[BostonCard]

With total deaths in New York passing 19,000, given a total population of New York of 19 million, the lower bound of the IFR in New York is 0.1%.  Of course, not all patients with COVID-19 who will eventually die have died yet (so that the numerator is likely higher than 19,000; New York is reporting several hundred COVID deaths a day), and it is unlikely that everyone in New York state has been infected (so the denominator is likely less than 19 million).

The question is how much higher is the numerator and how close to 19 million is the denominator?

For what it is worth, in New York City, the population is 8.4 million and there have been 10,000 deaths, so that pushes the lower bound of the IFR to .12%.

BC

[BostonCard]

With total deaths in New York passing 19,000, given a total population of New York of 19 million, the lower bound of the IFR in New York is 0.1%.  Of course, not all patients with COVID-19 who will eventually die have died yet (so that the numerator is likely higher than 19,000; New York is reporting several hundred COVID deaths a day), and it is unlikely that everyone in New York state has been infected (so the denominator is likely less than 19 million).

The question is how much higher is the numerator and how close to 19 million is the denominator?

For what it is worth, in New York City, the population is 8.4 million and there have been 10,000 deaths, so that pushes the lower bound of the IFR to .12%.

BC

[2006alum]

For what it is worth, in New York City, the population is 8.4 million and there have been 10,000 deaths, so that pushes the lower bound of the IFR to .12%.

BC

That's a conservative estimate, too, because it excludes probable deaths, which account for another ~4,000. At that point, you're well past the lower bound of .12% and getting closer to the upper bound of the Ioannidis Study's upper bound estimate of .2% IFR. 

And BTW, one other thing I have not seen mentioned anywhere is that NYC has no where near 8.4m people residing in the five boroughs right now; I would say approximately 50% of the people I know who live and work in NYC have not been sheltering in place in NYC for the past 2-5 weeks. That would further reduce the denominator and push up the IFR.

[2006alum]

For what it is worth, in New York City, the population is 8.4 million and there have been 10,000 deaths, so that pushes the lower bound of the IFR to .12%.

BC

That's a conservative estimate, too, because it excludes probable deaths, which account for another ~4,000. At that point, you're well past the lower bound of .12% and getting closer to the upper bound of the Ioannidis Study's upper bound estimate of .2% IFR. 

And BTW, one other thing I have not seen mentioned anywhere is that NYC has no where near 8.4m people residing in the five boroughs right now; I would say approximately 50% of the people I know who live and work in NYC have not been sheltering in place in NYC for the past 2-5 weeks. That would further reduce the denominator and push up the IFR.

[burger]

With total deaths in New York passing 19,000, given a total population of New York of 19 million, the lower bound of the IFR in New York is 0.1%.  Of course, not all patients with COVID-19 who will eventually die have died yet (so that the numerator is likely higher than 19,000; New York is reporting several hundred COVID deaths a day), and it is unlikely that everyone in New York state has been infected (so the denominator is likely less than 19 million).

The question is how much higher is the numerator and how close to 19 million is the denominator?

For what it is worth, in New York City, the population is 8.4 million and there've been 10,000 deaths, so that pushes the lower bound of the IFR to .12%.

At a minimum, the actual NY/NYC IFR has to be at least triple that lower bound, right?  Otherwise over 1/3 of the city's population would have to be infected, and there would have to be little undercount in the deaths.  I would save 5x those estimates is a safer bet for a lower bound.

Fun with bad serology studies:

Implied IFR from the Stanford study in Santa Clara Co.: 0.12%

Implied IFR from the LA study: 0.19%

So, BC's post is reason #642 to be skeptical of those serology studies.

[burger]

With total deaths in New York passing 19,000, given a total population of New York of 19 million, the lower bound of the IFR in New York is 0.1%.  Of course, not all patients with COVID-19 who will eventually die have died yet (so that the numerator is likely higher than 19,000; New York is reporting several hundred COVID deaths a day), and it is unlikely that everyone in New York state has been infected (so the denominator is likely less than 19 million).

The question is how much higher is the numerator and how close to 19 million is the denominator?

For what it is worth, in New York City, the population is 8.4 million and there've been 10,000 deaths, so that pushes the lower bound of the IFR to .12%.

At a minimum, the actual NY/NYC IFR has to be at least triple that lower bound, right?  Otherwise over 1/3 of the city's population would have to be infected, and there would have to be little undercount in the deaths.  I would save 5x those estimates is a safer bet for a lower bound.

Fun with bad serology studies:

Implied IFR from the Stanford study in Santa Clara Co.: 0.12%

Implied IFR from the LA study: 0.19%

So, BC's post is reason #642 to be skeptical of those serology studies.

[magnus]

NYC now has 14-20k (higher number includes probable case deaths).

Just below 0.2 to 0.25%.  Lombardy was 0.2% and Bergamo was 0.4% 3 weeks ago.

[magnus]

NYC now has 14-20k (higher number includes probable case deaths).

Just below 0.2 to 0.25%.  Lombardy was 0.2% and Bergamo was 0.4% 3 weeks ago.

[M T]

For what it is worth, in New York City, the population is 8.4 million and there've been 10,000 deaths, so that pushes the lower bound of the IFR to .12%.

At a minimum, the actual NY/NYC IFR has to be at least triple that lower bound, right?  Otherwise over 1/3 of the city's population would have to be infected, and there would have to be little undercount in the deaths.  I would save 5x those estimates is a safer bet for a lower bound.
Fun with bad serology studies:
Implied IFR from the Stanford study in Santa Clara Co.: 0.12%
Implied IFR from the LA study: 0.19%
So, BC's post is reason #642 to be skeptical of those serology studies.

The revised Stanford report estimated 23K infected (95CI 14K - 35K).  That was April 3-4. 
Considering those are antibody results, they may correspond to cases shortly after symptoms show.
So, corresponding deaths would be roughly 3 weeks later (Apr 24).  That was about 99 deaths.

99/23,000 = 0.43% IFR  (0.28% - 0.71%)

For NYC proper, 8.4M * 0.43% = 36K deaths

[M T]

For what it is worth, in New York City, the population is 8.4 million and there've been 10,000 deaths, so that pushes the lower bound of the IFR to .12%.

At a minimum, the actual NY/NYC IFR has to be at least triple that lower bound, right?  Otherwise over 1/3 of the city's population would have to be infected, and there would have to be little undercount in the deaths.  I would save 5x those estimates is a safer bet for a lower bound.
Fun with bad serology studies:
Implied IFR from the Stanford study in Santa Clara Co.: 0.12%
Implied IFR from the LA study: 0.19%
So, BC's post is reason #642 to be skeptical of those serology studies.

The revised Stanford report estimated 23K infected (95CI 14K - 35K).  That was April 3-4. 
Considering those are antibody results, they may correspond to cases shortly after symptoms show.
So, corresponding deaths would be roughly 3 weeks later (Apr 24).  That was about 99 deaths.

99/23,000 = 0.43% IFR  (0.28% - 0.71%)

For NYC proper, 8.4M * 0.43% = 36K deaths

[M T]

NYC publishes cases, hospitalization, and deaths for demographic information "per 100K".  I've determined that, for instance, for 0-17 year olds, the "per 100K" refers to per 100K of 0-17 year olds.   So from that, I generated approximate case counts (to make sure I had the interpretation right) from 2016 demographics.

I note that the incidence of cases is a little higher in 75+, but lower to much lower in the younger age group.  This suggests to me that we may be seeing under-counting in the younger age group (possibly asymptomatic or nearly asymptomatic).

REMEMBER: Due to (about a 3-week) latency in deaths, relative to cases, deaths are under-represented here.  The death percentages will go up (maybe by a factor of 2).  Hospitalizations will also go up.

Nearly 10% of the cases in 18-44 require hospitalization (but only 1% die).
Nearly a quarter of the cases in 45-64 require hospitalization and 5% die.  (This is close to the overall China numbers of 80% mild, 20% severe, 5% death)
Not quite half of the 65-74 cases require hospitalization.  One sixth have already died.
More than 60% of the 75+ cases require hospitalization.  More than one third have already died.

		per 100K			approx count
Age group	Cases	Hosp.	Deaths	Cases	Hosp.	Deaths
0-17	258.21	18.82	0	4640	338	0
18-44	1948.1	189.01	17.3	67870	6585	603
45-64	3188	746.92	161.28	66998	15697	3389
65-74	3184	1454.83	524.09	20841	9523	3430
75+	3748.6	2301.02	1312.67	18763	11517	6570
Total	2128.5	533.56	175.66	179112	43660	13993

[M T]

NYC publishes cases, hospitalization, and deaths for demographic information "per 100K".  I've determined that, for instance, for 0-17 year olds, the "per 100K" refers to per 100K of 0-17 year olds.   So from that, I generated approximate case counts (to make sure I had the interpretation right) from 2016 demographics.

I note that the incidence of cases is a little higher in 75+, but lower to much lower in the younger age group.  This suggests to me that we may be seeing under-counting in the younger age group (possibly asymptomatic or nearly asymptomatic).

REMEMBER: Due to (about a 3-week) latency in deaths, relative to cases, deaths are under-represented here.  The death percentages will go up (maybe by a factor of 2).  Hospitalizations will also go up.

Nearly 10% of the cases in 18-44 require hospitalization (but only 1% die).
Nearly a quarter of the cases in 45-64 require hospitalization and 5% die.  (This is close to the overall China numbers of 80% mild, 20% severe, 5% death)
Not quite half of the 65-74 cases require hospitalization.  One sixth have already died.
More than 60% of the 75+ cases require hospitalization.  More than one third have already died.

		per 100K			approx count
Age group	Cases	Hosp.	Deaths	Cases	Hosp.	Deaths
0-17	258.21	18.82	0	4640	338	0
18-44	1948.1	189.01	17.3	67870	6585	603
45-64	3188	746.92	161.28	66998	15697	3389
65-74	3184	1454.83	524.09	20841	9523	3430
75+	3748.6	2301.02	1312.67	18763	11517	6570
Total	2128.5	533.56	175.66	179112	43660	13993

[BostonCard]

I should hurry up and get this thing before my 45th birthday in November!

BC

[BostonCard]

I should hurry up and get this thing before my 45th birthday in November!

BC

[burger]

I should hurry up and get this thing before my 45th birthday in November!

BC

All of these studies that split the data into 5 age classes should be treating age as a 5th-degree polynomial instead.  Just as many degrees of freedom, and I'll bet the predicted mortality rates, etc. would fit the data better.

(Prediction: no one will do that)

[burger]

I should hurry up and get this thing before my 45th birthday in November!

BC

All of these studies that split the data into 5 age classes should be treating age as a 5th-degree polynomial instead.  Just as many degrees of freedom, and I'll bet the predicted mortality rates, etc. would fit the data better.

(Prediction: no one will do that)

[BostonCard]

I should hurry up and get this thing before my 45th birthday in November!

BC

All of these studies that split the data into 5 age classes should be treating age as a 5th-degree polynomial instead.  Just as many degrees of freedom, and I'll bet the predicted mortality rates, etc. would fit the data better.

(Prediction: no one will do that)

Obviously, I am making fun of reducing a continuous predictor into a bunch of categories.  Your point is well taken, but, more importantly, you don't have to preserve the degrees of freedom; a quick perusal of the data would tell you that a simple quadratic is probably a better descriptor of the data, and it has saved you three DF's.

And/but, as always, it depends on what you want to do with the results.  An nth degree polynomial (where n > 1) makes it hard to explain what is seen, since the terms higher than the linear term are hard to describe, unless they are all positive.  More importantly, while as a 44-year-old, it is pretty easy to look at my risk category and infer that I would likely be at the upper end of the range for the 18-44 year old category, while more precise, it would be much harder to calculate my risk with even a quadratic polynomial, absent a calculator.  So there is a tradeoff between precision and ease of understanding.

BC

[BostonCard]

I should hurry up and get this thing before my 45th birthday in November!

BC

All of these studies that split the data into 5 age classes should be treating age as a 5th-degree polynomial instead.  Just as many degrees of freedom, and I'll bet the predicted mortality rates, etc. would fit the data better.

(Prediction: no one will do that)

Obviously, I am making fun of reducing a continuous predictor into a bunch of categories.  Your point is well taken, but, more importantly, you don't have to preserve the degrees of freedom; a quick perusal of the data would tell you that a simple quadratic is probably a better descriptor of the data, and it has saved you three DF's.

And/but, as always, it depends on what you want to do with the results.  An nth degree polynomial (where n > 1) makes it hard to explain what is seen, since the terms higher than the linear term are hard to describe, unless they are all positive.  More importantly, while as a 44-year-old, it is pretty easy to look at my risk category and infer that I would likely be at the upper end of the range for the 18-44 year old category, while more precise, it would be much harder to calculate my risk with even a quadratic polynomial, absent a calculator.  So there is a tradeoff between precision and ease of understanding.

BC

[burger]

Obviously, I am making fun of reducing a continuous predictor into a bunch of categories.  Your point is well taken, but, more importantly, you don't have to preserve the degrees of freedom; a quick perusal of the data would tell you that a simple quadratic is probably a better descriptor of the data, and it has saved you three DF's.

And/but, as always, it depends on what you want to do with the results.  An nth degree polynomial (where n > 1) makes it hard to explain what is seen, since the terms higher than the linear term are hard to describe, unless they are all positive.  More importantly, while as a 44-year-old, it is pretty easy to look at my risk category and infer that I would likely be at the upper end of the range for the 18-44 year old category, while more precise, it would be much harder to calculate my risk with even a quadratic polynomial, absent a calculator.  So there is a tradeoff between precision and ease of understanding.

BC

I know you were being tongue in cheek. I was, too, sort of.

If I was doing this for publication, I'd try to get a more parsimonious model by 1) running polynomials of various degrees and comparing likelihoods, etc. or 2) using penalized splines.

And making it understandable to the public is easy.  Just make the r or matlab code available at github, and then anyone who's interested can download the data and run the models themselves.  Doesn't everyone know how to run a polynomial regression nowadays? ;)

[burger]

Obviously, I am making fun of reducing a continuous predictor into a bunch of categories.  Your point is well taken, but, more importantly, you don't have to preserve the degrees of freedom; a quick perusal of the data would tell you that a simple quadratic is probably a better descriptor of the data, and it has saved you three DF's.

And/but, as always, it depends on what you want to do with the results.  An nth degree polynomial (where n > 1) makes it hard to explain what is seen, since the terms higher than the linear term are hard to describe, unless they are all positive.  More importantly, while as a 44-year-old, it is pretty easy to look at my risk category and infer that I would likely be at the upper end of the range for the 18-44 year old category, while more precise, it would be much harder to calculate my risk with even a quadratic polynomial, absent a calculator.  So there is a tradeoff between precision and ease of understanding.

BC

I know you were being tongue in cheek. I was, too, sort of.

If I was doing this for publication, I'd try to get a more parsimonious model by 1) running polynomials of various degrees and comparing likelihoods, etc. or 2) using penalized splines.

And making it understandable to the public is easy.  Just make the r or matlab code available at github, and then anyone who's interested can download the data and run the models themselves.  Doesn't everyone know how to run a polynomial regression nowadays? ;)