Please Follow us on Gab, Minds, Telegram, Rumble, GETTR, Truth Social, X, Youtube, Instagram
Epidemiologist Eyal Shahar tears apart the CDC's annual estimates of the flu vaccine's effectiveness in his Brownstone Institute post. Every year the CDC publishes these estimates for the previous flu season. The NIH director recently criticized the test-negative design used to arrive at these numbers, and he was right to do so. The test-negative design for measuring vaccine effectiveness has the potential to introduce a second, potentially large, bias in addition to the bias it avoids by restricting the sample to individuals who sought medical care for their illness: collider bias. However, the net effect of this design is unknown.
That is not the only problem. Consider this big 2022-2023 flu vaccine study, from a season when the vaccine was well matched to the dominant flu virus. The study used the CDC-linked VISION Vaccine Effectiveness Network.
Confounding by background risk of infection. During the period from October 2022 to December 2022 (when risk was high) to January to March 2023 (when risk was low) the risk of infection changed. But vaccination occurred on a time scale that by end of December most people had been vaccinated. As a result those who were vaccinated had more time to be exposed during low risk periods and those who were not vaccinated had more time to be exposed during high risk periods. This imbalance alone could make the shot look effective, and indeed in an extreme example an apparent effectiveness of vaccination with saline given at start of low risk periods would be found.
The same bias was documented for the Covid vaccines during the pandemic, including in a study from Ontario, Canada. However, this bias has not been identified in the flu vaccines which are distributed as the season starts to rise and are typically completed by the time the peak of the season occurs.
The immortal time bias problem: the authors chose to exclude events that occurred within 14 days of vaccination of the cases in the vaccinated group. This well-known bias causes effects to be overestimated and has in fact halved the estimated effect size in the many Covid vaccine studies. Its removal here would bring the estimated effect size from the 30-to-40 percent range to 20 percent or lower. How many early events were in fact excluded? The study doesn't tell us - nearly 2,000 outpatient encounters were in fact dropped with no breakdown given.
Other findings simply do not add up. The adjusted effectiveness against hospitalization was almost always lower than against outpatient visits, the opposite of what you would expect, since more severe outcomes should show similar or greater protection. The authors call this "similar." It is not.
Another major problem to note is the healthy vaccinee bias. After controlling for a variety of confounders, the vaccine effectiveness in nearly every analysis went down, because, on average, the vaccinated individuals were healthier than the unvaccinated ones to start with. No matter how many variables a regression equation includes, it is impossible to fully control for this bias. To try to get around this problem, the authors employed some of the most complex methods available, including cubic splines, inverse-probability weighting, and a boosted regression tree, but they were described in such detail that it would be impossible for another researcher to replicate the analyses even with access to the data.
For the various age groups the results were also very unexpected. For older people the immune system is known to deteriorate with age and thus one would expect only weak protection by vaccination. Yet the study reports a much higher protection against hospitalization in older people (41%) than in younger people (23%). The authors explain these very unexpected results for older people by stating that they had been vaccinated with so-called "enhanced" vaccine products. Yet this excess of benefit is restricted to hospitalization and not found for outpatient visits.
The worst news was the high case fatality rate among elderly flu patients who were hospitalized. It was 50% higher among the vaccinated (4.0% vs 2.7%). The paper did not control for differences in baseline characteristics and simply called the outcomes "similar." We could interpret these numbers as saying that vaccination increased the likelihood of dying among these patients.
Instead of attacking the methodology in the abstract, the author goes through this one study in detail, because it shows so many of the flaws in CDC-based flu-vaccine research. Although randomized controlled trials are unlikely to be conducted, there are better observational designs than randomization: for example, cohort studies of people with and without symptoms of infection as well as regression discontinuity designs. So far, none of these designs have found the annual flu shot to be as reliable as the CDC estimates.
Get the latest on Georgia delivered straight to your mailbox.
