No, 1 in 6 People Do Not Have Long COVID
The ongoing story of bizarre overestimates of Long COVID rates
The question of how many people get Long COVID has always been contentious. The reason for that is pretty obvious—as COVID-19 has decreased in acute severity, causing far fewer hospitalizations and deaths, the argument has become almost entirely about Long COVID and the possible impacts it may be having on our health. If COVID-19 causes few deaths, but leaves lots of people sick with a new chronic disease, then it may still be something that we need to spend time on even now in 2026.
The newest estimate of Long COVID rates has just come out, and it’s remarkably high. A new paper has estimated that 1 in 6 people who catch COVID-19 end up with Long COVID, and that the number is actually increasing over time. If true, this would mean at least a billion people with long COVID worldwide, likely more.
The number is, however, absolutely ludicrous. There is absolutely no plausible way that this many people get Long COVID after an infection—even in 2020 the high-quality data showed much lower percentages.
The Study
The new paper that has caused a furore on Bluesky is called “Long COVID Persistence and Surveillance Gaps Across 58 US Hospitals”. It’s a fairly complicated calculation of Long COVID rates based on a dataset of nearly 500,000 people across 58 hospitals in four US states.
The authors developed what they describe as a “custom artificial intelligence algorithm” which identifies whether people had Long COVID or not. Based on this algorithm, they classified 16.28% of the patients in their cohort as having had Long COVID after their infection.
This estimate is one of the worst that I’ve seen in years.
The first issue is the algorithm. While the authors describe it as artificial intelligence, in practice it’s quite a simple mathematical calculation. The approach initially measures temporal correlations between various conditions. Basically, it identifies whether two things are likely related in the cohort in question, so for example whether chronic heart failure is related to chest pain.
The algorithm itself then uses these correlations to determine whether disease codes—the way that hospitals record the health problems that you have—are related to something an individual had before their COVID-19 infection. For conditions on the list of things that the authors have decided constitute Long COVID, if there is no other defined cause—so for example if someone has chest pain but no chronic heart failure beforehand—and they last for 2 months or more, they are considered to be Long COVID.
This sounds complicated, but in practice it’s really quite simple. There’s a list of conditions that the authors think are all part of Long COVID. If someone experiences one of these conditions for 2+ months after their COVID-19 infection, and it can’t be explained by another health condition that they had before their infection, it is considered Long COVID.
This means there is absolutely no assessment of causality in the study. This is just an estimate of the proportion of people who experience one of the conditions that the authors put on their list after having a COVID-19 infection. There’s a little bit more to it than that, but in essence that’s what the study shows.
In addition, there’s the list itself. While some of the things on the list make sense—for example, chronic fatigue—some of the conditions are truly bizarre. Here are some examples of International Classification of Diseases (ICD) codes that are included in the authors definition of “Long COVID”:
F14.180 Cocaine abuse with cocaine-induced anxiety disorder
O90.4 Postpartum acute kidney failure
D84.821 Immunodeficiency due to drugs
H53.50 Unspecified color vision deficiencies
E86.0 Dehydration
R19.6 Halitosis
E55.0 Rickets, active
J15.5 Pneumonia due to Escherichia coli
The list goes on. While you could argue in some cases that COVID-19 could be a contributor to some of these things, most of them are very straightforwardly not related to a coronavirus infection in any plausible way. Dehydration has a vast array of potential causes. Pneumonia due to E. Coli is, well, caused by E. Coli. Postpartum disease is by definition related to the pregnancy.
This is one of the inherent issues with Long COVID that I’ve been writing about for years. If you define a disorder as pretty much any symptom that could occur, you’re going to find a lot of that disorder in any population. But most of these things are almost certainly not related to COVID-19, and it makes no sense to include them in a disease definition for Long COVID.
On top of this, the authors have proven in their paper that the algorithm they used was terrible at estimating prevalence. The methods includes this paragraph:
Algorithm Validation Through Distributional Robustness
In the absence of site-specific EHR review, we assessed algorithm validity through distributional robustness testing.28,29 Consistent performance across demographically divergent populations provides evidence of robustness,30,31 whereas systematic miscalibration would be expected to manifest as divergent prevalence estimates across sites with different demographic compositions.32 The 58 hospitals across 4 regions represent substantially different populations in terms of racial and ethnic composition, comorbidity burden, and health care system scale, providing a rigorous test of algorithm generalizability beyond the development site.
In essence, they are validating their “artificial intelligence” algorithm by checking that different populations end up with similar numbers of Long COVID patients. If you have one group of people with COVID-19 where the algorithm finds a rate of 50%, and another where the rate is 5% then there are probably significant issues in how the algorithm is working.
The authors never again refer to this paragraph in the paper. Which is shocking, because the algorithm clearly fails the test of distributional robustness. Despite similarities in age and disease characteristics, nearly double as many people were classified as having Long COVID in California than Pennsylvania using the algorithm. In New England, only 5% of Long COVID ‘cases’ had endocrine issues, while in Texas 15% of ‘cases’ had these problems. The p-values for these differences—the likelihood that they would happen if the regions actually had the same distributions—are essentially 0.
In other words, the data fails the authors own stated test for robustness, which they never bothered to check. This makes the estimates even more dubious even as an interesting description of the diagnostic codes that people who once had COVID-19 eventually received.
COVID-19 Is On The Decline
I wrote about Long COVID rates using the existing best evidence way back in 2023. At that time, the data indicated that around 1% of people who were being infected with COVID-19 were going on to have long-term symptoms of the disease. The number appeared to be declining, but it was hard to say where exactly it would end up.
That analysis hasn’t really changed. The best evidence on Long COVID rates remains the high-quality studies conducted in the UK in 2021, 22, and 23, which suggested that around 10% of people in the initial 2020 outbreaks had long-term symptoms. This declined to around 1% of people by the Omicron waves, likely due to a combination of infection-derived immunity and vaccination.
I could try and estimate the current incidence of Long COVID—the percentage of infections that result in long-term symptoms. I would be surprised if it was higher than it was in 2023. But there’s really not that much point, because COVID-19 is on the decline anyway.
COVID-19 infections have plummeted in recent years. In 2022, there were around 1.5 infections per person, meaning almost everyone was infected once and many people had more than one infection. In 2023, the number was around 1. In 2024, we stopped recording really good statistics on the number of infections, but every indicator we had showed another huge decline. By 2026, we are looking at record low rates of COVID-19 across hospitalizations, deaths, wastewater data, and cases.
It’s hard to know precisely how many people have had a COVID-19 infection in the last 12 months, but we can look at the data to get a clue. The number of COVID-19 deaths has been halving every year since 2022, and is still falling. It’s likely that only 10,000 people will die of COVID-19 in the US in 2026, compared to half a million in 2021. There are currently no people hospitalized in Germany due to COVID-19, something that has not been true since before the virus initially broke out. Wastewater detections of COVID-19—the number of viral particles showing up in sewerage—show rates that are less than 0.1% of the 2022 waves in Denmark, the US, and elsewhere.
While we can’t put exact figures on the number of COVID-19 cases over the last year, we can say with a great deal of certainty that it is very low. And falling.
That makes all of this discussion about Long COVID incidence a bit pointless. It’s highly unlikely that even as many as 10% of people who catch COVID-19 will experience long-term symptoms, but even if the rate was that high it would still not be especially concerning with COVID-19 rates this low. More realistically, your risk of getting Long COVID today is quite close to 0%.
None of this is reassuring for people who have Long COVID from previous waves. I really feel for the community of people who have been permanently disabled by a disease that is no longer of great interest to most of the world. It sucks to have a condition that very few people want to spend money researching.
That being said, the ludicrously high estimates of Long COVID help no one. The 16.28% figure doesn’t pass the basic sniff test, and it’s clearly wrong. The true rate of Long COVID in 2026 is likely to be at least an order of magnitude lower, if not more.
Putting an accurate estimate of PASC cases together seems like an impossible task at this point. The overlap in symptomatology between PASC and ME/CFS is significant and often indistinguishable. If you consider that over a million people in the US were developing ME/CFS even before Covid, and the fact that virtually everyone has now had Covid, I’d venture to guess that many PASC diagnoses may simply represent patients who would have/are developing ME/CFS for all the reasons that existed before Covid. Many ME/CFS cases are not even tied to post-infection (such as my own). Good luck to anyone who wants to tackle this conundrum!
Niles Fox, Doctorate of molecular pathology, recovered (largely) ME/CFS patient, ME/CFS/PASC patient advocate.