Healthy vaccinee effect in practice

To the editor

We would like to thank Pietrzak et al¹ for responding to our commentary² on their original paper on COVID-19 vaccination effectiveness.³ In our letter, using data from the Czech Republic, we showed that the population opting for vaccination was generally healthier than the unvaccinated one, which causes a significant selection bias and makes any calculation of vaccine effectiveness problematic. In their response, Pietrzak et al¹ asked us for an explanation of a seemingly strange phenomenon, namely, that there was virtually no difference in all-cause mortality (ACM) among the unvaccinated individuals aged 50 to 79 years between the low- and high-COVID periods. This is actually an excellent point demonstrating the importance of careful interpretation of the data and how big a role the selection bias could play.

The key lies in understanding the group sizes and composition, and in taking into account the start of mass vaccination campaigns in individual age groups. In January 2021 (of note, the present paper analyzes solely 2021 data), unlimited mass vaccination started only in the age group of 80 years and above. In the subgroup of unvaccinated individuals of this age, there was a clear difference in ACM between the winter and summer months, as shown in our original response.² In this age group, the healthy vaccinee effect (HVE) was able to manifest itself from the start; that is, the most frail individuals (except for a relatively low number of nursing care home residents, who were all vaccinated) consisted a disproportionately large part of the unvaccinated group that grew smaller over time (Figure 1). Thus, ACM in the unvaccinated subgroup of the population aged 80 years and above was much higher than in the vaccinated subgroup during both the high- and low-COVID periods. Of note, the mortality was particularly high in March 2021 (this is true for all the analyzed age groups, as the pandemic wave peaked at that time), which increased the ACM in the high-COVID period, as shown in the previous paper.²

Let us now focus on the age groups explicitly mentioned by Pietrzak et al¹, as at first sight, the similar mortality in these groups in the high- and low-COVID periods might indeed seem strange. To illustrate the explanation, a month-by-month breakdown from January to September 2021 is provided for each of these groups in Figure 1. Note that the data used in this paper are the same as those analyzed previously.² Also, as in the previous paper, the population size in each month for each age group was determined as the mean of the population size at the beginning and end of the month. The vertical axis shows the mean daily ACM in the respective months to avoid data distortion due to the different numbers of days in the respective months.

As shown in Figure 1, the situation in the 50–79-year-old cohorts was quite different from that observed in the population aged 80 years and above. Mass vaccination rollout for the 70–79-year-old age group started only on March 1, 2021, whereas mass vaccination of the younger age groups began only after the end of the COVID-19 wave (April 14 for the group aged 65 years and above, April 23 for those aged 60 years and above, April 28 for those aged 55 years and above, and May 5 for those aged 50 years and above, respectively).⁴ Hence, in January and February 2021 (70–79-year-old group) and in January to March 2021 (60–69-year old group), the vaccinated subgroups were relatively small and included a disproportionately high number of frail individuals.

In the 70–79-year-old age group, ACM in the vaccinated individuals was higher than in the unvaccinated ones in January and February. It was also higher than the mean ACM in this age group in the prepandemic years, regardless of the vaccination status. This, however, did not mean that the vaccines were harmful, as claimed by vaccine skeptics. Rather, it meant that in these months, only the most infirm individuals in this age group received a vaccine, and the high mortality among the vaccinated individuals was a consequence of the selection bias—we could probably call it “the unhealthy vaccinee bias.” Once mass vaccination of this group started in March, the tide turned, and a reversed trend could be immediately observed, with ACM in the vaccinated subgroup plummetting (and remaining low throughout the low-COVID months) and ACM in the unvaccinated subgroup immediately increasing (and remaining high throughout the low-COVID months). This cannot be explained by any other mechanism than the HVE. It also answers the question raised by Pietrzak et al¹: the mortality rates in the unvaccinated subgroup were the same in the 2 periods of the pandemic because in the winter, the unvaccinated individuals constituted a majority in this age group, with relatively few of the most infirm persons being eligible for vaccination. On the other hand, in the low-COVID summer months, the most infirm individuals were largely included in the unvaccinated subgroup. These 2 factors led to the equalization of ACM between the low- and high-COVID periods.

The situation is very similar in the 60–69-year-old cohort, with the exception of the vaccination rollout starting later. Until the end of March, ACM in both the vaccinated and unvaccinated subgroups was above the long-term prepandemic mortality rate for this age group. As soon as mass vaccination of this age group started (April), ACM in the vaccinated subgroup dropped sharply, even though the pandemic wave was already ending, so the true protective effect of the vaccines must have had a low impact on ACM. Over time, ACM in the vaccinated individuals remained much lower than in the unvaccinated ones throughout the low-COVID months.

Lastly, the 50–59-year-old group was not at a high risk of death from COVID-19, so there was no reason to expect a massive difference in ACM between the low- and high-COVID periods. Note that the individuals who got vaccinated before the mass vaccination rollout in May were usually persons working in the critical infrastructure (hospitals, police, etc.), rather than the most infirm ones (ie, preselection of frail individuals for vaccination in this age group was not as obvious as in the older groups). Still, a minor HVE can also be observed in this age group.

We hope that the above explanation answered satisfactorily the question posed by Pietrzak et al,¹ and that this form of data presentation further clarifies the impact of HVE on any calculations of vaccine efficacy derived from observational studies. Should the readers wish to further explore this issue, we can refer them to our recent paper presenting HVE on more detailed data from 2 other health insurance companies in the Czech Republic.⁵