key: cord-0744855-cpkmk07m
authors: McGarry, Brian E.; Barnett, Michael L.; Grabowski, David C.; Gandhi, Ashvin D.
title: Nursing Home Staff Vaccination and Covid-19 Outcomes
date: 2021-12-08
journal: N Engl J Med
DOI: 10.1056/nejmc2115674
sha: 9d77e638f828dd2e374655fbdbacfde1f65cb29c
doc_id: 744855
cord_uid: cpkmk07m

nan

In May of 2021, CMS began requiring nursing homes to report resident and staff vaccination rates on a weekly basis with penalties imposed on facilities that were not in compliance by the week ending June 18, 2021. 2 Facilities report the percent of residents (who stayed at least 24 hours in the nursing home during the week of data collection) who received We obtained information about nursing homes, including overall quality ratings, profit status, and bed size from the 2021 Nursing Home Compare Provider Information dataset. 3 Additional information about nursing homes, including the racial/ethnic composition of residents, the share of residents with Medicaid, and chain affiliation were obtained from the 2017 Certification and Survey Provider Enhancement Reports (CASPER) system and 2017 Minimum Data Set (MDS) assessments, both through the National Institute on Aging-funded LTCFocus.org website. 4 Staff information was obtained from individual-level staffing records submitted by nursing homes during the second quarter of 2021 as part of CMS's mandatory payroll-based journal (PBJ) electronic staffing reporting system. 5, 6 Information on weekly county COVID-19 case rates were obtained from the NYTimes database. 7 

Outcome variables were the weekly cumulative number of confirmed COVID-19 cases among residents and staff and the number of COVID-19 deaths among residents. Outcomes were operationalized as rates per 100 nursing home beds.

The primary independent variable was the percent of healthcare staff with a completed COVID-19 vaccine as of the week ending June 13, 2021. This week was the first in which more than 75% of nursing homes reported staff vaccination rates, likely in anticipation of CMS penalties for non-compliance in the following week. Nursing homes were classified into quartiles of staff vaccine coverage using reported rates in this week. In using quartiles, we allow for nonlinear relationships between levels of staff vaccination coverage and the outcomes of interest while facilitating a meaningful comparison between facilities with "high" and "low" baseline staff vaccination rates. In a sensitivity analysis, we show the robustness of our findings to using continuous specification for staff vaccine coverage that include a second-degree polynomial to allow for non-linear relationships.

County COVID-19 prevalence was defined as the cumulative rate (per 100,000 population) of new COVID-19 cases for the weeks ending June 13, 2021 through August 22, 2021. Counties were classified into quartiles of COVID-19 prevalence using this measure.

Covariates included the following facility characteristics: profit status (including for-profit, non-profit, and government owned), overall 5-star quality ratings, the percent of nursing home residents who were non-white race (categorized into quartiles), the percent of nursing home residents with Medicaid (categorized into quartiles), facility size (categorized as 1-50 beds, 51-100 beds, 101-150 beds, 151-200 beds, and 200+ beds), overall staff size (defined as the average number of unique employees who worked in a nursing home per week, categorized into quartiles), part-time staff size (defined as the average number of unique employees who worked less than 35 total hours in a facility per week, categorized into quartiles), direct care staffing levels (defined as the number of direct care hours per resident day, categorized into quartiles), and direct care skill mix (defined as registered nurse hours per resident day, categorized into quartiles).

We also constructed two measures that capture nursing homes' prior experience with COVID-19 and the level of naturally acquired immunity among residents and staff. Specifically, we measured the cumulative COVID-19 case rate resident and staff between June 7, 2020 and May 30, 2021. Both measures were categorized into quartiles.

Finally, we account for time-variant community incidence of COVID-19 cases with a weekly measure of new COVID-19 cases per 100,000 population at the county level.

In a sensitivity analysis, we show the robustness of our findings to using continuous covariate controls rather than categorical measures reflecting quartiles. In these specifications, we utilize second-degree polynomials in each control to allow for non-linear relationships.

Additionally, we winsorize these continuous measures at the 1 st and 99 th percentile to address outliers. See results in Table S3. SAMPLE Among the 15,171 nursing homes in the NHSN data, we restricted the sample to those who reported staff vaccine coverage for the week ending June 13, 2021 and who reported COVID-19 outcomes in each week of the study period. These restrictions result in a final sample of 12,364 nursing homes (81% of total). We followed outcomes through the week ending August 22, 2021 to avoid confounding due to the Biden Administration announcing plans for a federal vaccine mandate and several large nursing home chains introducing staff vaccine mandates in late August.

To estimate the association between baseline staff vaccination rates and weekly COVID-19 outcomes, we fit linear regression models. The key independent variable of interest was an interaction term between staff vaccine coverage quartiles and county COVID-19 prevalence.

This term allows for the comparison of outcomes between nursing homes with differing levels of staff vaccine coverage given different levels of community spread. This interaction term was further interacted with indicators for the 11 weeks in our study window, allowing us to flexibly model the relationship of between staff vaccine coverage (for a given level of community COVID-19 prevalence) and outcomes over time. Models included the covariates described previously, as well as county fixed effects which restrict comparisons to nursing homes located within the same county. Robust standard errors were clustered at the facility level. Our approach is summarized in Equation 1.

, , = + 1 ( * * ) + 2 * + 3 + 3 , + , , Eq. 1 , , represents the COVID-19 outcome of interest for facility located in county in week .

represents a vector of indicators for each week in our study sample, is a set of indicators for quartiles of baseline staff COVID-19 vaccine coverage, and is a set of indicators for quartiles of county COVID-19 prevalence. is a vector of time-invariant facility-level controls and , represent the county COVID-19 new case rate in county in week , and is a county fixed effect term. In a sensitivity analysis, we replace the county fixed effect with a county-level random effect. See table S3 for results. We use heteroscedasticity-robust standard errors clustered at the facility level. , , is an error term that is allowed to be heteroscedastic, as well as correlated for observations within the facility. In other words, we use heteroscedasticityrobust standard errors that are clustered at the facility level.

We used our fitted model to produce the weekly adjusted estimates and 95% confidence bands presented in the manuscript Figure. To estimate the number of preventable cases and deaths, we again used the fitted model to predict the outcomes using facilities' observed levels of staff vaccine coverage and compared these estimates to predicted outcomes under the condition of all nursing homes being the top category of staff vaccine coverage. 

Estimates obtained from linear regression containing a 3-way interaction between weekly indicators, indicators for quartiles of county COVID-19 prevalence, and indicators for quartiles of baseline staff vaccination rates. Column (1) reflects the results presented in the letter Figure. Column (2) replace county-level fixed effects with random effects. Column 3 uses winsorized continuous covariates and second order polynomials (as opposed to quartile categories) for the following covariates: weekly county COVID-19 case rates, staff and resident prior COVID-19 infection rates, resident vaccination rates at baseline, the share of nursing home residents with Medicaid and who are non-White race, direct care and nursing staff-to-resident ratios, and overall-and part-time staff sizes. REFERENCES 1. Centers for Medicare & Medicaid Services. COVID-19 Nursing Home Data

Medicare requires nursing homes to report COVID vaccinations

Nursing Home Compare Data Web site

Daily Nursing Home Staffing Levels Highly Variable

High Nursing Staff Turnover in Nursing Homes Offers Important Quality Information

The New York Times. Coronavirus in the U.S