In insurance, there is a concept of Incurred claims But Not Reported (IBNR). These are claims for services that have been performed by a clinician but the bills have not hit the insurer’s claims system. IBNR is often predictable and can be used to inform reasonable guesses about what is still out there but not yet reported.
I think IBNR can be an analogue for COVID-19 as COVID-19 has predictable progressions. COVID-19 has Infected But Not Resolved (IBNR) cases.
A person infected today is highly unlikely to show up in today’s testing. Most likely they will show up in testing early next week.##
A person tested next week and found to be positive is unlikely to be admitted to a hospital on the day of the positive test. Instead, if they are ever in a hospital, they are likely to be in a hospital two weeks from today.
A person admitted to the hospital two weeks from today is likely to be discharged in late May or early June. However if they are to die, they are unlikely to die on the day of admission. Instead, they are likely to survive for several days before dying. **
A person infected today is unlikely to show up as a death until Memorial Day.
We looked at this basic math in March:
If we can safely add up median time from infection to symptoms and then symptoms to hospitalization, that sums to a back of the envelope span of 12 to 13 days.
We’ll have to add time from hospitalization to death. But this morbid math has a point. The next 7 to 10 days of deaths have almost entirely been baked into the cake as these are individuals who were infected before states started trying to open up again. We’re not going to get a reliable signal on mortality due to policy changes for at least another two weeks in the states that have been early and aggressive in re-opening. Right now, there are Infections incurred But Not Resolved that are working their way through the system and most of this IBNR is from the physical distancing policy and behavioral regime.
## DOI: 10.7326/M20-0504
** Lancet Infect Dis 2020 Published Online
March 12, 2020 https://doi.org/10.1016/S1473-3099(20)30195-X
PST
No one can hope to have the slightest understanding of this pandemic without taking these lags into account. It’s what makes infection rates and fatality rates so difficult to make sense of. For public policy purposes, it’s like steering an aircraft carrier. It takes a long time to turn that big ship, and people who rely on immediate feedback to conclude that masking up, for example, has done no good, or reopening restaurants and beaches has done no harm, shouldn’t be allowed at the tiller.
MattF
Another consideration is that infection is very uneven, with hot spots in time and space. This sort of bunching is a well-known property of Poisson statistical distributions, and is entirely random. And it complicates analysis— ‘Excess deaths’ appear to be a very useful metric, but what is the interval In space and time over which you take data? The data is likely to be in a hot spot, but if you don’t identify the hot spot accurately, you’ll get an underestimate.
davecb
This is very useful, as it is something we’ve dealt with before. For example, we plot the expected path of a hurricane like this https://blogs-images.forbes.com/marshallshepherd/files/2015/08/091256W5_NL_sm.gif?
I do similar things with week-over-week capacity planning plots, so readers can see what happened last week versus what is happening and predicted to happen this week
Amir Khalid
Malaysia’s Director-General of Health has been explaining at his media conferences that if you test positive for coronavirus here, you get an ambulance ride straight to hospital for isolation and treatment. I would have expected America to do at least that much.
dmsilev
So, “zero deaths by May 15th” is unlikely? The White House crack modeling team[*] will be very disappointed.
[*] “Models developed while high on crack”
MattF
@dmsilev: And negative deaths after May 15!! So don’t worry about death rates, we’ve got that problem solved!!
Amir Khalid
@dmsilev:
I hear the model they’re using is built on the President’s wishful thinking.
Ruckus
@Amir Khalid:
Rational people would come to that conclusion.
You can see how this differs greatly with a segment of the population in the US.
dmsilev
@MattF: Yes, but then zombies.
Of course, the White House will be safe; no brains.
The Thin Black Duke
I wonder how long it’ll take these geniuses to figure out that thousands of people dying every day is really really bad for the economy.
Ruckus
@MattF:
I have questions.
If we bring back people who died from COVID 19, will they die from it again? Or will they be cured by dying?
Can a person who is reanimated get insurance or will the government have to care for them?
This opens up an entire new concept of life doesn’t it?
Ruckus
@The Thin Black Duke:
Wouldn’t they have to understand how an economy actually works, besides the theft of capital?
The Thin Black Duke
@Ruckus: This is true, unfortunately. Vampirism is a lousy economic ideology, ain’t it?
Nora
@Ruckus: What kind of insurance are you talking about? Clearly life insurance wouldn’t work (did their heirs get a payout when the reanimated first died? Would they have to pay it back when the deceased are reanimated?). Would they need health insurance? Would they need health care? Would there be special reanimated hospitals designed for their particular problems and issues?
All these significant questions!
Gin & Tonic
I think I mentioned this once, but a long time ago I earned an income doing IBNR calculations (property-casualty, though, not life/health.) B-J is the only place I’ve been in the last 25+ years where I’ve heard the term. Weird.
scav
Amir is so cute with his pretense — polite, I would suppose — of continued belief in a rationally-tethered America. Either that or they’ve sutured his entire tongue to his interior cheek.
PST
@Gin & Tonic: Most of my career was spent representing medical malpractice carriers. Typically at the end of a year the IBNR is far more significant than the known claims — what they like to call a long tail.
Another Scott
@davecb: The IHME model plotted something like a “cone of uncertainty” early on, but the bounds were so large that it was hard to take seriously (or hard to know what to do based on the predictions).
It’s still a problem now, and not unexpectedly, given they based their model on Wuhan’s experience (cases falling to zero after an air-tight-weld-people-in-their-apartments lockdown) that no other jurisdiction has seriously duplicated after a major outbreak.
Yes, cones of uncertainty can be very valuable, but we still have far too little testing to have a handle on what the sensible ranges – and thus, potential outcomes – might be in the USA for COVID-19.
Cheers,
Scott.
Amir Khalid
@scav:
A bit of both, actually.
scav
@Amir Khalid: You speak remarkably clearly given your oral reorganization.
Fair Economist
It takes on average about 3 weeks from infection to death and generally a few days to report. The signal won’t even *start* until Memorial day, and it will take 2 more weeks for a change to become clear. It’s the inverse of lockdown effects, which generally took more than a month to be clear. It will be June before we know how much damage this has done.
Another Scott
@Fair Economist: I don’t know if it’s still the case, but around here in NoVA I heard a month or so ago about it taking 5 days to get a COVID-19 test result. I would hope that it’s shorter now… Sensible people who are able will isolate during that time, but many won’t be able to. That’s another variable time period that probably messes up all of these data sets.
Cheers,
Scott.
David Anderson
@Gin & Tonic: IBNR is a great concept that took me a while to get my head around when I was a young, bright eyed, bushy tailed bureaucrat.
Gin & Tonic
@David Anderson: It can also be used to play lots of accounting tricks, if one is so inclined.
David Anderson
@Gin & Tonic: OMG — if you need a good/bad year end close-out, minor changes to IBNR to move money from this reporting period to the next is a classic move.
Wapiti
Here in WA state, they’re definitely planning for the lag.
They have a four phase plan for opening back up. After one phase opens some stuff up, there will be at least 3 weeks before the next phase: 2 weeks for incubation time and 1 week for testing lag. If the cases rise, then we reverse course.
Rural counties have been thrown a bone: a county with less than 75k population and no new cases in the previous 3 weeks can apply to move forward into the next phase before the rest of the state. I assume that “can apply” means they need to explain their plan for testing and what triggers a reversal for them.
p.a.
@Ruckus: Haven’t you heard? It’s just a slow-rolled Rapture. Beta-Rapture as it were…
Dupe1970
@Amir Khalid: No. It’s built by the guy who predicted Dow 36,000. So it’s so much worse than we realize.
Brachiator
Nice bit of reasoning. This will be a good point to start looking at any increase in deaths. Assuming that states will report the information accurately. My inner cynic thinks that the Trump administration will try to suppress accurate reporting in his focus on opening things up.
Meanwhile, I see that a number of states are looking at May 15 as the date to review or end lock downs.
otmar
@Amir Khalid: we didn’t do that in Austria. A positive test means quarantine at home. Only go to the hospital if you have severe symptoms.
Shalimar
@Dupe1970: If only he had published a sequel in February when the Dow was over 29,000.
“Dow 36,000 2: I’m serious this time, guys”
Shalimar
@Amir Khalid: You know the US doesn’t have 960,000 available hospital beds. Quarantine in hotels for asymptomatic cases, sure. We should have been doing that.
otmar
@Wapiti: Austria is opening up in a similar way: one step, wait two weeks then take the next step.
All good so far. Under 1% of tests performed comes in positive and we’re at out 30 positive ones per day.
Brachiator
@David Anderson:
Reading the post, I immediately thought of accounts receivable and accounts payable. A timing thing.
Also, British scholar Kit Yates had touched on related issues on an episode of the very good BBC Radio show and podcast “More or Less.”
He also has an interesting post about modeling pandemics on The Conversation site.