Intended for healthcare professionals

CCBY Open access

Rapid response to:

Research

Covid-19 vaccine effectiveness against post-covid-19 condition among 589 722 individuals in Sweden: population based cohort study

BMJ 2023; 383 doi: https://doi.org/10.1136/bmj-2023-076990 (Published 22 November 2023) Cite this as: BMJ 2023;383:e076990

Linked Editorial

Does timely vaccination help prevent post-viral conditions?

Rapid Response:

Why the Lundberg-Morris et al. study shows puzzling, diverging results

Dear Editor,

The Lundberg-Morris et al. [1] study showing that more unvaccinated than vaccinated had post-covid-19 condition, or post covid, also shows that more vaccinated than unvaccinated died during the study period [2]. The diverging results are puzzling but may be due to strong imbalances in the study groups, which I elaborate on here.

21.8 times more unvaccinated were infected early when pre-Alpha- or Alpha variants dominated than vaccinated, and 6.0 times more vaccinated than unvaccinated were infected late when Omicron variants dominated. In parallel, 2.5 times more unvaccinated were hospitalized, and 5.2 times more were under intensive care than vaccinated (discussing reasons for this belongs elsewhere). The numbers inform that the groups are very different along at least two important dimensions where unvaccinated were predominately infected by pre-Alpha- or Alpha variants and had high morbidity, and vaccinated were predominately infected by Omicron variants and had low morbidity.

The study controls for the above and other issues. Still, it does not consider that the groups have relatively few overlapping data points along different dimensions, inducing extrapolation bias, according to King and Zeng [3]. As a remedy, they recommend matching, pruning – with the risk of data loss, and balancing, but Lundberg-Morris et al. did not do that.

However, they did analyze different strata separately without necessarily solving the problem of extrapolation bias. For instance, they found in the stratum not hospitalized that unvaccinated were more exposed to long covid than the vaccinated, in line with the study’s major conclusion. Still, the study compared the unvaccinated with the vaccinated stratum where the first group was predominately infected when pre-Alpha- Alpha variants dominated, and the last when Omicron variants dominated. According to the above discussion, comparing such different groups can induce bias, even with relevant controls.

It speaks to the study’s advantage that long covid was relatively widespread among the unvaccinated, regardless of the dominant variant. Still, since the group had a progressively higher proportion of people hospitalized and under intensive care than vaccinated, it is reasonable to assume that morbidity was also relatively high among the unvaccinated in the stratum not hospitalized. Consequently, this may have explained long covid as much as vaccination status.

References
1. Lundberg-Morris L, Leach S, Xu Y, et al. Covid-19 vaccine effectiveness against post-covid-19 condition among 589 722 individuals in Sweden: population based cohort study. BMJ 2023;383:e076990. doi: 10.1136/bmj-2023-076990
2. Lacout A, Azalbert X, Lesgards J-F, et al. Concern about Covid-19 vaccine efficiency related to biased analyses not captured by the review process Re: Covid-19 vaccine effectiveness against post-covid-19 condition among 589 722 individuals in Sweden: population based cohort study. Rapid Response BMJ 2023 doi: https://www.bmj.com/content/383/bmj-2023-076990/rr-0
3. King G, Zeng L. The Dangers of Extreme Counterfactuals. Political Analysis 2006;14(2):131-59.

Competing interests: No competing interests

03 December 2023
Jarle Aarstad
Professor, dr.oecon.
Western Norway University of Applied Sciences
PO Box 7030, NO-5020 Bergen