COVID-19 Shutdown Exit Strategy?

[audreyh1]

In my view, comparisons between states or between countries is of limited utility, because there are differences in demographics, testing, record keeping and even culture that can confound the data. What is crucial, I think, is for a country or a state to compare its own numbers to those in an earlier period, so that we can tell if we are making progress against the disease and detect whether our actions are helping or harming.

My own personal metrics of interest in my state (CT) are the daily percentage of new cases versus cumulative cases (4 day rolling average) and the net number of people in the hospital each day with COVID-19. The first (new case %) has been on a steady downward trend since March 18th and is now below 1%. During that period, daily testing has greatly ramped up, so I know that we are discovering more cases, which tells me the actual spread of the disease is abating even faster. This data is confirmed, with about an 8 day lag, in the percentage of new deaths to cumulative deaths. The second metric (hospitalization) has been on a steady downward trend since peaking on April 22nd, also indicating that the disease is abating. The website rt.live calculates that the effective reproduction number in CT is now at 0.87. R sub t of less than 1.0 is a sign the disease is fading rather than growing.

That’s fine and dandy until folks from other states with worse stats travel to your state.

 

[Leo1277]

My own personal metrics of interest in my state (CT) are the daily percentage of new cases versus cumulative cases (4 day rolling average) and the net number of people in the hospital each day with COVID-19. The first (new case %) has been on a steady downward trend since March 18th and is now below 1%. During that period, daily testing has greatly ramped up, so I know that we are discovering more cases, which tells me the actual spread of the disease is abating even faster. This data is confirmed, with about an 8 day lag, in the percentage of new deaths to cumulative deaths. The second metric (hospitalization) has been on a steady downward trend since peaking on April 22nd, also indicating that the disease is abating. The website rt.live calculates that the effective reproduction number in CT is now at 0.87. R sub t of less than 1.0 is a sign the disease is fading rather than growing.

Thanks much for the link to that site rt.live, this is very useful and interesting. Let's hope they compute that momentary reproduction rate correctly, like from hospitalizations, but if they do, this is the most important quantity you want to watch.

About your first measure (new cases divided by cumulative cases if I understand correctly), I think this would not be such a great indicator. For example, if you just have a steady state with R(t) = 1, then your new cases stay constant week by week, but your cumulative cases increase steadily; so your measure would fall steadily to zero over time. This may also be the reason why you report it was already going down from March 18th. I am pretty sure at that time the daily/weekly new cases were still rising rather rapidly at that time, it was just that since the disease was new, there weren't many cumulative cases earlier, and so your denominator got bigger quicker than the numerator.

But overall, all this is a rather high quality discussion, and I continue to be impressed by ER.org

 

[Gumby]

That’s fine and dandy until folks from other states with worse stats travel to your state.

They do now; we don't stop people from coming here. In fact, there has been an exodus of people from New York in recent months. The house across the street from me is for sale. All week long, I've seen people with New York license plates parked in the driveway next to the realtor's car as she shows them the house. The other house for sale on my street was just the subject of a bidding war by a bunch of people from New York.

The fact that they are from New York tells me nothing about them as people. I don't know what they have or haven't been doing as far as social distancing and use of PPE, so I can't conclude that they represent a greater risk to me when they come here than the risk that already exists in my neighborhood. Heck, the neighbor down the street caught COVID in March (she has since recovered).

 

[Gumby]

About your first measure (new cases divided by cumulative cases if I understand correctly), I think this would not be such a great indicator. For example, if you just have a steady state with R(t) = 1, then your new cases stay constant week by week, but your cumulative cases increase steadily; so your measure would fall steadily to zero over time. This may also be the reason why you report it was already going down from March 18th. I am pretty sure at that time the daily/weekly new cases were still rising rather rapidly at that time, it was just that since the disease was new, there weren't many cumulative cases earlier, and so your denominator got bigger quicker than the numerator.

The principal value of plotting the percentage increase in new cases is to monitor the rate of change rather than the actual number of cases. If the line slopes downward, you are no longer in an exponential growth phase. I fully expect that metric to asymptotically approach zero for the reason you have noted, as well as because we took actions to mitigate the spread, which is why I expanded the vertical scale of my graph at the end of April to more clearly see the variations. If we were to see an upward sloping line for week, that may indicate that our relaxation of the shutdown is having a negative effect.

 

[audreyh1]

They do now; we don't stop people from coming here. In fact, there has been an exodus of people from New York in recent months. The house across the street from me is for sale. All week long, I've seen people with New York license plates parked in the driveway next to the realtor's car as she shows them the house. The other house for sale on my street was just the subject of a bidding war by a bunch of people from New York.

The fact that they are from New York tells me nothing about them as people. I don't know what they have or haven't been doing as far as social distancing and use of PPE, so I can't conclude that they represent a greater risk to me when they come here than the risk that already exists in my neighborhood. Heck, the neighbor down the street caught COVID in March (she has since recovered).

I doubt the individuals traveling are anywhere near pre-Covid as a big part of the multi-state shutdown as most stopped business and personal travel except for essential services. I’m sure that helped limit the spread immediately and considerably.

A lot of infections were spread early on by individuals traveling within states and between states. As US mobility ramps back up, the problem returns and counties and states which have successfully lowered their infection rates are still going to have challenges from imported cases. It’s just part of an epidemic - you can’t get it down to zero without complete isolation. And you probably can’t have complete isolation in regions within our country except for Hawaii and maybe Alaska requiring 14 day quarantine on any visitors and returning residents.

So even if your county or state is doing great, unless the rest of the country is doing as well, you are still vulnerable.

 

[Leo1277]

The principal value of plotting the percentage increase in new cases is to monitor the rate of change rather than the actual number of cases. If the line slopes downward, you are no longer in an exponential growth phase. I fully expect that metric to asymptotically approach zero for the reason you have noted, as well as because we took actions to mitigate the spread, which is why I expanded the vertical scale of my graph at the end of April to more clearly see the variations. If we were to see an upward sloping line for week, that may indicate that our relaxation of the shutdown is having a negative effect.

That is very true, if you are in a pure exponential growth, then your number will stay constant. In the very beginning of an outbreak, this is a good measure to see whether you are having an impact with suppression methods and succeed in going sub-exponential.

But in the long run, for example where we are now, we are likely more interested in whether re-opening leads to new growth, and your curve will not show this. The R(t) curve does capture that part, and if R(t) were to go to, say, 1.1, then you would still see decreases in your measure. I think for such long term situations, you could modify your measure by purging historical cases by, for example, comparing the ratio of "new cases over the last week" and "new cases over the last month" instead of "new cases since I started counting". Then if R(t) would go beyond 1, you would see this in your model as its value would go up too.

 

[Gumby]

That is very true, if you are in a pure exponential growth, then your number will stay constant. In the very beginning of an outbreak, this is a good measure to see whether you are having an impact with suppression methods and succeed in going sub-exponential.

But in the long run, for example where we are now, we are likely more interested in whether re-opening leads to new growth, and your curve will not show this. The R(t) curve does capture that part, and if R(t) were to go to, say, 1.1, then you would still see decreases in your measure. I think for such long term situations, you could modify your measure by purging historical cases by, for example, comparing the ratio of "new cases over the last week" and "new cases over the last month" instead of "new cases since I started counting". Then if R(t) would go beyond 1, you would see this in your model as its value would go up too.

That is a very appropriate critique and a great suggestion. I think I may restart my calculations as of May 20, which was the date of our first opening phase.

 

[gerntz]

Cases are not down everywhere. Here in North Carolina we had the largest daily increase ever over the weekend. Also our hospitalizations are up.

Nationwide to me means across the country, not that every measuring spot sees lower cases. If you disagree on that meaning, OK.

 

[gerntz]

I think we’re underdetecting cases by a factor of at least 10 in the US based on several situations where everybody in a large group was tested.

I think one report concluded that 1 in 11 cases was being detected, so multiply by 11.

So you're saying if a thousand folks are tested and 2 are found +, that there are really 20+ in that thousand that are+; i.e., the tests are so bad that they miss 9 of 10?

 

[Leo1277]

It's true that our testing is woefully lacking in Ohio.
...

The irony is that we were one of the earliest and most aggressive states in shutting things down, but we're also reopening more aggressively than DeWine indicated we would in April. Follow the money or, more correctly, the lack of it.

Thanks for an interesting and very detailed summary. Indeed, as you say, the money will ultimately influence decisions much more than we would like.

 

[Leo1277]

So you're saying if a thousand folks are tested and 2 are found +, that there are really 20+ in that thousand that are+; i.e., the tests are so bad that they miss 9 of 10?

I think what she is saying (and I sadly now tend to agree), we ought to be testing 1000 folks, but we only have test kits/time/personnel/interest for testing 100. In these 100 we find 2, while in reality there may be 20 in the original group.

 

[Leo1277]

A bit more on rt.live

Just to report a bit more information on this very interesting website rt.live:

1) They are trying to estimate time dependent reproduction rate (average number of infections that one infected person generates) R(t), instead of the R0(t) which I had been talking about earlier. R(t) takes into account that a fraction of the population already had the bug and is immune, and although that fraction is small, R(t) is the more precise value compared to R0(t).

2) Unfortunately one cannot look at the early stages of the disease, when R(t) was still much greater than 1. Digging around in some of the references to their methodology, however, I could find some earlier data (which have much bigger error bars) and am including an example from NY. We can see that the R(t) before shutdown was around 3. I sent them an email and asked whether they couldn't extend their data to the more distant past. (Only in Corona Times does 3 months qualify as "more distant")

3) The error bars for the values are unfortunately very large: for example, for Hawaii, the estimate is R(t)=0.56 (a very good value, no wonder since they don't let any tourists in), but the 50% confidence level (usually considered a pretty low bar for accuracy) reaches from 0.3 all the way to 1.0.

4) I am still sifting through the details of the methodology they are using, but the key aspect is that it's still based on test results, with all the limitations these have and which have been discussed here in detail. I really hope somebody picks up Midpack's idea of hospitalization rates, as these can measure R(t) rather precisely.

 

[SteveNU]

In Connecticut they have ramped up the testing capability and anyone can get tested. They are having a tough time finding people who want to be tested now.

https://trib.al/MLi1V4q

"As part of the plan to safely reopen Connecticut, Gov. Ned Lamont set a goal to more than double the number of weekly COVID-19 tests administered between May 20 and June 20.

But over the past week, the state hasn’t found enough patients to fill its current testing capacity, let alone double it, raising questions about the state’s plans to move forward with opening gyms, movie theaters and other businesses on June 20."

 

[audreyh1]

So you're saying if a thousand folks are tested and 2 are found +, that there are really 20+ in that thousand that are+; i.e., the tests are so bad that they miss 9 of 10?

No, we simply aren’t testing enough to detect the other 10 of 11.

False negatives are another matter.

 

[gwraigty]

Thanks for an interesting and very detailed summary. Indeed, as you say, the money will ultimately influence decisions much more than we would like.

You're welcome.

Also, the lawsuits may play a part. When some gyms sued to reopen earlier than DeWine allowed, a judge ruled in their favor. Now other businesses are getting in line.

Jukebox, pinball machine and arcade game companies sue Ohio over coronavirus restrictions

The lawsuit notes the state’s order is inconsistent, since it exempts the Ohio Lottery kiosks at bars and restaurants that help raise revenue for the state’s coffers. It says jukeboxes, pinball machines, etc. can be operated safely and the companies, if allowed, would offer hand sanitizer, wipes and other means for people to clean their hands and the machines before and after using them.

And, the lawsuit says the people considered most at-risk to COVID-19 — the elderly — are unlikely to use their products.

 

[harllee]

You're welcome.

Also, the lawsuits may play a part. When some gyms sued to reopen earlier than DeWine allowed, a judge ruled in their favor. Now other businesses are getting in line.

Jukebox, pinball machine and arcade game companies sue Ohio over coronavirus restrictions

There are at least 3 lawsuits here in North Carolina about reopening-

1. Churches won their lawsuit on the basis of religious freedom and are free to reopen but so far only a handfull have reopened.

2. Strip clubs!!! lost their lawsuit and cannot reopen. The Judge said that the strip clubs did not prove they could open safely.

3. Fitness Center lawsuit is pending

 

[Chuckanut]

There are at least 3 lawsuits here in North Carolina about reopening-

1. Churches won their lawsuit on the basis of religious freedom and are free to reopen but so far only a handfull have reopened.

2. Strip clubs!!! lost their lawsuit and cannot reopen. The Judge said that the strip clubs did not prove they could open safely.

3. Fitness Center lawsuit is pending

Ultimately, people will have to make their own decisions. From what I have heard, and it's just what people in other states have told me, opening a bar is a big deal for a day or two and then attendance falls off. Most people are still shy about going to any place that is crowded.

As far as the pool party crowd goes, I am wondering about a few questions. If an employer finds out that Calamity Jane went to one of those mass gatherings would he or she be within their rights to not allow Jane to come to work for two weeks? Could a store refuse to admit Jane to the store? As an employee, could I refuse to work anywhere near to where Jane is or has been?

I can see the pool party crowd getting a lot of push-back from other citizens, especially if the virus cases start to rise significantly.

 

[cathy63]

No, we simply aren’t testing enough to detect the other 10 of 11.

False negatives are another matter.

There's also the issue that some of those 10 other people are asymptomatic and we mostly don't encourage people who feel just fine get tested for an illness they don't think they have.

There are a few exceptions, such as negative TB tests being required for some jobs; and I think Covid-19 is going to be like that for a while, but the Covid test is a lot more unpleasant than the TB test.

 

[gwraigty]

Ohio's exit strategy continues to move right along. The biggest news is that starting June 8th, outdoor visitation will be allowed for assisted living facilities, and intermediate care facilities for the developmentally disabled. DeWine said we're going to see how that goes before allowing visitation for nursing homes.

Ohio to allow visitors at assisted living facilities starting June 8

BTW, a robocall came in this afternoon from my son's workplace. A resident in the memory care unit tested positive and has been moved to the COVID wing. Starting this week, all staff who work directly in that area will be paid $4 more per hour. All other staff in the facility will be paid an extra $1 per hour. What a wonderful time to be phasing in increased visitation. Not.

County fairs may resume with the decisions to be made at the local levels. Guidance on reopening amusement parks and zoos will be announced next week.

 

[Gumby]

That is very true, if you are in a pure exponential growth, then your number will stay constant. In the very beginning of an outbreak, this is a good measure to see whether you are having an impact with suppression methods and succeed in going sub-exponential.

But in the long run, for example where we are now, we are likely more interested in whether re-opening leads to new growth, and your curve will not show this. The R(t) curve does capture that part, and if R(t) were to go to, say, 1.1, then you would still see decreases in your measure. I think for such long term situations, you could modify your measure by purging historical cases by, for example, comparing the ratio of "new cases over the last week" and "new cases over the last month" instead of "new cases since I started counting". Then if R(t) would go beyond 1, you would see this in your model as its value would go up too.

That is a very appropriate critique and a great suggestion. I think I may restart my calculations as of May 20, which was the date of our first opening phase.

To follow up, I have modified my spreadsheet to compute and track the ratio of new cases today to new cases one week ago, both on a 7-day rolling average. If that number goes above 1.0, it means cases are growing. I went back through the data to track hospitalizations the same way. Since the first phase of reopening on May 20, we have continued to reduce our hospitalized population at the rate of approximately 25% per week.

 

[Midpack]

Ultimately, people will have to make their own decisions. From what I have heard, and it's just what people in other states have told me, opening a bar is a big deal for a day or two and then attendance falls off. Most people are still shy about going to any place that is crowded.

As far as the pool party crowd goes, I am wondering about a few questions. If an employer finds out that Calamity Jane went to one of those mass gatherings would he or she be within their rights to not allow Jane to come to work for two weeks? Could a store refuse to admit Jane to the store? As an employee, could I refuse to work anywhere near to where Jane is or has been?

I can see the pool party crowd getting a lot of push-back from other citizens, especially if the virus cases start to rise significantly.

Did you mean Typhoid Mary (more apt), or Calamity Jane?

 

[audreyh1]

There's also the issue that some of those 10 other people are asymptomatic and we mostly don't encourage people who feel just fine get tested for an illness they don't think they have.

There are a few exceptions, such as negative TB tests being required for some jobs; and I think Covid-19 is going to be like that for a while, but the Covid test is a lot more unpleasant than the TB test.

That’s what the group testing brings out. When they test all the folks at a business, or prison, or shelter, or ship, etc. they find a bunch of asymptomatic folks that wouldn’t otherwise be tested except as part of contact tracing.

 

[Midpack]

I give up. "Deaths" and "hospitalizations" seem to be more reliable than "cases" or "testing." but what could anyone conclude from this? Deaths undoubtedly lag hospitalizations, but how they could run counter to one another for weeks is hard to explain. Effective treatments haven't been found that I heard about. Guess I need to wait and watch.

Not that anyone here is trying to, but it would be easy to make a convincing case for improving or deteriorating progress for the state simply by choosing charts that agree with your established views. Some/many (news) sources are actively doing so...

 

[donheff]

Midpack, it could be that hospitals are getting better at treating people so fewer are dying. I remember when NYC Docs discovered that it is important to keep people from lying on their backs.

 

[Leo1277]

I give up. "Deaths" and "hospitalizations" seem to be more reliable than "cases" or "testing." but what could anyone conclude from this? Deaths undoubtedly lag hospitalizations, but how they could run counter to one another for weeks is hard to explain. Effective treatments haven't been found that I heard about. Guess I need to wait and watch.

Not that anyone here is trying to, but it would be easy to make a convincing case for improving or deteriorating progress for the state simply by choosing charts that agree with your established views. Some/many (news) sources are actively doing so...

Thanks for sharing these plots, indeed this is interesting. But I think, if you look at these plots in the right way, it is not quite incompatible, and one can actually draw quite firm conclusions. First, with "right" I of course mean, right in the statistician's eye, and not in the eye of any person who wants to make a point one way or another. There are several things going on:

1) There are enormous day-to-day fluctuations in three of the four curves. Besides noisy data, which are most pronounced in the death rates because there are low count rates and hence very large statistical errors, I think there are significant "weekend effects" showing up: the three fluctuating curves have pretty pronounced dips with seven day periods. These seem likely to be due to reporting habits - testers are less active on the weekends, hence confirmed cases are going down with the same rhythm. Note that the whole testing machinery has been put together from scratch very recently, and it is not a well oiled machine yet. Also, reporting of weekend deaths may happen only on Monday when the coroner's office is open again; death reporting has always happened historically of course, but there is no urgency of reporting, and whether the weekend is bundled into Monday doesn't matter much in normal times.

2) The two top curves about confirmed cases and completed tests correlate quite strongly, both in their seven day average, and even in their individual single day spike pattern. The seven day pattern (which can average out weekend effects nicely, different from say a ten day pattern) just means testing has fallen off a bit since May 11. But if you compute the ratio of positive tests to performed tests, this number stays rather steady. So overall, the top two curves are a bit confusing, but nothing to be alarmed about, and just say that the percentage of positive tests is fairly constant and testing is decreasing a bit.

3) The hospitalization curve is much smoother than the others, and shows a rather clear upward trend. Well, different from the testers and the coroner, the virus knows no weekend, and ERs work 7 days a week and have done so for decades. So recorded influx of cases is much more steady than any testing. Also note that the hospitalization curve is smoothing itself, since what you are showing is the total hospitalized patients, not the new admissions, and if a person spends say a week in hospital on average, then this curve represents a seven day averaging of new cases and may not even need further smoothing.

4) Note that there is a 2-3 week lag between hospitalization and death. Your hospitalization curve seems to fall until around 5/14; and that's all you see in the pattern of deaths. The rise in the hospitalization curve starting around 5/14 - about a week after Phase 1 Reopen - simply isn't reflected yet in the death curve, since only two weeks have passed since that time, and the larger number of newly admitted folks haven't had a chance to die yet (excuse the morbid terminology)

5) The main conclusion that I would draw from this is that looking at hospitalizations as you proposed several times may indeed be the most meaningful way of understanding the trends, because of the proportionality of infected vs subsequent hospitalization, which depends only on the properties of this particular virus but not on human actions, and because hospitals have a machinery that has been in place for decades and that doesn't introduce a weekend effect.

Finally, let me try to go out on a limb, to be compared to other data over the next weeks, and venture an estimate of the current R(t) from this: in your hospitalization curve, we go from roughly 460 on 5/14 to 580 on 5/29; if we assume average time from a person's infection to the infection of his infectees of 1 week, this increase of roughly 20 to 25% then yields an R(t)=1.1, give or take. And the jump from perhaps R(t)=0.95 happens right around the time when Round 1 Reopen leads to new hospitalizations.