Protection of mRNA vaccines against hospitalized COVID-19 in adults over the first year following authorization in the United States

Abstract Background COVID-19 mRNA vaccines were authorized in the United States in December 2020. Although vaccine effectiveness (VE) against mild infection declines markedly after several months, limited understanding exists on the long-term durability of protection against COVID-19-associated hospitalization. Methods Case control analysis of adults (≥18 years) hospitalized at 21 hospitals in 18 states March 11 – December 15, 2021, including COVID-19 case patients and RT-PCR-negative controls. We included adults who were unvaccinated or vaccinated with two doses of a mRNA vaccine before the date of illness onset. VE over time was assessed using logistic regression comparing odds of vaccination in cases versus controls, adjusting for confounders. Models included dichotomous time (<180 vs ≥180 days since dose two) and continuous time modeled using restricted cubic splines. Results 10,078 patients were included, 4906 cases (23% vaccinated) and 5172 controls (62% vaccinated). Median age was 60 years (IQR 46–70), 56% were non-Hispanic White, and 81% had ≥1 medical condition. Among immunocompetent adults, VE <180 days was 90% (95%CI: 88–91) vs 82% (95%CI: 79–85) at ≥180 days (p < 0.001). VE declined for Pfizer-BioNTech (88% to 79%, p < 0.001) and Moderna (93% to 87%, p < 0.001) products, for younger adults (18-64 years) [91% to 87%, p = 0.005], and for adults ≥65 years of age (87% to 78%, p < 0.001). In models using restricted cubic splines, similar changes were observed. Conclusion In a period largely pre-dating Omicron variant circulation, effectiveness of two mRNA doses against COVID-19-associated hospitalization was largely sustained through 9 months.

The Coronavirus Disease 2019  pandemic led to an estimated 5.4 million deaths worldwide 2 through December 2021 [1]. Highly effective vaccines are available, and vaccination is the best tool to 3 control the impact of the pandemic [2,3]. In the United States, three licensed vaccines are available, 4 with most vaccinated persons receiving messenger RNA (mRNA) COVID-19 vaccine products including 5 mRNA-1273 (from Moderna) and BNT162b2 (from Pfizer-BioNTech) [4]. Vaccination has reduced the 6 burden of COVID-19 including COVID-19-associated deaths in the U.S. [5,6], with most severe COVID-19 7 illnesses and deaths occurring among unvaccinated persons [7,8]. 8 In countries with higher vaccination coverage, reductions in vaccine effectiveness (VE) with passage of 9 time prompted booster recommendations for COVID-19 vaccines [9]. In these vaccinated populations, 10 surges of COVID-19 complicate the understanding of the protective effect of vaccines and policy 11 discussions for several reasons. First, with increasing time since vaccination, protection has varied by 12 disease severity, with more sustained vaccine protection against severe disease as compared with mild 13 infections [10,11]. These infections in vaccinated individuals could be due to waning of antibodies [12, 14 13], particularly in the mucosal compartments at the site of infection, or from emergence of SARS-CoV-2 15 variants that might escape immune protection. In contrast, durable memory B cell and T cell responses 16 might provide sustained protection against more severe disease [14], possibly including heterotypic 17 protection against new variants. Second, protection may differ by underlying conditions such as 18 immunosuppression [15], vaccine product [16,17], and number of doses received [18]. Thus, as the 19 pandemic continues to evolve, disentangling factors of waning immunity, viral evasion of immunity, 20 number of doses and type of vaccine, and host immune responses have become increasingly complex. 21 Ongoing real-world VE studies in large, diverse populations can inform vaccination program goals in 22 terms of understanding of protection provided for different levels of disease severity, populations in 23 A C C E P T E D M A N U S C R I P T 5 whom booster doses may be most beneficial for the prevention of severe outcomes and timing of 1 booster doses, and the need for potential antigen updates in vaccines. 2 The Centers for Disease Control and Prevention (CDC) collaborates with the Influenza and Other Viruses 3 in the Acutely Ill (IVY) Network to monitor the effectiveness of vaccines for the prevention of COVID-19-4 associated hospitalizations among U.S. adults [3,16]. In this report, we evaluate the duration of 2-dose 5 mRNA vaccine protection against COVID-19 hospitalizations during the first year of the U.S. vaccination 6 program. Our primary goal was to examine VE over time by host factors such as age and underlying 7 conditions, vaccine product, and immunosuppression status to evaluate the durability of vaccine 8 protection to inform future vaccine strategies. 9

METHODS 10
We monitor the effectiveness of COVID-19 vaccines for the prevention of COVID-19 hospitalization 11 among U.S. adults (≥18 years of age) by enrolling adults at 21 U.S. medical centers. We assessed the 12 effectiveness of the mRNA vaccines over time in patients admitted March 11 through December 15, 13 2021 using a case-control design. Interim durability estimates including IVY Network enrollments 14 through July 14, 2021 were previously published (including 3089 hospitalized adults, with a median of 65 15 days between receipt of dose 2 and illness onset among vaccinated patients) [19]; this analysis adds five 16 additional months of enrollment data including a longer duration of follow-up since vaccination during a 17 period when the SARS-CoV-2 Delta variant predominated. 18 We considered immunocompetent and immunocompromised patients separately because of variation 19 in immune responses to COVID-19 vaccination in these patients. [15]. Immunocompromising conditions 20 were defined as having one or more of the following: active solid organ cancer (defined as treatment for 21 the cancer or newly diagnosed cancer in the past 6 months), active hematologic cancer, HIV infection 22 with or without AIDS, congenital immunodeficiency syndrome, previous splenectomy, previous solid 23 I P T   6   organ transplant, active immunosuppressive medication use, systemic lupus erythematosus, rheumatoid  1 arthritis, psoriasis, scleroderma, or inflammatory bowel disease. Enrollment methods have previously 2 been described [3,7]. In brief, COVID-19 case patients had COVID-19-like illness (CLI) and tested positive 3 for SAR-CoV-2 by molecular or antigen test within 10 days of illness onset. Two control groups of 4 hospitalized adults without COVID-19 were included: (1) a "test-negative" control group comprised of 5 patients hospitalized with CLI who tested negative for SARS-CoV-2 by reverse transcription polymerase 6 chain reaction (RT-PCR) and (2) a "syndrome-negative" control group comprised of patients hospitalized 7 without CLI who tested negative for SARS-CoV-2 by RT-PCR. VE using individual control groups was 8 highly similar, and therefore patients from both groups were combined into a single control group. Case 9 or control status was determined using SARS-CoV-2 clinical testing results and results from central RT-10 PCR testing of upper respiratory specimens collected at enrollment and tested at Vanderbilt University 11 Medical Center (Nashville, Tennessee). Patients enrolled as test-negative controls who subsequently 12 tested positive for SARS-CoV-2 were reassigned as a COVID-19 case-patient and syndrome-negative 13 controls with a subsequent positive SARS-CoV-2 test were excluded from the analysis. 14 COVID-19 vaccination status and vaccine product information were determined through self-report 15 during enrollment interviews with patients or their proxies and systematic review of source 16 documentation including hospital electronic medical records, state vaccine registry searches, and 17 vaccination record cards. Patients were considered vaccinated for this analysis if two doses of a single 18 mRNA vaccine product were documented or self-reported (with date and location) ≥14 days before a 19 reference date, defined as the date of symptom onset for cases and test-negative controls or five days 20 prior to hospital admission for syndrome-negative controls. If no COVID-19 vaccine was received prior to 21 the reference date, patients were considered unvaccinated. Patients who received 1 or more doses of a 22 mRNA vaccine but did not meet study criteria for full vaccination, who received mixed vaccine products, 23 or who received a non-mRNA vaccine were excluded, as well as patients who received more than two 1 doses of a mRNA vaccine with the third dose received ≥7 days before illness onset. 2 Patients were classified as being in a period of higher proportions of lineages other than Delta (pre-3 Delta) period if their admission date was before July 1 st 2021 [4]. Otherwise, patients with an admission 4 date on or after July 1 st , 2021 were classified as being in a period of higher B.1.617.2 and AY lineages 5 (Delta period). Information on patients' age, sex, self-reported race and ethnicity, and preexisting 6 chronic medical conditions were obtained through electronic medical record review and structured 7 enrollment interviews. 8 Logistic regression models were used to estimate VE by time since vaccination with different models 9 treating time as binary (<180 days vs ≥180 days between second dose and reference date) and as 10 continuous (applied a restricted cubic spline with number of knots determined by the lowest AIC of the 11 regression model tested with 3-7 knots). Briefly, we applied a spline to the daily time term due to the 12 non-linear nature of VE over time; in other words, the use of splines allowed the waning speed to 13 change over time opposed to a constant decline. Each logistic regression model used COVID-19 case 14 status as the outcome and vaccination status (vaccinated vs unvaccinated) as the predictor along with 15 the time since vaccination term (binary or continuous). Models included additional covariates for 16 calendar date of admission (in biweekly intervals), age (continuous years), sex, self-reported race and 17 ethnicity, presence of underlying chronic conditions, immunocompromised status, and US Health and 18 Human Services region of the admitting hospital. Unvaccinated patients were assigned a reference value 19 of zero days since vaccination. In binary time models, VE was estimated using logistic regression 20 comparing odds of case vs control outcome by a primary predictor of vaccination status (vaccinated 21 <180 days before symptom onset, vaccinated ≥180 days since symptom onset, or unvaccinated), using 22 the equation VE = (1 -aOR) × 100. In continuous time models, VE was calculated at each time since 23 vaccination t as VE(t) = (1 -aOR(t)) × 100, where aOR(t) is the estimated odds ratio of being a case 24 8 patient for vaccinated patients at t days since vaccination compared to an unvaccinated patient at 0 1 days adjusted for the specified covariates. 95% confidence intervals (CI) for these VE curves were 2 obtained using bootstrapping with 1000 replicates. Interaction terms were introduced to evaluate VE 3 over time stratified by characteristics of interests including age group (18-64 years or ≥65 years), 4 underlying chronic medical conditions (0 vs ≥1), vaccine product received (Pfizer BioNTech vs Moderna), 5 and baseline immunocompromising conditions. An additional model limited to patients with admission 6 date on or after July 1 st , 2021 was conducted to estimate VE over the Delta period. Separate models 7 were constructed for immunocompetent and immunosuppressed participants due to known effect 8 modification of VE by immune function status [15]. 9 VE across binary time since vaccination groups was compared with likelihood ratio chi-squared tests. P-10 values <0.05 were considered statistically significant. This activity was conducted as a public health 11 surveillance activity, with waiver of informed consent. 12

13
Of 12,513 patients enrolled through December 15, 2021, 2435 were excluded (1312 who were not 14 vaccinated with 2 doses of a mRNA vaccine or received a third dose; 606 who received a non-mRNA 15 vaccine or mixed products; and 517 who met other exclusion criteria). Of 10,078 patients included in the 16 analysis, 4906 (49%) were COVID-19 case patients and 5172 (51%) were COVID-19-negative controls 17 (Table 1). Among 4906 cases, 1119 (23%) were vaccinated and, among 5172 controls, 3229 (62%) were 18 vaccinated. Overall, median age was 60 years (IQR: 46 -70), 5675 (56%) were non-Hispanic White, 2198 19 (22%) non-Hispanic Black, and 1589 (16%) Hispanic of any race, 8203 (81%) had one or more chronic 20 medical conditions, and 1940 (19%) had an immunocompromising condition. COVID-19 case patients 21 were younger on average than controls (median 57 vs 62 years; p<0.001), were less likely to report 22 having prior laboratory confirmed infection with SARS-CoV-2 (3% vs 9%; p<0.001), and, among those 23 who were vaccinated, had a longer median time since receiving the second vaccine dose (median 163 vs 1 127 days, p<0.001). Among 4862 (99%) COVID-19 case-patients with hospital outcomes, 538 (11%) died 2 within 28 days of admission, 1919 (39%) were admitted to the intensive care unit, and the median 3 length of stay among those who survived and were discharged by day 28 (n = 3782) was 6 (IQR 3 -10) 4 initially (at 14 days since vaccination), increased to a maximum of 93% after 75 days, and then 1 decreased to 80% after 270 days (Figure 2), with time since vaccination being a significant factor in 2 estimating VE (p<0.001). Among immunocompetent adults, VE estimates similarly varied by time within 3 subgroups of interest. For the Pfizer-BioNTech vaccine, VE peaked at 92% after 74 days and decreased to 4 75% at 270 days, and for the Moderna product VE peaked at 94% after 83 days and decreased to 86% 5 after 270 days [ Figure 3], with time from vaccination being significant in each group (p<0.001). For those 6 aged 18-64, VE peaked at 94% after 84 days and decreased to a minimum of 86% at 198 days, and for 7 those 65 or older VE peaked at 92% initially at 14 days and decreased to 73% after 270 days [ Figure 4], 8 with time from vaccination being significant in each group (p<0.001). Models testing additional 9 interactions also showed a change in VE over time for both the group with no underlying conditions and 10 those with ≥1 underlying condition (Supplementary Figure 1) and for those in the Delta period 11 (Supplementary Figure 2). Immunosuppressed individuals showed overall lower VE compared to 12 immunocompetent patients over time (Supplementary Figure 3). 13

DISCUSSION 14
In this multicenter evaluation across 18 states over the first year following COVID-19 vaccine 15 introduction, we found that vaccination with two doses of an mRNA product provided protection against 16 COVID-19 hospitalization prior to predominant Omicron variant and subvariant circulation. Effectiveness 17 was generally sustained at ≥80% over a period of 270 days with some gradual decline after peaking 2-3 18 months after the second vaccine dose. This pattern of protection was similar across subgroups, such as 19 by vaccine product. Notably, overall protection was lower for older adults (≥65 years of age) compared 20 to young adults (18-64 years of age) and was modestly lower after 180 days (87% vs 78%), highlighting 21 the importance of additional vaccine doses in older adults who are at increased risk of severe COVID-19 22 illness. Both mRNA vaccine products provided a high level of protection, with modestly higher VE 23 observed for the Moderna compared to the Pfizer-BioNTech vaccine [16,17]  vaccines, may be a long-term strategy to durably reduce the impact of severe COVID-19 as SARS-CoV-2 5 continues to circulate globally. However, challenges remain in predicting new SARS-CoV-2 variants. 6 Our report has several limitations. We focused on hospitalized outcomes only. We did not include adults 7 who received more than 2 doses of an mRNA vaccine, first recommended for individuals with 8 immunocompromising conditions in August 2021 [29] and in the general adult population in November 9 2021 [9]. We did not include adults who received mixed vaccine products due to a limited number of 10 patients with heterologous vaccination. These data predated more recent predominance of the SARS-11 CoV-2 Omicron variant. We also could not control for some potential time-varying confounders such as 12 varying force of infection due to factors such as changes in mitigation measures. Although this analysis 13 included hospitalized adults from 18 geographically and demographically diverse states, patients may 14 not have been fully representative of the US adult population. In models evaluating patients with 15 immunocompromising conditions, diverse immunocompromising conditions associated with variable 16 degrees of immunosuppression and potentially with durability of vaccine protection were combined. 17 Lastly, a high proportion of these hospitalized adults had multiple chronic medical conditions, thus 18 reducing the generalizability to other populations with lower burden of chronic medical conditions. 19

Conclusions 20
In this multi-center US study, we found high and largely sustained protection against COVID-19 following The findings and conclusions in this report are those of the authors and do not necessarily represent the 7 official position of the Centers for Disease Control and Prevention. 8 Funding 9 This work was supported by the United States Centers for Disease Control and Prevention (CDC). 10