key: cord-0806455-utrherly
authors: McConeghy, K. W.; Bardenheier, B.; Huang, A. W.; White, E. M.; Feifer, R. A.; Blackman, C.; Santostefano, C. M.; Lee, Y.; DeVone, F.; Halladay, C. W.; Rudolph, J. L.; Zullo, A. R.; Mor, V.; Gravenstein, S.
title: Effectiveness of a SARS-CoV-2 mRNA vaccine booster dose for prevention of infection, hospitalization or death in two nation-wide nursing home systems
date: 2022-01-28
journal: nan
DOI: 10.1101/2022.01.25.22269843
sha: 711f3f4a175dde017e810e79b61827cc010046b7
doc_id: 806455
cord_uid: utrherly

Background A SARS-CoV-2 vaccine booster dose has been recommended for all nursing home residents. However, we lack effectiveness data on boosters preventing infection, death and hospitalization in this frail population. Methods We emulated nested target trials in two large nursing home systems in parallel to evaluate the effectiveness of a SARS-CoV-2 mRNA vaccine booster at preventing infection, hospitalization, or death. Residents who completed a 2-dose series of the mRNA vaccine and were eligible for a booster were included in from September 22, 2021 to November 5, 2021. Outcomes were measured through December 18, 2021, including test-confirmed SARS-CoV-2 infection, hospitalization, or death. The vaccine effectiveness at day 42 was estimated with a Kaplan-Meier estimator, both unadjusted and weighted with the inverse probability of treatment. Results The two NH systems were large and multi-state, System 1 included 200 NH (8,538 control and 5,721 boosted residents) and System 2 included 127 NHs (4,100 control and 2,291 boosted residents). Booster vaccination reduced infections by 50.4% (95% Confidence Interval [CI]: 29.4%, 64.7%) SARS-CoV-2 infections in System 1 and 58.2% (32.3%, 77.8%) in System 2. Boosted residents in System 1 also had a 97.3% (86.9%, 100.0%) reduction in SARS-CoV-2 associated death, but too few events for comparison in System 2. Conclusions: During a Delta predominant period, SARS-CoV-2 booster vaccination significantly reduced infection in two U.S. nursing home systems. In the larger System 1 a 97% reduction in SARS-CoV-2 related death was also observed. These findings strongly support administration of vaccine boosters to nursing home residents.

Background A SARS-CoV-2 vaccine booster dose has been recommended for all nursing home residents. However, we lack effectiveness data on boosters preventing infection, death and hospitalization in this frail population. Conclusions: During a Delta predominant period, SARS-CoV-2 booster vaccination significantly reduced infection in two U.S. nursing home systems. In the larger System 1 a 97% reduction in SARS-CoV-2 related death was also observed. These findings strongly support administration of vaccine boosters to nursing home residents.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

Nursing home residents have suffered significant SARS-CoV-2 morbidity and mortality. 1, 2 Resurgence in late summer/early fall 2021 with the Delta variant and waning immunity in NURSING HOME residents 3 led the Centers for Disease Control and Prevention (CDC) to recommend an additional "booster" dose for those vaccinated with a primary series. 4, 5 The Food and Drug Administration amended the emergency use authorizations for the BNT162b2 (Pfizer-BioNTech) and mRNA-1273 (Moderna) vaccines in September and October 2021, respectively, to include an additional dose at least six months after the primary series for 'high-risk' individuals, including those aged 65 years or older. 6 Both mRNA vaccines were also authorized in August 2021 to be used for additional doses in immunocompromised individuals, a description that applies to many long-term care residents. As of December 12, 2021 , approximately 55% of U.S. nursing home residents had received a booster. 7 Despite the high risk of SARS-CoV-2 morbidity and mortality in nursing home residents, limited data exist on the effectiveness of boosters in mitigating infection or death in this frail population. Vaccine trials did not include nursing home residents, and timely patient-level observational data are not widely available. Nursing homes are an optimal environment for measurement of vaccine effectiveness because of the residential stability, systematic documentation of immunizations (including boosters) and frequent testing for SARS-CoV-2. Additional clinical evidence of booster effectiveness would support efforts to increase booster distribution for this vulnerable population.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted January 28, 2022. To address this gap, we leveraged electronic health record data updated daily from two large, multi-state long-term care providers. Using a nested target trial emulation method 8 , we evaluated the vaccine effectiveness of an mRNA booster dose (versus no booster) among residents who completed an mRNA vaccination primary series. Our outcomes included any laboratory-confirmed SARS-CoV-2 infection, hospitalization, and death. We hypothesized that a mRNA vaccine booster would prevent these outcomes.

This study was designed to emulate a sequence of nested target trials comparing additional effectiveness of SARS-CoV-2 mRNA vaccine booster versus primary series only in nursing home residents. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted January 28, 2022. ; https://doi.org/10.1101/2022.01.25.22269843 doi: medRxiv preprint

The emulated target trial method involves designing a hypothetical trial and emulating it in the observational data through applying inclusion and exclusion criteria to a cohort, and identifying a well-defined intervention and outcome. 8, 9 Specific dates are selected as 'trial' dates where those meeting eligibility and assigned to treatment or not are eligible for inclusion. Our target trial is outlined in Table 1 for both nursing home systems. This represents the trial being emulated in the observational data. On each day residents were considered eligible for inclusion if they had been present in the home for at least 90 days with receipt of a 2-dose mRNA vaccine series at least 180 days earlier. Residents who received the Janssen vaccine as their initial vaccination were excluded due to small numbers. Residents were also excluded if they had a SARS-CoV-2 infection in the prior 90 days, planned discharge according to their most recent assessment, received monoclonal antibodies in the prior 90 days, were under hospice care, or had already received a third vaccine dose prior to the index date.

We examined four post-booster outcomes: (1) any incident SARS-CoV-2 infection; (2) hospitalization (3) death, and (4) a combined endpoint of hospitalization or death. All study outcomes were obtained via the electronic health record from the two systems.

polymerase chain reaction test occurring after the index trial date, and with no test confirming infection in the prior 90 days. Testing results were extracted from the laboratory results of either system. 10 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted January 28, 2022. SARS-CoV-2 associated hospitalization defined as transfer to an acute care hospital occurring within 21 days of (+) SARS-CoV-2 test. Transfers were identified through census records in system 1 and bed-section codes identifying acute care episodes in system 2, which has a single EHR system for both nursing homes and hospitals.

SARS-CoV-2 associated death defined as death occurring within 30 days of a (+) SARS-CoV-2 test. Deaths in System 1 were identified through transfers to funeral homes, or death records in the daily census. System 2 deaths were identified through a combination of census and vital status records.

Immunization records in each EHR were used to identify primary vaccination and booster events. 11 We indexed additional resident covariates separately for each eligible trial date (i.e. a person's baseline covariates may differ for each trial, as each trial represents a different index date). Data was obtained from the electronic health record in both systems. In system 1, additional information from the nursing home minimum dataset assessments (MDS) were included. 12 prior to index) for chronic condition information. These were classified using the Elixhauser comorbidity classification system. 13 We characterized residents as immunocompromised if they had received an immunosuppressive medication or a qualifying condition diagnosis (detailed in supplementary appendix Table S1 ). Finally, . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 28, 2022. ; https://doi.org/10.1101/2022.01.25.22269843 doi: medRxiv preprint we included measures of overall facility testing SARS-CoV-2 positivity rates in the 14day period prior to each index date.

Nested trial method The analytic approach follows a general daily nested target trial methodology. 8 The purpose of the approach was to provide a flexible method for assigning "time-zero" to unboosted residents, while being maximally inclusive of followup data for unboosted residents (up to the time they are boosted). When a resident receives a booster vaccination on a trial date, they are assigned to the booster arm.

Once a resident is boosted, they are ineligible for future trial dates, and therefore the booster arm does not contain repeated observations. Unboosted residents might be eligible as controls for multiple emulated trial dates up until they received a booster. We randomly selected one eligible target trial date per control resident. 8 

Both cohorts used the same analytical approach. Each event was modeled using the Kaplan-Meier estimator, with cumulative incidence (i.e., risk) measured as 1probability of event free survival at each time point (day). The relative ratio of the cumulative incidence curves between groups (boosted, unboosted) at each time point was computed, with 1 -relative risk reported as the conventional "vaccine effectiveness" . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 28, 2022. ; https://doi.org/10.1101/2022.01.25.22269843 doi: medRxiv preprint statistic. The cumulative incidence at day 42 (six weeks) was selected as the primary follow-up time of interest for reporting. We conducted a complete case analysis and did not impute missing data as the number of observations excluded for missing data was very small (<1% observations dropped). Final statistical models were selected by comparison of covariate imbalances between groups or a priori based on experience (i.e. gender, facility-level infection rates). We estimated a stabilized inverse probability weight for treatment (IPTW) adjusting for: the facilities' prior 90-day hospitalization rate, restricted cubic splines for the number of resident-level SARS-CoV-2 tests, and facility positivity rate in the prior 14 days, state (System 1 only), primary vaccination series manufacturer, month of trial date (September versus October, November), resident immunocompromised status, age, and race/ethnicity. System 2 also adjusted for a diagnosis of paralysis, alcohol, weight-loss and blood-loss. These variables were selected by comparing between groups using standardized mean differences (SMD), with an absolute SMD<0.1 after adjustment as a guide for achieving acceptable balance between groups. Weights were truncated at their 99% upper quantile. Sampling with replacement by resident (i.e. bootstrapping) with 2000 replications was used to generate 95% percentile intervals accounting for the estimated probability weights and crossover of treatment groups by resident. Initial data preparation was done with SAS Version 9.4 (Cary, NC) and STATA version 16 (Statacorp, TX). All analyses were performed using R statistical software, Version 4.0.1, Vienna, Austria.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 28, 2022. ;

Tables S2, S3 and Figure CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 28, 2022. ; https://doi.org/10.1101/2022.01.25.22269843 doi: medRxiv preprint were similar (standardized mean differences <0.1) by most observed characteristics in both System cohorts. The primary difference between treatment groups observed was at the facility level, where unboosted controls were more likely than boosted to be facilities with a SARS-CoV-2 infection in the prior 14 days (17.3% vs. 11.2%, aSMD 0.17). Supplementary figures S4 and S5 describe the propensity score and treatment weights.

In Figure 2 we present cumulative incidence curves for the four COVID-19 outcomes up to 6 weeks (day 42) for System 1. Table 3 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 28, 2022. 

This study evaluated the clinical effectiveness of an mRNA SARS-CoV-2 vaccine booster in two large multi-state populations of nursing home residents. We sought to estimate average booster vaccine effectiveness comparable to trial eligibility criteria. We found that those who received a booster dose had a substantial reduction in risk of SARS-CoV-2 infection and death in System 1, while in System 2 there was a similar reduction in infection but we were unable to exclude null effects for other endpoints. In the larger System 1, one SARS-CoV-2 associated death occurred following boosting, versus 15 deaths in non-boosted individuals, but smaller sample size and event counts made inference on death events in System 2 difficult. While some event rates are low, and magnitude of effects differ, the findings between cohorts are overall consistent in that those receiving boosters had a significant reduction in risk of SARS-CoV-2 infection.

To evaluate the robustness of vaccine effectiveness, we examined two significantly different nursing home populations; one a large, private U.S. provider of nursing home care and the other a public, national healthcare system caring specifically for elderly Veterans. These two health systems are demographically different (e.g. few women are present in VA CLCs), and represent two different administrative healthcare systems. In particular, the rates and reasons for acute hospitalization may be different between two populations due to differences in availability of hospice care, advanced planning and provider practice. It should also be noted the sample size for the VA CLCs is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 28, 2022. ; https://doi.org/10.1101/2022.01.25.22269843 doi: medRxiv preprint the VE or pooling of results is inappropriate. However, consistency in overall vaccine effectiveness to prevent infections in these two disparate cohorts lends credibility to the importance of boosters for the overall U.S. nursing home population.

There are few other studies of booster effectiveness in nursing home residents to date. Bar-On et al reported observational data in Israeli community-dwelling adults aged 60 and older boosted 5 months after primary vaccinations. 14 Comparing confirmed infections and adjusting for person-days at risk, older adults had about 88% fewer confirmed infections in the 12 to 42 days following the boost when the Delta variant was the predominant virus in circulation. In a second similar approach for comparison, Israelis aged 16 and older followed from 12 to up to 60 days following boost, experienced 9 to 17-fold fewer infections in person days at risk. 15 From the same period, Muhsen at al reported on over 41,000 Israeli nursing home residents aged 60 years and older boosted ~5 months after the primary series in a three-week campaign. 16 In that analysis, the contemporary comparison included 1.5 million community dwelling adults 60 and older, and younger groups aged 20-59 years and under 20 years. The nursing home residents had an average age of 81.8, but the average age of the 60 and older comparison group was not provided. Nonetheless, the boosted nursing home residents had a 71% lower rate of infections 80% lower hospitalizations than their unboosted community dwelling contemporaries. These ecological analyses suggest substantial protection by boosting against infection, hospitalization and death, but cannot account for confounding by access, indication or exposure risk in the outpatient comparison populations. Indeed, the outpatient elderly group reportedly had less risk reduction than the nursing home resident group, as did the younger comparison groups.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 28, 2022. ; https://doi.org/10.1101/2022.01.25.22269843 doi: medRxiv preprint

The present work evaluates VE by emulated trial design that takes advantage of comparable controls who differ primarily by when they received vaccine rather than by indication, exposure risk to circulating strains, and residents in facilities across a broad geographic and socioeconomic distribution. We report a slightly lower VE of 50-54% infection reduction comparing boosted and unboosted residents, and included all postbooster days in the exposure risk calculation during which boost-derived immunity may be incomplete. Also, the VE to prevent SARS-CoV-2 mortality in System 1 nursing home residents is substantially larger than reported by the Israeli studies, exceeding 90%. These differences could arise from differences in methods, population health, virus characteristics or other causes, but most likely from the first two.

Our study includes several significant limitations. To identify treatment effects, we assumed no unmeasured confounding and correct specification of probability weights in the Kaplan-Meier estimator. The relatively small differences between groups are encouraging and may arise from subjects' eligibility requirement for prior primary series vaccination, resulting in greater cohort homogeneity. Additionally, the plots of our cumulative incidence curves for infection show close overlap in the first 7-10 days for groups. This suggests that we face relatively low risk of residual confounding if the two groups have similar baseline event risk in the window after vaccination but prior to the quickly developing post-booster immunity. A significant limitation that we do not account for is interference, or the assumption that one resident's vaccination does not impact the effectiveness of vaccination for other residents. Subsequent work which accounts for this could lead to significantly different estimates as increasing proportions of boosted residents and staff protect unboosted residents from infection within a facility. Treatment . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 28, 2022. ; effects assuming no interference likely underestimate the total causal effect.

Additionally, deaths in the nursing home are easily identified, but for residents who are transferred, they may die after they are lost to follow-up, in which case we cannot account for them in our data. collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 28, 2022. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 28, 2022. Tables   . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 28, 2022. 

Kaplan-Meier estimator, cumulative incidence, vaccine effectiveness at specific timepoints, day 42 (6 weeks)

Description. This table outlines the hypothetical design of a randomized trial in nursing home residents, defining the intervention, inclusion/exclusion criteria and outcomes. The right-most columns highlights where the emulated data for each cohort deviates from this design.

.

It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 28, 2022. ) 28 (0.34) 10 (0.11) Description. VE -vaccine effectiveness, 1 -risk ratio. Calculated from Kaplan-Meier estimators (see Figure 1 and methods for more detail). a -K-M estimator is weighted with stabilized inverse probability treatment weights. b -Risk per thousand patient-days. c -Omitted 95% CI due to low event counts, inference is limited.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted January 28, 2022. ; https://doi.org/10.1101/2022.01.25.22269843 doi: medRxiv preprint

Nursing Home Staff Vaccination and Covid-19 Outcomes

COVID-19 Nursing Home Data. Last updated: December 23, 2021. Centers for Medicare and Medicaid Services

Significant reduction in vaccine-induced antibody levels and neutralization activity among healthcare workers and nursing home residents 6 months following COVID-19 BNT162b2 mRNA vaccination

The Advisory Committee on Immunization Practices' Interim Recommendations for Additional Primary and Booster Doses of COVID-19 Vaccines -United States

COVID-19) Update: FDA Takes Additional Actions on the Use of a Booster Dose for COVID-19 Vaccines. United States Food and Drug Administration

Nursing Home COVID-19 Vaccination Data Dashboard

Using Big Data to Emulate a Target Trial When a Randomized Trial Is Not Available

The value of explicitly emulating a target trial when using real world evidence: an application to colorectal cancer screening

Accuracy of ICD-10 codes for identifying skilled nursing facility residents with lab-confirmed COVID-19

Incident SARS-CoV-2 Infection among mRNA-Vaccinated and Unvaccinated Nursing Home Residents

The Minimum Data Set 3.0 Cognitive Function Scale

Identifying Increased Risk of Readmission and In-hospital Mortality Using Hospital Administrative Data: The AHRQ Elixhauser Comorbidity Index

Protection of BNT162b2 Vaccine Booster against Covid-19 in Israel

Protection against Covid-19 by BNT162b2 Booster across Age Groups

Effects of BNT162b2 Covid-19 Vaccine Booster in Long-Term Care Facilities in Israel

%) 5046 (35.5%) 3093 (36.2%) 1953 (34.4%) 0.04 Length of stay in days

SARS-CoV-2 history

5%) 961 (11.3%) 676 (11.9%) 0.02 Pulmonary disease

Diabetes mellitus (any), no. (%)

Cognitive-function scale score, median (Q1, Q3) c

SARS-CoV-2 (+) in prior 14 days 2116