COVID deaths & hospitalizations always lag cases. The lag has been demonstrated, is often 𝗺𝗼𝗿𝗲 than a month, and its timing can be predicted accurately (I have done it.) The #casedemic folks are just, well, wrong.

A thread explaining the lag with real-world examples.

1/n Image
I will show what causes the lag, and how I can predict it accurately. First, there are multiple causes behind it:

#1 clinical
#2 reporting
#3 age prevalence

I will explain these causes one by one

Lag #1 is the most obvious: clinically the mean infection-to-death time is 22.9 days (see pg 4:…)

So at minimum deaths will lag cases by a little over 3 weeks

Similarly, infection-to-hospitalization is 1-2 weeks

3/n Image
Lag #2 is reporting delays. Deaths may take 4 weeks or more to be reported.

For example in Florida the average lag from the date of death to the date the death is reported on the state's covid dashboard is currently 28.4 days:

Reporting delays affect hospital admissions as well. My locality—San Diego—warns hospital statistics are incomplete in the last 2 weeks. Indeed, the May 5th hospitalization peak took 15 days to be reported as the highest peak on the chart:

Lag #3 is caused by age prevalence

We know a 70-year-old is 100× more likely to die or be hospitalized due to COVID, compared to a 20-year-old: Keep this in mind. I will come back to it.

And we observe a surprisingly universal trend around the world: COVID outbreaks often start among the young, before propagating to older age groups

7/n Image
More examples of the propagation of COVID from the young to the elderly can be seen in my heatmaps thread:

When an outbreak starts among the young it will initially have a near-zero impact on hospitalizations & deaths (remember: 20 y.o. are 100× less likely to die/be hospitalized) until it propagates to older age groups

The propagation delay can be 4 weeks or more eg. see Tokyo:

9/n Image
The sum of all these lags is:

#1) 3 weeks infection-to-death
#2) 4 weeks deaths reporting
#3) 4 weeks propagation to older age groups

Total = it can take 11 weeks from increasing infections to increasing deaths.

However rising *infections* to rising deaths isn't exactly what we want to estimate. We want to estimate rising *cases* to rising deaths. So how long does it take for an infection to be reported as a case?

An infection occurs, symptoms appear days later, the patient is tested, and the lab or hospital reports the test result to public health authorities. Conservatively we could (over)estimate this delay to about 2 weeks.

So the lag from rising *cases* to risings deaths would be 9 weeks in this example (11 minus 2.)

And of course, to confirm retrospectively that deaths are increasing, one would need to observe deaths for a few weeks beyond the 9-week point.

See the chart below: the top panel shows Iran deaths up to 04 Jun.

Some people would say the declining trend of Iran deaths had not reversed by that point. It's only with more data (bottom panel) that it becomes clear the trend reversal started on 14 May

14/n Image
In my experience the #casedemic folks who are so focused on visual charts have—ironically— the hardest time reading charts.

Specifically they lack an intuitive sense about what constitutes a statistically significant trend reversal.

For example in June I argued with someone about this exact Iran chart but he could not be convinced the trend of deaths had reversed on 14 May until he had 5 weeks of data past that point:

Only with an extra 5 weeks he acknowledged it.

In this extreme scenario a person may need a total of 14 weeks (9-week lag + 5-week retrospective observation) to get a visually convincing chart that deaths were really rising.

14 weeks! That's why "wait 2 to 3 weeks" has become a meme!

In practice the lag is shorter than 9 weeks.

For example the summer COVID spike in Florida had a case-to-death lag of only 5 weeks. But with a retrospective observation of 1 week, it means 6 weeks were needed to show the #casedemic folks a chart of rising deaths.

I actually did build a model to predict this lag.

It forecast the timing of deaths in Florida by taking into account the 3 causes of lag: And it has been very successful. It predicted the 5-week lag with great accuracy:

This model was super-accurate because Florida is one of the few states that publish full line list information, including the age of every COVID case, which is required to account for lag #3 (age prevalence.)

You can read more about my model here:…

Without age information, we can still build decent guesstimates.

For example I also accurately predicted by how much deaths would rise in Spain, when many (such as @JamesTodaroMD) were wrongly claiming deaths wouldn't rise:

Having said all that, there are (only) two scenarios where increasing cases will NOT lead to increasing deaths:

Scenario #1: increasing case ascertainment rate.

Sometimes more tests are performed and catch more milder cases. But the epidemic is otherwise not growing. This occurred in Spring in Sweden. A reliable indicator of this is when the share of positive tests decreases

23/n Image
Scenario #2: when an outbreak grows among young age groups, but never propagates to the elderly.

This is uncommon. But it happened in September in Florida:

Bottom line, the case-to-death lag is real. It can be accurately predicted with the right data (age information)

It is certainly not on the order of 2-3 weeks, but can be up to 9 weeks or more, as I demonstrated.

Another very important point: the case ascertainment rate has changed dramatically between March-April and October.

In March-April many countries detected around 1 in 10 cases.

In October they detect 1 in 2, or 1 in 3.

What does this mean for cases and deaths?

This means a naive comparison of deaths per case, between March-April and today is going to be widely misleading.

Spain is a good example.

In October cases are spiking higher than March-April, but deaths are lower, so COVID has become less severe? Wrong.

27/n Image
The reality is that because the case ascertainment rate was terrible (1 in 10) in March-April, the peak of infections was truly much worse then.

That's why deaths were higher in March-April.

28/n Image
COVID has not become "less severe" over time.

Yes, treatments have gotten somewhat better at keeping people alive (), but the improving case ascertainment rate is, by far, the biggest factor.

Back to the topic of lag between cases and hospitalizations, or between cases and deaths.

In Sweden the lag also appears to be 5 weeks, as ICU admissions and deaths have just now started increasing:

Small correction: in Sweden, the case-to-death lag for the current Sep-Oct 2020 wave is about 4 weeks (not 5)

And the case-to-ICU admissions lag is about 3 weeks

See more details at


• • •

Missing some Tweet in this thread? You can try to force a refresh

Keep Current with Beͫvͣaͬnͨd

Beͫvͣaͬnͨd Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!


Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @zorinaq

29 Nov
A thread to refute the nonsense spewed by #covid deniers/minimizers...
#1 "It's just the flu"

No, vast majority of studies disagree:
#2 "Really, it's just the flu"

Really, no:
Read 14 tweets
27 Nov
This chart shows Sweden's mortality rate & excess deaths since 1900

While COVID-19 *seems* to be a small bump in mortality, 2020 has the most excess deaths of any year since the 1918 influenza pandemic

These ~4700 excess deaths are supported by SCB:

The line representing average expected mortality on the chart is a LOWESS regression

Normally demographers use more sophisticated statistical algorithm (eg. Farrington) to do so. LOWESS is kind of a sloppy technique, but it works well enough

For more accurate results, I didn't include 2020 data in the LOWESS regression. Instead I cut off the smoothing at 2019, and assume that without COVID-19 the expected mortality would have continued its generally improving trend of the last decades through 2020

Read 12 tweets
25 Nov
Things that happened in Sweden in last 2 months:

• family isolation
• ban public events of more than 8
• cinemas/museums/gyms closed
• nightlife curbed (alcohol ban)
• Tegnell: yes to face masks
• nursing home visit ban
• + many restrictions

Sources: see links below

Family isolation: household contacts of infected persons not allowed to go to work (since 01 Oct)…

Ban on public events lowered to 8 people since Nov 24…

Read 9 tweets
18 Nov
I dusted off my COVID-19 model (that predicted the Florida July wave) & applied it to Sweden

After today's data update from the Swedish Public Health Agency (FHM) I confidently forecast Sweden will surpass the peak of 100 COVID deaths/day they had in April

Hard to believe?

1/n Image
Specifically: by 25 December we will see Sweden has recorded 100 deaths/day around 11 December

(due to reporting delays, it takes up to 2 weeks past a given date to have a complete count of deaths on this date: )

My model is formally described in…

It predicts deaths from cases alone, but let me explain in layman's terms how it works...

Read 17 tweets
16 Nov
Covid lockdowns appear to reduce suicide rates

Suicide rates decreased during the lockdown in Victoria, Australia (paradoxically despite self-reported levels of depression increasing??)

Suicides (and other deaths) have also decreased in Peru:
Victoria chart is from @sometimes_data — thanks!

Who else has data for other countries?
Suicide rates have also decreased in Massachusetts during lockdowns:…

h/t @binaryanalogue
Read 5 tweets
25 Oct
͏@VoidSurf1 wrote a cool thread on Sweden excess deaths over the last few centuries. At first sight, his analysis seems correct... But there is a fatal flaw.

Despite the mortality data for 2020 being preliminary, he took great precautions to make it as accurate as possible. Good👍

Note how it is apparent that the month of April alone had 2000 excess deaths.

Read 8 tweets

Did Thread Reader help you today?

Support us! We are indie developers!

This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Too expensive? Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal Become our Patreon

Thank you for your support!

Follow Us on Twitter!