ASA Connect

I keep reading that because C19 is underreported due to lack of symptoms and shortage of testing that the fatality rates are more like regular flu.

My question is that aren't these underreporting issues similar for the regular flu?

Aren't there significant numbers of people who get the flu who never visit a doc or never get tested or die of other causes not listed as flue related?

Maybe and experienced virologist or biostats guy can claify my confusion on this.

------------------------------
Michael Mout
MIKS
------------------------------

Hi Michael,

Off the cuff (i.e., paraphrasing a very nice lady who actually does this modelling so I'm undoubtedly missing some nuances): my understanding is that these issues in estimation challenges are present in seasonal flu as well. Significant effort into estimating this latent variable.

So flu death estimates are a function of "flu death count" statistics and"flu incidence count" statistics but are heavily supplemented with ancillary predictive features like testing/sampling and also mortality statistics.

You can imagine that the techniques used to estimate "excess mortality" or competing risks spend significant effort accounting for the latent factors of the overall unobserved deaths as well as patients who would have died in the same time period anyway.




------------------------------
Glen Wright Colopy
DPhil Oxon
Data Scientist at Cenduit LLC, Durham, NC
------------------------------

There are a few different ways I would respond to that. (I am teaching an INtro to Stats class this summer and looked it up.) 

During the 2018-2019 flu season, 34,200 people died from the flu. https://www.cdc.gov/flu/about/burden/2018-2019.html That was during the entire year. 

So far, Covid-19 has killed 55, 417 (as of when I am writing this) in under 6 months in the US. (We could probably even make it 4 months in the US). 

There are a few rates I can think of to use here. But, the ones that really matter to me are: Deaths/population and deaths/month. In both cases, Covid-19 is much worse. 

There are many other issues too. First off, due to lack of testing, how do we know that there are 55,417 deaths due to Covid-19? If someone dies and they had the "symptoms" but they were not tested, how do we know? 

Also, how do we know that the death record has the right cause?  As long as the numbers are counted the same way, then you can raise the same issues with any reported cause of death. 

 https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5692167/

https://www.washingtonpost.com/nation/2019/09/12/there-may-have-been-dozens-more-deaths-linked-flint-water-crisis-than-previously-disclosed/

------------------------------
Andrew Ekstrom

Statistician, Chemist, HPC Abuser;-)
------------------------------

Flu estimates use a combination of lab-confirmed diagnoses and outpatient visits for influenza-like-illness (ILI) regardless of lab testing.  
https://www.cdc.gov/flu/weekly/overview.htm

As such, it is unclear how the current surveillance system will differentiate flu and coronavirus going forward.  

There are also surveillance methods that use factors like over the counter medication sales, school absenteeism rates, and ER admissions to detect possible outbreaks, but I am not sure if these methods are used for flu surveillance.  

The famous Google Flu Trends used search terms to estimate flu prevalence, and it was the favorite example of using big data for public health until it massively overestimated flu cases.  Now it is seen as a precautionary tale regarding the relative merits of big organic data versus well designed small data:  https://time.com/23782/google-flu-trends-big-data-problems/  



------------------------------
Deirdre Middleton
------------------------------

Original Message:
Sent: 04-27-2020 08:43
From: Andrew Ekstrom
Subject: Incidence and death rates C19 v. Flu

There are a few different ways I would respond to that. (I am teaching an INtro to Stats class this summer and looked it up.)

During the 2018-2019 flu season, 34,200 people died from the flu. https://www.cdc.gov/flu/about/burden/2018-2019.html That was during the entire year.

So far, Covid-19 has killed 55, 417 (as of when I am writing this) in under 6 months in the US. (We could probably even make it 4 months in the US).

There are a few rates I can think of to use here. But, the ones that really matter to me are: Deaths/population and deaths/month. In both cases, Covid-19 is much worse.

There are many other issues too. First off, due to lack of testing, how do we know that there are 55,417 deaths due to Covid-19? If someone dies and they had the "symptoms" but they were not tested, how do we know?

Also, how do we know that the death record has the right cause?  As long as the numbers are counted the same way, then you can raise the same issues with any reported cause of death.

 https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5692167/

https://www.washingtonpost.com/nation/2019/09/12/there-may-have-been-dozens-more-deaths-linked-flint-water-crisis-than-previously-disclosed/

------------------------------
Andrew Ekstrom

Statistician, Chemist, HPC Abuser;-)
------------------------------

Mortality rates are almost impossible to calculate without sufficient testing to calculate prevalence, but we can look at excess deaths to get a sense of the public health burden.  This article gives a good overview of how all-cause mortality compares in countries during the outbreak and during the same time periods in other years: 

"Mortality statistics show 122,000 deaths in excess of normal levels across these locations, considerably higher than the 77,000 official Covid-19 deaths reported for the same places and time periods. If the same level of under-reporting observed in these countries was happening worldwide, the global Covid-19 death toll would rise from the current official total of 201,000 to as high as 318,000. To calculate excess deaths, the FT has compared deaths from all causes in the weeks of a location's outbreak in March and April 2020 to the average for the same period between 2015 and 2019. The total of 122,000 amounts to a 50 percent rise in overall mortality relative to the historical average for the locations studied."

   https://www.ft.com/content/6bd88b7d-3386-4543-b2e9-0d5c6fac846c



------------------------------
Deirdre Middleton
------------------------------

Original Message:
Sent: 04-27-2020 10:16
From: Deirdre Middleton
Subject: Incidence and death rates C19 v. Flu

Flu estimates use a combination of lab-confirmed diagnoses and outpatient visits for influenza-like-illness (ILI) regardless of lab testing.
https://www.cdc.gov/flu/weekly/overview.htm

As such, it is unclear how the current surveillance system will differentiate flu and coronavirus going forward.

There are also surveillance methods that use factors like over the counter medication sales, school absenteeism rates, and ER admissions to detect possible outbreaks, but I am not sure if these methods are used for flu surveillance.

The famous Google Flu Trends used search terms to estimate flu prevalence, and it was the favorite example of using big data for public health until it massively overestimated flu cases.  Now it is seen as a precautionary tale regarding the relative merits of big organic data versus well designed small data:  https://time.com/23782/google-flu-trends-big-data-problems/



------------------------------
Deirdre Middleton
------------------------------


Original Message:
Sent: 04-27-2020 08:43
From: Andrew Ekstrom
Subject: Incidence and death rates C19 v. Flu

There are a few different ways I would respond to that. (I am teaching an INtro to Stats class this summer and looked it up.)

During the 2018-2019 flu season, 34,200 people died from the flu. https://www.cdc.gov/flu/about/burden/2018-2019.html That was during the entire year.

So far, Covid-19 has killed 55, 417 (as of when I am writing this) in under 6 months in the US. (We could probably even make it 4 months in the US).

There are a few rates I can think of to use here. But, the ones that really matter to me are: Deaths/population and deaths/month. In both cases, Covid-19 is much worse.

There are many other issues too. First off, due to lack of testing, how do we know that there are 55,417 deaths due to Covid-19? If someone dies and they had the "symptoms" but they were not tested, how do we know?

Also, how do we know that the death record has the right cause?  As long as the numbers are counted the same way, then you can raise the same issues with any reported cause of death.

 https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5692167/

https://www.washingtonpost.com/nation/2019/09/12/there-may-have-been-dozens-more-deaths-linked-flint-water-crisis-than-previously-disclosed/

------------------------------
Andrew Ekstrom

Statistician, Chemist, HPC Abuser;-)

So the answer is nobody really knows and hey are all just WAG's or maybe Intelligent Guesses?

------------------------------
Michael Mout
MIKS
------------------------------

Original Message:
Sent: 04-27-2020 10:26
From: Deirdre Middleton
Subject: Incidence and death rates C19 v. Flu

Mortality rates are almost impossible to calculate without sufficient testing to calculate prevalence, but we can look at excess deaths to get a sense of the public health burden.  This article gives a good overview of how all-cause mortality compares in countries during the outbreak and during the same time periods in other years:

"Mortality statistics show 122,000 deaths in excess of normal levels across these locations, considerably higher than the 77,000 official Covid-19 deaths reported for the same places and time periods. If the same level of under-reporting observed in these countries was happening worldwide, the global Covid-19 death toll would rise from the current official total of 201,000 to as high as 318,000. To calculate excess deaths, the FT has compared deaths from all causes in the weeks of a location's outbreak in March and April 2020 to the average for the same period between 2015 and 2019. The total of 122,000 amounts to a 50 percent rise in overall mortality relative to the historical average for the locations studied."

   https://www.ft.com/content/6bd88b7d-3386-4543-b2e9-0d5c6fac846c



------------------------------
Deirdre Middleton
------------------------------


Original Message:
Sent: 04-27-2020 10:16
From: Deirdre Middleton
Subject: Incidence and death rates C19 v. Flu

Flu estimates use a combination of lab-confirmed diagnoses and outpatient visits for influenza-like-illness (ILI) regardless of lab testing.
https://www.cdc.gov/flu/weekly/overview.htm

As such, it is unclear how the current surveillance system will differentiate flu and coronavirus going forward.

There are also surveillance methods that use factors like over the counter medication sales, school absenteeism rates, and ER admissions to detect possible outbreaks, but I am not sure if these methods are used for flu surveillance.

The famous Google Flu Trends used search terms to estimate flu prevalence, and it was the favorite example of using big data for public health until it massively overestimated flu cases.  Now it is seen as a precautionary tale regarding the relative merits of big organic data versus well designed small data:  https://time.com/23782/google-flu-trends-big-data-problems/



------------------------------
Deirdre Middleton


Original Message:
Sent: 04-27-2020 08:43
From: Andrew Ekstrom
Subject: Incidence and death rates C19 v. Flu

There are a few different ways I would respond to that. (I am teaching an INtro to Stats class this summer and looked it up.)

During the 2018-2019 flu season, 34,200 people died from the flu. https://www.cdc.gov/flu/about/burden/2018-2019.html That was during the entire year.

So far, Covid-19 has killed 55, 417 (as of when I am writing this) in under 6 months in the US. (We could probably even make it 4 months in the US).

There are a few rates I can think of to use here. But, the ones that really matter to me are: Deaths/population and deaths/month. In both cases, Covid-19 is much worse.

There are many other issues too. First off, due to lack of testing, how do we know that there are 55,417 deaths due to Covid-19? If someone dies and they had the "symptoms" but they were not tested, how do we know?

Also, how do we know that the death record has the right cause?  As long as the numbers are counted the same way, then you can raise the same issues with any reported cause of death.

 https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5692167/

https://www.washingtonpost.com/nation/2019/09/12/there-may-have-been-dozens-more-deaths-linked-flint-water-crisis-than-previously-disclosed/

------------------------------
Andrew Ekstrom

Statistician, Chemist, HPC Abuser;-)

It seems to me that the media most frequently reports the simple rate of deaths per size of the total population. I believe they have done this in most articles for all of the different types of infection outbreaks in the last decade. The 2nd most frequently reported rate is deaths per reported cases. Rarely do they report death rates within the infected population. Your question seems to relate to this last rate, which is difificut to calculate, & not well understood by lay persons. In addition, there appears to be serious debate concerning the best methodology to use for calculating this rate.

For the medical community, all 3 rates are important, depending on the issue. For political & social discussions, I think the simple death/population rate is a number that politicians & lay people can relate to. This rate tends to have a low level of uncertanty attached to it. In addition, this number relates more closely to the major political decisions rather than the medical decisions.The key issues for political decision makers is preventing the system from being overloaded &  minimizing the number of deaths.

Your question seems rhetorical to me. It is a statement rather than a question. I am not sure I understand what you are questioning.






------------------------------
Jack Lothian
Senior Survey Methodologist
Retired
------------------------------

The Washington Post just had an excellent article clearing up some of the popular misconceptions & erroneous comparisons.

https://www.washingtonpost.com/health/antibody-tests-support-whats-been-obvious-covid-19-is-much-more-lethal-than-flu/2020/04/28/2fc215d8-87f7-11ea-ac8a-fe9b8088e101_story.html?fbclid=IwAR3qGcIS1NiI103Uin0DPy8sh_yeQeKEiOkvHVQwiYuG5M0RfxGt27kjxTg

For instance, the Case Fatality Rate for the seasonal flu is usually about 0.1%, while the Case Fatality Rate for COVID-19 varies according to the condition of the patients and the preparedness of the medical system. In New York City, it has been about 7.4%
However, the CASE FATALITY RATE is how many people die of those whose symptoms are severe enough to bring them to a Doctor and be diagnosed.
When one expands the denominator to include infections without symptoms or only ignorable symptoms, that is known as the Infection Fatality Rate, a smaller number because it is the same number of deaths, but spread out over a larger number of people. For COVID-19, the IFR seems to be about 0.5-1% (which is also consistent with Chinese statistics in regions with extensive testing whose medical system were not overwhelmed). But the IFR and the CFR are apples & oranges statistics: the IFR of COVID19 is a totally different thing than the CFR of the seasonal flu. Best estimate of the flu IFR is about ½ to ¼ of the CFR, or about 0.025-0.05%.  So COVID-19 mortality as a proportion of infections (the IFR) is still some 20-fold higher than the seasonal flu, and furthermore, spreads much more rapidly.

------------------------------
Dennis Sweitzer
Principal Biostatistician
------------------------------