1,182,860 individuals infected): 1.4% IFR
Second round antibody testing: (17.9%, 1,491,070 individuals infected): 1.1% IFR
But it’s highly unlikely this is the prevalence of false positives accounted for in this testing, these calculations are simply theoretical to show the false-positive skew, or the base high-bound IFR.
__________________
Senast redigerad av OUFCompulsive 2020-04-28 kl. 17:03.
Senast redigerad av OUFCompulsive 2020-04-28 kl. 17:03.
