Author Topic: estimating the cfr  (Read 217 times)

gsgs

  • Administrator
  • Full Member
  • *****
  • Posts: 196
    • View Profile
estimating the cfr
« on: February 01, 2020, 02:40:31 am »
currently we have 2.2% = deaths/confirmed cases.
That would make ~1mllion deaths in USA within 1 year (when the vaccine may become available)
Assuming the spread is as in 2009
------------------------------------------------------------------
On February 12, 2010, the CDC released updated estimate figures for swine flu,
reporting that, in total, 57 million Americans had been sickened, 257,000 had been
hospitalised and 11,690 people had died (including 1,180 children) due to swine flu
from April through to mid-January.[128]

==========================================
but ...
I should really add here, that they have apparently many cases which are unconfirmed,
but less likely deaths that are unnoticed.
There were estimates about 100000 cases already.
These would probably still have developed immunity.
Then the CFR would be only ~0.3%
or 150000 deaths in USA with the assumptions above.
Twice as many as in the bad 2017/2018 flu-season
-------------------------------------------------------

also :
164 cases outside China with 0 deaths ; cfr=0
7153 cases in Hubei with 249 deaths , cfr=3.5%
3992 cases outside Hubei with 9 deaths ; cfr=0.2%

this supports the assumption that there are many nonconfirmed cases in Hubei
« Last Edit: February 01, 2020, 07:19:46 am by gsgs »

Share on Facebook Share on Twitter


epsilon

  • Newbie
  • *
  • Posts: 27
  • welcome to ncovinfo , here we discuss the pandemic threat from the new 2019 Coronavirus
    • View Profile
Re: estimating the cfr
« Reply #1 on: February 06, 2020, 06:43:58 am »
Unfortunately it is still too early for a reliable CFR estimate.

There are three major biases.

First the unknown number of actual infections vs. reported numbers of confirmed cases. There are estimates of an ascertain rate of about 0.1 in the literature (i.e. actual cases = ten times reported cases).
So this bias leads to a 10 times over-estimation of the true CFR.

Second: The time lag issue between symptom onset and death. For SARS/MERS like illnesses (with the typical clinical picture of worsening pneumonia only after 10 days, then progressing to ARDS and ICU/ventilator support as is also typically reported for nCoV ) it is well established in the literature that average time from symptom onset to death with SARS is about three weeks (and even longer for the younger age groups).
So we must not relate the number of deaths to the current number of cases but to the number of cases three weeks ago. Given the high case doubling rate of about 3-5 days, there were 10-100 times less cases three weeks ago. 
So this second bias leads to 10 to 100 times under estimation of the CFR.

Third there is the aspect of reliability of information from authoritarian state officials when it comes to such important and sensitive figures like death rates from a new epidemic. This is not to say that there is "large scale" intentional desinformation. All in all we should trust China to not completely misrepresent the big picture (in its own interest) but it would not be surprising if the numbers are indeed somewhat "tuned" towards being less panic/fear inducing for the general public.


Altogether there are so many sensitive parameters, assumptions and biases that it seems almost impossible to infer the true CFR at this time.

I have extensively researched the topic in the past weeks but still I can neither rule out a best-case scenario with <0.1% Flu-like CFR nor can I rule out a worst case scenario with 15% SARS-Like CFR.

We should get a better picture by end of February when the first 100-or-so closely observed "exported" cases outside china will have passed the 3-4 weeks mark after their symptom onset.   

Then we will have a reasonably unbiased sample of the final outcome of 100 closely monitored patients.








« Last Edit: February 06, 2020, 07:07:14 am by epsilon »

epsilon

  • Newbie
  • *
  • Posts: 27
  • welcome to ncovinfo , here we discuss the pandemic threat from the new 2019 Coronavirus
    • View Profile
Re: estimating the cfr
« Reply #2 on: February 07, 2020, 07:50:02 am »
Building upon the idea to only use the cases outside china (as a the most unbiased sample we have at this time),

[ Guests cannot view attachments ]


I thought about writing a little simulation program for calculating the probabilities (confidential intervals) of CFR given the known number of deaths among them (n=2 so far)  and the distribution of time-to-death which has been described as Weibull-like distribution in the literature so far:

[ Guests cannot view attachments ]

Luckily I found that this work has already been done by a Swiss research group who put their code and figures on github:

https://github.com/calthaus/ncov-cfr

They even regularly update their estimates based on new fatal cases outside china.

The most recent estimate is CFR = 2% (rightmost bar in the diagram)
Confidence: 5% probability that CFR is below 0.1% or above 8.8%

[ Guests cannot view attachments ]

https://github.com/calthaus/ncov-cfr/blob/master/figures/ncov_cfr.png

(Ironically: 2% is about the same number as given by the very naive and biased approach of the mass media to simply divide currently reported deaths by total cases)

Remaining biases are:
- Not counting for very mild/asymptomatic cases which would lead to CFR over-estimation.
- avg. time to death could be higher (for SARS it was over 3 weeks) which would lead to CFR under-estimation

Still, this is the first CFR estimate that I am reasonably confident in to be not too far off at least on the high side of the confidence interval. (regarding the low side I still hope that CFR will be lower in the end due to increasingly more very mild/asymptomatic infections)











 
« Last Edit: February 07, 2020, 08:31:45 am by epsilon »

gsgs

  • Administrator
  • Full Member
  • *****
  • Posts: 196
    • View Profile
Re: estimating the cfr
« Reply #3 on: February 08, 2020, 02:30:48 am »
take cases outside Wuhan or outside Hubei , that gives much more cases
and still relatively few deaths.

deaths in Hubei outside Wuhan started to climb on Jan29, since then ~12 deaths per day
=1% of cases.
deaths in Wuhan started ~Jan 25
deaths outside Hubei may be starting now slowly, 3,4,5 the last days
 
--------------------------------

yes, I should have considered that time lag. I thought it were just a few days

epsilon

  • Newbie
  • *
  • Posts: 27
  • welcome to ncovinfo , here we discuss the pandemic threat from the new 2019 Coronavirus
    • View Profile
UPDATE: Althaus CFR estimate now at 1.7%
« Reply #4 on: February 08, 2020, 07:20:47 am »
UPDATE: case fatality ratio of 2019-nCoV at

1.7%

(95% confidence interval: 0.1%-7.5%)

https://github.com/calthaus/ncov-cfr

epsilon

  • Newbie
  • *
  • Posts: 27
  • welcome to ncovinfo , here we discuss the pandemic threat from the new 2019 Coronavirus
    • View Profile
Re: estimating the cfr
« Reply #5 on: February 08, 2020, 07:29:23 am »
@gsgs:
Quote
take cases outside Wuhan or outside Hubei , that gives much more cases
and still relatively few deaths.

Yes. And whats even more interesting: Much more "discharged" cases too.
I did a very rough calculation CFR = fatal / (fatal + discharged) for all non-hubei provinces and it was only about 1%.

This seems roughly consistent with the Christian Althaus estimate (that I posted recently and that I have much confidence in his model and method) which is based only on non-china cases and is now at CFR=1.7%


This reminds me more and more of the situation with the 2009 H1N1 pandemic where the CFR was initially estimated 10% then 1% and in the end it was like less than 0.1%, i.e. really "just the flu".

The other lesson from 2009 H1N1, however, was that it could NOT be contained. The global attack rate was 50%. We were lucky that the CFR turned out so benign. Hopefully it will be similar this time.

However, there is a freightening scenario with the new Virus even if the final CFR would "only" be something like 0.5% and a 50% global attack rate which would still be a devastating pandemic with millions of deaths.
 






gsgs

  • Administrator
  • Full Member
  • *****
  • Posts: 196
    • View Profile
Re: estimating the cfr
« Reply #6 on: February 08, 2020, 09:57:40 pm »
thanks for the cfr-updates.
Yes, 1.7% is still pretty bad .

however the case numbers are going down !

gsgs

  • Administrator
  • Full Member
  • *****
  • Posts: 196
    • View Profile
Re: estimating the cfr
« Reply #7 on: February 10, 2020, 07:12:45 am »
10 February 2020 - Imperial College London‌
Report 4: Severity of 2019-novel coronavirus (nCoV)
(Download Report 4)‌

https://www.imperial.ac.uk/media/imperial-college/medicine/sph/ide/gida-fellowships/Imperial-College-2019-nCoV-severity-10-02-2020.pdf
this assumes a time-lag of 22.2 days from "confirmed" to "death"
and it assumes that all cases are confirmed ; CFR=deaths/(confirmed cases)

My data starts Jan.25, before that there were 572 confirmed cases in Wuhan,
157 in Hubei ex Wuhan, 558 in China ex Hubei.
Currently there are ~70 daily deaths in Wuhan, 18 in Hubei ex Wuhan, 7 in China ex Hubei.
If those cases pre-Jan.25 were all attributed to Jan.18 , then with the 22-day-lag
I get CFR=12.2 in Wuhan, CFR=11.5 in Hubei ex Wuhan, 1.3% in China ex Hubei.
Obviously many cases were unconfirmed, especially in Wuhan. There were those 5M people
leaving the city at that time.It is likely that severe cases had a greater probability
of being confirmed.
You would expect deaths climbing sharply after 10-20 days already - even if 22 were the average.
I do not see this. Deaths in Hubei ex Wuhan (where I'd assume not much underreporting)
were 2,6,12,12,12,13,15,16,16,18,5,14,18,18

And we clearly have no exponential increase in confirmed cases as they assume
--------------------------------------------
I'm unsure, that all doesn't make so much sense to me.
The CFR might be changing. The CFR might be different in Wuhan.
The 22-day-lag might be too long.
(CFR=deaths per all infections).
------------------------------------------------------
and then we have "severe cases" . These counts were up quite a lot since Feb.03,
but now they somehow have redefined "severe":
daily reported new severe cases in Hunan Jan.26-Feb.09 : 109,400,189,89,106,200,268,139,442,377,564,918,119 3,52,258
daily reported new severe cases in China ex Hunan , Jan.26-Feb.09 :
28,115,74,42,51,68,47,47,50,54,96,44,87,135,38 (212)

epsilon

  • Newbie
  • *
  • Posts: 27
  • welcome to ncovinfo , here we discuss the pandemic threat from the new 2019 Coronavirus
    • View Profile
Re: estimating the cfr
« Reply #8 on: February 10, 2020, 08:52:27 am »
"Obviously many cases were unconfirmed, especially in Wuhan. There were those 5M people
leaving the city at that time.It is likely that severe cases had a greater probability
of being confirmed."

Th big question is: How prevalent was the virus already in the general community when Wuhan Quarantaine begun ?
My gut feeling is at least 1% prevalence, maybe even 10% obviously (and hopefully) this would imply many more mild cases than assumed previously.
Most models based on exported cases put the pre-shutdown absolut case numbers between 25k and 150k, corresponding to roughly 100k/10M = 1% prevalence.
 
So there is, unfortunately, still large uncertainty of one order of magnitude (factor 10) about the actual pathogenity/severity of this virus.

What we really need are seroprevalence studies for 2019nCoV antibodies in every large community.

My hope is, that sero prevalence is very high already because this means that the vast majority of cases are mild or asymptomatic cases.
But until we have the data, this may be just wishful thinking.
 


« Last Edit: February 10, 2020, 08:53:59 am by epsilon »

gsgs

  • Administrator
  • Full Member
  • *****
  • Posts: 196
    • View Profile
Re: estimating the cfr
« Reply #9 on: February 11, 2020, 01:26:27 am »
you could compare the Wuhan deathrate with the deathrate outside Wuhan
to get ~ 2-3  times  more real cases in Wuhan

assuming no covered up deaths in Wuhan, assuming same real deathrates in and out Wuhan

~20 daily deaths in Wuhan, when it was quarantined. That's about the amount
in Hubei ex Wuhan now - 14 days later

so with that speed there had been ~400 cases in Wuhan ~Jan.10
redouble every 3-4 days to get 6400 cases on Jan24 , 100000 now

===========================

« Last Edit: February 11, 2020, 01:29:28 am by gsgs »

epsilon

  • Newbie
  • *
  • Posts: 27
  • welcome to ncovinfo , here we discuss the pandemic threat from the new 2019 Coronavirus
    • View Profile
Re: estimating the cfr
« Reply #10 on: February 11, 2020, 07:06:04 am »
I generally avoid drawing conclusions from official china numbers from Wuhan/Hubei (I have some more trust in the chinese numbers outside Hubei). This is not to say, the official lie intentionally, rather I think it is simply a problem of capacity and chaos in an near-overwhelmed health system)

Fortunately there are many studies now, that model the spread and severity of the epidemic using only non-china data and air travel models.

The other point we all should keep in mind is that you cannot simply mix numbers from before Jan 23 (i.e. before the drastic isolation measures) with the numbers after the shutdown.

Before the shutdown we probably had real exponential community transmission with case doubling of 3 days. (You can even use the official numbers to see that because for case doubling estimation it does not matter whether the numbers are off by a factor of ten)

Now, after the sutdown (plus one incubation period) i.e. roughly after February 1, we see about linear growth which means that the isolation measures work in so far as exponential spread is stopped (R_effective <= 1)
Unfortunately there is still no sign that the isolation measures are enough to not only slow down but actually END the epidemic (R_effective << 1)

This is crucial, because isolation cannot be maintained forever.










 

gsgs

  • Administrator
  • Full Member
  • *****
  • Posts: 196
    • View Profile
Re: estimating the cfr
« Reply #11 on: February 11, 2020, 08:01:11 am »
the numbers of new confirmed cases Hubei ex Wuhan and China ex Hubei
do really go down.
They did change the definition of confirmed cases somehow and hospitalizations
are still up , but -same argument- the trend should still show up,
if they underreport or change the definition abruptly.
-----------------------------
as for the CFR, I verified the 3 weeks lag , I was reading this :
https://republicanfreedom.com/2020/02/10/the-death-rate-is-up-to-5-the-harrowing-admission-of-a-wuhan-doctor/

found this while searching the link :
https://www.worldometers.info/coronavirus/coronavirus-death-rate/
 > Days from first symptom to death: 14 days

epsilon

  • Newbie
  • *
  • Posts: 27
  • welcome to ncovinfo , here we discuss the pandemic threat from the new 2019 Coronavirus
    • View Profile
Re: estimating the cfr
« Reply #12 on: February 11, 2020, 05:28:22 pm »
Quote
the numbers of new confirmed cases Hubei ex Wuhan and China ex Hubei
do really go down.

This is a good sign that control or even containmant works for prepared cities. They had a head sart of 2 weeks on Wuhan.
And they really implemented drastic measures (e.g. school closures) right from the start even when they had mainly imported cases fro Wuhan.

We should probably do the same now in our on cities/countries. But I'm afraid we are only managing this RE-actively instead of PRO-actively.

What I still not understand is why new infections in Wuhan do not decrease drastically. They live in maximum isolation for 3 weeks now. How could they continue infectiong each other. Airborne transmission ?

See also this very recent paper :

https://www.medrxiv.org/content/10.1101/2020.02.08.20021253v1

reproductive number R_c, i.e. the transmissibility of the virus AFTER isolation/quarantine measuers in China still larger than 1. (Epidemics can only be stopped if R_c < 1.)
My take: Even the draconic chinese isolation measures not yet sufficient.



gsgs

  • Administrator
  • Full Member
  • *****
  • Posts: 196
    • View Profile
Re: estimating the cfr
« Reply #13 on: February 12, 2020, 12:44:56 am »
no measures were really successfull in the past to stop pandemics
or even to isolate cities, countries,continents - keeping the virus out.
Even in times without air-travel, railroad.
And all the papers, the modelings said that it can't be stopped,
just delayed,mitigated to some degree.

So I was surprised when the Chinese numbers went down.
And I'm not sure, that it were the measures. There should be differences
from city to city but apparantly in no city, in no country except maybe Wuhan it did take off yet.

-----------------------------------------------------
https://flutrackers.com/forum/forum/internet-communication/avian-flu-diary/828940-covid-19-dealing-with-imperfect-numbers
« Last Edit: February 12, 2020, 08:16:30 am by gsgs »

epsilon

  • Newbie
  • *
  • Posts: 27
  • welcome to ncovinfo , here we discuss the pandemic threat from the new 2019 Coronavirus
    • View Profile
Re: estimating the cfr
« Reply #14 on: February 13, 2020, 12:25:44 pm »
apparantly in no city, in no country except maybe Wuhan it did take off yet.

Yes. That's a very interesting observation that I cannot explain at the moment.

Given the extremely explosive epidemiologic parameters (latest research: R0>4, doubling each 2.4 days!) we should be really seeing signs of large scale outbreaks some where else by now.  It's impossible that they perfectly identified an isolated all exported cases everywhere.

But all we see is small to mid sized, controlled clusters of 50-100 cases in Singapor and on the cruise ship.

Maybe its just too early.
Remember in Wuhan it was silently circulating from end of November until end of December when the first signs of pneumonia clusters revealed themselves.

The next 2-4 weeks are key !