Vaccines | Free Full-Text | The Effectiveness of the Third Dose of COVID-19 Vaccine: When Should It Be Performed?
1. Introduction
This study first aims at evaluating the potential effectiveness of booster dose administration to HCWs in a real-life setting after the completion of a two-dose vaccination-cycle completion. Data from registries of the main three different hospitals of the city of Milan were analyzed considering a follow-up period of 7 months from the start of the booster vaccination campaign.
2. Materials and Methods
2.1. Study Population
Our case series included all COVID-19-vaccinated HCWs of three large University Teaching Hospital networks in Milan (identified in the text as HN1, HN2, HN3) employing about 16,000 HCWs. All the considered hospitals’ networks have an agreement with the University of Milan to teach students of medicine and other health professionals and resident doctors.
Since the beginning of the pandemic, the Occupational Health Units of all involved hospitals have carried out intense activity in contact tracing and management of SARS-CoV-2 positive HCWs, following the national and regional legislative directives. Furthermore, from 28 December 2020, the physicians of the Occupational Units, supported by resident medical doctors, have also been actively involved in anti-COVID-19 vaccination. Vaccinations were performed using the BNT162b2 Pfizer–BioNTech© COVID-19 vaccine through the creation of an on-site vaccination center in the hospitals or sending a HCW to an external vaccine hub to ensure the utmost convenience and maximum adhesion. The vaccine was administered according to an initial two-dose schedule (21-day interval) and, approximately 8–10 months later, a subsequent booster dose. Therefore, booster doses were administrated between November 2021 and January 2022.
At the same time, the diagnosis of COVID-19 infection in HCWs was carried out by Occupational Health Units through the execution of PCR carried out on nasopharyngeal swabs in patients undergoing screening checks, in the presence of symptoms suggestive of infection or close contact with people with a confirmed infection in all three hospital networks assessed. After the administration of the third dose, no specific record about viral isolate variants was performed due to the high costs of analysis and the lack of specific utility in regard to the absence of differences in the treatment or management of the different variants.
2.2. Data Collection
Data related to the date of vaccinations and the number of doses administered within the three involved hospital networks were collected by Occupational Health Units within a specific computerized database. Moreover, demographic data were also collected such as gender, age, and anamnestic features with leading influences on vaccination. For those vaccinated elsewhere, information was completed by linking the cohort file with a regional vaccination database using Tax Code as a unique identifier. The specific computerized database in use at the Occupational Health Units also collected the number of nasal swabs performed with related results. Due to previous organization, HN2 and HN3 also collected clinical reports about signs and symptoms during COVID-19 infection to assess change in clinical findings among HCW with different vaccination doses. HN3 additionally assessed the hypothetical causes of infection such as close contact or professional exposition in a high-risk clinical ward. At the same time, HN1 collected professional features of HCWs infected by SARS-CoV-2 after vaccination, work position, and workplace.
2.3. Statistical Analysis
Since the measure of a real-life effectiveness of the booster dose administration was the primary endpoint of the study, this was evaluated by means of a time-to-event analysis. Here, the event of interest was SARS-CoV-2 infection. Namely, HCWs were observed in a period starting from the first day of the booster dose vaccination campaign to 7 months after that, when the follow-up administratively ended. During this period, HCWs could develop SARS-CoV-2 infection before or after booster dose administration or remain free of infection during the overall follow-up period. The latter were considered censored observations.
Given the presence of the censored observations and the follow-up time recorded in weeks, to quantify the effect of booster dose administration, regression analysis was performed on the discrete-time hazard function of infection. Essentially, this quantity measures the risk of SARS-CoV-2 infection in each week for a HCW, conditional on not having contracted the infection until that week. Booster dose administration was specified as a categorical time-dependent covariate. Indeed, to be administered with the booster dose, a subject must have been free of infection until the time of administration. Therefore, the times of infection of the group of subjects receiving the booster dose were not directly comparable with those of the group which did not receive it, as the time origin was different between groups.
According to a possible scenario of seasonal peaks in the conditional risk of SARS-CoV-2 infection during the follow-up time, a B-spline was utilized to smooth the possibly multipeaked baseline hazard function. A criterion of the minimization of the AIC was considered to choose the degree of its flexibility.
Also, to account for the within-HCWs difference in times from the completion of the two-dose vaccination cycle to the start of the follow-up, time from the completion of the two-dose vaccination cycle to the start of the follow-up was measured in weeks and regressed. This could be considered a proxy of the pre-existing level of immunity due to the vaccination cycle. Since the higher the time from the completion of the vaccination cycle to the start of the follow-up, the lower, potentially, the “booster” effect of the booster dose due to the lower pre-existing levels of the immune activity against the virus, the variable was regressed as having an interaction effect with the booster dose covariate.
In addition, to account for infections that occurred before the follow-up, a simple categorical covariate was considered, with two categories: “previously infected”, “not previously infected”. To account for a possible differential effect of the booster dose for subjects previously infected and subjects not previously infected, in this case, an interaction effect with the booster dose administration covariate was also specified in the regression model.
A sub-analysis of the NH1 data with complete information about demographic characteristics such as gender, age and occupation of the HCWs was performed to assess the effect of these other covariates on the hazard of infection. The same modeling procedure used for the whole cohort of HCWs was considered. In addition, to model a possible non-linear effect of the variable of age, a restricted cubic spline was utilized.
The exponentiated coefficients of the respective models related to the covariates offered the odds ratio (OR) of being infected with SARS-CoV-2 between covariate categories. For the sub-analysis, a forest plot represented the effect of the demographic categorical covariates, whereas the continuous non-linear effect of age was shown in a separate plot.
All the statistical analyses were performed with R software (version 4.2.2). The function Lexis from the package Epi was utilized to augment the data to fit the discrete-time hazard model.
3. Results
The databases of the three hospital networks exhibited great heterogeneity due to the difference in the data collected. Despite a relative uniformity of the demographic data such as gender, the heterogeneity chiefly concerned occupational data or the ways in which health data such as signs and symptoms were collected. One major challenge in retrieving data for the analysis was the lack of information about the demographic and occupational characteristics of all the HCWs. Namely, demographic information (age, occupation) about uninfected HCWs were not available for all the hospitals. To retrieve homogeneous information from the three HN and align it with the study objectives, a total cohort of 12,141 HCWs was considered. These individuals had complete information about the vaccination status (subjects must have completed the vaccination cycle) and time to an eventual infection of SARS-CoV-2 within the follow-up period was considered. Considering that usually, a period of 14 days is required from dose administration to complete the immunization process, subjects infected within 14 days from the booster dose administration were not included in the analysis.
Among the analyzed cohort of HCWs, 560 (5%) did not receive the booster dose within the specified time frame. In addition, 2563 (21%) HCWs became infected with SARS-CoV-2. The median time from completing the vaccination cycle to the start of the follow-up period was 34 weeks, with an interquartile range (IQR) of 4 weeks.
As expected, the weekly risk of SARS-CoV-2 infection during the follow-up exhibited a distinct multi-peaked shape, with a major peak of the risk of infection at 15 weeks from the start of the booster campaign (January–February 2022) and a minor peak at 25 weeks (April 2022).
Unexpectedly, in univariable analysis, receiving the booster dose was slightly associated to higher risk of infection with respect to not receiving it, OR = 1.18, 95% confidence interval (CI) [0.99, 01.41]. This effect was counterintuitive since the booster dose is expected to provide greater protection against the infection. However, in the multivariable adjusted analysis, the booster dose was found to have a strong protective effect against infection, OR = 0.43, 95% CI [0.26, 0.74]. Notably, the interaction effect between the time elapsed since completion of the vaccine cycle and the administration of the booster dose was highly significant when tested with the Wald statistic (p < 0.001).
The modulating effect on the effect of the booster dose was both quantitative and qualitative, leading to the conclusion that it was non-protective against infection when the comparison was made between subjects with longer time since completion of the two-dose vaccine cycle.
Regarding previous infections, the presence of an infection before the start of the follow-up was highly associated to the risk of infection, OR = 13.2 95% CI [8.00, 21.8]. Even though a previous infection might be associated to an increased protective effect of the booster dose, the differential effect of the booster dose in reinfected and not reinfected HCWs did not reach statistical significance.
HN1 Sub-Analysis
The multipeaked shape of the conditional weekly risk of infection was like that found in the whole cohort.
4. Discussion
Following an initial analysis, our findings indicated that the population receiving a booster dose had a higher risk of infection than the population receiving only two doses. This finding may give rise to many debates about the rationale behind these data, which may support pseudoscientific beliefs and vaccine reluctance. In order to explain our preliminary results, we hypothesize that the increase in infections after booster dose administration could be linked to the timeframe in which the primary vaccination cycle was performed in relation to the time when the third dose was administered. For this reason, during our analysis, we assessed the timing between first or second dose and booster administration, observing that the booster dose’s time distance was a strict predictor of its protective effect. Indeed, some healthcare workers who became infected after receiving the booster dose were administered the dose several weeks following the proper date or twelve weeks following prior infection. Despite these results, the characteristics of our study’s design prevent us from identifying a causal relationship—only an association—between the timeframe between vaccination dose administration and the likelihood of infection. Consequently, it is possible that other factors influenced the probability of infection during the observation period. However, our findings highlight the importance of administering vaccines at the appropriate time and may apply to all vaccinations. Given the significance of immunization, the need for an early intervention, and the efficacy of correctly administered vaccines, the results of this study could be a cornerstone in the prevention and treatment of a future pandemic by increasing knowledge on the topic and aiding in the reduction in vaccination hesitancy among both healthcare workers and the general population.
Despite the important results mentioned above, our study suffers several limitations. The evaluation of the real-life effectiveness of the booster dose is a relevant aspect in a public health perspective, but not an easy task to perform. Indeed, the real-life setting essentially leads to the analyses of observational data derived from registries with all the issues related to observational studies: the risk of unmeasured confounding and the interpretation of the results only in an associative way. As already explained, in our study, the registries were different in the quantity and quality of information recorded, making it difficult to find common confounding variables that might affect the results related to the effect of booster dose administration, even though these were confirmed when considering some additional demographic and occupational characteristics in the sub-analysis. In addition, some positive NasoPharyngeal Swab (NPS) tests performed outside the hospital networks may have been missed when HCWs did not communicate their COVID-19 positivity to Occupational Health Physicians, or some vaccinations could be performed outside Lombardy. Finally, as reported, this was an observational study that could not identify a causal relationship between different times of vaccination and the risk of infection. In order to confirm these results in a causal rather than just associative manner, future research must measure confounding as much and as homogeneously as possible.