Staffing Patterns in US Nursing Homes During COVID-19 Outbreaks

This cohort study quantifies changes in nursing home facility staffing during and after a severe COVID-19 outbreak.

eAppendix. Additional detail on data and methods eTable. Summary statistics of severe outbreak sample compared to full sample eFigure 1. Histogram of total outbreak cases per bed for all outbreaks, and definition of severe outbreak eFigure 2. New hires during a severe outbreak, contract and non-contract eFigure 3. Staffing measures during outbreaks, by outbreak decile eFigure 4. Staffing measures during a severe outbreak, using alternative definition of severe outbreaks (total cases) eFigure 5. Staffing measures during a severe outbreak, after adding additional controls of county COVID-19 prevalence eFigure 6. Staffing measures during a severe outbreak, estimated using Callaway Sant'Anna estimators This supplemental material has been provided by the authors to give readers additional information about their work.

Staffing data and staffing measures
We first cleaned the data by keeping the last period of continuous reporting with consistent employee identifiers for each facility. We keep the last continuous period of reporting for each facility to avoid any measurement issues related to facilities not reporting data in a particular quarter (which would prevent us from being able to accurately identify new hire, absence, and departure weeks). Reporting rates in most quarters are fairly high; however, due to the COVID-19 pandemic, CMS made reporting optional for the first quarter of 2020, resulting in 19% fewer submissions than the previous quarter. Our restriction means that for these facilities, we keep only the data starting in Q2 2020. We assume there has been a change in the assignment of employee identifiers as facilities where more than 75% of the present staff in a given week is newly hired, and the overall staff size has not changed by more than 25%. In this case, we again keep only the period following the change in employee identifiers, so that we are able to measure new hires, absences, and departures.
To avoid issues related to left-or right-censoring of the data at the beginning and end of the available data, we construct our new hire, absence, and departure measures using a 12-week lookback and a 12-week follow-up period. Thus, our final dataset contains these measures from April 1, 2017 through March 31, 2021 (12 weeks after the start of the available data to 12 weeks before the end of the available data). Specifically, we define new hires as someone who is working for the first time in 12 weeks, departures as someone who is working their last week for the next 12 weeks, and absences as people who did not work in a given week, but will return within 12 weeks from the last week in which they worked.

Case data and outbreak definition
We use data from the National Healthcare Safety Network (NHSN) COVID-19 Nursing Home Database to identify large outbreaks. This data consists of weekly reports of facility COVID-19 cases and deaths for residents and staff. The first data submission was on May 24, 2020, meaning that we are not able to identify outbreaks in the first two months of the pandemic. We also exclude the first submission of data because for that week, facilities had the option of either reporting retrospective data on cases and deaths from COVID-19 dating back to January 1, 2020, or simply reporting new cases and deaths from the past week. We thus have new weekly cases and deaths beginning with the week of May 31, 2020.
We define an outbreak as any period of consecutive weeks of positive COVID-19 (staff or resident) cases that is preceded and followed by at least two weeks with no new cases. This means that two periods of cases that are separated by only one week with no new cases are counted as the same outbreak, rather than two different outbreaks. We define the outbreak's start as the first week of cases within such a period. Because our analysis is indexed by the start of the outbreak, we exclude any outbreaks that were ongoing at the beginning of the available data (May 31, 2020). Thus the earliest outbreak start date in our analysis is June 14, 2020 (two weeks after the start of the data). We consider all outbreaks that started between this date and January 1, 2021.
For each of these outbreaks, we compute the total number of resident and staff cases at that facility during the outbreak. We label the top decile of these as severe outbreaks, and use this as our exposure measure.

Analysis sample definition
Using the NHSN data, we identify 39,485 outbreaks in 15,171 facilities. We restrict our analysis to the top decile of these outbreaks in terms of total cases per bed, corresponding to 3,939 outbreaks for 3,875 facilities. Of these, we keep only the first outbreak for each facility (dropping N=64 outbreaks) and also drop N=908 outbreaks for which we do not observe the facility in the PBJ staffing data in the week of the outbreak start. This results in a final analysis sample of N=2,967 outbreaks and N=456,029 facility-weeks between January 1, 2017 and March 31, 2021. Notes: Average outbreak, facility, and pre-outbreak weekly measures for full sample of outbreaks (column 1), severe outbreaks (column 2), analysis sample of severe outbreaks which is eFigure 1. Histogram of total outbreak cases per bed for all outbreaks, and definition of severe outbreak Notes: This histogram calculates total outbreak cases per bed for N=39,485 outbreaks using the NHSN data. The red line indicates the top decile of these outbreaks, which corresponds to the definition of a "severe outbreak" in the text (outbreaks that exceeded 0.98 total resident and staff cases per bed). We dropped 30 outliers with more than 5 cases per bed from this graph for ease of presentation, although they are included in the analysis sample.