The predictive value of COVID positivity

[Ausmumof3]

5 minutes ago, Corraleno said:

The 7 day average includes weekends, though. Each day the new deaths for that day replace the deaths for that day in the previous week, so what matters for the average is the difference between, say, this Saturday and last Saturday, not between Saturday and Friday. So the average tends to increase a little over the weekend, and then catch up rapidly during the week. That's why I generally use data from Fridays.

Makes sense.  Is it Friday there now?  I get lost with the time zones.

[Ausmumof3]

10 minutes ago, Corraleno said:

The reason I'm skeptical that average deaths will only be 1839 in two weeks is that deaths have increased by more than 300 in just one week, while this prediction means that deaths will only increase by a little over 400 in the next two weeks. That implies that the death rate will be slowing down while cases are rapidly increasing, which would only be true if positivity were going way down, when it actually seems to be going up, which is why the CFR is creeping up.  

I don't think there is currently a lot of testing of asymptomatic people, I think that was more common when colleges first opened, but a lot of them are sending kids home for Thanksgiving and going virtual thereafter, so probably a lot less testing of students from this point forward. And a lot of people were saying in September that although the case rate then reflected lots of asymptomatic young people, those people were going to be spreading it to their families within a couple of months, and I think we're already seeing that — and about to see even more of it next week.

Yeah I tend to think they’re accelerating rather than that.  The cases per day accelerated pretty steeply last week.

[Corraleno]

3 minutes ago, Ausmumof3 said:

Makes sense.  Is it Friday there now?  I get lost with the time zones.

Yes, but the final numbers for Friday aren't in yet, since it's only 3:00 pm on the west coast.

2 minutes ago, Ausmumof3 said:

Yeah I tend to think they’re accelerating rather than that.  The cases per day accelerated pretty steeply last week.

Average daily cases = 81K three weeks ago — those are the cases that are basically reflected in the current death rate of over 1400. Two weeks ago the cases were at 102K, then last week they jumped to 144K (those are the cases that will be reflected in the death count 2 weeks from now), and today it will be over 170K. And positivity has been going up, not down, so there's no reason to believe that the death rate is going to slow down while cases are rising so steeply. 

[Ausmumof3]

1 minute ago, Corraleno said:

Yes, but the final numbers for Friday aren't in yet, since it's only 3:00 pm on the west coast.

Average daily cases = 81K three weeks ago — those are the cases that are basically reflected in the current death rate of over 1400. Two weeks ago the cases were at 102K, then last week they jumped to 144K (those are the cases that will be reflected in the death count 2 weeks from now), and today it will be over 170K. And positivity has been going up, not down, so there's no reason to believe that the death rate is going to slow down while cases are rising so steeply. 

I think from what we discussed upthread the lead time between cases and deaths gets compressed when the testing is under pressure (maybe due to delays in seeking testing etc or delays in processing tests).  I’m not sure if that effect would have kicked in yet but likely would soon.

[Corraleno]

1 minute ago, Ausmumof3 said:

I think from what we discussed upthread the lead time between cases and deaths gets compressed when the testing is under pressure (maybe due to delays in seeking testing etc or delays in processing tests).  I’m not sure if that effect would have kicked in yet but likely would soon.

Yeah, and that's another factor that suggests that even the grim predictions based on case rate calculations could end up being low rather than too high.  ☹️

[Not_a_Number]

2 hours ago, Corraleno said:

Yes, that would be useful.

CFR of 1.7% x cases 3 weeks ago = 1380
CFR of 1.75% x cases 3 weeks ago = 1420 
CFR of 1.8% x cases 3 weeks ago = 1459

OK, here's how it goes. The minimum positivity in the last trough was 4%. The minimum number of deaths (7-day average) was 706. The positivity 2 weeks ago was 7.8%. 

Therefore, scaling, I predict that the average number of deaths around now is 

7.8/4*706 = 1376.7

That's pretty good as predictions go, I'd say. 

[Corraleno]

17 minutes ago, Not_a_Number said:

OK, here's how it goes. The minimum positivity in the last trough was 4%. The minimum number of deaths (7-day average) was 706. The positivity 2 weeks ago was 7.8%. 

Therefore, scaling, I predict that the average number of deaths around now is 

7.8/4*706 = 1376.7

That's pretty good as predictions go, I'd say. 

That's the exact same number I got last week using cases x a CFR of 1.7%.  But it looks like the actual figure today is going to be >1460, which corresponds to a CFR of 1.8.

I think using cases x CFR didn't make sense early on because basing the CFR on the crazy surges in the early spring when we had no testing would obviously throw things way off. But using the data just for Aug/Sep/Oct gives a CFR of 1.7% and that really does seem to be predictive — except that I would say the CFR is now up to 1.8%. 

Do you think that 1839 is really realistic for 2 weeks from now if today's average is >1460? That's an increase of less than 400 over a period of 2 weeks, when average deaths increased by 350 just in this last week alone. ☹️

[Ausmumof3]

49 minutes ago, Not_a_Number said:

OK, here's how it goes. The minimum positivity in the last trough was 4%. The minimum number of deaths (7-day average) was 706. The positivity 2 weeks ago was 7.8%. 

Therefore, scaling, I predict that the average number of deaths around now is 

7.8/4*706 = 1376.7

That's pretty good as predictions go, I'd say. 

What kind of impact would overwhelmed hospitals have in the calculations do you think?

[Not_a_Number]

6 minutes ago, Corraleno said:

That's the exact same number I got last week using cases x a CFR of 1.7%.  But it looks like the actual figure today is going to be >1460, which corresponds to a CFR of 1.8.

I think using cases x CFR didn't make sense early on because basing the CFR on the crazy surges in the early spring when we had no testing would obviously throw things way off. But using the data just for Aug/Sep/Oct gives a CFR of 1.7% and that really does seem to be predictive — except that I would say the CFR is now up to 1.8%. 

Do you think that 1839 is really realistic for 2 weeks from now if today's average is >1460? That's an increase of less than 400 over a period of 2 weeks, when average deaths increased by 350 just in this last week alone. ☹️

I am not sure 🙂 . It might not be realistic, I suppose, although if we look back 2 weeks, the slope in the positivity WAS highest 2 weeks ago. 

How constant has the CFR stayed? Has it really stayed at 1.7% the whole time the last few months? 

Edited November 21, 2020 by Not_a_Number

[Not_a_Number]

6 minutes ago, Ausmumof3 said:

What kind of impact would overwhelmed hospitals have in the calculations do you think?

That's a reasonable question. I suppose it'd drive the ratio up somewhat, although I do think on average deaths ARE mostly coming from overwhelmed regions. 

[Corraleno]

1 minute ago, Not_a_Number said:

I am not sure 🙂 . It might not be realistic, I suppose, although if we look back 2 weeks, the slope in the positivity WAS highest 2 weeks ago. 

How constant has the CFR stayed? Has it really stayed at 1.7% the whole time the last few months? 

1.7 is the average across the three month period. From Aug 1 to Oct 31, cases increased from 65K-80K and deaths decreased from 1200-850, so if anything that should underpredict deaths with cases rising so steeply now. It was spot-on in terms of predicting deaths for Nov 6th (956 predicted vs 947 actual) and 13th (1105 predicted, 1106 actual), but those deaths were based on new cases in October, when the numbers weren't rising as fast as they are now. So a CFR of 1.8 seems to accurately predict today's deaths of >1460 — and when that is projected forward, that gets really scary really quickly. ☹️

[Not_a_Number]

2 minutes ago, Corraleno said:

1.7 is the average across the three month period. From Aug 1 to Oct 31, cases increased from 65K-80K and deaths decreased from 1200-850, so if anything that should underpredict deaths with cases rising so steeply now. It was spot-on in terms of predicting deaths for Nov 6th (956 predicted vs 947 actual) and 13th (1105 predicted, 1106 actual), but those deaths were based on new cases in October, when the numbers weren't rising as fast as they are now. So a CFR of 1.8 seems to accurately predict today's deaths of >1460 — and when that is projected forward, that gets really scary really quickly. ☹️

But total tests were also MUCH more constant during that time. 

[Corraleno]

8 minutes ago, Not_a_Number said:

But total tests were also MUCH more constant during that time. 

Which means that 1.7% would underpredict going forward, right? Positivity ranged from 6.6-7.4 during that period, and it's now over 12%, so the actual deaths should be higher than predicted using 1.7, and the true CFR for cases that are testing positive right now could be 1.9 or even higher. Which is another reason that I think deaths are going to be waaaaay over 1839 two weeks from now.

Worldometer just ticked over (without Nebraska), and they have average deaths today at 1474 — almost 100 higher than predicted by positivity and 15 higher than predicted by cases x 0.018. Even just using a CFR of 0.017 x 143,700 cases last Friday gives an average of 2440 for deaths on Dec 4th — and x 0.018 = 2590. ☹️

[Not_a_Number]

2 minutes ago, Corraleno said:

Which means that 1.7% would underpredict going forward, right? Positivity ranged from 6.6-7.4 during that period, and it's now over 12%, so the actual deaths should be higher than predicted using 1.7, and the true CFR for cases that are testing positive right now could be 1.9 or even higher. Which is another reason that I think deaths are going to be waaaaay over 1839 two weeks from now.

Worldometer just ticked over (without Nebraska), and they have average deaths today at 1474 — almost 100 higher than predicted by positivity and 15 higher than predicted by cases x 0.018. Even just using a CFR of 0.017 x 143,700 cases last Friday gives an average of 2440 for deaths on Dec 4th — and x 0.018 = 2590. ☹️

Why would you expect it to underpredict as a result? I have a headache, so I'm not quite following. 

[Not_a_Number]

As far as I can tell, more tests mean we're picking up a bigger fraction of cases usually, so then the CFR should go down. 

ETA: you can see that this has generally happened as testing has gone up. Our CFR is much lower than it used to be -- the case graph and the death graph have never tracked. 

Edited November 21, 2020 by Not_a_Number

[Corraleno]

1 minute ago, Not_a_Number said:

As far as I can tell, more tests mean we're picking up a bigger fraction of cases usually, so then the CFR should go down. 

That would be true if the number of positive tests went up while the percent positive went down. But cases and positivity are both skyrocketing, which means that not only are the numbers of diagnosed cases increasing, the numbers of undiagnosed cases that we're missing are also increasing, which would be reflected in an increase in CFR. I just checked the positivity graph that was linked upthread, and its now showing 14% positivity for Nov. 17th (the most recent date they show), so its likely to be even higher than that right now. 

[Not_a_Number]

2 minutes ago, Corraleno said:

That would be true if the number of positive tests went up while the percent positive went down. But cases and positivity are both skyrocketing, which means that not only are the numbers of diagnosed cases increasing, the numbers of undiagnosed cases that we're missing are also increasing, which would be reflected in an increase in CFR. I just checked the positivity graph that was linked upthread, and its now showing 14% positivity for Nov. 17th (the most recent date they show), so its likely to be even higher than that right now. 

I work on the sampling assumption, to be honest. I tend to assume that's we're sampling people in a certain way, and that the positivity is a good estimate of the actual total number of people walking around with the virus. So that means total tests going up means that CFR goes down regardless of what else is going on. 

Do you see what I mean? If a positivity of 10% means roughly that 10% of people with symptoms are sick (I don't think that's any longer true, but it used to yield good estimates before all the college testing), then you could just figure out the total number of people who have COVID by doing something like 

Total number of people expected to be sick * 10%

So then when tests go up, you catch a bigger fraction of all those people, that's all, regardless of what the positivity is doing. 

Anyway, I've been tracking this for a while, and I have been finding it reliable, even during spikes. On the other hand, certainly things change, and I'll be very interested if it stops being a good model. So we'll see. 

[Not_a_Number]

On 11/20/2020 at 4:32 PM, Corraleno said:

I think that when the numbers are this huge and cases cover the entire US, cases x CFR is not a bad predictor. And obviously the CFR changes in line with positivity, so case-based predictions include positivity in that sense. 

Predictions for the first two weeks of November based on cases three weeks earlier x a CFR of 1.7% were absolutely spot on, and for this week the prediction was low — 1380, when we hit 1398 yesterday and will be well into the 1400s by the end of today. So that suggests that the CFR is creeping up, closer to 1.8, which implies we're going to be well over 2000 in 2 weeks (12/4). I think the numbers are going to be all over the place next week, with what's essentially a 4-day holiday weekend , but I think the numbers the week after are going to be really grim. 😕

Comparing notes, what's your model currently predicting? I'm underestimating deaths by a bit (like 150) right now. It's going to be hard to compare over the next few days, due to the long weekend, of course. However, I expect the increase in deaths to end shortly since the positivity has peaked. Of course, it may go back up with the holidays... 

Edited November 26, 2020 by Not_a_Number

[Corraleno]

My original prediction for Friday was 1730, based on a CFR of 1.7, but I think it's going to be higher than that. These were my ballpark figures for this week (actual figures in parentheses):

SAT 1510  (1516)
SUN 1540 (1548)
MON 1580  (1577)
TUE  1650  (1656)
WED 1700  (1708)
THU 1750
FRI 1800

That is based on a CFR of between 1.7 & 1.8, plus looking at patterns from previous weeks in terms of day to day changes.

Edited November 26, 2020 by Corraleno
added today's numbers since Worldometer ticked over

[Not_a_Number]

2 minutes ago, Corraleno said:

My original prediction for Friday was 1730, based on a CFR of 1.7, but I think it's going to be higher than that. These were my ballpark figures for this week (actual figures in parentheses):

SAT 1510  (1516)
SUN 1540 (1548)
MON 1580  (1577)
TUE  1650  (1656)
WED 1700
THU 1750
FRI 1800

That is based on a CFR of between 1.7 & 1.8, plus looking at patterns from previous weeks in terms of day to day changes.

I think I'm not following... how far back are you looking for the case numbers? 

Edited November 26, 2020 by Not_a_Number

[Not_a_Number]

If I use the cases from 3 weeks ago, I'm seeing 108,000 cases, and a CFR of 1.7 already predicts 1836 deaths. This also predicts that very shortly after that, deaths will go up to almost double that, right? 

[Corraleno]

7-day average on Friday Nov. 6th was 102K x 0.017 = 1730 predicted for Friday, Nov. 27* 

7-day average on Friday Nov. 13th was 143K x 0.017 = 2430 predicted for Friday, Dec. 4

*(it's possible there will also be a serious lag in reporting due to what is essentially a 4-day weekend, so there may be lower reported numbers on Friday and big jump the following week).

[Not_a_Number]

6 minutes ago, Corraleno said:

7-day average on Friday Nov. 6th was 102K x 0.017 = 1730 predicted for Friday, Nov. 27* 

7-day average on Friday Nov. 13th was 143K x 0.017 = 2430 predicted for Friday, Dec. 4

*(it's possible there will also be a serious lag in reporting due to what is essentially a 4-day weekend, so there may be lower reported numbers on Friday and big jump the following week).

Ah, you're using the average cases, of course. I wasn't looking at that for some very odd reason. 

So your prediction is that deaths will keep going up for quite a while, right? If my model is correct, they'll peak in about a week and then go down. I don't know exactly what they'll peak at, but possibly under 2,000. Possibly around 2,000, though -- the scaling factor has consistently been less predictable than the timing, and we seem to be considerably above the numbers I'm predicting right now. 

I do expect a serious reporting lag over the weekend and eye-popping numbers on Tuesday. However, it ought to all average out by the end of next week. 

Edited November 26, 2020 by Not_a_Number

[Corraleno]

I think deaths will continue to rise until at least the middle of December. What happens after that depends on what the case rate does after Thanksgiving. ☹️

[Not_a_Number]

15 minutes ago, Corraleno said:

I think deaths will continue to rise until at least the middle of December. What happens after that depends on what the case rate does after Thanksgiving. ☹️

Ah, see, I have a sunnier prediction than you, then 😉 . 

[Not_a_Number]

@Corraleno -- you should take comfort in my prediction, lol. I've followed this model through 4 peaks and troughs, I think, and it's been QUITE reliable on timing and decently reliable on numbers. It could, of course, break down at this point (I'm always willing to adjust models!), but I would guess it won't. And then we won't have as many deaths as you think! 

[Ausmumof3]

17 minutes ago, Not_a_Number said:

@Corraleno -- you should take comfort in my prediction, lol. I've followed this model through 4 peaks and troughs, I think, and it's been QUITE reliable on timing and decently reliable on numbers. It could, of course, break down at this point (I'm always willing to adjust models!), but I would guess it won't. And then we won't have as many deaths as you think! 

What do you think more people seeking extra testing for reassurance pre thanksgiving will do to positivity rates ?

[Not_a_Number]

6 minutes ago, Ausmumof3 said:

What do you think more people seeking extra testing for reassurance pre thanksgiving will do to positivity rates ?

That would drive them down. So it's possible that's why positivity has gone down, of course -- different sampling. In that case, we'd expect it to go right back up and this blip would be artificial. 

On the other hand, the number of tests has been going up linearly during the recent spike, and Thanksgiving didn't change that. It's possible there are enough symptomatic people getting tested that they are swamping the cautious people. 

Edited November 26, 2020 by Not_a_Number

[Ausmumof3]

58 minutes ago, Not_a_Number said:

That would drive them down. So it's possible that's why positivity has gone down, of course -- different sampling. In that case, we'd expect it to go right back up and this blip would be artificial. 

On the other hand, the number of tests has been going up linearly during the recent spike, and Thanksgiving didn't change that. It's possible there are enough symptomatic people getting tested that they are swamping the cautious people. 

Yep that may make sense.  I’m wondering if enough people getting precautionary testing might actually drive things down for a while.  The epidemiologist people on twitter don’t seem so optimistic they think that testing and data reporting is inaccurate this week and next week will be worse but I hope your model is right!  

[Not_a_Number]

25 minutes ago, Ausmumof3 said:

Yep that may make sense.  I’m wondering if enough people getting precautionary testing might actually drive things down for a while.  The epidemiologist people on twitter don’t seem so optimistic they think that testing and data reporting is inaccurate this week and next week will be worse but I hope your model is right!  

My model does make assumptions about sampling, and those assumptions could be wrong due to Thanksgiving! So I'm going to be a lot less surprised if the model turns out to be way off this time than I would have been without this point.

But I still hope it's right 😉 . 

[Corraleno]

14 hours ago, Not_a_Number said:

@Corraleno -- you should take comfort in my prediction, lol. I've followed this model through 4 peaks and troughs, I think, and it's been QUITE reliable on timing and decently reliable on numbers. It could, of course, break down at this point (I'm always willing to adjust models!), but I would guess it won't. And then we won't have as many deaths as you think! 

I really hope this is true. Never have I wished so fervently to be wrong. ☹️

[Not_a_Number]

7 minutes ago, Corraleno said:

I really hope this is true. Never have I wished so fervently to be wrong. ☹️

Fingers crossed... 

[Not_a_Number]

On 11/26/2020 at 12:17 PM, Corraleno said:

I really hope this is true. Never have I wished so fervently to be wrong. ☹️

As you can see, we aren't at 1800 deaths per day on average yet 😉 . We'll probably hit it in a day or two, but we aren't there yet. (Yes, we're playing catch up with missing weekend numbers, but this is the third day we're doing so, and we're probably mostly caught up.) 

[Not_a_Number]

On a grimmer note, the positivity has dipped for only a few days, and is now going back up. As Fauci says... a surge on top of a surge. I don't think we'll even have a similar mini-dip in the deaths, since the deaths ARE smoothed out compared to the actual cases, so at most we'll have a slowing down in the death increase, but not a peak. 

December is going to be brutal, if my model is correct. Absolutely brutal. 

Edited December 3, 2020 by Not_a_Number

[cintinative]

3 minutes ago, Not_a_Number said:

As you can see, we aren't at 1800 deaths per day on average yet 😉 . We'll probably hit it in a day or two, but we aren't there yet. (Yes, we're playing catch up with missing weekend numbers, but this is the third day we're doing so, and we're probably mostly caught up.) 

What is your average again? How many days? Thanks.

[cintinative]

Just now, Not_a_Number said:

 

December is going to be brutal, if my model is correct. Absolutely brutal. 

I saw we lost 3100 people which is more than we lost on September 11th. I am so sad that we are here.

[Not_a_Number]

5 minutes ago, cintinative said:

What is your average again? How many days? Thanks.

You mean how many days am I averaging over? I'm averaging over 7. Just using the Worldometer numbers, since they graph them nicely. 7 days is right, I think, since there are serious fluctuations over days of the week. But it does cycle every week, which makes it the right time to average over. 

Right now, according to Worldometer, we're averaging 1665 deaths per day, if you take the 7 day average. I expect that to go up to 1800 or above today and to keep going up. I don't know how long it'll go up for, since the positivity hasn't apparently actually peaked. 

Edited December 3, 2020 by Not_a_Number

[Corraleno]

I don't think we'll really be caught up until Monday or Tuesday, when the artificial dip after Thanksgiving will drop out of the 7 day average. Right now the average includes a 4 day holiday weekend and we've only started to see the numbers catch up in the last 2 days. The drop in deaths from 2332 the day before Thanksgiving to 1355 the day after is not reflective of a reduction in deaths, especially since the count on the previous Friday was over 2000 with a clear upward trend.

I definitely agree with you that December is going to be really brutal. The CDC and the IHME model are all predicting we'll see ~450K deaths by early March, which is 50-60K per month for the next 3 months. 😥

[Not_a_Number]

25 minutes ago, Corraleno said:

I don't think we'll really be caught up until Monday or Tuesday, when the artificial dip after Thanksgiving will drop out of the 7 day average. Right now the average includes a 4 day holiday weekend and we've only started to see the numbers catch up in the last 2 days. The drop in deaths from 2332 the day before Thanksgiving to 1355 the day after is not reflective of a reduction in deaths, especially since the count on the previous Friday was over 2000 with a clear upward trend.

If you look at previous long weekends, a few day cleared up the backlog. But anyway, we can wait a few days -- I don't think we'll go way over 2000. 

[Bootsie]

I just looked at the number of deaths for my county and realize that Worldometer is reporting 23.9% more deaths for my county than the official county numbers.  My county has numbers broken down for each town in the county and the unincorporated area of the county, the number of males and females, numbers by age group, and other demographic numbers--and those all add to the number the county is reporting for deaths (and matches cases-recovered-active cases).  I checked and the next county over has an even larger discrepancy when I check their official numbers and what is reported on Worldometer.  This makes me question the accuracy for any of the numbers for my state.

[Not_a_Number]

2 hours ago, Corraleno said:

I don't think we'll really be caught up until Monday or Tuesday, when the artificial dip after Thanksgiving will drop out of the 7 day average. Right now the average includes a 4 day holiday weekend and we've only started to see the numbers catch up in the last 2 days. The drop in deaths from 2332 the day before Thanksgiving to 1355 the day after is not reflective of a reduction in deaths, especially since the count on the previous Friday was over 2000 with a clear upward trend.

What's your model predicting for next Friday, by the way? 

ETA: just checked for this week, and I think I'm getting something like 3000 on average by next Friday. I really don't think so. I actually think we'll have a week much like this week.

As is, we aren't close to the 2430 predicted 2 weeks ago. We'll see how it goes tomorrow and over the weekend, of course, but weekends are rarely high times. 

Edited December 4, 2020 by Not_a_Number

[Corraleno]

32 minutes ago, Not_a_Number said:

What's your model predicting for next Friday, by the way? 

~2900

[Not_a_Number]

2 minutes ago, Corraleno said:

~2900

Yeah, I'd be very surprised if we climbed a thousand in a week! I do tend to assume all the data from the long weekend has been already dumped, though. 

[Corraleno]

9 minutes ago, kand said:

Im glad you bumped this. There was an article in The NY Times daily Covid email on Monday that I was saving to share here. Unfortunately, I can’t find where it appears in the paper, so I will paste the beginning of the email here. It made me think of your formulas in this thread, but it is based on cases:

The virus chart that forecasts the future

The number of Covid-19 deaths in the U.S. has dropped in the last few days, but there is reason to think the decline is a statistical mirage — and that deaths are on the verge of surging again.

Why? The relationship between confirmed new coronavirus cases and deaths has held fairly steady this fall. If you track the number of new cases, you can fairly accurately predict the number of deaths three weeks later. Every 100 new cases in the U.S. has led to an average of about 1.7 deaths, with that three-week lag. It’s not a precise equation, of course. The time between diagnosis and death in fatal cases is sometimes shorter than three weeks and sometimes longer. And the death rate is not exactly 1.7 percent. But that simple formula has done a striking job of describing the path of Covid deaths in recent weeks. The chart here shows the relationship — daily deaths compared with an index equal to 1.7 percent of newly diagnosed cases from three weeks earlier. The two lines have risen almost in tandem for the past three months:

Do you have a link to that article? Those are the exact same parameters (CFR of 1.7 + 3 week lag) I calculated and posted about three weeks ago.

ETA: Found it here

Edited December 4, 2020 by Corraleno

[Not_a_Number]

11 minutes ago, kand said:

Im glad you bumped this. There was an article in The NY Times daily Covid email on Monday that I was saving to share here. Unfortunately, I can’t find where it appears in the paper, so I will paste the beginning of the email here. It made me think of your formulas in this thread, but it is based on cases:

The virus chart that forecasts the future

The number of Covid-19 deaths in the U.S. has dropped in the last few days, but there is reason to think the decline is a statistical mirage — and that deaths are on the verge of surging again.

Why? The relationship between confirmed new coronavirus cases and deaths has held fairly steady this fall. If you track the number of new cases, you can fairly accurately predict the number of deaths three weeks later. Every 100 new cases in the U.S. has led to an average of about 1.7 deaths, with that three-week lag. It’s not a precise equation, of course. The time between diagnosis and death in fatal cases is sometimes shorter than three weeks and sometimes longer. And the death rate is not exactly 1.7 percent. But that simple formula has done a striking job of describing the path of Covid deaths in recent weeks. The chart here shows the relationship — daily deaths compared with an index equal to 1.7 percent of newly diagnosed cases from three weeks earlier. The two lines have risen almost in tandem for the past three months:

eta: I didn’t want to keep quoting more, but he goes on to explain that drop off at the end being due to the holiday weekend drop off and testing, and unlikely to be a trend.

Well, he can think so 😉 . I would guess it's a trend. At least, at the moment, positivity is being predictive and cases aren't. If you look at earlier long weekends, it didn't usually take more than a few days to catch up on the long weekend lag. 

[Not_a_Number]

1 minute ago, kand said:

His chart is from Monday, though. Are you saying you think that that drop off at the end is accurate and deaths are dropping? I might not be understanding what you’re saying.

No, it's not accurate, exactly, but I don't think it'll climb the way that chart does. It'll stay under it with a lower slope. 

[Not_a_Number]

@kand: I think the deceptive thing here is that the number of tests hasn't increased much this fall. So, assume positivity is predictive, if the number of tests stays more or less constant, the cases will be predictive. 

But tests have really jumped recently. So now, the prediction from tests and from cases is diverging. And my bet is that the positivity will win the prediction game. 

[Not_a_Number]

Of course, if I'm wrong, I'm wrong, and I'll start following the case counts and not the positivity 😉 . 

[vonfirmath]

I wouldn't have expected more than a 1 day dip in the numbers for Thanksgiving. My husband's team had half of them come in to work Friday in making sure tests were processed and half came in Saturday -- to make sure the test data was processed in a timely basis. They will have to do the same thing at Christmas (And are already working more Sunday hours than they ever had to before)

I mean -- people aren't going to not die just because it is Thanksgiving. It's just recording the death that might be late.

[Ausmumof3]

I saw a post recently about how hospitalisation percentages are going down.  That sounds like a good thing and that we are catching more mild cases.  But given the test positivity is still increasing its likely not that.  So the most likely explanation so that the bar for hospitalisation is increasing which is obviously not a good thing.

Edited December 4, 2020 by Ausmumof3