Study design and period

This retrospective observational study investigates the COVID-19 reinfection rate in Espírito Santo, Brazil, initially identifying suspected cases based on clinical and epidemiological criteria, subsequently confirmed through laboratory testing (PCR or rapid antigen tests). The study focused exclusively on laboratory-confirmed cases for the final analysis. This approach ensured diagnostic accuracy while maintaining the comprehensive surveillance approach of the e-SUS system, from September 2020 to February 2023. The analysis focuses on different age groups and biological sex (male/female as recorded in official health documentation – Notification Form); cases marked as ‘I’ (ignored) were retained in descriptive totals but excluded from subgroup-specific analyses. We also explored reinfection patterns across different waves of the pandemic, to identify key determinants. Temporal analysis was conducted to evaluate these factors over time, providing insights into reinfection trends throughout the pandemic period.

Data sources

Two primary databases were used: (1) the e-SUS Health Surveillance system, a comprehensive state-level database that tracks both mild to moderate and severe COVID-19 cases, and (2) the Vacina e Confia immunization system, a state-specific database for monitoring vaccination data. SIVEP-Gripe, a subset of e-SUS, exclusively tracks severe acute respiratory syndrome (SARS) cases and deaths, while e-SUS captures a broader spectrum of cases, making it more suitable for this study. By encompassing both mild/moderate and severe cases, e-SUS VS enables detection of reinfections occurring in outpatient and community settings, ensuring greater representativeness across demographic and geographic strata. Due to limited testing availability during the first three months of the pandemic, notification data from this period were excluded from the primary analysis. In addition to testing constraints, this early period coincided with ongoing modifications to the national surveillance infrastructure to incorporate COVID-19 as a notifiable disease, alongside training of healthcare staff for consistent data entry. These factors contributed to incomplete and heterogeneous data quality. Exclusion of this window follows established epidemiological practice to minimize detection bias during periods of limited surveillance capacity [31,32]. Although we cannot fully rule out the misclassification of early infections, the 90-day minimum interval requirement and the operationalization of systematic testing and contact tracing from September 2020 onward [14] reduce this risk [33]. The analysis begins in September 2020, when mandatory testing for both symptomatic and asymptomatic contacts was introduced [34].

With the end of the global public health emergency and the gradual introduction of pharmacy-based tests and self-tests, there was a significant decrease in the reporting of suspected COVID-19 cases. Mandatory testing for symptomatic and asymptomatic contacts, in place from September 2020, was progressively relaxed starting in mid-2022. From this period onward, case detection increasingly depended on individual initiative, as self-testing became common and positive results were no longer required to be reported. For this reason, the study focuses on the period from September 2020 to February 2023, ensuring greater accuracy and completeness of the information. While COVID-19 cases continued to be reported after March 2023, substantial underreporting occurred due to the transition to endemic surveillance and increased reliance on self-testing. Quantitative bias analysis was not feasible due to lack of reliable parameters to estimate the extent of underreporting in our specific population during this period. This temporal restriction is consistent with similar surveillance studies that focus on periods of systematic case detection [13,35].

Immunization data, which is recorded through the Vacina e Confia system, spans from January 2021 to February 2023 and tracks vaccination coverage throughout the state. Although data is available beyond February 2023, it was only used up to this point to align with the period covered by the notification data.

The Espírito Santo State Health Department operates two dedicated digital platforms: e-SUS Health Surveillance (VS) for mandatory case notifications and Vacina e Confia for immunization records. e-SUS VS was established to replace the SINAN model and was developed in partnership with the Pan American Health Organization (PAHO) to enable centralized, real-time reporting of diseases, conditions, and health events by both public and private providers across all 78 municipalities. Following statewide training of surveillance and primary care professionals in late 2019, the state health department assumed responsibility for consolidating municipal data, updating reporting instruments, providing technical support, and transmitting notifications daily to the Ministry of Health’s Health Surveillance Secretariat (SVS/MS), in line with national reporting deadlines.

Complementing the notification system, Vacina e Confia is Espírito Santo’s official immunization registry for all vaccines administered under the SUS. Developed by the State Department of Health (Sesa), the platform interacts daily with the National Health Data Network (RNDS), the National Immunization Program (PNI) database, and the National Health Registry. Its key features include digital card access via Citizen Access authentication, detailed records of vaccine administration routes and sites, ICD-10 coding for special risk groups, a real-time “Situation Room” dashboard for monitoring coverage and inventory, and centralized management of strategic immunobiologicals across the state’s 78 municipalities.

Importantly, both platforms are independently managed by the state and remained fully operational during the December 2021 cyberattack that disrupted federal health information systems. This ensured uninterrupted access to Espírito Santo’s notification and vaccination records during a period of major national data unavailability.

The databases used for this study do not include unique identifiers consistently across all records. As a result, it was necessary to perform record linkage using nominal information, such as name, date of birth, and other variables.

Data linkage

The record linkage process was conducted in three distinct stages. In the first stage, applied to the notification database, the objective was to identify records belonging to the same individual, perform deduplication, and define initial infections and reinfections, as well as identify the laboratory tests confirming these events. The second stage, applied within the vaccination database (Vacina e Confia), also involved deduplication to match records to the same individuals and determine how many doses each patient had received and when. The third stage involved merging both databases, integrating the notification and vaccination data. Although the e-SUS database contains fields for recording COVID-19 vaccinations, the data are incomplete, which made it necessary to perform linkage between the notification and vaccination databases.

The datasets were pre-processed for deterministic record linkage, which involved standardizing identification variables. This process included removing special characters, normalizing letter case, managing abbreviations, and cleaning address fields considered non-informative or inconsistent. Corrections were made to patient and mother names, as well as to address information, by fixing common typographical errors, expanding abbreviations with full terms, and addressing invalid or missing values.

Additionally, derived variables were created to aid the linkage process, such as Soundex codes for patient names and addresses. The deterministic linkage process involved applying sequential rules using sets of identification variables to identify duplicate or corresponding records in the notification and vaccination databases. This method, similar to those described in validation studies such as Oliveira et al. [36] and Pacheco et al. [37], was crucial for ensuring data quality and consistency, preventing distortions in the results of the analysis. These rules were applied step-by-step, with each stage based on a specific set of variables, such as CPF (Brazilian individual taxpayer registry), CNS (SUS health card number), patient name, date of birth, and address components. Records containing invalid or missing values were systematically reviewed and cleaned before linkage. This included exclusion of empty or placeholder values, correction of inconsistencies in dates (e.g., inverted digits in birth or vaccination dates), removal of implausible ages (e.g., negative or >125 years), and manual review of cases with conflicting information. No imputation techniques were used; data integrity was preserved by relying on logical corrections and internal consistency checks.

Linkage quality assessment

The deterministic linkage process between the e-SUS VS (COVID-19 case surveillance) and Vacina e Confia (vaccination) databases was conducted iteratively and incrementally, using multiple combinatorial rules involving identifying variables such as name, mother’s name, date of birth, CPF, CNS, sex, and address [38,39]. The process sought to maximize the accuracy of the matching, even in the face of heterogeneous and incomplete data.

Case Definitions

Throughout the study, several case definitions were used to classify individuals:

• Without Test Result (PCR and TRa): Patients without any recorded PCR (Polymerase Chain Reaction) or TRa (Rapid Antigen) test result during the study period. This category includes individuals who were not formally tested for COVID-19, with their condition identified without laboratory confirmation.
• With Test Result (PCR or TRa): Patients with at least one recorded result from a PCR or TRa test during the study period, regardless of whether the result was positive or negative.
• Not Infected: Patients with no positive test results throughout the study period. This definition is based on available surveillance data and does not imply regular or systematic testing. The category includes individuals with negative test results as well as those tested due to clinical suspicion who resulted negative.
• Infected: Patients with one or more positive test results (PCR or TRa) during the study period. This definition captures laboratory-confirmed infections regardless of symptoms. While systematic testing for asymptomatic individuals was not universally implemented, Espírito Santo’s surveillance protocol included testing of symptomatic and asymptomatic close contacts, healthcare workers, and individuals in outbreak settings [41].
• Reinfected: Patients with multiple positive test results (PCR and/or TRa) separated by at least 90 days, indicating a potential reinfection.
  • Reinfection Type 1: Patients with two or more positive PCR tests separated by at least 90 days, with at least one intervening negative PCR result.
  • Reinfection Type 2: Patients with two consecutive positive PCR tests separated by at least 90 days, without an intervening negative result.
  • Reinfection Type 3: Patients with two or more positive results (PCR or TRa) separated by at least 90 days, with at least one intervening negative test (PCR or TRa).
  • Reinfection Type 4: Patients with two consecutive positive results (PCR or TRa) separated by at least 90 days, without requiring an intervening negative test.
• Not Reinfected: Patients who did not meet the criteria for reinfection, either due to the time interval between positive tests being less than 90 days or the absence of an intervening negative test where applicable.

Pandemic waves

To provide a clearer understanding of the epidemiological context, the study identified specific time periods representing distinct pandemic waves. These waves were determined by the researchers based on temporal patterns observed in notification data, accounting for inherent delays in the infection-to-notification process.

Wave identification considered several temporal factors: (1) the median incubation period of SARS-CoV-2 (5.1 days, range 2–14 days) [42]; (2) testing delays, which varied from same-day to several days depending on healthcare system capacity; (3) notification delays from laboratories to surveillance systems; and (4) potential population heterogeneity in testing-seeking behavior.

To account for these delays, wave boundaries were defined using a 14-day moving average of case notifications to smooth short-term fluctuations, and a lag adjustment of 7–14 days was applied to align notification peaks with estimated infection peaks (diagnosis date) [43,44]. Waves were identified as intervals between successive epidemiological nadirs, with each wave characterized by a distinct rising and falling trend in smoothed case counts.

This methodological approach aligns with established epidemiological practices for wave identification during respiratory virus pandemics and provides a robust framework for contextualizing reinfection patterns throughout the study period [44,45].

Vaccination status

To determine the vaccination status, records from the “Vacina e Confia” system were utilized. Following deduplication and deterministic linkage with the main database, the vaccination doses associated with each patient along the time were identified. For the analysis, the highest dose administered up to 15 days prior to the date of the first infection or reinfection was considered. The 15-day window was chosen to account for the time required for the immune system to generate or boost an effective immune response following vaccination.

Special residential settings

Patient residence information was also analyzed, where separate variables were created to indicate if the patient resided in a prison, long-term care facility, or indigenous village during their first infection.

Statistical analysis

Descriptive techniques were employed to analyze the frequencies and proportions of reinfection cases. The reinfection rate was calculated by dividing the number of reinfection cases by the total number of infected cases, then multiplying by 100 to obtain a percentage. This calculation was performed for each of the four reinfection definitions. Analyses were stratified by age group, gender, and across different pandemic waves to explore potential patterns and determinants of reinfection. The analysis was conducted using Stata, version 17.

Ethical considerations

The study adhered to ethical and regulatory guidelines, ensuring patient privacy and confidentiality. After the record linkage steps, all data were anonymized, and results were presented in an aggregated manner without individual identification. The study was authorized by the Instituto Capixaba de Ensino, Pesquisa e Inovação em Saúde do Espírito Santo and approved by the Ethics Committee of the Instituto de Ensino e Pesquisa do Hospital Sírio-Libanês (CAAE Nº 45986821.9.0000.5461, December 2021), with a waiver for informed consent. In adherence to the STROBE guidelines for observational epidemiological studies, this manuscript addresses the 22-item checklist covering title, abstract, introduction, methods, results, and discussion sections, ensuring transparent and complete reporting.

Definitions and their implications

This study applied four hierarchical definitions of SARS-CoV-2 reinfection, requiring ≥90-day intervals between positive tests, with stricter criteria also demanding an intermediate negative result. Reinfection estimates ranged from 0.27% to 9.72% depending on the definition, underscoring how methodological choices directly affect observed rates. This variation mirrors findings by Chemaitelly et al. [4], who showed that stricter definitions, such as the CDC’s ≥90-day threshold, may underestimate reinfection. Some agencies, like the ECDC, use ≥ 60 days instead [7].

Long intervals may miss reinfections occurring within shorter windows, especially with more transmissible or immune-evasive variants. Conversely, short intervals risk capturing prolonged viral shedding as reinfection. Yamasaki and Moi [13] argue for combining clinical, temporal, and genomic criteria—where available—for more accurate classification.

Early in the pandemic, reinfection case definitions relied on genomic confirmation. However, as global case numbers surged, routine sequencing became impractical, prompting surveillance systems worldwide to adopt time-based criteria. In this context, our study employs a hierarchical classification system that, although not aligned with formal standards, offers a structured framework to assess how varying definitional stringency influences reinfection detection.

While the reinfection types defined in our study are not mutually exclusive—a recognized limitation—Type 4 serves as the primary outcome measure, capturing the broadest range of possible reinfections. Types 1–3, in turn, act as sensitivity analyses, reflecting progressively stricter criteria that enhance specificity at the expense of sensitivity. This strategy provides insight into how conservative versus inclusive definitions can shape reinfection estimates, enabling more meaningful comparisons across studies employing different methodologies.

Our findings support the growing consensus in the literature that reinfection rates must be interpreted in light of the definitions applied [13,46], especially during periods of high transmission or the emergence of new variants. Moving forward, surveillance systems should adopt a single, standardized case definition tailored to local testing capacity and public health goals, while relying on sensitivity analyses to evaluate robustness—rather than presenting multiple overlapping categories as equivalent outcome measures.

While early in the pandemic the definition of a reinfection required genomic confirmation, the scale of global infections quickly made sequencing infeasible. Time-based definitions were then adopted by surveillance systems worldwide. By applying multiple definitions in parallel, our study offers a comparative perspective and shows that the more inclusive definitions yield reinfection rates consistent with national data, such as ~8% reported by Fonseca et al. [47]. In contrast, strict definitions captured only a small fraction of events (~0.3%).

These findings reinforce calls in the literature to interpret reinfection estimates in light of the definitions used, especially during periods of intense viral circulation or variant emergence.

Gender distribution

Reinfection rates in Espírito Santo were higher among women across all methodological definitions. Women accounted for 527,250 (56.2%) of initial infections but represented a disproportionate majority of reinfection cases across all definitions, ranging from 61.6% to 66.3%. Under the broadest definition (Type 4), 57,939 of 527,250 infected women experienced reinfection (10.99%), compared to 33,244 of 410,855 infected men (8.09%). This pattern of female predominance in reinfections aligns with findings from the UK, where women represented 67% of reinfection cases despite accounting for only 53% of initial infections [7], and with a systematic review that identified higher reinfection risk in women [48].

This may reflect greater occupational exposure, as women constitute a large share of the health and education workforce—sectors with high transmission risk [3,17]. Biological factors have also been proposed: although men often have more severe primary infections, women may mount a stronger short-term immune response that wanes more quickly, increasing susceptibility to reinfection over time [28,49].

Additionally, health-seeking behavior differences may have contributed: women are more likely to seek medical care and adhere to testing recommendations, particularly after the relaxation of mandatory testing. This behavioral factor may have amplified reinfection detection among women in Espírito Santo. However, not all studies show gender differences—analyses from Qatar and Israel found similar reinfection rates between sexes [26,50]. In Brazil, both our findings and those of Fonseca et al. [47] suggest a modest female predominance, potentially shaped by exposure patterns, health-seeking behavior, and reporting practices. These results highlight the value of tailoring preventive strategies to occupational and social contexts, particularly in predominantly female environments like schools and health facilities.

Age distribution

Reinfection age-stratified risks followed a non-linear pattern, with adults aged 30–49 showing the highest reinfection rates (≈11–12% under the broadest definition), while children and older adults demonstrated lower proportions. This distribution pattern reflects complex interactions between exposure risk, vaccination coverage, and demographic behaviors.

Analysis of vaccination status by age stratification revealed a pronounced gradient across age groups. Older adults (60 + years) exhibited significantly higher proportions of multiple booster doses among reinfection cases (≈51% under the broadest definition) compared to all younger age groups, where proportions ranged from 4.0% in 10–19 years to 24.3% in 50–59 years (Supplementary S1 Table). This gradient likely reflects vaccination policy prioritization of elderly populations, age-related immunosenescence requiring additional doses, and greater adherence to booster campaigns among older adults [23].

Notably, cases with only first-dose vaccination remained consistently low across all age groups (4–7%), which may reflect the temporal progression of vaccination coverage during the study period, with most individuals having received additional doses by the time reinfections occurred. These findings align with international evidence suggesting that working-age adults face greater exposure due to occupational and social mobility patterns.

Some studies, like Mensah et al. [7], reported older age among reinfection cases—possibly reflecting delayed exposure or survival through earlier waves. In Espírito Santo, our findings may reflect both higher occupational and social exposure among working-age adults, as well as a survivorship bias, in which individuals who survived earlier waves remained in the population at risk for reinfection during subsequent waves. This survivorship effect is particularly relevant in longitudinal cohort studies, where the population available for follow-up consists primarily of those who survived previous infection episodes, potentially creating a selection bias toward individuals with certain immune or behavioral characteristics.

Although immunosenescence in the elderly may theoretically increase susceptibility to reinfection, greater vaccine coverage and precautionary behaviors in this population likely mitigated this risk [51] in our study. Conversely, children demonstrated lower reinfection rates, possibly due to reduced testing frequency and different immune response patterns compared to adults.

Our results demonstrate consistency with Latin American epidemiological data [25] and cohort studies among essential workers [3], reinforcing the critical need to prioritize booster campaigns and preventive measures for working-age adults while maintaining robust protection strategies for age extremes.

Regional distribution

Marked geographic differences were observed. While Greater Vitória registered the highest number of reinfections in absolute terms—consistent with its large population and intense transmission—the Central region of the state showed one of the highest proportional reinfection rates (~10–11%), suggesting that smaller municipalities also experienced multiple waves affecting the same individuals.

Reinfection rates were especially elevated in certain vulnerable communities, such as Indigenous villages, where they reached 16%. Similar patterns have been linked to limited isolation capacity and lower pre-existing immunity in remote populations [20].

High reinfection rates in populous areas reflect repeated epidemic waves, as seen in national data showing that the most populated states of São Paulo, Minas Gerais, and Rio de Janeiro concentrating most reinfections [47]. At the same time, localized surges in interior areas may result from household clusters or mass gatherings.

Although international literature rarely presents subnational detail, studies in South Africa and Mexico also found reinfection rates varied by region, reflecting differences in variant spread and wave intensity [23,25].

By highlighting these disparities within a single state, our findings emphasize the need for region-specific policies—such as enhanced surveillance in rural hotspots and tailored vaccination in underprotected areas, including Indigenous communities.

Pandemic waves and viral lineages

Our findings confirm that reinfections in Espírito Santo peaked during the second and fourth waves—periods dominated by the Gamma and Omicron variants, respectively. The second wave saw the introduction of Gamma, which partially evaded immunity from prior infections [51,52]. The fourth wave, driven by Omicron, marked an unprecedented spike in reinfections, with over 11% of cases involving individuals previously infected.

This pattern mirrors global data. In South Africa, Pulliam et al. [23] were among the first to report a sharp increase in reinfection risk during the Omicron wave. A meta-analysis by Chen et al. [30] estimated reinfection incidence of 3–5% during Omicron’s spread—far higher than during Beta or Delta waves.

In our data, reinfections remained rare in 2020, rose modestly during Gamma’s emergence, and surged during Omicron. Notably, Delta did not produce a marked reinfection increase, likely due to more limited immune escape compared to Gamma and Omicron [50,53]. Also, by late 2021, over a year had passed since initial infections, and waning immunity may have left many susceptible even without major antigenic shifts [29].

These temporal patterns suggest that reinfections may be influenced by both viral evolution and potentially waning immunity, though our observational data cannot definitively establish causal relationships between these factors. Local trends in Espírito Santo closely tracked lineage replacement, reinforcing the value of real-time genomic surveillance to anticipate reinfection waves and guide timely booster campaigns.

Vaccination status

Reinfections in Espírito Santo occurred across all vaccination groups, including individuals with booster doses. Among the vaccinated, those with only the primary two-dose series appeared to have higher reinfection rates than even some partially or unvaccinated groups. However, this pattern should not be interpreted as evidence of vaccine failure, as our study design does not allow for direct assessment of vaccine efficacy —rather, it reflects important timing and exposure biases.

By the time the major reinfection wave (Omicron) struck, a large share of the population had already been vaccinated. Thus, most reinfections occurred in people who had received at least two doses, simply because they constituted the majority. Additionally, individuals prioritized for early vaccination—such as healthcare workers and older adults—were also those at higher risk of repeated viral exposure or reduced immune responses due to age. These groups were more likely to face reinfection despite being vaccinated, not because of vaccine inefficacy, but due to their heightened vulnerability.

Furthermore, the reinfection risk increased as vaccine-induced immunity waned. Many individuals infected during Omicron had completed their primary series several months earlier and had not yet received a booster. International studies confirm that protection against infection declines over time, particularly without updated boosters, even as protection against severe disease remains high [4,10].

Brazil’s early immunization campaign also relied heavily on traditional platforms such as CoronaVac, which, while essential for coverage expansion, showed lower and less well-characterized effectiveness against immune-evasive variants compared to mRNA vaccine [54]. This context may have further influenced reinfection dynamics, especially in the absence of timely boosters.

Even so, global evidence supports the effectiveness of hybrid immunity and booster doses in reducing reinfection risk [12,25]. Our data suggest that the reinfections observed—particularly during the Omicron wave—were driven by a convergence of waning immunity, viral evolution, and sustained exposure rather than indicating reduced vaccine performance, though our study design does not permit direct evaluation of vaccine effectiveness.

These findings reinforce the critical importance of timely booster campaigns, especially in high-risk and high-exposure populations, and the need for continued genomic and epidemiologic surveillance to anticipate new waves of transmission.

Reinfection rates in specific settings

This study identified elevated reinfection rates in certain closed and vulnerable settings. In long-term care facilities, about 14% of residents experienced reinfection—well above the population average—likely due to advanced age, weaker immune responses, and prolonged exposure [55].

Among Indigenous communities in Espírito Santo, reinfections reached 16.5%, the highest rate observed. Successive outbreaks in these communities may reflect socio-environmental factors such as communal housing, limited isolation capacity, and delayed early exposure to the virus [20].

In contrast, reinfection rates among incarcerated individuals did not appear elevated in our data. However, this finding must be interpreted with caution. Due to data limitations, we could not distinguish between different types of incarceration facilities—such as short-term jails versus long-term prisons—which may differ in exposure and outbreak control. Moreover, underreporting is possible due to barriers to testing access, hospital transfers, and the frequent use of non-carceral addresses in notification records. These factors may have led to misclassification or underestimation of reinfection rates in this population.

While our study did not present specific reinfection estimates for healthcare workers, existing literature documents high incidence in this group, especially during the Omicron wave [3,17].

Overall, the findings highlight the need for tailored interventions in institutional and collective settings—including vaccination, testing access, and early outbreak response—particularly among long-term care residents and Indigenous populations. For correctional facilities, more granular data would be necessary to clarify reinfection risks and guide specific protective strategies.

Comparison with other reinfection studies

Our findings align with global and national evidence showing that SARS-CoV-2 reinfections became more frequent over time, particularly during the Omicron period. Early studies, such as Hall et al. [46] in the UK and Abu-Raddad et al. [50] in Qatar, reported low reinfection rates (<1%) during the pre-Omicron phase—consistent with our strictest criterion (0.27%).

Later studies showed increasing reinfection rates. Chen et al. [30] estimated a global prevalence of 3–5% during Omicron. Our broader criterion (~9.7%) may appear higher, but it includes antigen tests and does not require a negative intermediate test—unlike stricter definitions used in some international studies [13,56]. When aligned methodologically, our estimates converge with international findings.

In Brazil, Fonseca et al. [47] found a reinfection rate of ~3.6% (≥60-day interval) by mid-2022, rising to ~8% during Omicron. These figures closely match our data from Espírito Santo, despite different sources (notifications vs. private labs), underscoring consistency. Both studies reported that over 90% of reinfections occurred after late 2021.

Estimates from high-risk groups also align: Cegolon et al. [3] reported ~13% reinfection in Italian healthcare workers, and Michael et al. [55] observed ~18% among residents in long-term care facilities. Our 10% overall rate is consistent with this range, considering population risk heterogeneity.

Importantly, some international studies count multiple reinfections per person; our analysis considered only the first. Nonetheless, multiple reinfections remained rare in our dataset, reinforcing global findings.

In terms of risk factors, our results corroborate prior studies highlighting exposure (e.g., healthcare, education), lack of boosters, and variant emergence as key drivers [12,57]. Additionally, our study offers novel insights by detailing subnational and vulnerable group patterns—particularly in Indigenous communities, which are often absent from international analyses.