1. They compare same-day PCR against self-administered/interpreted rapid tests.
For most ppl, that's not reality. It's RAT today or PCR result in 1-3 days (or do both).
A fairer comparison would be a RAT today versus a PCR yesterday (since you'd have to wait for results).
2. They excluded 13% of the sample who were "confirmatory testers" (weird term) -- basically, people who had just tested RAT+ and were coming in for a PCR.
Basically, they excluded the people for whom the test easily worked. Drug trialists pull this stuff all the time.
3. The study excluded ppl with symptoms. Some people use RATs before socializing to reduce risk (hopefully with other Swiss Cheese slices). BUT many use RATs when symptomatic. I don't think the study fairly represents how RATs are being used.
4. As RATs became more common, many novices started using them without following instructions well. It would have been helpful if the researchers also administered and interpreted a RAT. I suspect the diminished utility is more "user error" than a property of the test. Fixable.
5. RATs detect most closely the infectious window. A PCR will flip positive a litter earlier & stay positive longer. If a goal is to prevent forward transmission, rather than merely diagnosis, then studies like this are not testing hypotheses with real-world significance.
Mina:
Overall, I see RATs as an important piece of mitigation. Even in this anti-RAT study, positive & negative predictive values were 85%-98% across tests (Table S2).
That's about as efficacious an N95 among a novice wearer. All of our shields have holes, so we combine them.
RAT efficacy estimates are much higher than many "too good to be true" interventions you see pushed heavily: SSRIs, enovid, nasal sprays. 🤣🤣🤣
The evidence that those protect against COVID is so poor as to harm by distraction. Go buy some bittrex.
🚩🚩🚩
As a vigorous defender of #CDC data, their switch from using normalized to non-normalized COVlD wastewater surveillance data today harms data quality.
"Normalizing" means accounting for basic confounders like rain levels. It is a choice to use worse data.
1/5🧵
Historically, the CDC data have correlated near-perfectly with similar metrics, such as Biobot's wastewater estimates (still active) or the IHME true case estimates (through mid-2023).
The changes reduce those correlations. It's like going from an A+ to a B.
2/5🧵
You can readily see the loss of data quality in the PMC "whole pandemic" graph (preview shown, subject to change) with choppier waves, caused by the CDC adding extra noise to the data and applying retroactively from BA.1 Omicron to present.
U.S. CDC numbers just released. Good news (for those not in Louisiana). "Only" a 5% national increase.
2025 has closely tracked with summer 2023 transmission. A 12-13% increase would have been expected based on those numbers. That said...
real-time data have been prone to retroactive corrections. This is frustrating, of course, because it leaves people making decisions based on data that are only of good quality when 2 weeks old.
If we saw a 12% increase this week, I'd say look at 2023 for a glimpse...
at the future. Instead, I would consider these plausible scenarios:
🔹Wave still similar to 2023
🔹Later wave with schools more implicated
🔹Something temporarily much better
COVlD is surging in 7 states, according to the CDC.
🔹Hawai'i (Very High)
🔹California (High)
🔹Nevada (High)
🔹Texas (High)
🔹Louisiana (High)
🔹Florida (High)
🔹South Carolina (High)
2. PMC COVlD Dashboard, July 21, 2025 (U.S.)
Western surge:
🔹California: 1 in 63 actively infectious, much higher in LA & Bay areas
🔹Hawai'i: 1 in 35 actively infectious
🔹Nevada: 1 in 63 actively infectious
These are wastewater derived estimates, not from individual tests
3. PMC COVlD Dashboard, July 21, 2025 (U.S.)
Southern surge:
🔹Texas: 1 in 56
🔹Louisiana (New Orleans): 1 in 65
🔹Florida: 1 in 66
🔹South Carolina: 1 in 71
Again, wastewater estimates (wise indicator), not individual testing (low-quality data).
We estimate 1 in 148 Americans are actively infectious. This equates to 2.3 million infections/week, expected to result in >100,000 new #LongCOVID conditions & >800 deaths.
A room of 100 people is a coin toss of an exposure.
2) PMC COVlD Dashboard, July 14, 2025 (U.S.) 🧵
Transmission (red) is closely tracking the path of 2 years ago (yellow). However, the incoming data are spotty. >20% of CDC states have limited/no data, & Biobot hasn't reported in weeks.
Could be MUCH worse or slightly better.
3) PMC COVlD Dashboard, July 14, 2025 (U.S.) 🧵
Our model formalizes the mathematical assumptions in those predictions. If transmission follows what we know in terms of how waves grow or slow generally and historical patterns, this is what we'd expect.
The spottiness of the current real-time data reduce precision substantially. Retroactive corrections can make the forecast jump around from better to worse from one week to the next. Expect the worst. Hope for the best.