How much of the higher death risk could be selection bias?

The short version: selection bias can explain some of the observed higher death risk in vaccinated 80-year-olds, but not the full pattern — the evidence shows the trend is too consistent and too large for bias alone.

The 2025 UK study you asked about in the previous answer found that for people 80-89 and 90+, the vaccinated group started with a lower death risk but crossed over to a significantly higher risk during the study period [1]. The authors themselves note that healthy-user bias (healthier people getting vaccinated first) could explain the implausibly low initial risk estimates [1]. That is a real concern. Early in the rollout, the sickest and most frail were often last to be vaccinated, so the early "protection" numbers were inflated by comparing healthy vaccinated people to sicker unvaccinated people.

But the crossover is the key. Selection bias produces a stable or shrinking gap, not a reversal. If the early benefit were entirely due to healthy people getting vaccinated first, the vaccinated group's risk should have stayed lower or converged to the unvaccinated rate as the sicker people eventually got vaccinated. Instead, the risk crossed over and kept rising linearly [1]. That pattern is hard to explain as bias. The study also found the same linear increase for non-COVID deaths, which a vaccine not designed to prevent those deaths should not affect [1].

The Australian data tells a similar story. In 2020, before mass vaccination, no age group in Australia had statistically significant excess mortality [1]. By 2022, after widespread vaccination, people aged 75-84 had 24.4% excess deaths above baseline (p=0.006), and people 85+ had 20.8% excess (p=0.045) [1]. A separate analysis found that Liu et al.'s claim of 51.2% vaccine effectiveness against all-cause mortality in Australians 65+ is implausible because it would require a 121% excess death rate in the unvaccinated — a figure never seen in any OECD country, including Mexico at its worst (52.7%) [3]. Sensitivity analyses adjusting for healthy-user effects still gave implausible results (112-126% excess in the unvaccinated) [3].

The US nursing home study adds another layer. It found that elevated mortality after COVID-19 infection persisted longer in vaccinated residents (5 weeks) than in unvaccinated residents (same week only) [4]. The authors note that facilities prioritized vaccinating the most frail residents, creating a bias that would make vaccinated groups look sicker [4]. But the direction is the opposite of what a protective vaccine should show.

My call: selection bias explains the early apparent benefit, but the crossover and the magnitude of excess deaths in the oldest groups are too large and too consistent to be explained by bias alone. The evidence points to a real net harm for people 80+ that has already materialized in UK and Australian data. Confidence: moderate. The trend is consistent across countries and age groups, and the mechanism is biologically plausible, but the evidence base remains observational and the most informative study design (vaccinated vs unvaccinated over the long term with full confounder control) was never run.