Post

More from @michael_hoerger

Mike Hoerger, PhD MSCR MBA

@michael_hoerger

Sep 30

PMC COVID-19 Forecasting Model, Sep 30, 2024
🧵 1/5

COVID transmission remains extremely high, but we're entering a "lull" in the U.S. sooner than anticipated.

Among all summer/fall waves, the 22% 1-week drop in transmission is steepest all-time.

Details:

In Friday's data release, the CDC retroactively corrected the prior week's numbers downward 6% (for Sep 14). This is a big correction, bigger than average, but nothing nefarious. The most recent week's numbers show an additional 1-week decline in transmission of 22% (from Sep 14 to Sep 21) on top of that 6% correction (for Sep 14), so the transmission estimates have fallen quickly.

Looking back, we estimate that the 1-week drop of 22% is the largest decline on the back end of any summer/fall wave in the U.S.

*If* these numbers hold against future retroactive corrections, it means people have about 5 weeks of similar transmission from today through Nov 7.

Why might we have seen a record decline in post-peak transmission?

Several hypotheses:

1) Reporting Error: The 22% decline could be driven in part by errors in real-time reporting. These average is 5%, based on our analyses of Biobot wastewater data. We do not have long-term data on the accuracy of Verily/CDC real-time reports versus retrospectively corrected values. In the updated graphic, we have added 95% confidence intervals for the real-time values based on Biobot data, which show that 95% of real-time errors fall within 8.33%. Note that the dotted lines do not show 95% confidence intervals for the forecast, merely how the best estimates would change if a large error in the real-time reports of +/- 8.33%. It’s possible that next week the data will get corrected upward, and the forecast will more resemble the top dashed line. In the next Tweet, we show the forecast for our old (Biobot-based model); it’s still showing a slower decline, but they update their data about 5 days slower than the CDC, so it is unclear whether it’s a big real-time reporting error at the CDC or just that the CDC is ahead of the game.

2) Unprecedented School Transmission: This is the largest wave during the August back-to-school period. It’s possible transmission disproportionately affected school children and their families, and in being more targeted than typical transmission, the wave went down faster that what is normative thus far in the pandemic.

3) Laissez Faire Public Health: Public health guidance has weakened (e.g., 1-day isolation policy, not strongly pushing additional mitigation), which likely pushed the peak of the 9th wave higher, which could have led to a faster-than-usual resolution. The model accounts for these changing dynamics reasonably well, but with the school issue noted in hypothesis #2, it is possible the weakened public health guidance disproportionately hit a subset of the population, which altered the back side of the wave.

4) Missing Data: There are no widespread state-level instances of missing data, as is often the case. It is possible that specific areas did not report this week, and if there is a bias toward higher transmission in those places, the numbers will get retroactively corrected upward. This is one example of the issues that contribute to #1.

5) Politics: There is no evidence to suggest the CDC is modifying transmission data for political reasons. We put deep trust in the fundamental scientists doing the critical work translating wastewater into meaningful downloadable data. The inferences, agency graphs, and guidance can be influenced by politics, but the data are sound, given the limitations noted under #1.

Overall, if the data hold or anything reasonable within the ballpark, which is likely, this means a more prolonged and slightly lower lull than previously anticipated. Those putting off medical appointments and other risky activities may see this as a slightly broader window (today through Nov 7) to get things done. Note that even under the most optimistic forecasting scenarios, transmission remains very high in absolute terms, even if low in relative terms.

PMC COVID-19 Forecasting Model, Sep 30, 2024
🧵 2/5

Note that while the CDC data show an unprecedentedly steep decline, Biobot data suggest more gradual changes, along the lines we have been forecasting the past several weeks.

This graph shows a resurrected version of our discontinued model 1 (Biobot-only) forecast.

I would not be surprised if the CDC data get retroactively corrected upward a little bit (this is common, not nefarious), AND the Biobot data go down a bit faster (they lag the CDC data by about 5 extra days). There is regional variation between the data sets, which means that BOTH models could actually be accurate without corrections.

Nonetheless, anything in the ballpark of these two forecasts suggests a "lull" from about now through Nov 7. It's a very high lull, but people find these dates useful for risk-based planning.

PMC COVID-19 Forecasting Model, Sep 30, 2024
🧵 3/5

The United States is coming down from a 9th Covid wave and will soon transition into a 10th.

We are still #DuringCOVID.

The peak will likely be around New Year's Eve, though some models suggest slightly later.

A 3-month forecast is extremely volatile. I mention it now because people are already making travel plans.

Hospitals should formalize their plans for requiring universal masking if they have not already done so. Many have missed the mark during the ongoing summer/fall wave due to reliance an outdated zeitgeist of "respiratory virus season" that treats cold, flu, and Covid at interchangeably equal. Covid peaks twice annually and is more disabling long-term. Nonetheless, even hospitals relying on outdated models will be more likely to take the forthcoming wave more seriously.

As we have published, universal masking during Covid waves is a key indicator of healthcare quality. Yes, this is true based on 2024 data. Unfortunately, it's mostly only the very best-of-the-best health centers. Consider printing and mailing this article to places where you receive care.
jamanetwork.com/journals/jaman…

Read 5 tweets

Mike Hoerger, PhD MSCR MBA

@michael_hoerger

Sep 24

As an expert in psychological assessment who has testified in court on cognitive assessments I've conducted, people should interpret carefully the new eClinicalMedicine paper on Covid-related cognitive changes.

Quick 25-pt thread⚡️
🧵1/25

The field of clinical psychology has developed, implemented, and evaluated normed broadband tests of cognition for the past 119 years.

These are highly specialized instruments with carefully selected tests to cover the breadth of key areas of cognition.

2/25

The study did not use one of the well-established, normed, broadband tests of cognition. Instead, it used a novel app-based hodgepodge of tests with little empirical history.

People should be very cautious in interpreting results. There is no vast literature on the tests.
3/25

Read 25 tweets

Mike Hoerger, PhD MSCR MBA

@michael_hoerger

Sep 23

PMC COVID-19 Forecasting Model, Sep 23, 2024
🧵 1/8

The U.S. continues to see an estimated 1.1 million daily COVID infections with 2.2% of the population actively infectious as we descend from the peak of a 9th Covid wave.

Transmission will remain very high the rest of 2024.

PMC COVID-19 Forecasting Model, Sep 23, 2024
🧵 2/8

Looking at the year-over-year graph, note that we're in uncharted territory for this point in the fall. Expect a very high "lull" in early November before the winter surge sets in.

As we have noted previously, our current estimates are likely slight underestimates given elevated school-based transmission and the CDC data standardization process.

Kids: They are smaller, and make smaller "contributions" to wastewater. Basically, it takes more sick kids to produce the average amount of wastewater virus to equate to an average infection (mostly in adults). This issue evens out over time, but it means the model may underestimate during back-to-school periods.

CDC: They describe the details of their process for standardizing data over time. It's very strong, much better than what most localized wastewater orgs or WWS do. It's a bit Dunning-Kruger to question basic wastewater scientists in doing the most fundamental components of their jobs, and the data continue to correlate >.90 with other metrics like Biobot. In comparing with other data, I take their estimates as spot on, but could also see arguments that they may underestimate transmission by 0-5%. It's something we always keep an eye on. Trust, but verify.

Transmission is magnitudes higher than much of the public realizes, so we focus on the big-picture view rather than quibbling over such issues that tend to balance out over time.

I hear @jlerollblues is considering weighting a model based on estimates of the proportion of infections in children. That's actually tougher to estimate than it sounds at first glance. You should key an eye on his models too and the work he and others are doing with the WHN.

PMC COVID-19 Forecasting Model, Sep 23, 2024
🧵 3/8

Zooming in from the big picture, here's the past year of the pandemic.

🔹1.1 million daily infections
🔹Bimodal peak (Aug 10 & 24) of 1.3 million daily infections
🔹50-60% of transmission happens on the back end of waves

Continue to educate family, friends, and co-workers. Look at how the wave descends much more gradually than is arose. Many infections to come, and to try to prevent! Also, most people are not monitoring wastewater. They keep track of the cumulative "anecdata" of people they know sick or diagnosed with Covid recently. In their mind, the peak will feel like late October, when the cumulative count has really built up. You may find people more open to listening during the next month.

Read 8 tweets

Mike Hoerger, PhD MSCR MBA

@michael_hoerger

Sep 10

PMC COVID-19 Forecasting Model, Sept 9, 2024
🧵1/7

Nationally, we appear to have passed the peak of our late-summer wave, and it's not pretty.

At peak:
🔹>1.3 million daily infections
🔹2.8% (1 in 36) actively infectious
🔹Transmission higher than 90.5% of the pandemic

We are showing a peak around Aug 10, but as you look closely at the graph and in later Tweets, you'll see it was bimodal, with near-identical transmission on Aug 10 and Aug 24.

The CDC consistently corrects historical data, so in hindsight, we might expect the official peak date to flip to the 24th, or for the stats on the 10th to jump higher.

We had expected that Friday's data release might show this was the largest summer peak all-time (by the slimmest of margins), but the prior week's data were retroactively corrected downward by about 5%.

This is a common occurrence, which is why it's important to focus on the big-picture forecast (very bad transmission the remainder of 2024) as opposed to minute details.

Let's walk through the details in this Thread....

PMC COVID-19 Forecasting Model, Sept 9, 2024
🧵2/7

The current year-over-year graph on Covid transmission is troubling. We just had the worst August of Covid transmission in the U.S.

We are likely to have our worst September, worst October, and potentially worst November of transmission.

We expect to bottom out around 850,000 daily infections in early November, before the winter surge picks up.

These new monthly records for Covid transmission are the consequence of #LaissezFairePublicHealth, especially the 1-day isolation policy, but more generally that public health officials are not describing transmission frankly and the need for multi-layered mitigation.

PMC COVID-19 Forecasting Model, Sept 9, 2024
🧵3/7

Zooming in, see the forest for the trees: About 74 days in a row with 1 million daily infections.

The peak date is somewhat arbitrary, and either Aug 10 or 24 may be estimated the peak in hindsight.

Read 7 tweets

Mike Hoerger, PhD MSCR MBA

@michael_hoerger

Sep 2

PMC COVID-19 Forecasting Model, Sept 2, 2024
🧵1/8

We've headed from a false summit toward a larger peak in back-to-school transmission. Next week we'll know whether this is the largest or 2nd largest summer wave all-time. Expect 1.4-1.5 million daily infections at the peak.

This secondary peak is larger than anticipated previously, even when accounting for patterns of back-to-school transmission in prior years.

One, the West and South were peaking earlier, and the Northeast and Midwest have transmission picking up much faster than its falling off elsewhere. These are atypical regional differences.

Two, the 1-day isolation policy and general decline in school-based mitigation have unleashed transmission at higher levels than would be anticipated in prior years, even when accounting for the already-high levels of transmission as children went back to school.

Three, our model uses a combination of CDC and Biobot (former CDC contractor) data, using a 60% versus 40% mix for current case estimation. Biobot has not updated their data the prior two weeks without explanation. They were running much cooler than the CDC data, whether reflective of reality, or merely reporting delays. We have downgraded them to 20% in the model, and they will be downgraded further to 0% if they do not update their data this week.

You can take points 1 and 2 to account for the secondary and higher peak (shape of transmission), when models accounting for prior years would have suggested transmission slowing already, and point 3 accounting for a shift toward a slightly higher overall estimate of transmission.

PMC COVID-19 Forecasting Model, Sept 2, 2024
🧵2/8

This graph shows year-over-year transmission. It clarifies the unusual shape of the transmission distribution. A "peak" or false summit, followed by a very slight decline in transmission (negligible), then a 2nd (true) peak.

PMC COVID-19 Forecasting Model, Sept 2, 2024
🧵3/8

Presently, we estimate a peak at 1.4 million daily infections. Next week, we will have a much better sense. It could be in the 1.3-1.5 million range. There's about a 50% chance it will be the largest summer wave.

Read 8 tweets

Mike Hoerger, PhD MSCR MBA

@michael_hoerger

Aug 27

PMC COVID-19 Forecasting Model, August 26, 2024
🧵1/8

We're in uncharted territory during the back-to-school period.

1.2 million people in the U.S. are getting infected per day.

1 in 41 people (2.5%) are actively infectious.

PMC COVID-19 Forecasting Model, August 26, 2024
🧵2/8
The year-over-year graphs shows that we are experiencing much high transmission than during previous back-to-school periods.

This is a much higher and wider wave than Y1, Y2, and Y4 of the pandemic. Year 3's summer wave was very bad, but much earlier.

The 1-day isolation period, decline in most other mitigation, and a culture of denialism ("post pandemic," "Covid is over,") will fuel many infections among young people, teachers, school staff, and parents.

Expect absences at schools. Expect substantial workforce problems related to sick parents.

PMC COVID-19 Forecasting Model, August 26, 2024
🧵3/8
Here's a zoomed in version of current transmission and the forecast.

Notice that transmission will likely remain >1 million infections per day for the next month and counting.

Note that the national peak appears just behind us. The CDC graphs may suggest that we're currently peaking, but note their data are >1 week old. The forecast provides an updated best estimate for today. The peak and shape of the observed wave could change as CDC and Biobot retroactively correct numbers (perhaps a bigger, earlier spike). Alternatively, Biobot has been slow to update their data, and their data were suggesting a later peak. However, I suspect the current picture won't change terrible much in hindsight.

We're seeking a less spiky, more plateau-like peak because there is substantial regional variation. The West and South regions have extremely high transmission and have likely peaked. The Midwest and Northeast are lower but rising.

Keep an eye on regional or local data, and overall remember that local peaks could occur weeks before or after the national peak. Also, 50-60% of transmission commonly occurs on the back end of waves.

Read 8 tweets

Share this page!

Enter URL or ID to Unroll

Mike Hoerger, PhD MSCR MBA

Try unrolling a thread yourself!

More from @michael_hoerger

Mike Hoerger, PhD MSCR MBA

Mike Hoerger, PhD MSCR MBA

Mike Hoerger, PhD MSCR MBA

Mike Hoerger, PhD MSCR MBA

Mike Hoerger, PhD MSCR MBA

Mike Hoerger, PhD MSCR MBA

Did Thread Reader help you today?

Don't want to be a Premium member but still want to support us?

Send Email!