Main

The emergence and spread of the B.1.1.529 lineage of SARS-CoV-2, the cause of COVID-19, has caused an unprecedented number of infections worldwide in a short period1,2. Emerging evidence suggests that Omicron causes less severe disease than previous variants of concern (VOCs), probably due to lower virulence, infection-acquired immunity and higher vaccination coverage3,4,5,6. However, its high transmissibility and ability to partially evade the immune response induced has been associated with a substantial increase in severe COVID-19 cases globally2. The absolute number of pediatric hospital admissions has also surpassed previous waves4,7,8, straining healthcare systems even further. The increase may be related to higher transmissibility of Omicron, less use of face masks in children and, especially concerning, lower vaccination rates among children.

Policymakers urgently need evidence of the effectiveness of vaccines in preventing severe clinical presentations of COVID-19 in children to balance the costs and benefits of mass vaccination campaigns in this population. Although the risk of severe COVID-19 in healthy children is substantially lower than among adults, vaccinating children may reduce community transmission, avoid potentially life-threatening presentations such as multisystemic inflammatory syndrome in children (MIS-C) or pediatric inflammatory multisystem syndrome (PIMS) and prevent long-term consequences of SARS-CoV-2 infection9. Although many countries are vaccinating children, few have authorized COVID-19 vaccines for children under 5 years of age, and some have restricted vaccines for children older than 12 years10. Evidence of the efficacy or effectiveness of COVID-19 vaccines in children is limited, primarily related to mRNA vaccines, and only two studies were conducted during the Omicron outbreak11,12,13,14,15. To our knowledge, there is no published evidence of vaccine effectiveness against COVID-19 in young children under 5 years of age. Furthermore, recent research suggests that several COVID-19 vaccine platforms provide limited protection against infection and symptomatic disease caused by the Omicron variant but were more effective against severe disease16,17,18. These studies have examined vaccine protection against Omicron in adult populations but are consistent with preliminary, unpublished results from a study in children aged 5–12 years13.

Leveraging a population-based cohort of children aged 3–5 years, we estimated the effectiveness of the complete primary immunization schedule (two doses, 28 days apart) of an inactivated SARS-CoV-2 vaccine, CoronaVac, to prevent laboratory-confirmed, symptomatic COVID-19, hospitalization and admission to an ICU. We estimated vaccine effectiveness using inverse probability-weighted survival regression models to estimate hazard ratios of complete immunization (starting 14 days after the second dose) over the unvaccinated status, accounting for time-varying vaccination exposure and available clinical, demographic and socioeconomic confounders at baseline.

Our study cohort included 516,250 children aged 3–5 years affiliated with the Fondo Nacional de Salud (FONASA), the public national healthcare system of Chile. In total, 490,694 children were included in the final study population; 194,427 had received two doses of CoronaVac, 28 days apart between 6 December 2021 and 26 February 2022; and 189,523 had not received any COVID-19 vaccination by the end of the follow-up period. On 25 November 2021, the Public Health Institute of Chile authorized the emergency use of CoronaVac on young children (3–5 years) and began vaccinating on 6 December 2021. CoronaVac was the only COVID-19 vaccine authorized for young children during the study period. We excluded children who had probable or confirmed COVID-19 according to reverse transcription polymerase chain reaction (RT–PCR) assay for SARS-CoV-2 or antigen test before 6 December 2021, reported to the Ministry of Health (Fig. 1). The cohort characteristics are described in Extended Data Tables 1 and 2. We found statistically significant differences (P < 0.001) in the incidence of COVID-19 and according to vaccination status by children’s sex, age group, comorbidities, nationality, region of residence and insurance category, which justify their inclusion in the models. Vaccination rollout was organized through a public schedule; children needed to show up at their nearest vaccination site with their national ID card (Extended Data Fig. 1). The study period overlapped with that of the Omicron outbreak in Chile (with the Omicron BA.1.1 sublineage predominant), defined by whole-genome sequencing of a sample of the infecting variants circulating over time (Extended Data Tables 3–5 and Extended Data Fig. 2).

Fig. 1: Study participants and cohort eligibility, 6 December 2021 through 26 February 2022.
figure 1

Participants were 3–5 years of age, affiliated to the FONASA, the public national healthcare system, and received two doses of CoronaVac, 28 days apart between 6 December 2021 and 26 February 2022 or did not receive any COVID-19 vaccination. We excluded children who had probable or confirmed COVID-19 according to RT–PCR assay for SARS-CoV-2 or antigen test before 6 December 2021.

Full size image

The estimated adjusted vaccine effectiveness for CoronaVac in children aged 3–5 years during the Omicron outbreak was 38.2% (95% CI, 36.5–39.9) for the prevention of COVID-19, 64.6% (95% CI, 49.6–75.2) for the prevention of hospitalization and 69.0% (95% CI, 18.6–88.2) for the prevention of COVID-19-related ICU admission (Table 1). We did not estimate vaccine effectiveness against fatal outcomes because only two deaths were observed in the unvaccinated group as of 26 February 2022, the study end.

Table 1 Effectiveness of the CoronaVac COVID-19 vaccine in preventing symptomatic COVID-19 outcomes in children 3–5 years of age in the study cohort according to immunization status, 6 December 2021 through 26 February 2022a
Full size table

Our estimates provide evidence of vaccination effectiveness in children aged 3–5 years during the Omicron outbreak in Chile (Table 1 and Extended Data Fig. 3). These results are substantially lower than recent preliminary estimates of the effectiveness of two-dose vaccination of CoronaVac in children 6–16 years, in a period when B.1.617.2 (Delta) was the predominant circulating SARS-CoV-2 variant14. In that study, the estimated effectiveness in children 6–16 years was 74.5% (95% CI, 73.8–75.2) for the prevention of COVID-19, 91.0% (95% CI, 87.8–93.4) for the prevention of hospitalization and 93.8% (95% CI, 87.8–93.4) for the prevention of COVID-19-related ICU admission. The estimates for the subgroup of children aged 6–11 years were 75.8% (95% CI, 74.7–76.8) for the prevention of COVID-19 and 77.9% (95% CI, 61.5–87.3) for the prevention of hospitalization14. Although the estimates are not directly comparable, the lower estimated vaccine effectiveness in this study could be due to Omicron or because the cohort included younger children. Vaccine effectiveness was estimated shortly after vaccination. In light of recent evidence suggesting that the effectiveness of a two-dose COVID-19 vaccination against infection and symptomatic disease wanes over time19, our estimates of protection for children aged 3–5 years may be at their highest level.

Recent research suggests that currently available vaccines may be less effective against Omicron. Consistent with our results, an unpublished study in New York13 found that the effectiveness of two BNT162b2 vaccine doses for the prevention of COVID-19 and hospitalization decreased from 66% to 51% and from 85% to 73% for children aged 12–17 years, respectively. The drop was more considerable among children aged 5–11 years; protection against COVID-19 fell from 68% to 12%; and protection against hospitalization fell from 100% to 48%13. Preliminary, unpublished results from a large cohort of children aged 3–11 years in Argentina show that two doses of Sinopharm’s inactivated SARS-CoV-2 vaccine BBIBP-CorV were 59% effective against hospitalization when Omicron was the predominant variant and 83% effective when Delta and Omicron circulated15. Results among adults tell the same story. Early data from South Africa reported that BNT162b2 protection against COVID-19-related hospitalization decreased from 93% to 70% among adults16. Among adults in the United Kingdom, two doses of ChAdOx1 nCoV-19 provided no detectable protection against the Omicron variant after 20 weeks, and two doses of BNT162b2 were only 8.8% effective against Omicron after 25 weeks17. The study suggests that a BNT162b2 or mRNA-1273 booster substantially increased protection against Omicron17. Similarly, a study that evaluated serum neutralization against Omicron or the D614G variant among adult participants with the mRNA-1273 vaccine primary series observed neutralization titers 35 times lower for Omicron18.

Children’s age could also potentially affect vaccine effectiveness estimates for severe disease, as suggested by older children in recent unpublished studies in New York13 and Chile14. Clinical trials for Moderna’s mRNA-1273 and Pfizer-BioNTech’s BNT162b2 in children 6 months of age to under 5 years of age are being conducted. Preliminary results for two 3-µg doses, 21 days apart, of the BNT162b2 in children 2 years of age to under 5 years of age did not produce an adequate immune response, although the immune response of children between 6 months of age and 2 years of age was similar to that of young adults20. Data from the mRNA-1273 vaccine in children have not yet been released.

Observational studies have limitations. Selection bias could affect vaccine effectiveness estimates if the vaccinated and unvaccinated groups are systematically different. We partially addressed this issue by adjusting our estimates with observable confounders that may affect vaccination and the risk of COVID-19. However, we do not have data to assess whether vaccinated and unvaccinated children or their caregivers differ in some unobservable characteristics, such as compliance with COVID-19 behavioral guidelines. Another limitation in our study relates to genomic surveillance capabilities. The Ministry of Health’s strategy has focused on detecting VOCs through traveler and community surveillance but uses non-probabilistic sampling (Extended Data Fig. 2 and Extended Data Tables 3–5). There were few child admissions to the ICU associated with SARS-CoV-2 infection during our study period, which led to wide CIs in our estimates. Finally, because laboratory-confirmed COVID-19 cases depend on the patients’ healthcare-seeking behavior, it is possible that asymptomatic or mildly symptomatic cases were missed in our study. Although this may occur in both groups, immunized children may be more likely to develop mild symptoms due to vaccine-induced protection than unvaccinated children. If so, we might have overestimated protection against symptomatic infection. However, this potential bias would not have affected our effectiveness estimates for protection against COVID-19-related hospitalization and ICU admission. Our study examined the effectiveness of a two-dose CoronaVac schedule; the results may be different with a homologous or heterologous booster dose, as shown for adults.

Strengths of the study include that data were collected during the Omicron outbreak, with the highest transmission rates since the beginning of the pandemic. Vaccination rollout in Chile was quick and had high uptake (Extended Data Fig. 1). Our estimated vaccine effectiveness reflects a ‘real-life’ situation by including the challenges public health officials face in the field, such as a more diverse set of participants (for example, with underlying conditions), schedule compliance, logistics and cold chains. These estimates may be essential for decision-making as a complement to controlled clinical trials.

Overall, our results show that the effectiveness of a complete primary immunization schedule with CoronaVac in children 3–5 years of age against symptomatic COVID-19 during the Omicron outbreak was limited. However, vaccines were effective against severe disease in young children. These results support the vaccination of children 3–5 years of age to prevent severe illness and associated complications; however, they underscore the importance of maintaining layered protections against SARS-CoV-2 infection in this population. Important next steps include examining how long vaccine protection lasts and whether booster shots will be necessary. We hope that the results from this study inform policymakers in countries considering child vaccination against COVID-19.

Methods

Outcomes

The Ministry of Health in Chile requires that all suspected COVID-19 cases are immediately notified to health authorities through Epivigila, an online platform that centralizes all case notification and test results and represents the case count source used for this study. Suspected COVID-19 cases require laboratory testing with RT–PCR assay or antigen tests. We estimated the vaccine effectiveness of CoronaVac for children aged 3–5 years using three primary outcomes: laboratory-confirmed symptomatic SARS-CoV-2 infection (COVID-19), hospitalization and admission to the ICU associated with SARS-CoV-2 infection. We considered the time to the onset of symptoms from the beginning of the follow-up, 6 December 2021, as the endpoint of each outcome. We used the onset of symptoms as a proxy for the time of infection. We classified participants’ status into two categories along the study period: unvaccinated and fully immunized (≥14 days after receipt of the second dose with CoronaVac). A child was excluded from the unvaccinated group when she or he received the first vaccine dose. The period between the first dose administration and 13 days after the second dose was excluded from the at-risk person-time in our analyses.

Statistical analyses

We used Bonferroni-adjusted Pearson’s χ2 to compare descriptive data and control for multiple comparisons. To estimate hazard ratios, we used extensions of the Cox hazard model that allowed us to account for the time-varying vaccination status of participants21,22,23. We adjusted for differences in observed individual characteristics by inverse probability of treatment weighting as in marginal structural models24, estimating the weights non-parametrically25. Vaccine effectiveness was estimated based on the hazard ratio between the treated and non-treated status. We reported hazard ratio estimates adjusted for age, sex, region of residence, nationality, health insurance category (a proxy of household income) and underlying conditions (Extended Data Tables 1 and 2) under the standard and stratified versions of the Cox hazard model.

Let Ti be the time-to-event of interest, from 6 December 2021, for the i-th individual in the cohort, \(i = 1, \ldots ,n\). Let \(x_i,i = 1, \ldots ,n\) be a p-dimensional vector of individual-specific characteristics, such as age and sex, and let zi(t) be the time-dependent treatment indicator. The model assumes that the time-to-events are independent and with probability distribution given by

$$T_i|x_i,z_i\sim f\left( {t|x_i,z_i} \right),i = 1, \ldots ,n,$$

where

$$\begin{array}{l}f\left( {t|x_i,z_i} \right) = \lambda _0\left( t \right)exp\left\{ {x_i^\prime \gamma + \beta _{z_i(t)}} \right\} \times \\ exp\left\{ { - exp\left\{ { - x_i^\prime \gamma + \beta _{z_i(t)}} \right\}\mathop {\int}\limits_0^t {\lambda _0} \left( u \right)du} \right\},\end{array}$$

with \(\gamma \in {\Bbb R}^p\) being a vector of regression coefficients, \(\beta _k \in {\Bbb R}\) being the regression coefficient measuring the effectiveness of the kth vaccine and λ0 being the baseline hazard function

$$\lambda _0\left( t \right) = \mathop {{\lim }}\limits_{h \to 0} \left\{ {\frac{{P_0\left( {t \le T \le t + h|T \ge t} \right)}}{h}} \right\},$$

where P0 is the baseline probability distribution. A Cox model with time-dependent covariates compares the risk of the event of interest between immunized and non-immunized participants at each event time but re-evaluates which risk group each person belonged to, based on whether they had been immunized by that time.

We also fitted a stratified version of the model26, where the time-to-event distribution is given by

$$\begin{array}{l}f\left( {t|x_i,z_i} \right) = \lambda _{x_i,0}\left( t \right)exp\left\{ {\beta _{z_i(t)}} \right\} \times \\ exp\left\{ { - exp\left\{ {\beta _{z_i(t)}} \right\}\mathop {\int}\limits_0^t {\lambda _{x_i,0}} \left( u \right)du} \right\},\end{array}$$

with \(\beta _k \in {\Bbb R}\) being the regression coefficient measuring the effectiveness of the kth vaccine, and λx,0 is the predictor-specific baseline hazard function. We fitted a stratified version of the extended Cox proportional hazard model to test the robustness of our estimates to model assumptions. Under the stratified Cox model, each combination of predictors has a specific hazard function that can evolve independently.

We estimated the vaccine effectiveness as \(100\% \cdot \left( {1 - exp\left\{ {\beta _k} \right\}} \right)\). We show the adjusted vaccine effectiveness results, including covariates as controls (age, gender, region, nationality, health insurance category and comorbidities). We show the results for the standard and stratified versions of the Cox hazard model using inverse probability of treatment weighting. We computed standard 95% Wald CIs for the estimates. Inference was based on a partial likelihood approach27. Recall that the effectiveness estimate for the COVID-19 vaccines in the Cox model with time-dependent vaccination status compares the risk of an event for children who received the vaccine and those who were unvaccinated at each event time. The risk group is determined by whether the child had received the vaccine shot or not in a specific calendar time, and the comparison of the risk of an event is made at the same calendar time. Each term in the partial likelihood of the effectiveness regression coefficient corresponds to the conditional probability of an individual to express the outcome of interest from the risk set at a given calendar time.

Under the standard Cox model, all individuals at risk are included in the risk set, and their contribution is weighted based on their covariates (as shown in Extended Data Table 1). Under the stratified version of the Cox model, each stratum has a different risk set determined by the covariates.

We conducted the analysis with the survival package28 of R, version 4.0.5 (ref. 29).

Ethics statement

The research protocol was approved by the Comité Ético Científico Clínica Alemana Universidad del Desarrollo (Santiago, Chile). No human health risks as a result of our study were identified because we analyzed administrative datasets. The study was considered exempt from informed consent.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.