Great to see estimates of the prevalence of chronic SARS2 from scientists working with the remarkable UK @ONS Covid infection Survey. Chronic prevalence (>60 days) was at least 0.1-0.5% in people infected before July 2022 (thru BA.1/2).
Gonna use this moment to grind an axe...🧵
twitter.com/Mahan_Ghafari/status/1760338169864618124
The interesting and frustrating thing is it was possible to guess this 2 years ago. How?
1) take patient self-reported symptoms seriously @ahandvanish ncbi.nlm.nih.gov/pmc/articles/PMC8280690/
2) take the ONS long-covid symptom survey seriously (the best survey in the world for this).
Here's three slides I made in June 2022 stitching it together.
The first gives my biological interpretation of the symptom clusters.
1: looks like long tail of normal infection
2: looks like organ damage after acute infection
3: looks like new-onset autoimmunity or ME/CFS
the second slide (repeated here) maps that onto the ONS results for post-covid onset symptoms lasting 12 or more weeks.
The data on symptom correlations isn't available, but one can eyeball that cluster 1 ("looks like persistent infection") is about 0.1-0.5%.
We also get estimates of the long-covid prevalence in cluster 2 ("looks like organ damage") of 0.5-1.5% and cluster 3 ("looks like autoimmunity and ME/CFS") of 0.25-0.75%. To my knowledge, these numbers, derived with the back of envelope in 2022, haven't been reported elsewhere.
The similarity between long covid that looks like chronic infection and observed chronic infection is striking.
It's actually a bit too striking. As the new ONS data reports that only 5-10% of the confirmed chronic infected have LC. Why am I still excited?
The authors of the new work are clear it's a lower bound. It's based on needed CT<=30 for sequencing (which isn't common at the tail of infections) from a nasal swab (which ignores other reservoirs, especially including the gut).
And in the opposite direction, the ONS symptom survey doesn't have a control group, and so the percent estimates are likely upper bounds for self-reported long covid symptoms due to long covid. (I know I'm poking a bear here, but *taps sign*).
twitter.com/famulare_mike/status/1689804196214943745
Albeit self-reported LC likely under-estimates LC because of social desirability bias to not be sick and lack of knowledge that the condition could exist.
Taken together, the sequence-based and crude LC survey-based estimates of LC prevalence due to chronic infection are, as reported, within a factor of ten, and the biases in both point to the middle. So I think they are roughly consistent.
This is great! Here's a picture of three different long covid syndromes and their prevalence (at least among those infected through mid-2022).
LC that looks like
- chronic infection of 0.1-0.5% of all infected
- organ damage of 0.5-1.5%
- autoimmunity and ME/CFS of 0.25-0.75%.
So why did I say I had an axe to grind at the top? I sat on these estimates for nearly 2 years, only showing them to some folks internal to BMGF. Why? Because there is no good venue to publicly share reasoned back-of-the envelope estimates in biomedical science.
Twitter doesn't work for this anymore. If this thread gets more than 20 likes, I'll be surprised.
The effort to go from some basic reasoning to a journal article is enormous. And I don't have the primary data anyway, so there isn't any real pathway to doing it even if it was worthwhile.
Anything resembling the blogosphere of 2008 is completely atomized.
Can we please have a journal of reasonable speculation? A place for clear thinking that can help move mental models in a field forward?
This isn't the most careful analysis, and I never intended it to be. But two years later, I still see people arguing about the one cause of long covid when it's multiple syndromes! I still see trials designed to test one mechanism of intervention when its multiple syndromes!
I still see insane estimates of long covid prevalence drawn from carelessly mining EMRs--things that do not fit naked eye observations of the fraction of newly disabled people in the workplaces, IRL friends and family, and schools of the people publishing the papers.
We need to do better. And it's not just that we need "better quantitative science." We need better reasoning about the world in biomedical science. And we need a venue to share better reasoning.
End of my rant. Now your turn to pile on!