Comparative Effectiveness of the Bivalent (Original/Omicron BA.4/BA.5) mRNA COVID-19 Vaccines mRNA-1273.222 and BNT162b2 Bivalent in Adults With Underlying Medical Conditions in the United States

Data Source

This study leveraged electronic health record (EHR) data from the Veradigm Network EHR linked to healthcare claims sourced from Komodo Health spanning March 1, 2020, through May 31, 2023. This integrated dataset has been previously characterized and used previously in COVID-19 epidemiology and VE research [10,12–14]. The dataset used in this study contains only de-identified data as per the de-identification standard defined in Section §164.514(a) of the Health Insurance Portability and Accountability Act of 1996 (HIPAA) Privacy Rule. As a noninterventional, retrospective database study using data from a certified HIPAA–compliant de-identified research database, approval by an institutional review board was not required.

The Komodo data contains claims sourced directly from payers as well as other sources, such as revenue cycle management platforms and claims clearinghouses. For the purpose of this analysis, the closed claims analysis leveraged all claims captured during a period of continuous health plan enrollment, regardless of claim source. By contrast, the open claims sensitivity analysis leveraged all claims during the study period regardless of health plan enrollment. When multiple claims appear for the same encounter, the claim from the payer superseded the claims from any other source.

Study Design and Study Population

This retrospective, observational cohort study was designed, implemented, and reported in accordance with Good Pharmacoepidemiological Practice, applicable local regulations, and the ethical principles laid down in the Declaration of Helsinki.

Individuals ≥18 years of age with ≥1 underlying medical condition associated with an increased risk of severe outcomes from COVID-19 as defined by the CDC [15] who had received either the mRNA-1273.222 (50 mcg) or BNT162b2 Bivalent vaccine (30 mcg) between August 31, 2022, and February 28, 2023, were eligible for inclusion in the study. Underlying medical conditions were identified based on medical encounters in the 365 days before vaccination. Codes used to identify the underlying medical conditions are listed in Supplementary Table 1. Codes used to identify vaccination with bivalent mRNA vaccines are reported in Kopel et al. [12]. The index date was defined as the date of receipt of the bivalent vaccine, and the cohort entry date (CED) was defined as 7 days after the index date (Figure 1). The follow-up period for outcomes of interest for each individual participant began on the CED and continued until the first occurrence of an event of interest, disenrollment from medical/ pharmacy plan, receipt of an additional COVID-19 vaccination, or the end of available data (May 31, 2023), whichever occurred first.

• Download figure
• Open in new tab

Figure 1. Study Design

CED, cohort entry date. ^a The index date is the vaccination date. ^b Begins 365 days before vaccination with a bivalent vaccine

Individuals were included in the main analysis if they had a minimum of 365 days of continuous medical and pharmacy claims enrollment in a contributing health plan and ≥1 contact with a health service provider in the previous 365 days. Patients with evidence of COVID-19 infection or additional vaccination between the index date and CED, <1 day of follow-up available, or missing birth year/sex were excluded. The two cohorts were mutually exclusive, and the assignment was based on the first vaccine received during the vaccination period.

Objectives

The primary objective of this study was to assess the rVE of mRNA-1273.222 versus BNT162b2 Bivalent in preventing COVID-19-related hospitalization. The secondary objective was to assess the rVE of mRNA-1273.222 versus BNT162b2 Bivalent in preventing COVID-19-related outpatient encounters. Both primary and secondary objectives included individuals with any CDC-defined underlying medical condition associated with an increased risk of severe outcomes from COVID-19.

The exploratory objective of this study was to assess the primary and secondary objectives (rVE of COVID-19-related hospitalizations and outpatient encounters) among 5 pre-specified subgroups of patients with underlying medical conditions: cardiovascular disease, chronic kidney disease, chronic lung disease, diabetes, and immunocompromised. These five groups were selected based on their prevalence among the adult population and the risk for severe outcomes [10,15,16]. The specific underlying medical conditions included in these subgroups are defined in Table 1.

Outcome Measures

The primary outcome of interest was COVID-19-related hospitalizations, which were identified from hospitalization claims with documentation of COVID-19 diagnosis in any position. The secondary outcome of interest was COVID-19-related outpatient encounters, which were identified from outpatient EHR and medical claims with a COVID-19 diagnosis. Outpatient encounters included visits in various clinical settings, such as emergency department visits, urgent care visits, office visits, and telemedicine visits. The codes used to identify outcomes are reported in Kopel et al. [12].

Covariates

The study utilized patient characteristics as covariates to adjust for any baseline differences between vaccine cohorts (see statistical analysis). The following demographic variables were captured at the index date: age, sex, race, ethnicity, insurance type, region, and month of index. COVID-19 vaccination and infection history were assessed using a lookback period that started on March 1, 2020, and included primary series vaccination, time since the last COVID-19 monovalent vaccination in months, time since the last COVID-19 infection in months, and data source of recorded vaccination. Healthcare resource utilization variables were assessed during the 365-day baseline period and included the number of prior hospitalizations and the number of prior outpatient encounters.

The following underlying medical conditions were captured in the 365 days prior to, and inclusive of, the index date: asthma, cancer, cerebrovascular disease, chronic kidney disease, chronic liver disease, chronic lung disease, cystic fibrosis, dementia, diabetes mellitus, disability, heart conditions, human immunodeficiency virus (HIV), mental health conditions, obesity (body mass index > 30), physical inactivity, primary immunodeficiencies, respiratory tuberculosis, smoking, solid organ or stem cell transplant, and use of select immunosuppressive medications. Pregnancy was captured in the 301 days following the index date. The codes used to identify patients with underlying medical conditions are listed in Supplementary Table 1.

Statistical Analysis

Statistical analysis followed the methodology described in Kopel et al. [12]. Briefly, propensity scores predicting receipt of mRNA-1273.222 or BNT162b2 Bivalent were calculated using a multivariable logistic model, adjusted for all covariates outlined above. Stabilized and truncated weights were used to re-weight the study sample using the inverse probability of treatment weighting (IPTW) methodology. Standardized mean differences (SMD) were calculated to assess sample balance before and after IPTW. SMDs with absolute values >0.1 indicated covariate imbalance.

Covariates were reported descriptively before and after weighting. Mean and standard deviation (SD) are reported for continuous variables, while number (N) and percent are reported for categorical variables. Follow-up time in days is captured and reported as median and interquartile range (IQR).

Unadjusted hazard ratios (HRs) were reported for the unweighted sample, and adjusted HRs were reported for the weighted sample. Unadjusted HRs were estimated using Cox regression models with exposure as the only predictor. Adjusted HRs were estimated using a multivariable Cox regression model that included the exposure and any baseline covariates with SMD >0.1 after weighting. The rVEs were calculated as 100 × (1 − HR) for both unadjusted and adjusted estimates with 95% confidence intervals (95% CI).

All analyses were performed in the overall population of adults ≥18 years of age with ≥1 underlying medical condition to address the primary objective and secondary objective and then repeated separately for each of the prespecified five cohorts with specific underlying medical conditions of interest (cardiovascular disease, chronic kidney disease, chronic lung disease, diabetes, and immunocompromised).

All statistical analyses were performed using SAS 9.4 or R Statistical Software (v4.1.3) survival (v3.2-13) package.

Sensitivity analyses

The main analysis was restricted to patients with more than 365 days of continuous health plan enrollment with medical and pharmacy benefits. As this approach may be biased towards patients with stable health insurance, we conducted a sensitivity analysis of the primary and secondary objectives using an open claims approach. The methodology used was the same as the main analysis, except there were no requirements for continuous health plan enrollment. Instead, we required that individuals have at least one medical or pharmacy claim in the 365 days preceding the index date.

A second sensitivity analysis examined the primary and secondary objectives in a shorter follow-up period ending on February 28, 2023. Apart from the shorter follow-up period, all analytical methods were consistent with the main analysis.

Main Analysis

In the overall study cohort, post-weighting, 2,360 (0.31%) individuals who received mRNA-1273.222 and 4,198 (0.35%) who received BNT162b2 Bivalent were hospitalized for COVID-19 during the follow-up period. Whereas, 26,185 (3.5%) individuals who had received mRNA-1273.222 and 42,866 (3.6%) who had received BNT162b2 Bivalent had an outpatient encounter related to COVID-19.

For both the primary and secondary endpoints, mRNA-1273.222 was significantly more effective at preventing the outcomes of interest compared to BNT162b2 Bivalent. Specifically, the rVE (95%CI) against COVID-19-related hospitalization was 10.9% (6.2%– 15.2%, p<0.001), and the rVE against outpatient COVID-19 was 3.2% (1.7%–4.7%, p<0.001) (Figure 3).

• Download figure
• Open in new tab

Figure 3. Adjusted rVE estimates of mRNA-1273.222 vs. BNT162b2 Bivalent for adults with ≥1 underlying medical condition

For all subgroups, mRNA-1273.222 was significantly more effective at preventing COVID-19-related hospitalizations and outpatient encounters compared to BNT162b2 Bivalent (Figure 4). The rVE (95% CI) of mRNA-1273.222 vs. BNT162b2 Bivalent against COVID-19-related hospitalization was 15.1% (8.7%–21.0%) in patients with diabetes, 14.7% (9.0%– 20.1%) in those with cardiovascular disease, 11.9% (5.1%–18.2%) in those with chronic lung disease, 15.0% (7.2%–22.2%) in immunocompromised patients, and 8.4% (0.5%–15.7%) in CKD patients. Point estimates for rVE against outpatient COVID-19 were lower but statistically significant and ranged from 3.1% in immunocompromised patients to 7.6% in patients with CKD. Unadjusted and adjusted hazard ratios and 95% CI are reported in Table 4.

• Download figure
• Open in new tab

Figure 4. Adjusted rVE estimates of mRNA-1273.222 versus BNT162b2 Bivalent among subgroups with specific underlying medical conditions

Sensitivity analyses

The results of the sensitivity analysis were consistent with the findings of the main analysis (Tables 4 and 5). Specifically, in the open claims dataset, we identified 1,960,185 adults ≥18 years old with at least one underlying medical condition who received mRNA-1273.222 and 2,969,597 who received BNT162b2 Bivalent vaccine during the vaccination period. Study subjects met all inclusion criteria as in the main analysis except that the requirement for continuous enrollment in claims ≥365 days prior to the index date was replaced with a requirement for at least one claim within 365 days of the index date. In this population, the rVE (95% CI) of mRNA-1273.222 versus BNT162b2 Bivalent against COVID-19-related hospitalization was 12.0% (9.3%–14.7%), and the rVE against outpatient COVID-19 was 5.0% (4.0%–5.9%). Similarly, the use of a shorter follow-up period (ending February 28, 2023) in the closed claims database resulted in rVE estimates of 13.1% (7.8%–18.2%) and 3.8% (2.1%–5.4%) against hospitalized and outpatient COVID-19, respectively. Baseline characteristics of individuals included in the sensitivity analyses before and after weighting are reported in Supplementary Tables 7 and 8.

Strengths and limitations

The primary strength of this analysis is the depth and breadth of the linked EHR and claims dataset used in this analysis. By integrating sources of patient information across the continuum of care, we increase the likelihood of capturing exposures (vaccination events), covariates (patient characteristics and underlying medical conditions), and outcomes (COVID-19-related hospitalizations and outpatient encounters). Additionally, as information on exposure, outcome, and covariates was collected from patient records in a consistent manner across all cohorts, there is a reduced likelihood of differential misclassification. Residual confounding between cohorts was addressed using propensity score weighting and multivariable regression.

This study is subject to several limitations which are inherent to many non-interventional studies utilizing real-world data. Specifically, this retrospective observational study, which relies on routinely collected data on insured patients, might include data entry errors and cannot be generalized to the uninsured population. Additionally, the current primary analysis was restricted to closed claims, which provide a comprehensive overview of a patient’s healthcare interactions but may not capture all cases and is restricted to patients with stable healthcare insurance. However, the sensitivity analysis performed using an open claims approach was consistent with the main analysis, which supports a broad generalizability of our findings. Furthermore, we required that study subjects have at least one healthcare encounter during the previous 365 days, which would exclude healthy patients; however, as this study was restricted to individuals with underlying medical conditions, the effects of this bias are minimized.

Finally, there may have been differences between the two vaccine groups that were not fully accounted for by the pre-defined covariates and residual confounding may be present between the cohorts. For example, differences in the timing and availability of the two vaccines may have impacted which patients were more likely to receive one vaccine over the other. In our previous analysis that included the overall adult population, we found no difference in the proportion of patients vaccinated with each vaccine over time [12].