Introduction

The protection conferred by the ancestral-strain mRNA vaccines, BNT162b21 (Pfizer-BioNTech) and mRNA-12732 (Moderna), against severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection and severe coronavirus disease 2019 (COVID-19) has been well established through randomized clinical trials1,2 and real-world observational studies3,4,5,6. However, waning immunity following the primary vaccination series7,8,9,10, along with the emergence of immune-evasive variants11,12,13,14,15, has necessitated repeated booster doses to sustain protection5,14. Understanding the durability of protection conferred by booster doses is essential for guiding vaccine development and public health strategies.

This study aimed to contribute to the existing body of evidence by evaluating the effectiveness of a fourth ancestral-strain mRNA vaccine booster in preventing SARS-CoV-2 infection and severe (acute-care hospitalization)16, critical (intensive-care-unit hospitalization)16, or fatal17 COVID-19, compared to the protection provided by three booster doses. A matched cohort design, emulating a target trial18,19, was used to assess these outcomes. The study also examined the durability of protection over time following the fourth dose.

Although these vaccines are no longer in use due to viral evolution—particularly following the emergence of the omicron variant and the availability of updated vaccines better matched to circulating strains—this study was conducted to help complete the evidence base on the real-world effectiveness of first-generation vaccines, thereby contributing to efforts aimed at guiding the development and deployment of more effective future vaccines. The study provides insights from one of the most extensively studied national populations for COVID-19 vaccine effectiveness3,4,5,8,20,21,22,23.

Methods

Study population, data sources, and vaccination

This study was conducted on Qatar’s population, analyzing data from February 10, 2021, the earliest recorded third-dose vaccination, to May 13, 2024, the study’s conclusion. The analysis utilized national, federated databases that include COVID-19 testing, vaccination, hospitalization, and mortality data, sourced from Qatar’s integrated nationwide digital health platform (Supplementary Section S1). These databases cover comprehensive SARS-CoV-2-related data with no missing information since the onset of the pandemic, including all polymerase chain reaction (PCR) tests performed, regardless of location or facility, as well as rapid antigen tests conducted at healthcare facilities from January 5, 2022, onward (Supplementary Section S2). By national policy, all COVID-19 data were captured using a centralized, advanced digital health platform, ensuring that no healthcare encounters were missed across any facility.

Until October 31, 2022, Qatar implemented an extensive testing strategy, with 5% of the population being tested weekly, primarily for routine purposes such as screening or travel-related requirements8,20. After November 1, 2022, this testing rate was scaled down to less than 1% of the population per week24. The majority of infections in Qatar were detected through routine testing rather than symptomatic presentation (Supplementary Section S1)8,20.

Qatar’s mass vaccination campaign began in December of 2020, with the BNT162b2 vaccine, followed by the introduction of mRNA-1273 three months later25 (Supplementary Section S1). Vaccines were offered free of charge to all individuals and administered solely through the public healthcare system25, with prioritization based on risk factors, including frontline healthcare workers, individuals with chronic conditions, and those aged 50 years or older8.

Immunization with both mRNA vaccines followed the protocol approved by the United States Food and Drug Administration (FDA)1,2, with high adherence across the study population8,9. Nearly all individuals were vaccinated according to the approved schedule, or with only minor deviations from it.

Demographic data was retrieved from the national health registry. Qatar’s demographic structure is unique, with only 9% of the population aged 50 or older and 89% consisting of expatriates from over 150 countries26. Further details on Qatar’s population and COVID-19 databases have been previously published5,8,15,20,21,26,27,28.

Study design

This study assessed the incidence of SARS-CoV-2 infection and severe16, critical16, or fatal17 COVID-19 among individuals who received a fourth dose of the BNT162b2 or mRNA-1273 vaccines compared to those who received only three doses. A matched retrospective cohort design, emulating a randomized controlled trial (target trial design)18,19 and informed by our earlier cohort studies on Qatar’s population5,21,22,23,29,30,31,32,33,34, was used to evaluate vaccine effectiveness in preventing SARS-CoV-2 infection and severe COVID-19.

Incidence of infection was defined as the first PCR-positive or rapid antigen-positive test after the start of follow-up, regardless of symptoms. The classification of infection severity followed the World Health Organization (WHO) guidelines for COVID-19 case severity16, criticality16, and fatality17 (Supplementary Section S3). Infection severity assessments were conducted by trained medical personnel, independent of the study investigators, through individual chart reviews35.

Following national protocol, every individual with a SARS-CoV-2-positive test and a concurrent COVID-19 hospital admission underwent infection severity assessments every three days until discharge or death, regardless of the duration of hospital stay35. Individuals whose infections progressed to severe, critical, or fatal COVID-19 were classified according to their worst recorded outcome, with priority given to COVID-19 death17, followed by critical disease16, and then severe disease16 (Supplementary Section S3).

Cohorts’ eligibility and matching

Eligibility for inclusion required individuals to have received three doses of the same mRNA vaccine for the three-dose cohort or four doses of the same vaccine for the four-dose cohort. Individuals who received other types of vaccines or mixed vaccines were excluded. Individuals were excluded if they had a documented SARS-CoV-2-positive test within 90 days prior to the start of follow-up.

Individuals in the four-dose cohort were matched exactly one-to-one with those in the three-dose cohort based on sex, 10-year age group, nationality, number of coexisting conditions (ranging from 0 to ≥ 6; Supplementary Section S4), vaccine type (BNT162b2 or mRNA-1273), prior infection status (categorized as no prior infection, prior pre-omicron infection, prior omicron infection, or both prior pre-omicron and omicron infections), and the calendar week of the third vaccine dose.

Exact matching here refers to the pairing of individuals in the four-dose and three-dose cohorts based on identical values of the matching factors, ensuring that each matched pair shared precisely the same characteristics. Infections before December 19, 2021, the onset of Qatar’s first omicron wave36, were classified as pre-omicron, while those on or after this date were considered omicron infections. Individuals with documented infections both before and after this threshold were categorized as having experienced both pre-omicron and omicron infections.

Matching was implemented using an iterative algorithm that selected individuals from the reference cohort (vaccinated with three doses), ensuring that at the start of follow-up, they were alive, had the same prior infection status as their match, and had no documented SARS-CoV-2 infection within the previous 90 days. The matching algorithm was implemented using the ccmatch command in Stata, supplemented with conditions to retain only controls that met these eligibility criteria. It was iterated using loops with as many replications as needed until exhaustion (i.e., no more matched pairs could be identified).

In this study design, individuals in the matched three-dose cohort contributed follow-up time before receiving a fourth dose, meaning they were initially analyzed as part of the three-dose cohort. If they later received a fourth dose and were rematched, they subsequently contributed follow-up time as part of the four-dose cohort.

The matching strategy aimed to balance observable confounders that might affect risk of infection across exposure groups26,37,38,39,40. The selection of matching factors was informed by findings from earlier COVID-19 studies on Qatar’s population4,8,9,25,41.

Prior infection was defined as a SARS-CoV-2-positive test occurring at least 90 days before the start of follow-up. The 90-day threshold was established to prevent misclassification of prior prolonged infections as new infections42,43,44.

Cohorts’ follow-up

For each matched pair, follow-up began on the day the fourth vaccine dose was administered. Consequently, the start of follow-up for each individual in the three-dose cohort was determined by that of their matched counterpart in the four-dose cohort following the matching process.

To maintain exchangeability5,45, both members of each matched pair were censored at the earliest occurrence of receiving an additional vaccine dose. Therefore, individuals were followed until the first occurrence of one of the following events: a documented SARS-CoV-2 infection (regardless of symptoms), fourth-dose vaccination for individuals in the three-dose cohort (with matched-pair censoring), fifth-dose vaccination for those in the four-dose cohort (with matched-pair censoring), death, or the administrative end of follow-up on May 13, 2024.

Notably, this study design ensured that there was no delay in follow-up between cohorts, with both groups initiating follow-up at the same time. As a result, both were equally exposed to the same circulating viral variants, healthcare practices, testing policies and accessibility, public health measures, and epidemiological context throughout the follow-up period. This design also avoids survival bias, as both individuals in each matched pair were alive and eligible for follow-up at the start of the same calendar date.

Oversight

The institutional review boards of Hamad Medical Corporation and Weill Cornell Medicine–Qatar approved this retrospective study with a waiver of informed consent. All methods were performed in accordance with the relevant guidelines and regulations. The study was conducted and reported in accordance with the guidelines outlined in the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE; Supplementary Table S1).

Statistical analysis

The eligible and matched cohorts were described using frequency distributions and measures of central tendency, and comparisons were made using standardized mean differences (SMDs). An SMD of ≤ 0.1 indicated adequate matching46.

The cumulative incidence of infection (or severe, critical, or fatal COVID-19), defined as the proportion of individuals at risk whose primary endpoint during follow-up was an infection, was estimated using the Kaplan-Meier estimator method.

The incidence rate of infection (or severe COVID-19 outcomes) in each cohort, defined as the number of identified infections divided by the number of person-weeks contributed by all individuals in the cohort, was estimated along with the corresponding 95% confidence interval (CI) using a Poisson log-likelihood regression model with the Stata 18.0 stptime command.

Overall adjusted hazard ratios (AHRs) with corresponding 95% CIs were estimated using Cox regression models to compare the incidence of infection between cohorts. Analyses were conducted in Stata using the stcox command, with adjustments for matching factors to ensure precise and unbiased variance estimation47. The adjustment was performed by including the matching factors (e.g., sex, nationality, and calendar week of the third vaccine dose) as covariates in the model.

The overall hazard ratios were further adjusted for testing rate to account for potential variations across cohorts. Individuals were grouped into three categories based on their testing frequency during the study period: low testers (0 to 1 test per person-year), intermediate testers (more than 1 to 2 tests per person-year), and high testers (more than 2 tests per person-year). This additional adjustment was necessary because most SARS-CoV-2 tests in Qatar were conducted routinely rather than in response to symptoms8,20. Differences in testing rates could lead to variations in infection detection between cohorts if routine testing practices differed across groups.

A sensitivity analysis was also performed in which, instead of adjusting for testing behavior during follow-up, adjustment was based on testing frequency prior to baseline. Specifically, testing frequency was measured over the 360 days preceding the start of follow-up and included as a baseline covariate. Based on testing frequency during this pre-baseline period, individuals were classified as low, intermediate, or high testers. This approach avoids conditioning on a post-treatment variable while still accounting for baseline differences in testing propensity and healthcare-seeking behavior.

A subgroup analysis was conducted to assess AHRs for infection at 3-month intervals from the start of follow-up. Separate Cox regression models were fitted for each interval, restricting “failure” (infection) to occurrences within the specified time frame. The analysis included only individuals who remained at risk at the beginning of each interval and had not been previously censored.

Relative vaccine effectiveness (rVE), along with the associated 95% CI, was derived from the aHR as 1 - aHR if the aHR was < 1, and as 1/aHR − 1 if the aHR was ≥121,48. This approach ensured a symmetric scale for both negative and positive effectiveness, ranging from − 100 to 100%, allowing for a meaningful interpretation of effectiveness regardless of whether the value was positive or negative.

Notably, the proportional hazards (PH) assumption was not satisfied in the context of the Cox regression models, as confirmed by Schoenfeld residuals and log-log survival plots. However, this assumption is inherently violated in the presence of waning immunity and is rarely upheld for any non-null effect49. As such, the AHRs derived in this study should be interpreted as weighted averages of time-varying AHRs49, with the corresponding vaccine effectiveness estimates reflecting average effectiveness over the follow-up period21. Moreover, AHRs were estimated for shorter time intervals following vaccination—intervals during which the PH assumption is less violated and more likely to hold—and vaccine effectiveness was accordingly reported by time since vaccination.

Statistical analyses were performed using Stata/SE version 18.0 (Stata Corporation, College Station, TX, USA).

Results

Figure 1 illustrates the study population selection process, while Table 1 outlines the characteristics of both the eligible and matched cohorts. Since the study was conducted on Qatar’s population, it reflects the country’s internationally diverse and predominantly male demographic.

Fig. 1
Fig. 1The alternative text for this image may have been generated using AI.
Full size image

Cohort selection for investigating the effectiveness of four-dose booster vaccination versus three-dose booster vaccination against SARS-CoV-2 infection and against severe forms of COVID-19. COVID-19 denotes coronavirus disease 2019 and SARS-CoV-2 severe acute respiratory syndrome coronavirus 2. *This cohort includes everyone with at least three mRNA vaccine doses, therefore it also includes those who eventually received a fourth dose some time after the third dose. †Cohorts were matched exactly one-to-one by sex, 10-year age group, nationality, number of coexisting conditions, vaccine type, prior infection status, and calendar week of the third vaccine dose. ‡These persons became part of the four-dose cohort (n = 9,673) after receiving Dose 4.

Table 1 Baseline characteristics of the eligible and matched cohorts.
Full size table

Between February 10, 2021, and May 13, 2024, a total of 689,088 individuals received at least three doses of the ancestral-strain mRNA vaccine, of whom 9,673 received also a fourth dose. The median dates for the administration of the first, second, third, and fourth doses were April 10, 2021, May 7, 2021, January 19, 2022, and August 25, 2022, respectively.

The median time between the first and second doses was 22 days (interquartile range [IQR], 21–28 days). The median time between the second and third doses was 254 days (IQR, 229–287 days), and the median time between the third and fourth doses was 259 days (IQR, 202–322 days).

Each of the matched cohorts consisted of 7,463 individuals. The median follow-up time was 606 days (IQR, 533–693 days) for the four-dose cohort and 607 days (IQR, 538–692 days) for the three-dose cohort (Fig. 2). Nearly the entire follow-up period occurred during the circulation of different omicron subvariants in the population20,24,50,51,52.

Fig. 2
Fig. 2The alternative text for this image may have been generated using AI.
Full size image

Cumulative incidence of SARS-CoV-2 infection in the matched four-dose and three-dose vaccination cohorts. IQR denotes interquartile range and SARS-CoV-2 severe acute respiratory syndrome coronavirus 2.

In the four-dose cohort, 652 infections were recorded after the start of follow-up, none of which progressed to severe, critical, or fatal COVID-19 (Fig. 1; Table 2). In the three-dose cohort, 582 infections were recorded, with only one case progressing to severe COVID-19. The cumulative incidence of infection six months after the start of follow-up was 5.6% (95% CI: 5.1–6.2%) for the four-dose cohort and 5.5% (95% CI: 5.0–6.1%) for the three-dose cohort (Fig. 1).

Table 2 Hazard ratios and relative vaccine effectiveness for SARS-CoV-2 infection incidence after four versus three doses of ancestral-strain mRNA vaccines.
Full size table

The overall AHR for SARS-CoV-2 infection in the four-dose cohort compared to the three-dose cohort, adjusted for sex, age, nationality, coexisting conditions, vaccine type, prior infection status, calendar week of the third dose, and testing rate, was 0.91 (95% CI: 0.81–1.02) (Table 2). The corresponding rVE was 9.2% (95% CI: − 1.7 to 18.9%). An overall AHR for severe, critical, or fatal COVID-19 could not be estimated due to an insufficient number of cases; thus, rVE against severe disease could not be assessed.

The subgroup analysis, conducted at 3-month intervals after the start of follow-up, showed that the rVE of the fourth dose relative to the third dose was highest within the first 3 months, at 35.0% (95% CI: 20.6–46.8%) (Table 2; Fig. 3). However, protection declined rapidly over time and diminished after the first 3 months.

Fig. 3
Fig. 3The alternative text for this image may have been generated using AI.
Full size image

Relative vaccine effectiveness against SARS-CoV-2 infection following four versus three doses of ancestral-strain mRNA vaccines. SARS-CoV-2 denotes severe acute respiratory syndrome coronavirus 2.

In the sensitivity analysis adjusting for pre-baseline rather than post-baseline testing behavior, the overall rVE of the fourth dose compared to the third dose was estimated at 2.0% (95% CI: − 8.8 to 12.5). When stratified by time since vaccination, the rVE was highest within the first 3 months of follow-up, at 30.8% (95% CI: 15.9 to 43.1%). Between 3 and 6 months, the rVE declined to − 16.6% (95% CI: − 33.2 to 4.0%), and to − 13.6% (95% CI: − 27.9 to 3.4%) beyond 6 months. These findings are consistent with those of the main analysis (Table 2; Fig. 3).

Discussion

The results indicate that a fourth dose of the ancestral-strain mRNA vaccines provided only modest additional protection against infection compared to three doses. Moreover, this protection waned rapidly and was lost after three months. These findings align with the broader body of evidence on the effectiveness of ancestral-strain mRNA vaccines following the emergence of the omicron variant, which demonstrated only modest and short-lived protection against infection5,6,7,8,9,10,14,53.

The study findings can be attributed to waning vaccine-induced immunity in the context of a mismatch between ancestral-strain vaccines and the omicron subvariants circulating during the follow-up period11,12,13,14,15,54,55. Although the follow-up duration was relatively short—limiting the extent of viral evolution during the study period—the circulating variants were already highly evolved omicron subvariants relative to the ancestral virus. Previous studies have shown that such mismatch between vaccine- or infection-induced immunity and circulating variants can accelerate the waning of immunity14,32, underscoring the complex and interconnected nature of waning immunity and immune evasion—two phenomena that are difficult to disentangle14,32. The findings further underscore the need for periodically updated vaccines that better align with circulating variants30, as well as the development of next-generation vaccines capable of providing more durable protection.

The modest protection observed may also be explained by the characteristics of the study population, which differed from the general population of Qatar (Table 1). The four-dose cohort predominantly consisted of individuals aged 50 years or older and those with coexisting conditions—factors that necessitated vaccination with these ancestral-strain boosters before updated vaccines became available. Older adults56,57,58,59,60 and individuals with comorbidities61,62 often exhibit weaker immune responses compared to the general population, which may have contributed to the modest protection and the more rapid waning of immunity observed after vaccination.

This study has limitations. The study could not assess vaccine effectiveness against severe COVID-19 due to the very low number of such cases in the observed cohorts. With the accumulation of population immunity over time, the severity of infections has declined substantially35. Consequently, evaluating effectiveness against severe disease would have required much larger cohorts than those available in this study. However, in the context of findings from other vaccine effectiveness studies6,63, including those evaluating updated vaccines6, it must be emphasized that effectiveness against severe disease is likely to have been substantially higher than the modest effectiveness against infection observed in this analysis.

The study compared documented SARS-CoV-2 infections between cohorts; however, some infections may have gone unrecorded. Home-based rapid antigen tests, which were not documented, were not included in the analysis. Nevertheless, there is no indication that these factors would have differentially affected the cohorts in a way that would alter the study’s conclusions. The matching process accounted for key socio-demographic characteristics, which may also have helped control for differences in infection documentation between cohorts.

Matching was performed based on the number of coexisting conditions rather than the exact type of conditions to preserve the size of the matched cohorts for meaningful analysis. However, this may have led to some differences in the types of coexisting conditions across cohorts, potentially affecting the study estimates.

Matching was conducted using 10-year age groups rather than narrower age bands to minimize matching attrition and preserve cohort size. Despite this broader categorization, age was well balanced between the cohorts, with an SMD of only 0.03 (Table 1). This negligible imbalance indicates that age-related differences are not likely to have influenced the study findings.

Prior testing behavior (before baseline) was not included in the matching algorithm. Instead, individuals were categorized according to their testing propensity during the study period—defined as their average testing frequency over the study period—and this was incorporated as a baseline covariate in the Cox regression model. This strategy was adopted to avoid introducing time-varying confounding influenced by the exposure, which would otherwise require advanced causal inference methods. By using a fixed classification of testing behavior, we aimed to control for systematic differences in testing frequency between cohorts without conditioning on post-exposure variables.

While immune imprinting effects may influence vaccine effectiveness20,21,30,33,34,64,65,66,67, accounting for specific sequences of immune history in the matching process was not feasible in this study due to matching attrition. Incorporating such detailed immune exposure profiles would have required extensively larger cohorts to enable adequate matching without compromising statistical power. Nonetheless, immune imprinting is not likely to have materially affected the findings of this study. Although matching was not performed on precise immune sequences, the immune histories of individuals in both cohorts were largely comparable. This is because both cohorts were subject to the same national vaccination policies, experienced a rapid vaccine rollout, and were exposed to the same waves of infection during the pandemic.

The study was conducted in a specific national population, limiting the generalizability of findings to other populations. The relatively small cohort sizes constrained the statistical power needed to produce reliable effectiveness estimates over shorter timeframes than those presented. The AHRs were estimated both overall and at 3-month intervals from the start of follow-up. Notably, interval-based analyses can in principle be susceptible to changes in the composition of the study population over time.

As an observational study, the lack of blinding and randomization introduces the potential for unmeasured or residual confounding, which may bias the estimated outcomes68. Although robust matching was employed to control for known confounders, data limitations precluded matching on certain factors such as geography, occupation, or other social factors and behavioral patterns that may influence infection risk69. However, Qatar’s status as a city-state, with infections broadly distributed across neighborhoods, may have reduced the impact of geographic variation. Additionally, nationality, age, and sex serve as effective proxies for socio-economic status in Qatar26,37,38,39,40, and matching on these factors may have at least partially accounted for differences in other socio-economic factors.

Furthermore, the matching procedure used in this study has been previously evaluated in studies on the same population of Qatar, using different epidemiologic designs and control groups to test for null effects4,8,9,25,70. These prior studies demonstrated that the matching procedure effectively balances differences in infection exposure, allowing for accurate estimation of vaccine effectiveness4,8,9,25,70.

This study has strengths. It was conducted on Qatar’s population, encompassing a diverse range of national backgrounds. It leveraged extensive, validated databases from numerous prior COVID-19 studies, ensuring robust data quality. The availability of an integrated digital health information platform also enabled data collection on potential confounders, allowing for rigorous matching based on socio-demographic and health factors and prior infection statuses.

In conclusion, a fourth dose of the ancestral-strain mRNA vaccines provided only modest and short-lived additional protection against infection, with immunity waning rapidly and disappearing after three months. This underscores the importance of developing updated vaccines tailored to circulating variants and next-generation vaccines that offer more durable protection.