This is the first study examining the comparative effectiveness of three COVID-19 vaccines used in Russia. Our study results assure that Gam-COVID-Vac is highly effective against symptomatic SARS-CoV-2 and severe COVID-19 pneumonia during the Delta VOC spread. We have shown that Gam-COVID-Vac provides at least 56% protection against symptomatic infection caused by the Delta VOC. However, the effectiveness is likely to be higher as it is difficult to account for all past SARS-CoV-2 infections in the Russian population. Furthermore, past COVID-19 is associated with decreased vaccine uptake in Russia. Our study’s apparent strength is the attempt to account for past infections in both cases and controls and show the resulting direction of possible bias. Assuming natural immunity is protective against re-infection, failure to account for it in populations with high seroprevalence would bias VE estimates downwards. Another possible representation of this is the change in Gam-COVID-Vac VE by age group after accounting for the history of confirmed COVID-19.
We have shown that the bias related to a significant number of unvaccinated individuals with a history of COVID-19 will likely lead to underestimating the effectiveness of vaccination from observational data. However, we remain uncertain about the possible magnitude of this bias. The individual-level data on past asymptomatic infection is challenging to obtain, if possible at all. Preliminary results show that seroprevalence in unvaccinated may be more than 75% in October, 2021 [17] in St. Petersburg. In countries with higher vaccine uptake and lower seroprevalence VE estimated in observational studies is higher [5], but these estimates are not directly comparable with our results. The Omicron VOC spread will likely make the interpretation of the VE studies even more difficult. Direct comparison similar to the Hungarian study will be needed for the new variants [11].
Our study included information on three vaccines built using different platforms; however, it was not powered to provide any intergroup comparisons. E.g. while the difference between VE point estimates of CoviVac (40%) and Gam-COVID-Vac (56%) may reflect the superior effectiveness of Gam-COVID-Vac, it could have resulted from the random error.
Both vaccines, EpiVacCorona and CoviVac, were relatively rare in the population of St. Petersburg, and our study was underpowered for them. Therefore, our study provided only preliminary findings, given the relatively small sample size, especially for these two vaccines. More studies are needed to assess the VE of EpiVacCorona and CoviVac against new variants of SARS-CoV-2 for them to be used in the ongoing vaccination programme. Unfortunately, efficacy data is currently available only for Gam-COVID-Vac [22]. Booster campaigns now gaining more scientific support should only utilise vaccines with proven efficacy and effectiveness [23,24,25].
It is also worth mentioning that while the VE for CoviVac was beyond the VE for Gam-COVID-Vac, the estimate for EpiVacCorona VE was negative. The efficacy is not likely to be negative, so our results have two realistic explanations. First, individuals could change their behaviour after vaccination, but more likely negative VE is a marker for the bias arising from the undercounting of past COVID-19 in controls.
This is our second VE study in St. Petersburg, Russia, and it provides a promising independent and timely framework for assessing COVID-19 vaccines in Russia. Population-based case-control studies represent a critical post-registration tool to monitor VE against emerging SARS-CoV-2 VOCs. The Omicron VOC pandemic has not involved Russia by the end of 2021, but there are few doubts that it will affect the course of the pandemic in Russia as previously the Delta VOC had [16]. The lack of real-world evidence may be one of the reasons behind the modest uptake of vaccination in Russia. The majority in Russia does not deny the idea of vaccination but is hesitant [26].
Despite the wide use of case-control studies to assess VE, researchers should be aware of all possible biases arising from this study design. Unfortunately, the golden standard to estimate VE — randomised trials — are not applicable in the rapidly changing epidemiological situation, and we have to rely on observational study design. The varying VE against different SARS-CoV-2 variants is an example of a lack of generalizability for the results of randomised trials. Our study underlines the biases related to the population under study, but additional biases arise from the misclassification of exposure, e.g., vaccination status [27].
The self-reported vaccination status is an important limitation of our study. Several survey participants included in the control group have not reported the exact date of vaccination. While the overall number of such individuals was low, we assumed that the vaccination date for such individuals is likely to be several months from the interview date. However, we assigned them a “non-vaccinated” status in our sensitivity analysis, and the estimates were only slightly affected. Our definition for full vaccination status was also very conservative, as we decided to accept a minimum of six days between the second vaccine dose and study inclusion. While our decision was driven by the idea that we should not exclude participants without an exact date of vaccination, we do not think that this assumption would significantly bias the results. However, most of the studies choose 14-day period [5], and that should be taken into account when comparing our results to other studies.
We have undertaken additional attempts to identify cases (patients with symptomatic SARS-CoV-2 in October, 2021) who had the history of confirmed COVID-19 more than two months before the current episode. We were able to identify only two cases of re-infection. While underreporting may occur, it is also likely that a patient with re-infection that requires additional diagnostic follow-up is an infrequent event. Absolute risks of re-infection, especially of severe disease, are low for the Alpha, Beta, and Delta VOCs [28, 29]. However, more studies are needed to observe the risk of re-infection with new Omicron VOCs, as it is likely to be higher [30]. Overall, the risks may still be lower in absolute terms than for primary infection.
In our study, the VE in the Sputnik Light group was similar to two-dose Gam-COVID-Vac. However, the correction for the history of confirmed COVID-19 did not move the Sputnik Light VE upwards. Single Gam-COVID-Vac vaccination labelled as Sputnik Light was used as a booster after the COVID-19, so it is likely that the prevalence of past COVID-19 is higher in this group. The VE for Sputnik Light could represent the combination of single-dose boosted natural immunity mixed with single-dose vaccine.
Some of these limitations are inherent to observational study design. Still, other difficulties can be overcome by establishing a pre-existing framework for real-time assessment of vaccine effectiveness as a part of epidemiological surveillance. We did not attempt to assess the duration of protection. Additional studies are needed to explore the waning effectiveness of Gam-COVID-Vac and other vaccines. Our sample size, risk of misclassification related to self-reported vaccination status, and possible selection did not allow its valid estimation. The cohort study with a large sample size and registered vaccination status is probably the best tool to assess waning effectiveness [31].
In conclusion, our preliminary results show that Gam-COVID-Vac effectiveness against symptomatic SARS-CoV-2 infection caused by Delta VOC is at least 56%, but is likely to be higher. However, estimating effectiveness is difficult due to the high prevalence of natural immunity in the population. Nevertheless, Gam-COVID-Vac significantly outperforms two other Russian vaccines whose effectiveness against symptomatic SARS-CoV-2 infection caused by Delta VOC is yet to be shown.