9/15 Weekly covid19-projections.com Update:

We forecast 24,000 (13-39k) additional reported deaths in the US by November 1 (~500/day).

Nov 1 Total Deaths Forecasts:

Today: 219k (208-233k)
Last week: 219k (207-236k)
2 weeks ago: 219k (205-240k)
4 weeks ago: 225k (203-255k)
The 7-day average for deaths increased yesterday and will spike up today due to the Labor Day effects.

But there is no cause for alarm.

The average should fall again starting on Wednesday and we'll likely reach an average of around 750 deaths per day by the weekend.
It's possible that cases may flatten over the next few weeks, while the test positivity and hospitalizations continue to decrease.

We expect the downward trend in deaths to continue.

Reporting delay for deaths may be peaking in states like Florida:

Here is how our US daily deaths forecasts from 3 weeks ago have done thus far, compared with the most-cited model in the media.

20 days in, the true results have already fallen out of the confidence interval of the IHME model. They are now projecting out until 2021 (108 days).
According to our most recent evaluation, covid19-projections.com continues to be a top-performing model.

You can find more evaluations, methodology and plots here: covid19-projections.com/about/#histori…
It appears that IHME now has COVID forecasts for 18 months in the future.

It’d makes more sense for them to focus on accurately forecasting 18 days in the future before focusing on 18 months.

• • •

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

Keep Current with Youyang Gu

Youyang Gu 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 @youyanggu

25 Aug
In our update this week, we've lowered our COVID-19 projections by ~10%. We're now forecasting 42,000 (25-68k) additional deaths & 220,000 total deaths (202-244k) by November 1.

See our latest projections at: covid19-projections.com

Below are some additional observations.

According to our open-source evaluations, covid19-projections.com continues to be a top-performing model for both US and state-by-state recent projections.


There are still many media reports saying that there might be 300,000 deaths by December.

Using our projections, the chances of that are very low.

I think it's important that the media provides a realistic view of the pandemic, rather than one shaped by fear and headlines.

Read 12 tweets
12 Aug
There has been a bit of skepticism for my model's estimate that 20% of Florida (& other states) have been infected.


While I'm not claiming that this is necessarily true, here are some common mistakes I've seen people make when estimating true infections.
1) If we use cases and infection-to-case ratio to estimate true infections, the result we get is an estimate of the true infections from *~2 week prior*, not *right now*. This is because cases lag infections.

This matters when 1-3% of the population are being infected weekly.
1a) Also note that the infection-to-case ratio was >10x in March/April due to lack of testing, skewing the overall ratio upwards.

Test positivity plays a role as well. At 10-20%, there is a high chance that we are missing more infections than usual.

Read 11 tweets
10 Aug
There's been increasing discussion about the role of immunity vs behavior vs interventions in the spread of COVID-19.

Louisiana is a good case study because it's the only state that had two significant waves affecting a large proportion of its population.

Here's a closer look.
On the surface, it seems like all of Louisiana was heavily impacted after reopening despite already experiencing a significant first wave.

But if you break it down by parishes (counties), the data tells a different story.

Plots source: @LADeptHealth ImageImage
The two parishes that were most significantly impacted in March/April, Orleans Parish (New Orleans) and Jefferson Parish, did not see a major second wave after reopening.

The majority of new cases & deaths in Louisiana come from parishes that were largely spared in March/April. ImageImage
Read 22 tweets
5 Aug
Here's a thread that contains my findings regarding the relationship between true infections, reported cases, test positivity rate, and infection fatality rate for COVID-19.

Full write-up: covid19-projections.com/estimating-tru…

I hope to get some review/feedback/discussion from #epitwitter.
These are the main conclusions I drew:

1) The virus is more prevalent now than in March/April

2) Current infection fatality rate is lower (~0.25%) mostly due to lower median age of infection

3) Herd immunity threshold is lower (~10-35%) due to lower rate of transmission (Rt)
1) Higher prevalence

I computed the true daily new infections using 3 separate methods, and they all point towards a higher prevalence in July.

The estimated peak is ~450k new infections per day in July compared to ~300k/day in March.
Read 22 tweets
31 Jul
Here's my estimate of the number of true new daily infections in some of the most affected states, using reported case counts and test positivity rates.
One can then combine these estimates with the daily reported deaths to compute what I call the "implied infection fatality rate" (IIFR). The IIFR for the US is currently at ~0.25%.

Note that there is a lag of about 4 weeks between infection and reported death.
Here is the current implied IFR for states with more than 1,000 reported cases per day.

Tennessee, Florida, Georgia, Nevada, and North Carolina all have an implied IFR of less than 0.2%.
Read 7 tweets
24 Jul
Florida either has an incredibly low fatality rate (IFR), or may be underreporting deaths.

I computed the implied IFR for every state based on my analysis of reported cases, test positivity rate, and reported deaths.

According to my calculations, Florida's implied IFR is 0.15%.
Here are the implied IFR for all states with more than 1,000 cases/day this past week (chart above). The implied IFR for US as a whole is currently 0.27%.

0.15%: FL
0.16-0.20%: GA, NV, TN, LA, AZ
0.21-0.25%: NC, TX, SC, CA
0.26%-0.30%: AL, MS
0.31-0.45%: OH, IL
For states with lower than expected implied IFR, I can come up with a few hypotheses:

- Better treatment
- Lower median infection age
- Lower test positivity than reported
- Significant delay in death reporting
- Underreporting of 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!