Jump to content

Menu

The predictive value of COVID positivity


Not_a_Number
 Share

Recommended Posts

8 minutes ago, Pen said:

The biggest mental disconnect I have is that using OurWorldinData graphs, new daily deaths in many places including USA seemed to be only climbing a little or more or less leveled, while new daily cases have been climbing a lot.   

That's because we're running more tests. That's why you want to do percent positivity, and not total cases. 

Link to comment
Share on other sites

11 minutes ago, Pen said:

Or Wyoming and Mississippi have similar positivity rate, and similar number of tests per 1000 population, but very different case numbers so presumably will have quite different death numbers in a few weeks. 

 But you don't expect same positivity to lead to the same number of deaths. You have to scale by the size of population. And even then it's not perfect, because different states have different testing protocols -- as @Matryoshkanoted, some places are testing more asymptomatic people than other states, so then that messes with the ratio. 

Overall, though, it does seem to average out 🙂  .

Link to comment
Share on other sites

22 minutes ago, Not_a_Number said:

That's because we're running more tests. That's why you want to do percent positivity, and not total cases. 

 

So, working backwards using chart below - you believe CV19  IFR  is the same now as it ever has been and that there were actually far more total cases back in March or so (3 weeks before peak deaths)  than total cases 3 weeks before now? 

USA daily deaths from Worldometer. 

 

I cannot disprove that there could actually have been more than twice as many cases in early spring than recently. It is possible when many people weren’t yet realizing there was a problem and were still going about regular daily life, but it doesn’t make sense to me that that was actually the case. 

I think it was much more already here than we realized much earlier than we realized, but actually more than twice the cases then as now is not matching my general sense of people I know (or know by a close degree of separation) getting sick.

  

My sense is it is much more widespread in population now.  (But less outbreaks in places like Nursing Homes.) 

 

58C31EEE-F752-4959-8D15-01F91AAEB91D.png

Edited by Pen
Link to comment
Share on other sites

1 minute ago, Pen said:

I cannot disprove that there could actually have been more than twice as many cases in early spring when many people weren’t yet realizing there was a problem and were still going about regular daily life as there have been recently, but it doesn’t make sense to me that that was actually the case. 

Well, that's certainly what the antibody testing in NY suggests. A sixth of the people in NYS seemed to have antibodies according to their sampling. A fifth in the city. 

Link to comment
Share on other sites

11 minutes ago, Not_a_Number said:

Well, that's certainly what the antibody testing in NY suggests. A sixth of the people in NYS seemed to have antibodies according to their sampling. A fifth in the city. 

 

Doesnt that then go back to idea that stats from a small (geographic) part of USA are largely driving the national stats?

 

Edited by Pen
Link to comment
Share on other sites

 

Are you sure they aren’t now?

Like the now lower rates in NYC area particularly and Northeast in general aren’t what is holding the more or less steady looking graph despite flare ups here and there elsewhere in country —but that the elsewhere peaks and dips aren’t enough to offset the relatively stable graph line now ? 

 

Iow theres a huge population which had massive outbreak early on and now is down to low and pretty steady daily deaths - not just NY but also NJ, CT etc

then elsewhere some places are still doing various forms of “lockdowns” or having “opening” or having flare ups locally addressed and doused so that on a big national average the whole is relatively steady.    But the 20% and 15% etc positivity rates don’t much affect the national statistics because of small populations. 

And if that’s so, does it mean that measures like masks are Not actually helping as much as I had hoped? Or not much? That ifr is just as high with implication that reducing viral load if people get sick at all just as readily leads to death. No lesser viral load less severe case correspondence?

 

Edited by Pen
  • Like 2
Link to comment
Share on other sites

7 minutes ago, Pen said:

 

Are you sure they aren’t now?

Like the now lower rates in NYC area particularly and Northeast in general aren’t what is holding the more or less steady looking graph despite flare ups here and there elsewhere in country —but that the elsewhere peaks and dips aren’t enough to offset the relatively stable graph line now ? 

Yes, I'm sure. You can just look at the number of tests coming out of each state and you can see that right now it's close to proportional to population. 

  • Like 1
Link to comment
Share on other sites

9 minutes ago, Pen said:

Are you sure they aren’t now?

Like the now lower rates in NYC area particularly and Northeast in general aren’t what is holding the more or less steady looking graph despite flare ups here and there elsewhere in country —but that the elsewhere peaks and dips aren’t enough to offset the relatively stable graph line now ? 

You can take a look at Worldometer tests/million population column here: 

https://www.worldometers.info/coronavirus/country/us

Link to comment
Share on other sites

 

9 minutes ago, Pen said:

if that’s so, does it mean that measures like masks are Not actually helping as much as I had hoped? Or not much? That ifr is just as high with implication that reducing viral load if people get sick at all just as readily leads to death. No lesser viral load less severe case correspondence?

 

I added above  as our posts crossed

Link to comment
Share on other sites

11 minutes ago, Pen said:

And if that’s so, does it mean that measures like masks are Not actually helping as much as I had hoped? Or not much? That ifr is just as high with implication that reducing viral load if people get sick at all just as readily leads to death. No lesser viral load less severe case correspondence?

I don't think masks by themselves help that much. It's like, one of the things you can do, but you have to do quite a few of them to make a good dent in the rate of transmission. 

I'm obviously wearing masks everywhere, so I do think they HELP. But we're still avoiding the indoors entirely, and I think that helps a lot more. 

  • Like 3
Link to comment
Share on other sites

1 minute ago, Not_a_Number said:

You can take a look at Worldometer tests/million population column here: 

https://www.worldometers.info/coronavirus/country/us

 

Yet the list I found seems to show a good bit of variation in tests per 1000 population for recent testing . 

with NY doing around 3x as many per 1000 as California . Some states near 1 test per 1000 population others above 9 tests per 1000 population.  

Link to comment
Share on other sites

2 minutes ago, Pen said:

Yet the list I found seems to show a good bit of variation in tests per 1000 population for recent testing . 

with NY doing around 3x as many per 1000 as California . Some states near 1 test per 1000 population others above 9 tests per 1000 population.  

It's true there are differences. But it's much less stark than it was back in the spring. And much more geographically varied. 

Link to comment
Share on other sites

3 minutes ago, Pen said:

Yet the list I found seems to show a good bit of variation in tests per 1000 population for recent testing . 

with NY doing around 3x as many per 1000 as California . Some states near 1 test per 1000 population others above 9 tests per 1000 population.  

Can you show me this list, by the way? I'd be curious to see. 

Link to comment
Share on other sites

 

I am wanting to think that measures matter and are not just window dressing.

At same time I do realize things like that if my nephew had gotten his confirmed sickness with CV19 while at camp a year ago it would just have been seen as a 48 hour respiratory virus, for which no one would have been tested or quarantined or contact traced.

And mildness of his case and other family not catching it may have had nothing to do with any carefulness. 

And there could have been a massive amount of that sort of thing in January to March. 

  • Like 4
Link to comment
Share on other sites

7 minutes ago, Not_a_Number said:

I've seen basically no evidence for that. And I'm going to live my life as if that's false until I see some evidence. 

 

Well, same here. Still using masks, still distancing etc.

But no change in IFR idea —if true and I don’t really think it is true— suggests that nothing helps, not even “protecting the most vulnerable” or else ifr should be lower than it was when lots of NH etc had outbreaks. 

  • Like 1
Link to comment
Share on other sites

12 minutes ago, Not_a_Number said:

I've seen basically no evidence for that. And I'm going to live my life as if that's false until I see some evidence. 

 

MAnd while I don’t really think we could have been like Sweden, it is looking pretty good at the moment

Worldometer daily deaths Sweden:

FC120B97-9DDB-438C-8CAA-F6E47BEC6236.png

Link to comment
Share on other sites

6 minutes ago, Pen said:

Well, same here. Still using masks, still distancing etc.

But no change in IFR idea —if true and I don’t really think it is true— suggests that nothing helps, not even “protecting the most vulnerable” or else ifr should be lower than it was when lots of NH etc had outbreaks. 

And do nursing homes currently not have outbreaks? I have the impression that the Northeast actually did a decent job with its nursing home outbreaks, all things considered. Lots of states have had lots of nursing home outbreaks and deaths: 

https://data.cms.gov/stories/s/COVID-19-Nursing-Home-Data/bkwz-xpvg/

Edited by Not_a_Number
Link to comment
Share on other sites

1 minute ago, Pen said:

 

MAnd while I don’t really think we could have been like Sweden, it is looking pretty good at the moment

Worldometer daily deaths Sweden:

FC120B97-9DDB-438C-8CAA-F6E47BEC6236.png

Yeah, Sweden is doing fine right now. Of course, they still have restriction on large gatherings, right? And they didn't have high schools or universities open. 

 

Link to comment
Share on other sites

36 minutes ago, Pen said:

But no change in IFR idea —if true and I don’t really think it is true— suggests that nothing helps, not even “protecting the most vulnerable” or else ifr should be lower than it was when lots of NH etc had outbreaks. 

I'm not following your logic at all here.  Wearing masks or not nor any other preventive measure is going to affect the IFR of the virus short of mutation or a miracle cure.  IFR is how many people die of those infected, not of the total population.  Masks might mean that many less people get infected in the first place, but aren't necessarily going to mean you're less likely to die from it if you do get it (which is all IFR measures).  Well, maaaybe if you get a lower viral load, but it does seem pretty darn random which people get hit hard by this...

But the idea is to infect less people in the first place, and stop spread.  Masks aren't perfect (especially if inside over an extended period) but that doesn't mean they don't help cut down the overall spread.

If a person manages to avoid getting it by avoiding being inside with others, distancing, wearing masks, washing hands, etc., then that person's IFR is Zero.  Because no Infection.

So, for example, if 10,000 people in somewhere get it without preventive measures, and the IFR were 1%, 100 people die.  If with preventive measures only 1000 people get it with the same IFR, 10 people die.  Same IFR, less death.  That's why we have so many more fatalities per capita here than anyplace else - more people per capita have caught it.  The other countries don't have a lower IFR, they have been taking better preventive measures.

Edited by Matryoshka
  • Like 3
Link to comment
Share on other sites

30 minutes ago, Matryoshka said:

I'm not following your logic at all here.  Wearing masks or not nor any other preventive measure is going to affect the IFR of the virus short of mutation or a miracle cure.  IFR is how many people die of those infected, not of the total population.  Masks might mean that many less people get infected in the first place, but aren't necessarily going to mean you're less likely to die from it if you do get it (which is all IFR measures).  Well, maaaybe if you get a lower viral load, but it does seem pretty darn random which people get hit hard by this...

But the idea is to infect less people in the first place, and stop spread.  Masks aren't perfect (especially if inside over an extended period) but that doesn't mean they don't help cut down the overall spread.

If a person manages to avoid getting it by avoiding being inside with others, distancing, wearing masks, washing hands, etc., then that person's IFR is Zero.  Because no Infection.

So, for example, if 10,000 people in somewhere get it without preventive measures, and the IFR were 1%, 100 people die.  If with preventive measures only 1000 people get it with the same IFR, 10 people die.  Same IFR, less death.  That's why we have so many more fatalities per capita here than anyplace else - more people per capita have caught it.  The other countries don't have a lower IFR, they have been taking better preventive measures.

I wasn't entirely sure what she meant, either. However, it's possible that she was talking about decreasing viral load changing IFR? I'm not really sure. I'm mostly wearing a mask to not catch it, period. 

Link to comment
Share on other sites

Just now, fifiruth said:

Not_a_Number, could you explain your theory again, and how it differs? Are you getting your info from the CDC?

By the way, I clearly see that I cannot hold a candle to you and the others in your statistical analysis! I appreciate the determination to stick to facts and actual data. I’ll just try to follow along. 

My theory is that current rates of positivity are the best measure of the actual number of people infected 🙂 . I’m testing this by comparing positivity 2-3 weeks ago to deaths today, since the actual NUMBER of infections is quite well correlated to deaths 2-3 weeks later, at least in our current regime of a lot of spread.

Link to comment
Share on other sites

2 minutes ago, Happymomof1 said:

 

Then I don't understand. I thought that was the whole point with the graphs and such how the positivity can predict the death rate.

And Texas as a whole won't truly help. You have El Paso.  You have schools that are ONLY meeting virutally. You have schools like mine that meet with no masks for elementary. So many variable. 

It can predict it, yeah, but I can’t promise it works for every single state. Texas testing data is very spiky and unreliable.

The fact that the data is different in different places is fine, though. It averages out. That’s how stats work. But I’m not sure Texas is reporting its tests on time, by the looks of its graph...

If I had to make a prediction using the relatively bad data Texas is providing, I’d say deaths will inch up slightly in the coming weeks. But that’s not what I’d call a certainty.

Link to comment
Share on other sites

8 minutes ago, Happymomof1 said:

 

Then I don't understand. I thought that was the whole point with the graphs and such how the positivity can predict the death rate.

And Texas as a whole won't truly help. You have El Paso.  You have schools that are ONLY meeting virutally. You have schools like mine that meet with no masks for elementary. So many variable. 

If you look at the Texas positivity graph, you’ll see what I mean:

https://coronavirus.jhu.edu/testing/individual-states/texas

Link to comment
Share on other sites

8 hours ago, Happymomof1 said:

Yeah. I've seen the graph but I don't understand it. I don't see how they get the line they show. I just want regular stats that tell me what will happen. I was hoping you could give it to me.

Rough estimate — take the population of the region in millions and divide by three. Multiply that by the positivity rate today to get the death count two weeks from today.  
 

You should use the 7-day rolling average for both positivity rate and death count to cancel out weekend reporting effects.

 

If you have past data for your region you can test that. You might find out that 2.5 or 3.5 is closer. But if it’s fairly constant over time you can use it to make an educated guess about two weeks from now.

  • Like 1
Link to comment
Share on other sites

18 minutes ago, Danae said:

Rough estimate — take the population of the region in millions and divide by three. Multiply that by the positivity rate today to get the death count two weeks from today.  
 

You should use the 7-day rolling average for both positivity rate and death count to cancel out weekend reporting effects.

 

If you have past data for your region you can test that. You might find out that 2.5 or 3.5 is closer. But if it’s fairly constant over time you can use it to make an educated guess about two weeks from now.

Thanks! That's much better than my waffling 😉 .  

Let's see how well that works with Texas, one sec... if you try to predict today's deaths, that gets you 7*29/3 or so, which is about 68. Pretty good! 

So we expect that to go up a bit but not a ton to 

9*29/3 = 87,

which you should really just treat as "about 90" in a few weeks. 

Edited by Not_a_Number
Link to comment
Share on other sites

16 minutes ago, Danae said:

Rough estimate — take the population of the region in millions and divide by three. Multiply that by the positivity rate today to get the death count two weeks from today.  
 

You should use the 7-day rolling average for both positivity rate and death count to cancel out weekend reporting effects.

 

If you have past data for your region you can test that. You might find out that 2.5 or 3.5 is closer. But if it’s fairly constant over time you can use it to make an educated guess about two weeks from now.

Texas data really is a mess, though. They report SUCH an uneven number of daily tests that it's obvious they are not reporting tests by day in any way. At least Minnesota seems to stay more or less caught up weekly. 

Link to comment
Share on other sites

  • 2 weeks later...
On 10/26/2020 at 10:38 AM, Happymomof1 said:

Yeah. I've seen the graph but I don't understand it. I don't see how they get the line they show. I just want regular stats that tell me what will happen. I was hoping you could give it to me.

OK, we're now about 2 weeks later, a little less. The prediction was 87, and right now, the 7-day average is 92. So as you can see, it's not bad. 

@Danae, I've found that the death:positivity ratio has crept up, probably due to all the testing of asymptomatic people we're doing nowadays. So you may want to adjust your formula slightly. 

Link to comment
Share on other sites

1 minute ago, Not_a_Number said:

OK, we're now about 2 weeks later, a little less. The prediction was 87, and right now, the 7-day average is 92. So as you can see, it's not bad. 

@Danae, I've found that the death:positivity ratio has crept up, probably due to all the testing of asymptomatic people we're doing nowadays. So you may want to adjust your formula slightly. 

I was thinking about using the  average actual ratio over the previous 30 days as the multiplier for predictions.  That would take into account both changes in testing and any improvements in treatment.  

I've noticed looking back that, while always close to the predicted curve, the actual death line tends to be slightly below the prediction when the curve is flat or decreasing and above it when it's increasing.  Right now it's increasing steeply.  We're going to be back where we were in April fairly soon.  😞

  • Like 1
Link to comment
Share on other sites

On 11/6/2020 at 9:56 PM, kand said:

I lost this thread and have had a browser tab open for a couple weeks to share a Twitter thread that make me think of this thread. This guy isn’t using positivity, but I still thought you might find his analysis interesting. 
 

 

 

I'm not sure about this analysis. I skimmed, to be honest, but the thing that makes me skeptical is that I haven't found that in countries that keep good track of their cases, there's such a long lag between cases and deaths. It's usually 2-3 weeks. Here's South Korea, for example: 

https://www.worldometers.info/coronavirus/country/south-korea/

I would guess part of the reason for that is that cases aren't reported on the day they are tested, anyway. I would also guess that case numbers themselves aren't actually indicative of the cases walking around in the population... so, for instance, he mentions a lack of Florida September spike, but if you look at Florida in September, the positivity was at the lowest levels in ages. In fact, it has stayed lowish until recently, and now it's going up again... and I expect deaths will now go up.

To be clear, I expect that positivity IS actually a measure of case numbers, and our measure of case numbers is woefully inadequate. It's not like I think positivity is magic in some way 😉

 

  • Like 1
Link to comment
Share on other sites

On 11/6/2020 at 9:53 PM, Danae said:

I was thinking about using the  average actual ratio over the previous 30 days as the multiplier for predictions.  That would take into account both changes in testing and any improvements in treatment.  

That's a great idea. You should do that. 

 

On 11/6/2020 at 9:53 PM, Danae said:

I've noticed looking back that, while always close to the predicted curve, the actual death line tends to be slightly below the prediction when the curve is flat or decreasing and above it when it's increasing.  Right now it's increasing steeply.  We're going to be back where we were in April fairly soon.  😞

Yeah, the positivity is going way up. We're in for a rough time, especially since half the country has decided not to worry about it... 

  • Sad 1
Link to comment
Share on other sites

On 11/6/2020 at 9:53 PM, Danae said:

I was thinking about using the  average actual ratio over the previous 30 days as the multiplier for predictions.  That would take into account both changes in testing and any improvements in treatment.  

I've noticed looking back that, while always close to the predicted curve, the actual death line tends to be slightly below the prediction when the curve is flat or decreasing and above it when it's increasing.  Right now it's increasing steeply.  We're going to be back where we were in April fairly soon.  😞

Hey, any interest in calculating the different multipliers for the different states? Make it the multiplier per population, so they are actually measuring the same thing for different states. I'm curious how similar the ratios are between different states. (I'd guess at least somewhat different, depending on sampling procedure, but I'm very curious.) 

Link to comment
Share on other sites

1 minute ago, Not_a_Number said:

Hey, any interest in calculating the different multipliers for the different states? Make it the multiplier per population, so they are actually measuring the same thing for different states. I'm curious how similar the ratios are between different states. (I'd guess at least somewhat different, depending on sampling procedure, but I'm very curious.) 

Possibly next week.  I’m working on application materials for a tenure-track position at the moment and can’t start any new projects until they’re done.  (Kind of a long shot, and don’t know what me going full-time will mean for homeschooling, but . . . )

Link to comment
Share on other sites

2 minutes ago, Danae said:

Possibly next week.  I’m working on application materials for a tenure-track position at the moment and can’t start any new projects until they’re done.  (Kind of a long shot, and don’t know what me going full-time will mean for homeschooling, but . . . )

Oooh, good luck! That's stressful. 

Link to comment
Share on other sites

On 10/26/2020 at 7:26 PM, Danae said:

Rough estimate — take the population of the region in millions and divide by three. Multiply that by the positivity rate today to get the death count two weeks from today.  
 

You should use the 7-day rolling average for both positivity rate and death count to cancel out weekend reporting effects.

 

If you have past data for your region you can test that. You might find out that 2.5 or 3.5 is closer. But if it’s fairly constant over time you can use it to make an educated guess about two weeks from now.

Let me see if I have this right, using a couple of examples......

Current positivity rates are: for SD, an utterly eye-popping 54% - it's been around 50% for at least a week; Iowa, 48%; and Idaho, nearly 40%.....would I then take the population of the entire state in each case? Or would I need to break it down by region & try to find the regional positivity rate?

Because using the population of SD (895,000) / 3 = 295,000 x 0.50 positivity rate = 147,500 dead people.

Obviously, I'm not doing something right.

Link to comment
Share on other sites

6 minutes ago, Happy2BaMom said:

Let me see if I have this right, using a couple of examples......

Current positivity rates are: for SD, an utterly eye-popping 54% - it's been around 50% for at least a week; Iowa, 48%; and Idaho, nearly 40%.....would I then take the population of the entire state in each case? Or would I need to break it down by region & try to find the regional positivity rate?

Because using the population of SD (895,000) / 3 = 295,000 x 0.50 positivity rate = 147,500 dead people.

Obviously, I'm not doing something right.

No, population in millions. So for SD that'd be 0.9 or so, not what you did. And you leave the positivity number as how many percent it was. So then you get 

0.9/3*50 =15, 

which seems pretty plausible, I gotta say. 

Edited by Not_a_Number
Link to comment
Share on other sites

1 minute ago, Not_a_Number said:

No, population in millions. So for SD that'd be 0.9 or so, not what you did. And you leave the positivity number as how many percent it was. So then you get 

0.9/3*50 =15, 

which seems pretty plausible, I gotta say. 

Ah, "millions". Yes, that does affect a different result, doesn't it?

Well, good think I'm no longer teaching math.

I thought 147,500 was a little draconian. Entire state'd be wiped out in less than a week. 😛

  • Haha 1
Link to comment
Share on other sites

5 minutes ago, Happymomof1 said:

Ok, the rate I am seeing for SD is 23.2 percent.  So divide 895,000 by 3 and you get 298,333.  Now you  multiply that by 23.2 and you get 69,213.   Now that should be in about a month if I understand it.  I was using this to find the percentage:  https://covidactnow.org/?s=1304314

I use John Hopkins. Here is what they show for SD - well over 50%.

And here is a summary of positivity rates by state, based on JH data.

Not looking to quibble over it, just backing up my stats above.

Edited by Happy2BaMom
  • Like 2
Link to comment
Share on other sites

On 10/18/2020 at 6:38 AM, Pen said:

 

MAnd while I don’t really think we could have been like Sweden, it is looking pretty good at the moment

Worldometer daily deaths Sweden:

FC120B97-9DDB-438C-8CAA-F6E47BEC6236.png

Slowly ticking back up now unfortunately although they are introducing some measures to slow it down.

 

And here is the cases.  Deaths lag is 4-6 weeks of more so this could be quite scary in weeks to come.  Obviously the earlier spike is probably underrepresented due to lack of testing.

 

 

 

 

 

 

 

C28CAB25-71EF-4789-9744-8EE8708007BB.jpeg

C9A5A7A1-4B20-4D38-A067-D38710DF1115.jpeg

Edited by Ausmumof3
Tidying up my screen shots
Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

 Share

×
×
  • Create New...