Based on a national programme prospectively investigating all contacts of a confirmed SARS-CoV-2 infection, we were able to quantify viral infection and direct and indirect vaccine effectiveness in real-life settings over a 14-month period. With a SAR of around 50%, household settings had the highest infection attack rates irrespective of index cases’ and contacts’ vaccination statuses, while occupational and school settings showed lower rates of infection. Such a difference may be due to infection pressure (duration and type of contact), as well as non-pharmacological measures such as mask wearing or social distancing, which are difficult to implement in households [29, 30].

Infection in household settings was higher than previously described: one meta-analysis estimated SAR out of 135 studies and reported 29.7% and 42.7% during the delta and omicron BA.1&2 periods, respectively [31]. Such differences may be explained by several factors: first, studies testing and collecting all contacts’ test results regardless of symptoms measured similarly high SAR, even in the pre-delta period [15]; second, Monaco is the most densely populated country in the world, with commuting workers doubling the local population every day, thus intensifying viral circulation. If the inclusion of asymptomatic contacts improves the generalisation of our findings, it also illustrates the intensity of viral circulation regardless of clinical presentation — and could lead to multiple importations into these settings.

In addition, the modelled aVE against infection was significant in delta and omicron BA.1&2, but low, 32% and 27%, respectively. Although this adjusted vaccine effectiveness is low, at the population level this indicates that a large number of vaccinated contacts would still be protected against infection during the delta and omicron BA.1&2 periods. The aVE for reducing transmission from index cases to contacts was substantially lower and was only significant during the omicron BA.1&2 period, though the lack of significance during the delta period may be caused by the smaller sample size. The aVE against both infection and transmission were statistically insignificant during the omicron BA.4&5 period, for all vaccination histories. The absence of vaccine impact during this latter period may be attributed to a combination of increased immune evasion for these variants and the fact that a large proportion of the population -including the unvaccinated—had by then been infected with SARS-COV-2 antigen by previous delta and in particular omicron BA.1&2 waves. Regardless of the vaccination scheme (i.e. one dose with a previous SARS-CoV-2 infection; two-dose regimen; or at least one booster dose), aVE was constantly calculated below 50% and reached its maximum in case of a previous SARS-CoV-2 infection, highlighting the importance of mucosal immunity in virus transmission [32,33,34]. Fully vaccinated contacts appeared to be more protected than cases (Fig. 2), and when analysing distinct vaccination schemes, the most protective strategy was provided amongst contacts with a single vaccine dose and a previous infection during the delta period (Fig. 3). During the omicron BA.1&2 period, fully vaccinated cases and contacts were significantly protected, and this protection was also conferred to the combination of a single dose and a previous infection. A booster dose was also providing a significant protection during the omicron BA.1&2 period, but only to contacts and not cases. There are only a few other studies that have screened all contacts in large prospective cohorts, and these similarly observed the limited impact of the vaccine on viral circulation [9,10,11,12,13,14,15]. However, many of these studies either did not adjust for important factors such as age, previous SARS-CoV-2 infections, number of vaccine doses including boosters, or inconsistently included both index cases and contacts without symptoms. Furthermore, none has studied the omicron BA.4&5 period. In the studies which adjusted for these factors, the authors similarly found significant aVEs for reducing both transmission and infection risk, but these findings were limited to the delta variant period [13, 14]. When adjusted for vaccination status, the delta and omicron BA.4&5 periods were found to have a lower risk of transmission and infection than the omicron BA.1&2 period. This may be due to the increased transmissibility of omicron BA.1&2 relative to delta, and the high levels of immunity generated during the omicron BA.1&2 period increasing population-level immunity during the omicron BA.4&5 period.

The strength of this analysis is the access to routine surveillance and immunization individual data on index cases and contacts for SARS-CoV2. This dataset includes several levels of disaggregation (age, gender, presence of symptoms, various dates) to produce vaccine effectiveness outputs in various settings. Initially, data quality check was operated manually; however, we defined processes and scripts to operate automated data quality checks and validation rules to improve the timeliness of analysis and reports. This process is reproducible and could be implemented on a continuous basis for real-time monitoring and evaluation of surveillance and immunization indicators for SARS-CoV.

By the end date of the study, the proportion of the population of Monaco that remained unvaccinated was 21%. In contrast, the proportion of unvaccinated index cases in this study is approximately 44%. This reflects the increased risk of becoming an index case if unvaccinated, which is a combination of increased risk of infection and increased risk of symptoms if infected (which increases the probability of the infection being detected and thus becoming an index case). Among contacts, the vaccination uptake differs by setting. In occupational settings, 26% of contacts were unvaccinated. This matches the population distribution, since the vaccine status of occupational contacts are independent of case vaccine status, and identification of contacts is independent of contact vaccine status. In household settings, however, the proportion of unvaccinated contacts was approximately 47%. This did not reflect the vaccine distribution among the general population, instead matching the vaccine distribution among index cases (and the case detection rate is expected to be higher in the unvaccinated part of the general population). This is because the vaccine status of household contacts will be dependent on the vaccine status of the corresponding index cases, since individuals within a single household will follow similar behaviour around vaccination. In the multivariate analysis, we adjust for the vaccine status of both index case and contact, which controls for this correlation in vaccine status.

Nevertheless, our analysis presents several limitations. First, positively tested contacts may have been infected by another index case than the one assumed due to other exposures outside the household. Our modelled analysis focused on households, where the likelihood to link index cases with contacts is higher than in other settings. Also, contacts who tested positive on the same day as their index case were considered co-primary cases, especially those with the same date of symptom onset (if any). Yet, we cannot exclude that some household contacts may be the actual index case. Second, only a small number of the 30,535 index cases and contacts were sequenced, so SARS-CoV-2 variant periods only represent an indicator of predominance. Further, we could not analyse each omicron subvariant separately, because of an insufficient number of sequenced cases and frequent co-circulation (e.g. SARS-CoV-2 omicron BA.1 and BA.2 co-circulated between February 2022 and March 2022, see Additional Fig. S2). Third, only contacts with a screening result were included in the analysis. Asymptomatic contacts may be less likely to get tested, which could lead to fewer vaccinated contacts getting tested due to reduced disease severity. It is therefore possible that we underestimated the number of infected contacts if vaccinated. Fourth, there was a large volume of missing data in the household analysis. This reduced the sample size which can be used in the statistical multivariate modelling. Since most missing data relates to the target variable, contact test result, or the main explanatory variable, vaccination status, we did not perform imputation. If the missingness is not at random, there is a risk that the exclusion of these data may introduce bias into the study. However, the majority of the missing values corresponded to contacts where the test was “not done”. Therefore, these are likely missing at random, and should not introduce much bias into the study. Most individuals removed due to missing vaccination data were due to poor data quality around vaccination dates. The collection of these data is likely uncorrelated with the infection status of the individual, so these are most likely missing at random. Therefore, the risk of bias from the missing data in this study is minimal. Another limitation of this study is the potential for self-selection bias. The statistical analysis focused on household settings, where high/low-risk behaviour is likely to have lower variance relative to external contacts. However, there may be some remaining behavioural biases. For example, if vaccinated individuals have milder symptoms, they may feel less inclined to isolate from household members prior to testing positive, which may lead to a higher risk of infection, reducing the observed protective effect of vaccination. On the contrary, unvaccinated individuals may be less health conscious, and therefore choose not to isolate from household members, which would bias the results in the other direction. Self-selection may also be present in the distribution of vaccines across the population. If vaccine uptake is higher in at-risk groups, this could affect the study. However, although individuals at higher risk may be more likely to become infected given an infectious contact, they may also undertake stricter isolation/protection measures, which could reduce the risk of becoming infected. These factors bias the results in opposite directions. We reduced this bias by adjusting for age, which accounts for a lot of multimorbidity associated with COVID-19 risk. However, without full multimorbidity data, we could not fully adjust for this. Last, we did not include the mRNA bivalent COVID-19 vaccine targeting both the original Wuhan-1 and omicron BA.4&5 strains, since our analysis ended prior to its launching in Monaco. Whether this vaccine is more efficient at limiting viral transmission and infection needs to be investigated, knowing that recent evidence reported limited neutralisation activity on the latest omicron subvariants [35,36,37].

The implication of such significant yet low direct and in particular indirect VE should be viewed from a public health perspective: although it was not known how effective COVID-19 vaccines were in preventing transmission some national campaigns promoted COVID-19 vaccine as a protective measure for “protecting others”, which may have created potential distrust, undermining population adherence to future immunisation recommendations [38]. Emphasising the utility of mRNA COVID-19 vaccines on severe disease and mortality and the role of non-pharmacological measures on transmission could help the population to better understand, and therefore accept, public health interventions [39,40,41].