Vaccination is the safest means of protection against coronavirus disease (COVID-19). The level of protection for a population depends on the achieved coverage and the effectiveness of the vaccines. Vaccine effectiveness (VE) against COVID-19 is not constant over time due to the emergence of new SARS-CoV-2 variants [1] and waning of vaccine-derived immunity [2, 3]. Therefore, it is crucial not only to estimate the effectiveness of vaccines against COVID-19, but also to timely monitor changes in their protective effect in the population. We describe a simple approach using surveillance data, based on the screening method, to timely monitor the actual protection of all vaccines applied in the population.

We analyzed data on mandatory notifications of laboratory-confirmed COVID-19 infections in Berlin from April until December 2021 [4]. Laboratory confirmation required detection of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) nucleic acid by PCR.

We restricted our analysis to symptomatic cases to reduce possible bias through underdiagnosis of asymptomatic vaccinated cases and to exclude cases that have not yet been investigated by the local public health authority (LPHA). A symptomatic case was defined as a notified person with laboratory-confirmed COVID-19 infection for whom the LPHA has entered at least one symptom compatible with COVID-19 [5], or an onset of disease in the notification software.

A vaccine breakthrough infection (henceforth: a fully vaccinated case) was defined as laboratory-confirmed COVID-19 infection in a person with onset of disease ≥14 days after completion of a full vaccination cycle with any of the vaccines approved in Germany (see Supplementary Material). If the date of onset was missing, we used the date of laboratory confirmation or, if also missing, the notification date.

Booster breakthrough infections were defined as infections in fully vaccinated cases with an additional third dose of Comirnaty or Spikevax.

As of 15 December, 23,079 breakthrough infections have been detected, of which 236 have been detected in persons who had received a booster vaccination. In total, 19,467 had a symptomatic infection, 292 were admitted to hospital, and 87 died (Table 1).

We calculated 14-day notification rates by dividing for each day all symptomatic cases of the last 14 days by the respective population of Berlin, thus creating a 14-day rolling time series of notification rates (Figure 1A). As population for the fully vaccinated cases (n _fullvac ), we used those who received their second (in case of Janssen, first) vaccination 21 days prior to the notification date [6], as this represented the number of fully vaccinated people at the midpoint of the 14-day rolling window. The population (pop) for the cases with a booster vaccination (n_booster) was defined as those who received their third vaccination 21 days prior to the notification date. For the 14-day notification rate of unvaccinated cases (n _unvac ), we excluded everyone with at least one dose of a vaccine (n _vac ) and additionally those who recovered in the past 6 months (n _inf6mo ) from the population of Berlin [7]. n u n v a c ( t ) =   p o p   −   n v a c ( t − 21 )   − n i n f 6 m o ( t )

To express differences in notification rates according to vaccination status, we divided the 14-day notification rate of unvaccinated cases by the 14-day notification rate of 1) fully vaccinated cases (including those with booster vaccinations), 2) fully vaccinated cases but without booster vaccinations, and 3) all cases that received a booster vaccination, i.e., we computed for each day t relative risks RR(t) for the unvaccinated (formula for case 1, additional formulas can be found in the supplement). R R ( t )   = ∑ i   =   t − 13 t c a s e s u n v a c ( i ) / n u n v a c ( t ) ∑ i   =   t − 13 t c a s e s f u l l v a c ( i ) / n f u l l v a c ( t − 21 )

To estimate the protective effect (henceforth: vaccine protection (VP)) against symptomatic infection (Figure 1B), we used:   V P ( t ) = ( 1 −   1 R R ( t ) ) × 100 , analogous to the computation of vaccine effectiveness. Because our estimate reflects an actual snapshot of the combined effectiveness of all vaccines applied in the population, unadjusted for potential confounders (e.g., age), we coined it VP to distinguish it from VE (which usually applies for an adjusted estimate of a single vaccine). Its value lies primarily in observing changes over time, not in accurately estimating VE once.

By doing so, we made the following observations: First, there is a substantial drop in VP in June, which coincides with the emergence of the SARS-CoV-2 variant of concern (VOC) Delta. The correponding VPs are in line with VE estimates during the emergence of Delta in other countries [1, 8, 9]. Second, for both the time in which VOC Alpha dominated and for the first months in which VOC Delta dominated, there is a moderate increase in VP, which might result from the decreasing average age of the vaccinated population. Vaccines were given first to those with the highest risk of developing severe COVID-19 particularly the older population [10], this might have led initially to an underestimation of VP [11]. Third, since mid-August, the VP constantly decreased from around 84% in August to around 60% in November. This probably signals waning immunity [9, 12, 13]. Time since vaccination is systematically longer for older age groups (Figure 2), potentially increasing the effect of waning immunity visible in our data. Fourth, additional booster vaccinations markedly increased overall VP. When calculating VP only for those who received a booster vaccination, VP increased to around or even higher than the protective level initially observed when VOC Alpha was dominant.

Our estimates apply for the combined effect of all vaccines given to the Berlin population (Table 1) and might be affected by changes over time in the individual contribution of the vaccines. However, when restricting our analysis to the predominantly applied vaccine in Berlin, Comirnaty, we saw the same pattern (data not shown). In general, VP estimates derived from surveillance data may suffer from several biases. Most notably, they are incomplete. Completeness of COVID-19 cases might differ by vaccination status, due to differential health-seeking behavior. Additionally, the accuracy of vaccine coverage data might be limited. Notwithstanding inherent limitations, our VP estimates are congruent to published VE estimates. Notably, interpretation of our analysis should focus on changes over time rather than single-point estimates. As such they can complement more accurate studies on vaccine effectiveness by monitoring VP in the population over time, provided that biases in the surveillance data remain stable over time.