I think you would have to pick a fixed date for the start, and take your leveling off line and fit a curve that gets you from initial infection to total infection by the projected death leveling off point minus some days between infection and death.

Then you can look at the slope of that line for an R.


You could also look at how the R changes by playing with the initial infection date, but we really have different patient 0s in different parts of the country at different times.

BiochemAg97 said:

---

I think you would have to pick a fixed date for the start, and take your leveling off line and fit a curve that gets you from initial infection to total infection by the projected death leveling off point minus some days between infection and death.

Then you can look at the slope of that line for an R.


You could also look at how the R changes by playing with the initial infection date, but we really have different patient 0s in different parts of the country at different times.

---

and this is one of the biggest issues with anything we are trying to do here. all of the equations for r and growth are "simple" models that assume a fully random population and a clean start date. What we actually have is multiple populations with some level of "contamination" between them, and varying start dates, and sometimes multiple introductions into the same community at different times. so anything we do is just going to be a rough guess. anything better will require much more complex modeling and better access to data than most of us have. it would look similar to what was done here:
https://wwwnc.cdc.gov/eid/article/26/7/20-0282_article

What biochem is saying is what I mean. Start at Jan 15, use the observed slope of deaths for maybe the first 30-45 days - before any big mitigation happened - then generate new estimated cases by assumed CFR, and those will give you Rt and the corresponding population.

OP Plots updated. Added a bunch of info to the last major plot too, taking a look at what the implied R0 might be for each CFR curve and what the final death toll would be in that scenario had it been allowed to simply run its course. Basically for each assumed infection curve I said, if we assume the first cases in the US were 1/19/20 and there were 15 initial cases, what would the R0 be if you assumed perfect exponential growth from that point through that last calculated infection number that's based on real data. i then used that R0 to calculate a death estimate for that CFR had we just let it ride up until herd immunity (1-1/R0). All of this info is in the legend. The projected final infection numbers are based on the curves continuing to bend over since we didn't just let this run.

BlackGoldAg2011 said:

---

Since so many people seem so insistent on wanting to compare this disease to flu but none seem willing to actually do the work to see that comparison and would rather just throw virtual stones in spite of all the knowledge our TexAgs medical community keeps providing, I decided to just go ahead and give them the numbers they so think they desire. All data is based on 2018-2019 US flu season, data sourced here

4/1/20 update notes: Iooking through the data i realized i had been using the wrong flu numbers for total case load and estimated death count. I was using number of cases that sought medical attention not symptomatic cases. This took the 2018-2019 flu totals from 16.5MM to 34.1MM cases and deaths from 16.5k to 34.1k. I have updated all of the graphs to reflect this info. Also, i adjusted the COVID curves in all of the graphs to use daily data expressed in fractional weeks to smooth the curves better rather than taking partial weeks and applying the data to the end of the week.

First up is total cases by week. Total lab confirmed flu cases reported to the CDC in the provided data fell well short of their 34 MM case estimate for the season. So for these plots I took the lab confirmed cases for each week and multiplied them to match the total case load for the year.


(updated 4/17/2020)
This is probably the least fair look because it inflates the Flu lab confirmed cases but leaves the COVID cases as just lab confirmed. Even this though shows the need to "flatten the curve". Looking to the log scale plot, if left exponential, in 3 weeks COVID would overtake total estimated flu cases and in a month would surpass the entire estimated 2018-2019 flu case load.


Since the first look didn't quite seem like a fair comparison, especially since we are seeing our resident medical professionals confirm what we know, that a lot of cases are not getting lab tests, this second look is just looking at total lab confirmed cases submitted to the CDC from all labs


(updated 4/17/2020)
This look is better but still not ideal because this includes all the local/small commercial labs that to date have been unable to run tests on COVID cases. I think so far everyone agrees we are vastly under testing COVID cases to understand our full case load. But even with this issue, COVID will overtake total lab confirmed flu cases in a week. Also worth noting, since flu is not novel and is technically always around, the flu curve starts at 285k cases in the first week, COVID being novel and having an actual start date starts at 15.
4/1/20 update notes: total lab confirmed cases have officially passed 2018-2019 lab confirmed flu cases at equivalent points in the season, and in another few weeks should have eclipsed the season total
4/3/2020 update notes: lab confirmed COVID has officially passed lab confirmed flu at just shy of 7 weeks into the season despite flu having a 20k case head start
4/17/2020 update notes: COVID confirmed cases have so far surpassed Flu confirmed that this graph is about to outlive its usefulness.

Third look is only flu tests run by CDC labs. This should be a closer comparison in the terms of percentage of cases tested since so far CDC has been the only one really able to test COVID cases. It may not be completely fair to flu though because there may be less of a push to have CDC test flu cases since there is less need to do so. That may not be true but is possible and must be acknowledged

(updated 3/27/2020)
No log scale needed for this one... COVID will pass total CDC flu labs fr the whole 2018-2019 season by the end of this week.
3/27/2020 update notes: So it has been brought to my attention I may bee looking at this plot wrong. What I am calling "CDC labs" the fluview site actually is labeling "all public health labs so this might actually include more than just CDC. But we also started including clinical labs in the COVID numbers here in the last week. so due to the discrepancy i'm just going to cut the data off on this graph at the end of week 5 (3/20/2020) and stop updating it. But even at the end of week 5 the point is still clear


ok, ok, i hear you, but this is just case load and tells us nothing about true impact, what about deaths? we all know due to under testing cases our CFR is super inflated right? flu kills so many more people after all. well glad you asked, i've got those numbers for you too. Now I couldn't find deaths by week so i took the total composite CFR from 2018-2019 flu of 0.1% (technically 1 in 1009 but i rounded) and multiplied each weeks case load by that CFR to estimate deaths by week. the actual weeks may be off slightly but the total numbers should be right. Since there are solid arguments for why our COVID CFR may be artificially inflated due to under testing we will be looking at total deaths in the US

First up, the estimated deaths calculated from the total estimated flu cases.


(updated 4/17/2020)
lots of reasons this may not be a fair comparison, but even with its slow start COVID will pass total estimated flu deaths within a week.
3/27/2020 update notes: one point worth noting here, as as of today there still one full day missing from COVID's "week 6" data so by the end of today, COVID will be brushing up against the total estimate flu deaths
4/1/2020 update notes: at this point, basically no matter what we have done to the curve, confirmed COVID deaths will pass total estimated flu deaths at the equivalent point in the season by the end of this week despite Flu's massive head start.
4/3/2020 update notes: COVID confirmed deaths have officially passed flu at just shy of 7 weeks into the season. and depending on your opinion on the correct forecast trend on death count, will pass total flu deaths for 2018-2019 flu season sometime in the next 1-2.5 weeks.
4/17/2020 update notes: As of today, total COVID deaths (34,617) have surpassed total estimated flu deaths from last season (34,200) with no signs of stopping anywhere close to flu


One last look. What I am seeing from our resident TexAgs medical staff is that we are seeing deaths in "unconfirmed" COVID cases due to the lag/shortage in testing, so for a final look lets look at total COVID death count vs the estimated weekly flu deaths calculated from the total lab confirmed flu cases rather than the total estimate:

(updated 4/1/2020)
from this data set, COVID will pass the flu death count estimate by the end of saturday 3/28/2020
3/27/2020 update notes: with one full day of data still missing from week 6, confirmed COVID deaths have already surpassed my calculated estimate of "confirmed flu deaths" for the entire 2018-2019 flu season.
4/1/2020 update notes: I will stop updating this one going forward too as COVID has so surpassed the flu curve that this plot is no longer relevant.

4/1/2020 update notes: I decided to add another look to approximate what our actual real caseload is assuming several different CFRs since we all agree we don't actually know the "denominator". Here is that plot:
4/3/2020 update notes: my original methodology for calculating an estimated true case load for the most recent 2 weeks broke down with the most recent data, so i adjusted how i handled that and changed the methodology description below to match the new graphs
4/8/2020 update notes: I updated my death forecast methodology to try to better capture what a true forecast looks like. this eliminates the need to have two different looks. here is that combined plot as well as supporting methodology and graphs below
4/17/2020 update notes: I updated my CFR curve assumptions to be 21 day average lag between initial infection and death to be more in line with the latest papers being published


(updated 4/17/2020)
Methodology: To come up with these estimated true case load curves, I assumed the death count trails the true case count by 2 weeks, so I took the death total for each day, and use that and the assumed CFR to estimate what the total true case load was 21 days earlier. For the 3 most current weeks, where 3 weeks out death numbers are not available, I used forecasts for the death count. For the death count forecast I started by forecasting the daily death count. This broke into 3 sections. initially the data is in an exponential growth, then it transitions to a 2nd order polynomial to get through the peak and back into a decline, and then moves into exponential decay. I also added a CFR of 2% since that is what S. Korea is trending towards at nearly 1% of their population tested and 60% of their total cases having resolved already. and my observations from this look are:

• First, 0.022% is currently the absolute lower limit for possibilities, because this means that every single person in the USA will have been infected by COVID-19 by 3 months into the US outbreak. So not the lowest feasible CFR, but the lowest that is physically possible.
• I dropped the .022% curve and changed it to .037% CFR because that would mean 60% of the US will be infected in the first 3 months. This should represent the highest "reasonable" number.
• in all but the really deadly CFR scenarios, we have already surpassed last year's flu numbers in estimated true case counts 7 weeks into the season despite flu having a 600k case head start.
• I also added some population % lines at points that are relative to thresholds of infection percentage estimates from past pandemics.
• if my death curve is even close to right, our "bend the curve" efforts are proving effective.

here are some of my supporting plots used in the construction of the above two plots




i'll keep these graphs updated on a regular basis in this first post, and will indicate if i add any observations or change any conclusions.

---

https://www.nationalreview.com/2020/04/coronavirus-kills-more-americans-in-one-month-than-the-flu-kills-in-one-year/

Quote:

---

Although there is still much we don't know about the coronavirus, we know enough to say that it is far more dangerous and deadly than the flu. It took twelve months and 61 million infections for the H1N1 swine flu to kill 12,500 Americans in 200910. The Centers for Disease Control estimated that the seasonal flu killed 34,200 Americans during the 201819 flu season. In 2019, car crashes killed 38,800 Americans.

As for the new coronavirus? On March 20, the death toll in the United States was 225. By April 20, the coronavirus had killed more than 42,000 Americans.

---

Yeah but the flu had a vaccine. /S

PJYoung said:

---

---

Amazing how scary you can make data trends look when you take portions and limited scopes vs. totals and averages.

Some age data from Massachusetts.

I'm working on revamping all of my plots but here's a "fun" one using NY as a case study.

Did the same thing I did for the country level. The relevant takeaway here is that the absolute lowest CFR/IFR can be is 0.094% or roughly equal to that of the flu. and that would mean that as of today, based on projected deaths over the next 3 weeks, 98% of the population of the state of New York have been infected with COVID.

From the article posted above:

Quote:

---

The Wall Street Journal reported that confirmed coronavirus cases in the Italian province of Bergamo (population 1.1 million) had killed 0.2 percent of the entire population in one month. The true percentage may be higher: There were 4,000 more deaths in Bergamo in March 2020 than the average number of deaths in March in recent years, but only 2,000 of those deaths were attributed to confirmed COVID-19 cases.

---

NASAg03 said:

---

PJYoung said:

---

---

Amazing how scary you can make data trends look when you take portions and limited scopes vs. totals and averages.

---

Rather than just throwing stones, why don't you explain why you believe this data presentation is misleading. To me it looks like a useful presentation showing how a novel cause of death is rapidly overtaking other established causes of death, many of which are essentially in steady state. I see a presentation that doesn't tell the entire story, but is a useful look at the data to paint a portion of the picture.

BlackGoldAg2011 said:

---

NASAg03 said:

---

PJYoung said:

---

---

Amazing how scary you can make data trends look when you take portions and limited scopes vs. totals and averages.

---

Rather than just throwing stones, why don't you explain why you believe this data presentation is misleading. To me it looks like a useful presentation showing how a novel cause of death is rapidly overtaking other established causes of death, many of which are essentially in steady state. I see a presentation that doesn't tell the entire story, but is a useful look at the data to paint a portion of the picture.

---

To start with, the average per week lines for heart disease and cancer continue adding deaths all 52 weeks of the year (because it is a weekly average). Since COVID19 is a viral infection, it follows the rules of viruses in a population where the deaths will peak and then drop off. Presumably, it is peaking about now, at just about the average of heart disease, and then will fall, but heart disease and cancer will keep killing people. In other words, COVID-19 may exceed heart disease for a couple of weeks and then fall off again while cancer and heart disease keep racking up high numbers of victims.

Is COVID-19 going to be "no big deal" again once the weekly deaths drop down to well below cancer and heart disease?

This is similar to the fallacy that occurred when people were comparing total flu deaths per year to the instantaneous COVID-19 deaths early on.

BiochemAg97 said:

---

BlackGoldAg2011 said:

---

NASAg03 said:

---

PJYoung said:

---

---

Amazing how scary you can make data trends look when you take portions and limited scopes vs. totals and averages.

---

Rather than just throwing stones, why don't you explain why you believe this data presentation is misleading. To me it looks like a useful presentation showing how a novel cause of death is rapidly overtaking other established causes of death, many of which are essentially in steady state. I see a presentation that doesn't tell the entire story, but is a useful look at the data to paint a portion of the picture.

---

To start with, the average per week lines for heart disease and cancer continue adding deaths all 52 weeks of the year (because it is a weekly average). Since COVID19 is a viral infection, it follows the rules of viruses in a population where the deaths will peak and then drop off. Presumably, it is peaking about now, at just about the average of heart disease, and then will fall, but heart disease and cancer will keep killing people. In other words, COVID-19 may exceed heart disease for a couple of weeks and then fall off again while cancer and heart disease keep racking up high numbers of victims.

Is COVID-19 going to be "no big deal" again once the weekly deaths drop down to well below cancer and heart disease?

This is similar to the fallacy that occurred when people were comparing total flu deaths per year to the instantaneous COVID-19 deaths early on.

---

True, but the purpose of that plot is not to compare total deaths. It is to compare rate of deaths. And what you see from that plot is that even in their peak weeks, none of these other viruses even come close to touching the average weekly mortality of the leading two causes of death. This is a classic case of taking a model or data presentation and trying to use it in a way it was never intended. Of course this is a terrible way to look at total mortality by cause, but that's not the purpose of this plot.

Here is what that plot looks like updated through today by the way:

on a weekly basis COVID has blown all other causes of death from past years out of the water in spite of shutting the country down to try to prevent it.

edit to add:
if you want to look at totals here is your plot. and what this shows is that when comparing against the average Annual Totals from the last 10 years, at less than 2 months from the 10th death in this country, COVID would already be the #10 leading cause of death if you stopped today and will likely end up being #7 or #8 in spite of the country wide shutdown.

It will be interesting to see the effect of COVID-19 on heart disease and other mortalities associated with poor health choices and old age since COVID-19 is preferentially killing the demographic most likely to die from those mortalities in the next year or so.

Updated plots. probably one of my last few updates on this OP, at least updating them all, as the NY case study has basically all but proven the absolute lower limit on the severity for this is the Flu. I'll keep a couple of my favorite graphs updated, but otherwise, if that isn't enough data to convince someone that this is not "just another flu" i don't think there will be any convincing them regardless of the data presented.

BlackGoldAg2011 said:

---

Since so many people seem so insistent on wanting to compare this disease to flu but none seem willing to actually do the work to see that comparison and would rather just throw virtual stones in spite of all the knowledge our TexAgs medical community keeps providing, I decided to just go ahead and give them the numbers they so think they desire. All data is based on 2018-2019 US flu season, data sourced here

4/1/20 update notes: Iooking through the data i realized i had been using the wrong flu numbers for total case load and estimated death count. I was using number of cases that sought medical attention not symptomatic cases. This took the 2018-2019 flu totals from 16.5MM to 34.1MM cases and deaths from 16.5k to 34.1k. I have updated all of the graphs to reflect this info. Also, i adjusted the COVID curves in all of the graphs to use daily data expressed in fractional weeks to smooth the curves better rather than taking partial weeks and applying the data to the end of the week.

First up is total cases by week. Total lab confirmed flu cases reported to the CDC in the provided data fell well short of their 34 MM case estimate for the season. So for these plots I took the lab confirmed cases for each week and multiplied them to match the total case load for the year.


(updated 4/22/2020)
This is probably the least fair look because it inflates the Flu lab confirmed cases but leaves the COVID cases as just lab confirmed. Even this though shows the need to "flatten the curve". Looking to the log scale plot, if left exponential, in 3 weeks COVID would overtake total estimated flu cases and in a month would surpass the entire estimated 2018-2019 flu case load.
4/22/2020 update notes: As of today, this is the only metric in which COVID has not surpassed the flu.


Since the first look didn't quite seem like a fair comparison, especially since we are seeing our resident medical professionals confirm what we know, that a lot of cases are not getting lab tests, this second look is just looking at total lab confirmed cases submitted to the CDC from all labs


(updated 4/22/2020)
This look is better but still not ideal because this includes all the local/small commercial labs that to date have been unable to run tests on COVID cases. I think so far everyone agrees we are vastly under testing COVID cases to understand our full case load. But even with this issue, COVID will overtake total lab confirmed flu cases in a week. Also worth noting, since flu is not novel and is technically always around, the flu curve starts at 285k cases in the first week, COVID being novel and having an actual start date starts at 15.
4/1/20 update notes: total lab confirmed cases have officially passed 2018-2019 lab confirmed flu cases at equivalent points in the season, and in another few weeks should have eclipsed the season total
4/3/2020 update notes: lab confirmed COVID has officially passed lab confirmed flu at just shy of 7 weeks into the season despite flu having a 20k case head start
4/17/2020 update notes: COVID confirmed cases have so far surpassed Flu confirmed that this graph is about to outlive its usefulness.

Third look is only flu tests run by CDC labs. This should be a closer comparison in the terms of percentage of cases tested since so far CDC has been the only one really able to test COVID cases. It may not be completely fair to flu though because there may be less of a push to have CDC test flu cases since there is less need to do so. That may not be true but is possible and must be acknowledged

(updated 3/27/2020)
No log scale needed for this one... COVID will pass total CDC flu labs fr the whole 2018-2019 season by the end of this week.
3/27/2020 update notes: So it has been brought to my attention I may bee looking at this plot wrong. What I am calling "CDC labs" the fluview site actually is labeling "all public health labs so this might actually include more than just CDC. But we also started including clinical labs in the COVID numbers here in the last week. so due to the discrepancy i'm just going to cut the data off on this graph at the end of week 5 (3/20/2020) and stop updating it. But even at the end of week 5 the point is still clear


ok, ok, i hear you, but this is just case load and tells us nothing about true impact, what about deaths? we all know due to under testing cases our CFR is super inflated right? flu kills so many more people after all. well glad you asked, i've got those numbers for you too. Now I couldn't find deaths by week so i took the total composite CFR from 2018-2019 flu of 0.1% (technically 1 in 1009 but i rounded) and multiplied each weeks case load by that CFR to estimate deaths by week. the actual weeks may be off slightly but the total numbers should be right. Since there are solid arguments for why our COVID CFR may be artificially inflated due to under testing we will be looking at total deaths in the US

First up, the estimated deaths calculated from the total estimated flu cases.


(updated 4/22/2020)
lots of reasons this may not be a fair comparison, but even with its slow start COVID will pass total estimated flu deaths within a week.
3/27/2020 update notes: one point worth noting here, as as of today there still one full day missing from COVID's "week 6" data so by the end of today, COVID will be brushing up against the total estimate flu deaths
4/1/2020 update notes: at this point, basically no matter what we have done to the curve, confirmed COVID deaths will pass total estimated flu deaths at the equivalent point in the season by the end of this week despite Flu's massive head start.
4/3/2020 update notes: COVID confirmed deaths have officially passed flu at just shy of 7 weeks into the season. and depending on your opinion on the correct forecast trend on death count, will pass total flu deaths for 2018-2019 flu season sometime in the next 1-2.5 weeks.
4/17/2020 update notes: As of today, total COVID deaths (34,617) have surpassed total estimated flu deaths from last season (34,200) with no signs of stopping anywhere close to flu


One last look. What I am seeing from our resident TexAgs medical staff is that we are seeing deaths in "unconfirmed" COVID cases due to the lag/shortage in testing, so for a final look lets look at total COVID death count vs the estimated weekly flu deaths calculated from the total lab confirmed flu cases rather than the total estimate:

(updated 4/1/2020)
from this data set, COVID will pass the flu death count estimate by the end of saturday 3/28/2020
3/27/2020 update notes: with one full day of data still missing from week 6, confirmed COVID deaths have already surpassed my calculated estimate of "confirmed flu deaths" for the entire 2018-2019 flu season.
4/1/2020 update notes: I will stop updating this one going forward too as COVID has so surpassed the flu curve that this plot is no longer relevant.

4/1/2020 update notes: I decided to add another look to approximate what our actual real caseload is assuming several different CFRs since we all agree we don't actually know the "denominator". Here is that plot:
4/3/2020 update notes: my original methodology for calculating an estimated true case load for the most recent 2 weeks broke down with the most recent data, so i adjusted how i handled that and changed the methodology description below to match the new graphs
4/8/2020 update notes: I updated my death forecast methodology to try to better capture what a true forecast looks like. this eliminates the need to have two different looks. here is that combined plot as well as supporting methodology and graphs below
4/17/2020 update notes: I updated my CFR curve assumptions to be 21 day average lag between initial infection and death to be more in line with the latest papers being published


(updated 4/22/2020)
Methodology: To come up with these estimated true case load curves, I assumed the death count trails the true case count by 2 weeks, so I took the death total for each day, and use that and the assumed CFR to estimate what the total true case load was 21 days earlier. For the 3 most current weeks, where 3 weeks out death numbers are not available, I used forecasts for the death count. For the death count forecast I started by forecasting the daily death count. This broke into 3 sections. initially the data is in an exponential growth, then it transitions to a 2nd order polynomial to get through the peak and back into a decline, and then moves into exponential decay. I also added a CFR of 2% since that is what S. Korea is trending towards at nearly 1% of their population tested and 60% of their total cases having resolved already. and my observations from this look are:

• First, 0.022% is currently the absolute lower limit for possibilities, because this means that every single person in the USA will have been infected by COVID-19 by 3 months into the US outbreak. So not the lowest feasible CFR, but the lowest that is physically possible.
• I dropped the .022% curve and changed it to .037% CFR because that would mean 60% of the US will be infected in the first 3 months. This should represent the highest "reasonable" number.
• The curve showing a CFR lower than flu has been dropped (4/22/2020) due to the fact that currently, to have a CFR even equal to flu, New York State would need to have greater than 100% of its population infected. So everything below 0.1% has been ruled out in my mind.
• in all but the really deadly CFR scenarios, we have already surpassed last year's flu numbers in estimated true case counts 7 weeks into the season despite flu having a 600k case head start.
• I also added some population % lines at points that are relative to thresholds of infection percentage estimates from past pandemics.
• if my death curve is even close to right, our "bend the curve" efforts are proving effective.

here are some of my supporting plots used in the construction of the above two plots






i'll keep these graphs updated on a regular basis in this first post, and will indicate if i add any observations or change any conclusions.

---

BiochemAg97 said:

---

It will be interesting to see the effect of COVID-19 on heart disease and other mortalities associated with poor health choices and old age since COVID-19 is preferentially killing the demographic most likely to die from those mortalities in the next year or so.

---

That will be really interesting to track, especially in the hardest hit areas, because you are right, theoretically this should cause a decrease in those causes of death over the next few years by killing them "early". Now, that's also not considering the possibility that someone who survives this with an underlying condition who was not going to die in the next few years, could have just had their timeline shortened by this.