My thoughts on long covid

Everyone in my social group is suddenly arguing about the risk associated with long term complications from covid. I don’t really get why it’s happening now. We knew omicron was coming and almost all of the evidence people are citing is at least a couple months old. But here we are.

Prevalence of symptoms

Al-Aly, et al: Nature article from April 2021 looks at health records from people enrolled in healthcare from US Veterans Affairs. They look at symptoms 6 months out from acute covid. The most obvious issue to me is that people who enroll in VA health stuff might be abnormal on various axes. For example, they might have worse mental health on average.
Caspersen, et al: a pre-print from October 2021 looks at a large Norwegian cohort, drawn exclusively from unvaccinated people who got covid during 2020. The numbers I use here are based on the 12 months reporting. I can’t find evidence that it was ever accepted to a journal, so I’m somewhat hesitant to take it too seriously. But, unlike most studies, it does have a control group.
Havervall, et al: JAMA paper from April 2021 looks at Swedish healthcare workers. Scott covers this one in his big post from September 2021. It still seems pretty informative to me. I used both the 4 months and 8 months figures.
Chowdhury, et al: A huge British pre-print, also from April is another study that Scott wrote about in September. They report on symptoms after 6 months. I find it bothersome that it doesn’t seem to have been accepted for publication anywhere.
UK ONS has estimates for how many people have ongoing covid symptoms, and this group at Cambridge have estimates for how many people in England have been infected. I have not looked in any detail at either source’s methods, but nothing stands out as particularly bad. Based on the Cambridge infection numbers, it looks like as of October 2021, about 40% of people in the UK had been infected.

I collected the statistics from these studies for memory problems, brain fog/concentration symptoms, and fatigue in this spreadsheet. Not all four studies had the same definitions for symptoms or severity. Here is a summary for my best guess about the risk ratio and excess risk there is for each of these, for any early covid case and for any “mild” early covid case (I imported this table from Google docs and it didn’t transfer well, sorry for the bad formatting):

Symptom	RR (any covid)	RR (mild covid)	Excess risk (any covid)	Excess risk (mild covid)	Percent of UK infections
Memory problems	4	2	5-10%	1-8%	0.8%
Brain fog	3.5	2.5	5-15%	2-10%	1%
Fatigue	4	2.5	3-15%	2-7%	1.4%
Activity limited a lot					1%

The first four sources only looked at people who were infected with covid in 2020, before vaccines were available and before we really knew what we were doing, in terms of treatment. It is also unclear what exactly people are experiencing when they show up as suffering from one of these symptoms on these studies. It seems likely that this includes a very wide range of badness. Anecdotally, I remember taking a survey for a covid study in 2020 and they asked me if I had “trouble concentrating” and I answered “yes” because it was maybe a 20th percentile week for me, which was 70% of my average. I was not experiencing anything that I would consider a major detriment to my life in general, though having that in perpetuity would be annoying.

The UK prevalence seems a lot less ambiguous, except there’s no control and it’s hard to say how long people have been having the symptoms. If we take the risk ratios from the other studies, this suggests that around 50% to 80% is due to covid, suggesting excess risk in the vicinity of 0.5% to 1% from Covid. The prevalence for any self-reported long covid drops by half when you look at “longer than 12 months” vs “longer than 12 weeks”, so maybe .25% to .5% is a reasonable guess for years-long symptoms.

Estimating risk for myself and people similar to me

It seems unlikely to me that my risk from getting covid this month is as bad as a naive reading of those studies might suggest.

Vaccination: Vaccines reduce severity, and severity is closely tied to long covid. This UK report from July 2021 shows about a 50% reduction in symptoms lasting more than four weeks. Those were mostly people with AstraZeneca, which was substantially worse at preventing death than the mRNA vaccines. I’ve seen people suggest that vaccines do something to help with long covid beyond its effect on symptoms, but I haven’t seen anything all that convincing. Still, I think they probably do more than just pushing risk down to the published values for mild acute cases. Anecdotally, “mild” cases in 2020 were still pretty bad, compared to breakthrough cases in the vaccine era. Everyone I know who had it in 2020, most of whom are young and healthy, described it as very unpleasant, fairly debilitating, and much more flu-like than cold-like. Most people I know who got a breakthrough case following vaccination had a pretty easy time, though I do know one 30-something who had a pretty rough time. My guess is that in practice, mRNA vaccination with a booster brings the “mild covid” numbers down by a factor of 2-5.

Age: I’m in my 30s, and there seems to be a pretty strong age-related effect for severity and prevalence for long covid. Still, the UK data suggests that being in your mid thirties puts you pretty close to the average. I’m not going to make any adjustment for myself, but if you’re in your 20s, you can probably estimate a 40% reduction in risk vs the general population.

Omicron: The current variant seems to be low severity, at least for people who have been vaccinated. I haven’t dug into this too deeply, but this pre-print from January 11, 2022 that Zvi shared seems to show a factor of 2-10 reduction in severity vs delta, depending on which kinds of outcomes you look at. On the other hand, I guess Delta was more severe than the strains that generated most of the long covid data? I’m not sure, so I’ll say that omicron is half as bad as Covid Classic® and 25% as bad as Delta. Systematically trying to adjust the UK long covid prevalence for Delta vs earlier strains seems pretty time-consuming, so I’ll just say omicron is a factor of 2 less dangerous on top of other considerations.

Overall estimate of risk: If I apply a 2x reduction each for vaccines and omicron to the excess risk numbers for mild cases, and assume that only half of those were something substantially bad that I should worry about (like “activity limited a lot” for more than a year), this gives a risk of about 0.2% to 1%. If I apply just the vaccination and omicron reductions to the 0.5% “activity limited a lot” prevalence I described in the last section, this gives a risk of 0.12%. But I have a huge amount of model uncertainty here, and I assign about 20-30% credence to this all being off by some large amount. I’ll go with an estimate of 1% chance of a 50% reduction in the overall value of my life. This works out to about 75 days of my life gone, in expectation. Note this does not yet account for the difference in expected outcome of getting it now vs maybe getting it later.

What does the world look like when a small but not vanishing fraction of people are suffering from something?

Presumably for any level of badness/visibility, there’s a level of prevalence for which a public health problem is pretty hard to miss. Diabetes is hard to miss as a problem, because it affects a lot of people in a big, somewhat visible way. On the other hand, Aarskog-Scott syndrome is something I would never have found out about if I hadn’t just grabbed it from the top of an alphabetically-sorted list of very rare diseases, with fewer than 100 cases reported in the literature. Being a victim of abuse is very common, but not all that visible, so it’s easy for people to underestimate the overall size of the problem.

For the most part, I share the intuition with others that if long covid really were life-ruinous for a single-digits percent of people who get covid, it would be pretty clear what was going on by now. But I’m not sure. The prevalence we’re worried about is just low enough and the badness of chronic fatigue or brain fog are invisible enough that it’s hard to say confidently that the overall public health burden is not getting lost in the noise of the pandemic. For example, it’s hard to estimate the effects on people working because the labor market is pretty wacky right now for other reasons.

To try to make a guess about what the prevalence-badness frontier of visibility looks like, I looked up the rates for various things that are easy or not-so-easy to notice about someone. Here is a list, with percentages for Americans in most cases:

Chronic diseases in similar reference class:

Chronic fatigue syndrome – ~.6% in men and 1.2% in women
Chronic Lyme disease – 0.02% to 0.6%

Very noticeable things:

Being taller than 6’2” for a man or 5’8” for a woman – 3%
Have lost a limb – .6%
Using a wheelchair – .85%

Maybe not noticeable unless you know the person:

Rely on insulin – 2.2%
Carry an epi-pen – 0.4%
Have a pacemaker – 0.5%
Are a cancer survivor – 5%
Have sleep apnea – 3-9%
Broke a bone in the last year – 2%
Have celiac disease – 0.7%
Have a twin – 0.7%
Identifies as LGBT – 5%
Born on Christmas – 0.15%

Among people who took the 2019 EA survey:

Attended Oxford or Cambridge – 8.7%
Work in government – 5.6%
Studied medicine – 3.7%
Identify as Black – 1.4%
From Canada – 4.4%

I put very little effort into these numbers so if they are somehow cruxy for you, I recommend verifying them. I tried to choose things that are relatively uniform across Americans, and not things that my social group is unusually likely to have, but I’m not sure if I succeeded. For example, I wouldn’t be surprised if people in my social circle are more likely to carry an epipen than typical US residents. Looking at this, it seems to me that a 5% signal is absolutely unmistakable, even for things you might not notice unless you know someone fairly well, but a 0.1% signal might be hard to notice, unless it is something that is extremely visible. I would say that a 1% prevalence for life ruining outcomes for every covid infection seems high enough that I’m pretty dubious it is happening, but it is hard to rule it out. I think 5% is pretty unlikely and 0.1% seems completely reasonable.

Am I just going to get it anyway?

I have no idea. I also think this is not quite the right question to ask. What I really want to know is if getting it now is substantially worse than whatever will happen if I don’t get it now. Maybe the risk conditional on getting covid will drop precipitously after this spike because of new treatments or even less dangerous strains or whatever. Or maybe omicron is an unusually safe way to get some additional immunity, and the next few strains will be more dangerous. Or maybe my risk from each wave will be mostly independent of what happens in previous waves, and there’s nothing to be gained long term from getting it now.

My wildly uninformed guess is that getting omicron now, while my December 2021 booster is still relatively fresh, is net-beneficial, if I do not have long term complications (and, of course, if I do not have a bad acute case, though that does seem quite unlikely). I don’t take this guess all that seriously, but I do think it nudges me slightly toward less caution.

I guess I’ll go with about 40% that if I don’t get omicron during this spike I’ll incur a similar cost in the next few years anyway. This drops the expected cost of getting omicron now down to 44 days.

Costs of being cautious

I think many, many people I know are underestimating the costs associated with avoiding covid. There is some typical-minding here, because I am a social creature who does poorly in isolation, but nonetheless some of the costs are external. Avoiding each other is costly for the same reason that building Lightcone and Constellation are beneficial. Spending time worrying about it is costly for similar reasons to why the base rates on long-covid-like symptoms are so high during the pandemic.

Sitting outside in the cold sucks. Worrying about whether you made someone feel unsafe sucks. Breathing stale air all day sucks. Being socially divided on how to avoid covid sucks. Not seeing friends and coworkers sucks. Missing out on all the valuable conversations you may have had at the office sucks. Going to a social event where everyone is trying to follow uncertain rules and spends half their mental energy on trying to solve the sociologistical puzzle of talking to each other without doing the wrong thing sucks. These costs are no joke, and they may be worth it, but I think too many people view them as trivial.

All in all, I feel like covid has reduced the value of the last two years of my life by around 20-30% and the overall value of my output, broadly construed, by at least 50%. I know that it was pretty good for some people, at least in terms of happiness and work output. I’m happy about this, but I would guess most of those people had substantially smaller positive externalities than they would have, absent all the things we did to avoid covid.

That said, I think I can, at least occasionally, isolate myself pretty well for a few days at a time without incurring too much of a cost. I get lonely, but my work output can be pretty high.

More speculatively

Hanging out in Constellation without a mask when everyone is testing and nobody has substantial symptoms may be a good way to get some exposure to the virus, with very low levels of viral load. This applies to covid caution in general, so that being too cautious might be pretty bad, but being too reckless (e.g. letting people with positive antigen tests yell at you in a small room) is also bad. I don’t know if this is true, but it seems plausible to me.

Galaxy brain speculation

Omicron is so epidemiologically convenient and the evidence that it was engineered is so compelling that we should assume it was created by a competent actor for the purpose of inoculating the public, and it is in our interest to get it. (I do not actually endorse this view)

What I will probably do in light of all of this

Absent model uncertainty, my estimated cost for me getting omicron now is something like 4 days, which I’m willing to tolerate and I’m not willing to seriously modify my life for, for more than a few days. Incorporating model uncertainty, it is about 40 days. Even given my views on the high cost of caution, I think it is probably worth some additional caution until the spike dies down.

Unfortunately, part of the cost of caution is deciding what to do all the time. For the next week or two I will probably be more judicious about wearing a well-fitted N95 on BART and I will probably be more careful around the office, in terms of masking, distancing, and limiting face-to-face conversations.