One thing I don’t get: the study excludes the first 6 months after vaccination to avoid immortal-time bias. But if people died right away due to the vaccine (hypothetically), wouldn’t this design exclude those deaths?

>As unvaccinated individuals were alive between their random index date and November 1, 2021, corresponding to a maximum duration of 6 months, starting follow-up from the index date would have introduced immortal time bias. Therefore, we initiated the study of long-term mortality 6 months after the index date, while mortality within 6 months after vaccination was analyzed separately using SCCS models. While the SCCS models are well-suited for short-term vaccine safety studies, they are less appropriate for long-term mortality, particularly due to limited control periods among vaccinated individuals, and age differences within the same individual across risk or control periods, as age is by far a major risk factor for death. For both analyses, we introduced the calendar period as an adjusting factor to account for the infection dynamics, baseline mortality trends, and the varying propension to get vaccinated.

> But if people died right away due to the vaccine (hypothetically), wouldn’t this design exclude those deaths?

Yes. That's what we have plenty of other studies for, including the clinical trials that led to the vaccines being approved in the first place.