This retrospective cohort study provides evidence of significant protection elicited by mRNA-1273.815 against COVID-19-related hospitalization and medically attended COVID-19 through more than 3 months post-vaccination among adults active in the insurance system and with the healthcare system in the US, featuring a median of nearly 3 months of follow-up. Notably, this protection was consistently observed across various subgroups, including older adults and immunocompromised patients, and those with other specific underlying medical conditions that may have increased their risk of severe COVID-19. These findings could serve as additional evidence for supporting recommendations by CDC and other health authorities for COVID-19 vaccination with the most updated vaccine and underscore the benefit of receiving mRNA1273.815 for the adult population and among those who were at higher risk for COVID-19-related outcomes.

The follow-up duration of this study overlaps with the emergence of the JN.1 subvariant [23]. While estimates of VE have been shown to wane slightly over time, it is difficult to separate out the timing of the variant predominance from the time since vaccination [24, 25]. No analyses were done specifically to untangle these effects, as a limitation of the database is that variant-specific data are not available, and the declining rate of vaccination during the study period made it difficult to assess VE in patients vaccinated later when the JN.1 subvariant was predominant. However, this underscores the need to highlight the observed protection over a long study period despite the changes in circulating variants. We demonstrated that the VE for up to approximately 2 months post-vaccination was 52% (95% CI 49–55%) and was 32% (5%, 52%) for the time period beginning approximately 3.5 months after vaccination, indicating the vaccine is still effective in this time period, even with waning and against the background of changing variants. Our findings are consistent with neutralizing antibody data from individuals vaccinated with mRNA-1273.815, which showed significant increases in antibody levels against various SARS-CoV-2 variants [26]. Data from a US CDC IVY study assessing the effectiveness of the updated 2023–2024 COVID-19 vaccines also suggest that the Omicron XBB.1.5-containing vaccine formulation offers protection from a range of XBB variants, including newer variants in the XBB lineage, like EG.5 and HK.3, but waned over time [27]. Despite changes in the circulating variant, mRNA-1273.815 has been shown to protect well against severe COVID-19 outcomes.

The overall VE of 51% (95% CI 48–54%) against COVID-19 hospitalization in our study aligns with other studies on the 2023–2024 XBB.1.5 updated COVID-19 vaccines, which report effectiveness estimates ranging from 23 to 76%, despite differences in study designs, data sources, outcome definitions, and duration of follow-up [24, 27,28,29,30,31,32,33,34,35,36,37,38,39,40,41]. The CDC most recently reported that incremental VE (i.e., receipt versus no receipt of updated vaccination) against hospitalization with COVID-19 among adults aged ≥ 18 years was 42% (95% CI 37–46%) with a median time (IQR) since last dose 79 (44–119) days in the VISION Network, just slightly shorter than the median follow-up of 84 (58, 101) days in this study [27]. This VE estimate may be lower than reported in the current study because of different study designs and patient populations, including the potential for residual confounding and selection bias. Conversely, a recent cohort study in a large electronic health record (EHR)-linked claims database in the US reported a VE of 60.2% (95% CI 53.4%, 66.0%) against COVID-19-related hospitalization over a median follow-up of 63 (IQR 44 to − 78) days [42]. While the shorter follow-up time would be expected to result in a slightly higher VE, it is also possible that lengthening the follow-up in this study included the emergence of new variants, possibly affecting VE. Differences in population demographics [mean (SD) age in this study was 63 (16) years vs. 54 (16) years in the current study] and study design (i.e., this study matched on the timing of the most recent healthcare interaction and/or claim whereas the current study matched on frequency of healthcare claims in the previous year) could also lead to differences in VE, although it is worth noting that the confidence intervals of these two studies overlap.

The strengths of this study include the comprehensive inclusion of all eligible patients from a large, geographically representative database, with vaccination status determined through healthcare claims [21]. The study’s robust design, featuring direct matching and inverse probability of treatment weighting, ensured well-balanced cohorts of vaccinated patients compared to a non-active control group. Direct matching on key demographic and clinical characteristics, including engagement with the healthcare system, led to identification of an appropriate non-active comparator, and IPTW was able to further mitigate confounding and estimate average treatment effect among the full population. Matching the unexposed to the exposed on the index date mitigated any potential selection bias that would be imposed by a “ever/never (vaccinated)” study design approach. The large size of the database also contributed to a narrow confidence interval, adding confidence and precision to our reported effect estimates.

There are several limitations of this study. Although we have identified and adjusted for many baseline confounders, because of the lack of randomization, there could be potential residual confounding. For instance, it is possible that the exposed patients had otherwise healthier lifestyles not captured in the database (i.e., beyond what is captured with healthcare utilization and measures of comorbidity), contributing to their lower rates of hospitalization with COVID-19. This is an inherent limitation in any observational study design. The reliance on an insurance claims database means the study population represents a subset of the population, potentially excluding individuals less engaged with the healthcare system because of barriers such as trust, socioeconomic status, or accessibility. The outcome definition of hospitalization with COVID-19 was chosen to partially address these limitations, where a severe outcome would be less subjective and less influenced by these factors than a more subjective outcome where a person may or may not seek medical attention. Nevertheless, it is important to emphasize that these external factors limit generalizability of this study to a broader population. While the study population has similar age and gender distribution to the HealthVerity dataset, the population aged ≥ 65 is underrepresented compared with the US population. However, similar vaccine effectiveness was observed across different age groups, suggesting that this underrepresentation is unlikely to affect the overall VE reported.

All variables, including exposure, outcomes, and covariates, were derived from the claims databas;, thus, measurement errors are possible. Most individuals in the database are missing race/ethnicity data, so estimates could not be assessed based on this characteristic. As in all claims and/or EHR datasets, vaccination is likely under-captured; this may have led to a bias toward the null, resulting in the true VE being higher than reported. Consistent with product labeling, the unexposed population may have previously received a 2023–2024 COVID-19 vaccine (non 0.815) up to 60 days prior to their index date. Although this represented < 2% of the population and is unlikely to have a major effect on the VE, they may have protective effects from the prior vaccination, which may bias the VE toward the null. Administrative claims data are valuable for large-scale healthcare research; however, there are inherent limitations associated using these data, including potential bias due to misclassification of the outcome and exposure in this study. Diagnostic and procedural codes used to define vaccination and COVID-19-related hospitalizations may be inaccurately recorded, leading to incorrect classification of patient conditions or treatments. Additionally, claims data may not capture all hospitalizations or vaccination records, particularly among specific subpopulations such as uninsured individuals, underrepresented groups, or those receiving care outside the insurance network. These data gaps may result in underestimating the number of vaccinated individuals, as well as the number of COVID-19 hospitalizations, and limit the generalizability and accuracy of research findings. A validation study by Kadri et al. [43] reported a high positive predictive value (PPV) of 98% for the U07.1 code, used for COVID-19 diagnoses, based on clinical documentation and laboratory results. The primary outcome of hospitalization with COVID-19 in this study included the diagnosis identified in any position, which may limit the specificity of identifying COVID-19 hospitalizations (i.e., “due to” vs. “with” COVID-19). We conducted a sensitivity analysis examining a more specific definition of COVID-19 hospitalization, with higher PPV, developed by Kluberg et al. [22], which required either U07.1 or B97.29 (other coronavirus as the cause of diseases classified elsewhere) to define COVID-19 AND at least one diagnosis of pneumonia/sepsis or acute respiratory distress syndrome. The sensitivity analysis was hypothesized to result in less outcome misclassification and fewer events observed with a more specific definition. Indeed, the sensitivity analysis examining a more specific definition of COVID-19 hospitalization demonstrated an 80% decrease in the observed number of events across both groups and higher VE (63%, 95% CI 57–69%) compared to the primary analysis.

Since the start of the COVID-19 pandemic, most of the adult population has developed immune responses against the SARS-CoV-2 virus, whether through previous infections, vaccinations, or a combination of both [44]. The current study included individuals regardless of their vaccination and infection history, which were included as covariates. Most individuals in both cohorts received a previous bivalent vaccine. Thus, the results of this study should be interpreted in the context of the incremental protection provided by the most updated COVID-19 targeting the circulating variants in a real-world setting regardless of exposure and/or vaccination history, highlighting the protection against hospitalization. In an environment of suboptimal vaccination coverage rates across all ages, it is essential to communicate the additive protection to increase vaccine confidence among clinicians and the general population.