Poll results can be skewed or biased in a few ways. First, only people who are interested in answering for one reason or another are going to answer, which leads to participation bias or nonresponse bias, amongst other forms of bias. People with particular experiences are also the most likely to respond to certain surveys. In this case, it may be that people with strong feelings about COVID vaccines are the most likely to respond to the survey.

The survey question in this case doesn’t ask for verification on cause of death, so they aren’t asking if someone knows anyone whose official cause of death was the vaccine. The question asks if the respondent believes they know someone whose death was caused by a vaccine. As we know from looking at social media, many of those beliefs aren’t based on any medical fact, but rather on misinformation and misunderstanding.

The concern with this poll isn’t that vaccines are causing deaths, but that there is enough misinformation still perpetuating that those who are responding to these polls believe these vaccines are causing these deaths.

A computer engineer should not fall victim to a Table 2 Fallacy, but here we have such a situation.

Often, epidemiologists need to pool together data to come up with an estimate of risk–known as an adjusted effect estimate. A Table 2 Fallacy results from multiple adjusted effect estimates from a single model presented in a single table. Problems arise when different types of estimates are presented on the same table.

In this case a table presents the results of multiple doses of a vaccine. It’s important to adjust for prior vaccination doses, but we can’t then use that same adjustment to draw conclusions about the efficacy of those prior doses.

The actual results of the study conclude that “the bivalent COVID-19 vaccine booster was 30% effective in preventing infection, during the time when the virus strains dominant in the community were represented in the vaccine.” This conclusion is supported by another analysis showing that bivalent boosters had an effectiveness of 59% against hospitalization and 62% against hospitalization or death.

While it’s accurate to say that the spike protein of SARS-CoV-2 can damage blood vessels, causing a condition called vascular leak, it does so in conjunction with the virus in its entirety and is not correlated with vaccination.

The paper cited in the tweet used retrospective data and looked at the incidence of retinal vascular occlusion (RVO) for two years post-vaccination. Several limitations were noted, including the concern that underprivileged populations, who are also more likely to be unvaccinated, are also less likely to seek care. Additionally, rates of COVID infection weren’t accounted for, and RVO is a known complication of COVID infection. Another study, following over 3 million people who received an mRNA COVID vaccine, found an RVO incidence of 0.003%, a rate similar to other vaccinations. No evidence of an association between newly-diagnosed RVO and COVID vaccination was found.

The original study highlighted in the tweet ends its conclusions with: “The number of reported ophthalmic complications has remained low, and vaccine-related retinal vascular occlusion is very rare, although the number of COVID-19 vaccinations is enormous…We recommend that individuals without a history of severe allergic reaction to any component of the vaccine be vaccinated to protect against COVID-19.”