COVID Vaccines and Arrhythmia?

A recent preprint alleges a strong connection between COVID vaccination and cardiac arrhythmia.

Preprints come and go, but some have the opportunity to cause major damage to our public health. Shi et al. (2022) is a relatively new preprint that alleges COVID-19 vaccinations cause cardiac arrhythmia. If their numbers are to be believed, vaccination causes almost as many cardiac arrhythmias as getting COVID!

Patone et al. (2022)

This urges caution, because if true, it means COVID vaccination carries a substantial burden for arrhythmia, which may lead to an increased risk of coronary artery disease and, ultimately, death.

Luckily, the meta-analysis used to support Shi et al.’s conclusions overestimates.

---

First, here is their forest plot:

Some estimates quickly stand out; namely, Mansanguan et al. (2022), El-Shitany et al. (2022), Ortiz-Prado et al. (2022), Farhat et al. (2022), Alghamdi et al. (2021), and Alghamdi et al. (2022). Their estimates are extreme and their sample sizes are also comparatively small. The number of events next to the total also does not add up to a conclusion of a rate of more than 2,000 arrhythmia events per million people. In fact, just taking (199,645/830,585,553) * 1,000,000, we get 240.37 events per million people.

So how did they get nearly ten times as many? The answer has to do with how studies are weighted in meta-analyses. They’re weighted by the inverse of their variance, so after a point, it doesn’t really matter how much larger your study gets because the scaling is unfavorable. A sample of 3,000,000 sounds much more impressive than a sample of 300,000, but if they both have unit variance, the weight attached to a sample of 3,000,000 is only three times as great as the weight attached to the ten times smaller sample. An extreme estimate from a small sample size can easily drag a meta-analytic estimate into insane territory when only a large sample would do.

So why are these studies extreme? The answer seems to be that they shouldn’t have been included. Below I’ve put the studies in the same order and made a risk of bias traffic light plot for them. The domains here are, in order, whether the study tested for or asked about arrhythmia at all, whether there was baseline data, whether the sample was a general population one, whether there was self-reported measurement, and whether the ages represented in the sample were diverse. Additionally, it should be noted that Rahman et al., Almufty et al., Ortiz-Prado et al., Abdel-Qader et al., Basavaraja et al., Tavakoli et al., Wong et al., Chen et al., Wang et al., and Abu-Hammad et al. used non-Western vaccines like Sinovac and Sputnik-V.

Some of these studies (e.g., Mansanguan et al., 2022) were included in the meta-analysis with normal variation treated as if it indicated vaccine-induced arrhythmia. Others used VAERs, or treated things like self-reported palpitations (e.g., Alghamdi et al., 2021, 2022; Abu-Hammad et al., 2021) or self-reported fast or irregular heartbeat (e.g., Farhat et al., 2022) as if they were solid evidence of arrhythmia, and some of them had their numbers misreported by Shi et al (e.g., Kant et al., 2022; Tavakoli et al., 2022). Finally, some used the same data sources, so people were counted multiple times (e.g., Lai et al. vs Wong et al., and Cari et al. vs Montano et al.). When those came up, I just went with the newer, larger one.

The final dataset is going to be rather biased for several reasons. In most cases, there was no pre-measurement, but many datasets are also biased because resources like VAERs and EudraVigilance were used despite having become repositories for misreporting. Finally, some others were didn’t work because their samples were restricted to young ages (e.g., Mansanguan et al., 2022), and young people tend to have high rates of arrhythmia that disappear with age, and they are highly resilient to things like mild myocarditis, so the numbers could be misleading anyway.

The baseline meta-analytic results are a fixed effect estimate of 340 arrhythmia events per million and a random effect estimate of 1120 per million. Trim-and-fill suggested adding three studies, reducing the random effect estimate to 480 per million.

But meta-analyses have a garbage in-garbage out problem that meta-analytic corrections often can’t overpower, so for a more accurate picture, what is the rate in the combined sample? There were 190,679 recorded events and 563,249,955 cases, so the rate was actually pretty close, at 338.53 events per million administered shots.

This is probably still an overestimate; the studies simply didn’t tend to be very good. At least they don’t seem to indicate much evidence for a larger effect with second or third doses.

A final question to ask is how this estimate compares to catching COVID. As it turns out, it holds up pretty well! Instead of a 1:1 arrhythmia risk, it looks like COVID is actually around seven times more likely to cause arrhythmia. Given that many studies seem to support substantially reduced arrhythmia risk with COVID for those who do get infected if they’re vaccinated, the vaccine is likely aggregately protective.

---

The question of if COVID vaccines are likely to cause cardiac arrhythmia is answered with a swift “no”, but the question of how well they compare to other vaccines is mostly unanswered. The quality of the existing literature is low to modest, but the randomized, controlled trials are in agreement that the rate is at least very low, and the vaccines still certainly confer net benefits with COVID as prevalent as it is. For most people, there won’t even be more than typical vaccine symptoms like a mild fever, aches, and injection site numbness. But, for some number of people, there will be mostly mild cardiac symptoms. Whether vaccination is worth the risk is up to individuals, but for the old, infirm, and - indeed - population members in general, it is almost certainly safer to be vaccinated.

I might make another post on cardiovascular mortality at a later date.