Advertisement

Associations Between Daily Nurse Staffing Levels and Daily Hospitalizations and ED Visits in Nursing Homes

Open AccessPublished:August 07, 2022DOI:https://doi.org/10.1016/j.jamda.2022.06.030

      Abstract

      Objectives

      Although many prior studies have shown that high average levels of nurse staffing in nursing homes are associated with fewer hospitalizations, some studies have not, suggesting that the average nursing level may mask a more complex relationship. This study examines this issue by investigating the associations of daily staffing patterns and daily hospitalizations and emergency department (ED) visits.

      Design

      Retrospective analyses of national Payroll Based Journal (PBJ) staffing data merged with the Minimum Data Set.

      Setting and Participants

      A total of 15,718 nursing homes nationally reporting PBJ data during 2017–2019, their staff, and residents.

      Methods

      We estimated facility-day-level models as conditional facility fixed-effect Poisson regressions with robust standard errors. The dependent variables were daily numbers of hospitalization and ED visits and the independent variables of interest were the number of registered nurse (RN), licensed practical nurse (LPN), and certified nurse assistant (CNA) hours on the same and prior days.

      Results

      The daily number of hospital transfers averaged 0.28 (SD 0.21). Daily total direct-care staffing hours averaged 288.7 (SD 188.2), with RNs accounting for 35.0, LPNs for 68.7, and CNAs for 185.0. Higher staffing was associated with more hospitalizations on the concurrent day. Higher staffing on the day prior was associated with fewer hospitalizations. The effect size was larger for RNs and LPNs (same day = ∼2%; prior day = approximately −0.7% to −0.9%) than for CNAs (same day <1%; prior day < −0.5%). ED visits not leading to hospitalizations, and analyses for subsamples exhibited similar findings.

      Conclusions and Implications

      Our findings suggest that staff can address developing problems and prevent admissions the next day and identify emergent problems and hospitalize the same day. They also underscore the complex array of nursing home factors involved in hospitalization and ED visits, including the influence of daily staffing variation, suggesting the need for further research to better understand the associations between staffing and appropriate resident transfers to the hospital or the ED, and the potential implications for quality metrics in these domains.

      Keywords

      Nursing homes are expected to provide most of the medical care their residents require in-house and minimize hospitalizations. Although some conditions, like acute myocardial infarction, almost always require hospitalization, many not only can be treated in the nursing home, but hospitalizing the resident may actually increase the risk of iatrogenic events and discontinuity of care contributing to further deterioration in the resident's condition, functional decline, and delirium.
      • Creditor M.C.
      Hazards of hospitalization of the elderly.
      • Dwyer R.
      • Gabbe B.
      • Stoelwinder J.U.
      • Lowthian J.
      A systematic review of outcomes following emergency transfer to hospital for residents of aged care facilities.
      • Ouslander J.G.
      • Lamb G.
      • Perloe M.
      • et al.
      Potentially avoidable hospitalizations of nursing home residents: frequency, causes, and costs: [see editorial comments by Drs. Jean F. Wyman and William R. Hazzard, pp 760–761].
      Therefore, a low hospitalization rate has long been considered a marker of high-quality nursing home care. Measures of appropriate hospitalizations were included in the Centers for Medicare and Medicaid Services (CMS) nursing home value-based payment demonstration
      CMS.gov
      Nursing home value-based purchasing demonstration.
      and measures of hospitalizations are included as quality indicators in the Nursing Home Care Compare report card published by CMS.,
      • Saliba D.
      • Weimer D.L.
      • Shi Y.
      • Mukamel D.B.
      Examination of the new short-stay nursing home quality measures: rehospitalizations, emergency department visits, and successful returns to the community.
      Factors influencing hospitalizations have been studied extensively, with many studies examining staffing levels as explanatory variables, because staffing, whether registered nurses (RNs), licensed practical nurses (LPNs), or certified nurse assistants (CNAs), provide direct day-to-day patient care and are likely to be intimately involved in the care processes and decisions leading to hospitalizations. These studies, summarized by several recent reviews,
      • Laging B.
      • Ford R.
      • Bauer M.
      • Nay R.
      A meta-synthesis of factors influencing nursing home staff decisions to transfer residents to hospital.
      • Backhaus R.
      • Verbeek H.
      • van Rossum E.
      • Capezuti E.
      • Hamers J.P.
      Nurse staffing impact on quality of care in nursing homes: a systematic review of longitudinal studies.
      • Spilsbury K.
      • Hewitt C.
      • Stirk L.
      • Bowman C.
      The relationship between nurse staffing and quality of care in nursing homes: a systematic review.
      • Grabowski D.C.
      • Stewart K.A.
      • Broderick S.M.
      • Coots L.A.
      Predictors of nursing home hospitalization: a review of the literature.
      • Castle N.G.
      Nursing home caregiver staffing levels and quality of care: a literature review.
      • Bostick J.E.
      • Rantz M.J.
      • Flesner M.K.
      • Riggs C.J.
      Systematic review of studies of staffing and quality in nursing homes.
      revealed mixed evidence about the relationship between staffing and hospitalization, with some failing to identify the expected association of higher staffing levels with lower hospitalization rates. The authors of these reviews identified a set of common limitations as likely reasons for the failure to find the expected associations between staffing and hospitalizations. These limitations include reliance on staffing data from the CMS Online Survey Certification and Reporting (OSCAR) System or the system that replaced it, the Certification and Survey Provider Enhanced Reporting (CASPER), both of which collect staffing data once a year and may be inaccurate,
      • Kash B.A.
      • Hawes C.
      • Phillips C.D.
      Comparing Staffing Levels in the Online Survey Certification and Reporting (OSCAR) system with the Medicaid cost report data: are differences systematic?.
      ,
      • Feng Z.
      • Katz P.R.
      • Intrator O.
      • Karuza J.
      • Mor V.
      Physician and nurse staffing in nursing homes: the role and limitations of the Online Survey Certification and Reporting (OSCAR) system.
      the cross-sectional nature of most studies, and the lack of statistical control for other variables that might be important, such as presence of other providers paid by the facility (eg, nursing administrators, nurse practitioners, medical directors), and inability to distinguish between employed and contract staff.
      We present analyses that address some of these prior limitations by using a new dataset, the Payroll Based Journal (PBJ), which provides daily staffing data based on each nursing home's payroll information, and is more accurate and more detailed than prior data. We take the novel approach of analyzing day-to-day patterns of hospitalizations as they relate to the daily staffing levels of RNs, LPNs, and CNAs for all nursing homes in the United States, as well as facilities at the top and bottom of the staffing distribution, offering a complementary perspective to prior studies. We test the hypothesis that the observed associations between average higher staffing and fewer hospitalizations will also be observed between same-day and prior-day staffing and hospitalizations.

      Methods

      Sample

      The sample included all 15,718 Medicare- and Medicaid-certified nursing homes in the country during the period January 2017 through August 2019. The analysis for emergency department (ED) visits not followed by a hospitalization was based on outpatient claims files, which are limited to Medicare Fee-for-Service (FFS) residents and included 15,608 nursing homes (99.3%).

      Data Sources

      The PBJ includes the number of hours paid daily for each staff type, including RNs, LPNs, CNAs, therapy staff, physicians, administrators, and others, by nursing home. It also provides daily patient census.
      We merged the PBJ data with resident-level data, calculated from the Minimum Data Set (MDS) 3.0, at the facility-day level. The MDS includes assessments for all nursing home residents with information about age, gender, case mix [Resource Utilization Group (RUG) IV], cognitive impairment, Alzheimer or related dementias (ADRD) diagnoses, and dates of hospitalization and death. For some analyses we also used the Medicare outpatient claims for ED visits not followed by a hospital admission. We also merged in the Master Beneficiary Summary File, to identify residents enrolled in Medicare Advantage during their nursing home stays, the Nursing Home Five-Star Quality Rating data, and facility characteristics reported in Long-Term-Care Focus.
      LTC Focus, Brown University
      Facts on Care in the US - Create Custom Reports on Long-Term Care.

      Analyses

      The study was approved by the University of California, Irvine, institutional review board.
      Analyses were conducted at the facility-day level. Time-dependent variables were calculated for each nursing home, as counts or averages for each day.

      Dependent Variables

      The dependent variables were the count of daily discharges from the nursing home to an acute care or psychiatric hospital as reported on the MDS discharge or transfer to the ED not followed by a hospital admission as reported on the Outpatient claim.

      Daily-Varying Independent Variables

      There were 6 independent variables of interest: RN, LPN, and CNA hours on the transfer day to the hospital (or the ED) and RN, LPN, and CNA hours on the day before transfer. For example, when examining the number of transfers on July 1, we measure staffing on that same day (July 1) and on the preceding day (June 30).
      Daily control variables included percent male residents in the facility on the day of transfer; percent residents who were younger than 65, 65 to 74, and 75 to 84, with 85+ as reference, and percentage of residents enrolled in Medicare Advantage on the transfer day. Case mix on the day of transfer was controlled by 66 variables. Each variable corresponds to 1 of the 66 RUGs categories and indicates the number of residents classified into that RUG on that date based on their most recent MDS assessment. Also included were a set of monthly indicator variables controlling for time trends.

      Stratification Variables for Subanalyses

      We performed stratified analyses to examine hypotheses about different patterns of associations between staffing and hospitalizations in facilities with specific characteristics, as follows: (1) facility quality measured by the 5-Star Quality Measures (QMs) ratings and the 5-Star Survey ratings published in Nursing Home Care Compare and averaged over the study period; (2) average total direct-care nursing (sum of RNs, LPNs, and CNAs) hours over the study period; (3) average hospitalizations over the study period; (4) for-profit and nonprofit ownership; (5) percentage of residents covered by Medicare; and (6) percentage of residents with ADRD and cognitive impairment.

      Estimated Models

      We estimated a facility-day-level model, with hospitalization on each day as the dependent variable and nursing hours by type (RN, LPN, and CNA) on the same day and on the day prior as the 6 independent variables of interest, controlling for all other variables described previously: age, sex, case-mix count, all of them calculated for the transfer day, and the month. The model was estimated as a conditional facility fixed-effect, Poisson regression, with robust standard errors. The facility fixed-effects approach controls for all time-invariant differences between nursing homes (eg, ownership).
      We estimated a base case model on the full sample. We then stratified the sample by facility characteristics as described previously, and estimated models for subsamples defined by the bottom and top 30th percentiles of the distribution for each characteristic. We also estimated a full sample model that separated employed and contract nursing staff and a full sample model that added other personnel types (if salaried by the facility) including hours of director of nursing-RN, RN administrators, LPN administrators, medical directors, other MDs, nurse practitioners, nurse aides in training, and medication aides.
      We present the regression results for the staffing variables as a change in the percentage of daily hospitalizations associated with a 1 full-time equivalent (ie, 8 hours) increase in the staffing variable, holding all other variables constant. (To calculate this percentage, we used the formula %Δ(y|x) = 100∗(exp(β∗Δx), where %ΔE(y|x) = expected % change in daily hospitalizations.) The P values reported are for the actual coefficients.

      Sensitivity Analyses

      To test the robustness of our findings to different assumptions we performed several sensitivity analyses:
      • To account for possible mismatches in dates between the hospital and ED data and the nursing home data, we estimated models in which transfer was matched more liberally, within 1 day of the discharge from the nursing home.
      • To investigate sensitivity to the number of prior days' staffing included, we estimated models with (1) no variables for staffing on prior days, and (2) 2 variables for staffing 1 and 2 days prior.
      • To test the sensitivity of our stratified analyses to the definition of the strata threshold, all analyses were repeated with thresholds set to 85% and 15%.
      • To test whether our results are driven by the known phenomenon of lower staffing on weekends
        Centers for Medicare and Medicaid Services
        Design for care compare nursing home five-star quality rating system: technical users’ guide-January 2022.
        or on Mondays and Fridays (as observed in our data, see later in this article), we performed an analysis limited to hospitalizations during midweek only (Tuesdays, Wednesdays, and Thursdays).

      Results

      Table 1 presents descriptive statistics for the analytic sample. The average daily number of hospital transfers was 0.28 (SD 0.21) and transfers to the ED not followed by a hospitalization was 0.11 (SD 0.07). Average daily total direct-care staffing hours (including employed and contract staff) was 288.7 (SD 188.2). Of those, RNs accounted for 35.0, LPNs for 68.7, and CNAs for 185.0 hours, mostly accounted for by employed staff.
      Table 1Descriptive Statistics
      Number of Nursing HomesMeanSD
      Dependent Variables: Measured Daily
       Discharges from nursing home to acute care or psychiatric hospital15,7180.280.21
       ED visits not followed by a hospitalization15,6080.110.07
      Independent Variables: Nursing Homes and Resident Variables Measured Daily
       RN hours per day15,71835.033.3
      Salaried15,71834.032.2
      Contract15,7181.05.6
       LPN hours per day15,71868.748.8
      Salaried15,71866.447.0
      Contract15,7182.37.6
       CNA hours per day15,718185.0124.8
      Salaried15,718180.0121.0
      Contract15,7185.017.9
       Total direct-care nurse staffing - per day15,718288.7188.2
       Director of nursing (RN) per day15,7185.01.7
       RN Administrator per day15,71811.712.5
       LPN administrator hours per day15,7186.08.7
       Nurse aide in training hours per day15,7182.67.3
       Medication aide hours per day15,7186.614.8
       Medical director hours per day15,7180.40.7
       Other MD hours per day15,7180.22.0
       Nurse practitioner hours per day15,7180.31.2
       Daily resident census15,71885.252.8
       % of residents in Medicare Advantage15,71826.119.0
       % of residents who are male15,71835.711.8
       % of residents younger than 6515,71815.814.7
       % of residents aged 65–7415,71818.58.0
       % of residents aged 75–8415,71827.06.8
       % of residents aged 85 and older15,71838.718.0
      Facility Characteristics Measured Quarterly or Annually
       Average QMs 5-star rating (1–5)13,4223.81.0
       Average survey 5-star rating (1–5)13,4552.81.1
       % of residents with Medicare payer14,96013.313.4
       % of residents with Medicaid payer14,96060.123.5
       % of residents with ADRD and cognitive impairment diagnoses15,71847.716.2
      Number of Nursing HomesPercent
      For-profit ownership10,47970.1
      Nonprofit ownership448130.0
      Figure 1 depicts daily staffing variations. It shows, separately for RNs, LPNs, and CNAs, the average number of hours per resident-day for each day of the week as a percent of the number of hours per resident-day averaged over the week, thus identifying days with staffing above and below average. All 3 staff types exhibit the same pattern, with the highest in midweek—Tuesday through Thursday—with approximately 2 to 5 percentage points decline on Friday and Monday, and the largest decline on the weekend, of more than 10 percentage points from the midweek high. The biggest decline is for RNs and the smallest for CNAs. The figure also shows the daily variation in hospitalizations as percent of its weekly average. Hospitalizations follow a similar pattern to that of staffing, with a lower percent during the weekend compared with weekdays. The range of the differences between weekdays and weekend is larger for hospitalizations than for staffing, at approximately 25–30 percentage points.
      Figure thumbnail gr1
      Fig. 1Staffing per resident-day and hospitalizations by day of the week as percent of the weekly average. The weekly average is calculated separately for each of the 4 variables depicted in the chart. Example: On Sunday, hospitalizations are 81% of the hospitalization weekly average, RNs are at 91% of the RN weekly average, LPNs are at 95.5% of the LPN weekly average, and CNAs are at 96% of the CNA weekly average.
      Table 2 presents the results of the regression models, examining the relationships between staffing and hospitalizations for the base case, which includes the full sample, all nursing homes and all residents, followed by results of 2 models with different staffing specifications and then the stratified models. The table presents, for each model, the daily percent change in number of hospitalizations for a 1 full-time staff (FTE) increase. The full base case model is provided in Supplementary Table 1.
      Table 2Summary of Findings: Base Case and Stratified Analyses–Percent Change in Daily Hospitalizations Due to an Increase of a 1 Full-Time Equivalent Staff Position
      Each line reports findings from a separate regression model that also controls for age, gender, case mix, seasonality, and facility fixed effects. Models differ by either staffing variables (eg, salaried vs contract), or facility characteristics (eg, low quality vs high quality).
      Same DayPrevious Day
      RNsLPNsCNAsRNsLPNsCNAs
      Base case, full sample
      2.3% [2.1, 2.4] (<.001)2.1% [2.0, 2.2] (<.001)0.6% [0.5, 0.7] (<.001)−0.8% [−0.9, −0.7] (<.001)−0.8% [−0.9, −0.7] (<.001)−0.4% [−0.4, −0.3] (<.001)
      Models with different specifications, full sample:
       (1) Separate staffing variables for employed and contract staff by type in same model, full sample
      Employed staff2.3% [2.2, 2.4] (<.001)2.1% [2.0, 2.3] (<.001)0.6% [0.6, 0.7] (<.001)−0.8% [−0.9, −0.7] (<.001)−0.8% [−0.9, −0.8] (<.001)−0.4% [−0.4, −0.3] (<.001)
      Contract staff1.7% [1.3, 2.1] (<.001)1.4% [1.2, 1.6] (<.001)0.3% [0.2, 0.5] (<.001)−0.4% [−0.7, −0.1] (.017)−0.0% [−0.2, 0.1] (.698)−0.1% [−0.2, 0.1] (.383)
       (2) Adding variables for non-nursing FTE providers to the model,
      Additional providers included director of nursing-RN, RN administrator, LPN administrator, medical director, other MD, nurse practitioner, nurses aid in training, medication aid.
      full sample
      1.5% [1.3, 1.6] (<.001)1.5% [1.4, 1.6] (<.001)0.0% [−0.0, 0.1] (.149)−0.3% [−0.4, −0.2] (<.001)−0.2% [−0.3, −0.1] (<.001)−0.1% [−0.2, −0.1] (<.001)
       (3) NH Quality measured by 5-Star QMs stratification
      High average quality 5-star rating (≥4.75)2.4% [2.1, 2.7] (<.001)2.1% [1.8, 2.5] (<.001)0.5% [0.3, 0.7] (<.001)−0.9% [−1.2, −0.6] (<.001)−0.8% [−1.0, −0.6] (<.001)−0.3% [−0.4, −0.1] (.001)
      Low average quality 5-star rating (<2.5)2.0% [1.6, 2.4] (<.001)2.0% [1.7, 2.4] (<.001)0.8% [0.7, 1.0] (<.001)−0.6% [−0.9, −0.3] (<.001)−0.8% [−1.0, −0.6] (<.001)−0.3% [−0.5, −0.1] (.001)
       (4) NH quality measured by 5-star survey stratification
      High average survey 5-star rating (≥4)2.4% [2.0, 2.7] (<.001)2.0% [1.6, 2.3] (<.001)0.5% [0.2, 0.7] (0.003)−0.7% [−0.9, −0.4] (<.001)−0.7% [−1.0, −0.5] (<.001)−0.3% [−0.5, −0.1] (.001)
      Low average survey 5-star rating (<2)2.3% [2.1, 2.5] (<.001)2.2% [2.0, 2.4] (<.001)0.7% [0.6, 0.8] (<.001)−0.9% [−1.1, −0.8] (<.001)−0.9% [−1.0, −0.8] (<.001)−0.4% [−0.5, −0.3] (<.001)
       (5) Total nursing level stratification
      High average nursing level (above 70th percentile)2.3% [2.1, 2.5] (<.001)2.0% [1.8, 2.2] (<.001)0.4% [0.2, 0.5] (<.001)−0.9% [−1.1, −0.7] (<.001)−0.8% [−1.0, −0.7] (<.001)−0.2% [−0.3, −0.1] (<.001)
      Low average nursing level (30th percentile and below)2.1% [1.9, 2.4] (<.001)2.2% [2.0, 2.5] (<.001)0.9% [0.8, 1.1] (<.001)−0.7% [−0.9, −0.5] (<.001)−0.9% [−1.0, −0.7] (<.001)−0.5% [−0.6, −0.4] (<.001)
       (6) Hospitalization-level stratification
      High hospitalization rate (above 70th percentile)2.4% [2.2, 2.6] (<.001)2.3% [2.1, 2.5] (<.001)0.7% [0.5, 0.8] (<.001)−0.9% [−1.1, −0.8] (<.001)−0.9% [−1.1, −0.8] (<.001)−0.5% [−0.6, −0.3] (<.001)
      Low hospitalization rate (30th percentile and below)2.0% [1.7, 2.4] (<.001)1.6% [1.3, 1.9] (<.001)0.4% [0.3, 0.5] (<.001)−0.6% [−0.7, −0.4] (<.001)−0.5% [−0.7, −0.3] (<.001)−0.2% [−0.3, −0.1] (<.001)
       (7) Nonprofit vs for-profit nursing homes
      Nonprofit nursing homes1.9% [1.7, 2.1] (<.001)1.4% [1.2, 1.6] (<.001)0.5% [0.4, 0.6] (<.001)−0.7% [−0.9, −0.5] (<.001)−0.6% [−0.8, −0.5] (<.001)−0.2% [−0.3, −0.1] (<.001)
      For-profit nursing homes2.5% [2.3, 2.6] (<.001)2.4% [2.3, 2.5] (<.001)0.7% [0.6, 0.8] (<.001)−0.8% [−1.0, −0.7] (<.001)−0.8% [−0.9, −0.8] (<.001)−0.5% [−0.5, −0.4] (<.001)
       (8) High vs low Medicare census nursing homes
      High Medicare nursing homes (above 70th percentile)2.4% [2.2, 2.6] (<.001)2.2% [2.0, 2.4] (<.001)0.7% [0.5, 0.8] (<.001)−0.8% [−1.0, −0.7] (<.001)−0.9% [−1.1, −0.8] (<.001)−0.4% [−0.6, −0.3] (<.001)
      Low Medicare nursing homes (30th percentile and below)2.0% [1.8, 2.3] (<.001)1.8% [1.6, 2.0] (<.001)0.6% [0.5, 0.7] (<.001)−0.6% [−0.8, −0.4] (<.001)−0.6% [−0.8, −0.5] (<.001)−0.4% [−0.5, −0.3] (<.001)
       (9) Nursing homes with high and low percentage of residents with ADRD and cognitive impairment
      High ADRD (above 70th percentile)2.1% [1.9, 2.3] (<.001)1.9% [1.7, 2.2] (<.001)0.5% [0.4, 0.6] (<.001)−0.7% [−0.8, −0.6] (<.001)−0.7% [−0.9, −0.6] (<.001)−0.3% [−0.4, −0.2] (<.001)
      Low ADRD (30th percentile and below)2.7% [2.5, 2.8] (<.001)2.4% [2.3, 2.6] (<.001)0.8% [0.7, 0.9] (<.001)−1.0% [−1.1, −0.8] (<.001)−1.0% [−1.1, −0.9] (<.001)−0.6% [−0.7, −0.5] (<.001)
       (10) Residents transferred to the ED but not followed by a hospitalization
      Based on outpatient claims files that include FFS residents only1.4% [1.3, 1.6] (<.001)1.5% [1.3, 1.6] (<.001)0.1% [0.1, 0.2] (<.001)−0.4% [−0.5, −0.3] (<.001)−0.4% [−0.5, −0.3] (<.001)−0.1% [−0.1, −0.0] (.023)
      Bold values are statistically significant.
      Each line reports findings from a separate regression model that also controls for age, gender, case mix, seasonality, and facility fixed effects. Models differ by either staffing variables (eg, salaried vs contract), or facility characteristics (eg, low quality vs high quality).
      Additional providers included director of nursing-RN, RN administrator, LPN administrator, medical director, other MD, nurse practitioner, nurses aid in training, medication aid.
      All models exhibit the same general findings. On the same day as the hospitalization, more staff is associated with more hospitalizations, typically with a stronger effect for RNs and LPNs, often approximately 2%, compared with CNAs, typically with less than 1% and often less than 0.5%. More staff on the day prior is associated with fewer hospitalizations, again mostly with stronger effects for RNs and LPNs, mostly between −0.7% and −0.9% and smaller effects for CNAs, less than −0.5%.
      All effects are significant at the 0.001 level, except for 2 models. The model with separate covariates for employed and contract staff has very low previous day effects with P values of 0.02, 0.7, and 0.4 for contract RNs, LPNs, and CNAs, respectively. The model with additional types of non-nursing staff has a very low same day and nonsignificant (P = .15) effect for CNAs.
      The sensitivity analyses described previously have similar findings. A model without any lags shows a positive association between staffing and hospitalization on the same day. Adding 2-days lags show a positive same-day effect and negative effects 1 and 2 days before the transfer. Supplementary Table 2 shows summary results for dyads defined with different thresholds and for the midweek days sample.

      Discussion

      In this study, we evaluated the association between daily staffing and hospitalization and found that it depends on the day of transfer to the hospital. High staffing was protective against hospitalization on the following day, but high staffing measured on the concurrent day with the hospitalization was associated with more hospitalizations. These findings were very robust, observed not only in the full sample and base case, but also in all other model specifications, when estimated separately for employed and contract staff, when non-nursing staff were added to the model, and when the model was estimated on stratified samples.
      Although the protective effect of high staffing on hospitalization has been observed in many previous studies,
      • Laging B.
      • Ford R.
      • Bauer M.
      • Nay R.
      A meta-synthesis of factors influencing nursing home staff decisions to transfer residents to hospital.
      • Backhaus R.
      • Verbeek H.
      • van Rossum E.
      • Capezuti E.
      • Hamers J.P.
      Nurse staffing impact on quality of care in nursing homes: a systematic review of longitudinal studies.
      • Spilsbury K.
      • Hewitt C.
      • Stirk L.
      • Bowman C.
      The relationship between nurse staffing and quality of care in nursing homes: a systematic review.
      • Grabowski D.C.
      • Stewart K.A.
      • Broderick S.M.
      • Coots L.A.
      Predictors of nursing home hospitalization: a review of the literature.
      • Castle N.G.
      Nursing home caregiver staffing levels and quality of care: a literature review.
      • Bostick J.E.
      • Rantz M.J.
      • Flesner M.K.
      • Riggs C.J.
      Systematic review of studies of staffing and quality in nursing homes.
      evidence for high staffing increasing the risk of hospitalization for nursing home residents is novel, to our knowledge. We believe that these unexpected findings are because for the first time we are able to analyze daily staffing data. All prior studies of the associations between staffing and hospitalizations relied on staffing data averaged over large periods of time, which likely masked this phenomenon.
      What might be the explanation for these 2 different associations between staffing and hospitalizations? The same-day increased hospitalization with higher staffing is likely due to staff's increased ability to monitor residents and identify those residents who may require immediate transfer. We note that the same-day effect is more than 3 times larger for RNs and LPNs than for CNAs. As the former are the ones likely to perform assessments and make transfer recommendations, this offers further support for the explanation. The protective effect of the prior-day high staffing on hospitalization may also be explained by staff having increased ability to monitor and identify residents with clinical issues requiring interventions. In these cases, the interventions available in-house might avert further decline and prevent the need for hospitalization when given time. Hence, increased staffing coupled with more time (ie, the additional day or two) mean that some hospitalizations can be prevented, such that on net, the effect of increased staffing is to lower hospitalizations.
      Comparing the magnitude of the associations across strata dyads in many cases offers further support for these explanations. For example, we find higher effect sizes for employed staff who know the residents better and are, therefore, more likely to ascertain their needs than contract staff. We find higher effect sizes for RNs in nursing homes with higher 5-star ratings for both the QMs and the Survey, as well as facilities that have higher total nursing levels. The high Medicare facilities, which have more post-acute patients who are more likely to have higher patient acuity and skilled needs that might require rehospitalizations also exhibit larger effects than nursing homes with fewer Medicare patients. Nursing homes with higher census of patients with ADRD diagnoses have lower effect size compared with nursing homes with fewer patients with these diagnoses, possibly reflecting a more judicious approach toward hospital transfer for patients with ADRD, as high-quality care practices would suggest. Although another less benign possibility is that residents with ADRD are less likely to express themselves and staff may not recognize the need to hospitalize as often.
      We note that the short-term patterns we observe are not likely to be due to long-term staff shortages. Our sensitivity analysis comparing nursing homes with high and low average staffing found the same staffing-to-hospitalization associations among both, suggesting that long-term staffing differences do not explain away these patterns. We also note that we did not control for turnover because it is likely one of the mechanisms for daily staffing variations, and including it would mask the effect of interest. Future studies examining factors leading to daily staffing patterns should include turnover.
      Several limitations of this study should be mentioned. Because of data limitation, our estimates of ED transfers were limited to FFS residents only. Second, the associations we found are not necessarily causal; however, we used a strong facility fixed-effects design to control for confounding by all time-invariant facility-level characteristics. Third, our staffing measures are at the facility level and we cannot identify the staff hours received by individual residents or groups of residents, such as those with ADRD, or whether or not residents had private duty nurses or family involvement. Finally, these quantitative data cannot establish the mechanisms driving the associations we find. For example, we did not have data about nurses’ training to start intravenous lines in the nursing home, manage total parenteral nutrition or tube feeding. Furthermore, we focused on staffing only and had no information about availability of other resources, such as laboratory and x-ray results within 4 to 8 hours, specialists, hospice services, and 7 days-per-week therapy services, all of which are considered important for preventing hospitalizations.
      • Ouslander J.G.
      • Lamb G.
      • Perloe M.
      • et al.
      Potentially avoidable hospitalizations of nursing home residents: frequency, causes, and costs: [see editorial comments by Drs. Jean F. Wyman and William R. Hazzard, pp 760–761].

      Conclusions and Implications

      This is the first study, to our knowledge, to examine short-term patterns of staffing in nursing homes and their associations with hospitalizations and ED transfers not followed by a hospitalization. The richness of the PBJ data allowed us to examine daily staffing patterns vis-à-vis daily outcomes, and revealed patterns that have not been observed before. Our findings reaffirm the importance of staffing to hospitalization and ED outcomes, even when inspected on a microlevel, and offer a new perspective for understanding the roles of different staffing types. Specifically, our findings suggest that staffing patterns, especially for RNs and LPNs, play a key role in assessing and triaging residents, both in identifying those residents who require immediate hospitalization, thus contributing to the higher number of hospitalizations on the same day, and in identifying those in which hospitalizations can be averted if appropriate action is taken early enough.
      An indirect implication of our study is that a simple count or rate of hospitalizations may not be a good marker for quality, and a more nuanced approach should be considered. The use of hospitalization and rehospitalization rates has become ubiquitous as a quality metric. At the same time, it has long been recognized that some hospitalizations are necessary, and that failing to transfer may be a sign of poor quality. Our study reinforces the need to remember this caveat and continue to refine quality metrics to try to disentangle appropriate and inappropriate transfers.
      This study offers new observations about the associations between staffing and hospitalizations. Its strength is its reliance on national data. This, however, is also its weakness. It cannot get into the details of the underlying processes of care, which lead to the outcomes and associations we observe, and some of the questions we raised previously. Future studies, relying on other data sources and different methods, should address these questions, as well as the mechanisms leading to daily staffing variations.

      Acknowledgments

      Debra Saliba is an employee of the Veterans Health Administration. The views presented here do not represent those of the Department of Veterans Affairs. The sponsor had no role except for funding.

      Supplementary Data

      Supplementary Table 1Full Regression Model of Base Case: Full Sample, Dependent Variable: Number of Hospitalizations from the Nursing Home
      Conditional fixed-effects Poisson regression
      Number of observations = 14,620,886
      Number of groups = 15,718
      Observations per group:
      min = 27
      avg = 930.2
      max = 1002
      Wald χ2 (110) = 11,866.77
      Log pseudolikelihood = −8,927,513.8
      Prob >χ2 = 0.0000
      (SE adjusted for clustering on id)
      CoefficientRobust SEzP > z95% CI
      RN hours0.0027957.79E-0535.87<.0010.0026420.002947
      LPN hours0.0026036.94E-0537.48<.001.0024660.002739
      CNA hours0.0007744.62E-0516.76<.001.0006830.000864
      1 day lag RN hours−0.001026.05E-05−16.81<.001−.00114−0.0009
      1 day lag LPN hours−0.001015.14E-05−19.69<.001−.00111−0.00091
      1 day lag CNA hours−0.000463.82E-05−11.95<.001−.00053−0.00038
      Percent male0.0034630.00020716.73<.001.0030580.003869
      Percent age 65 or less0.0055150.00036814.98<.001.0047940.006236
      pctage65_740.0061170.00029320.86<.001.0055430.006692
      pctage75_840.0043480.00023718.38<.001.0038840.004811
      Percent Medicare Advantage−0.001730.000231−7.5<.001−.00218−0.00128
       Jan-17−0.005820.017593−0.33.741−.04030.028666
       Feb-170.1545820.00539328.67<.001.1440120.165151
       Mar-170.1746780.00798321.88<.001.1590320.190324
       Apr-170.1282580.0088214.54<.001.1109710.145545
       May-170.1150960.00891712.91<.001.0976180.132573
       Jun-170.0991640.00880711.26<.001.0819040.116424
       Jul-170.0653090.0088537.38<.001.0479580.082661
       Aug-170.0871470.0087499.96<.001.0699990.104295
       Sep-170.0905460.00859410.54<.001.0737010.107391
       Oct-170.0967210.00859811.25<.001.0798680.113573
       Nov-170.0995470.00865911.5<.001.0825750.116518
       Dec-170.1520270.00832718.26<.001.1357070.168347
       Jan-180.2316860.00792429.24<.001.2161550.247217
       Feb-180.1742710.00795921.9<.001.1586720.18987
       Mar-180.1423610.00802717.74<.001.1266290.158093
       Apr-180.1226410.00819514.97<.001.1065790.138703
       May-180.0936860.00825711.35<.001.0775030.109869
       Jun-180.0786960.0081399.67<.001.0627440.094648
       Jul-180.0700250.0080368.71<.001.0542750.085775
       Aug-180.0838190.00795110.54<.001.0682360.099403
       Sep-180.0781090.0079469.83<.001.0625360.093683
       Oct-180.0917450.00772811.87<.001.0765990.10689
       Nov-180.078340.00780810.03<.001.0630370.093644
       Dec-180.0941320.00761112.37<.001.0792150.109049
       Jan-190.1415380.00739819.13<.001.1270380.156038
       Feb-190.1426290.00700920.35<.001.1288930.156366
       Mar-190.1385520.00683520.27<.001.1251570.151948
       Apr-190.1137810.00682416.67<.001.1004070.127155
       May-190.0855390.00678512.61<.001.0722420.098836
       Jun-190.0613540.0067659.07<.001.0480940.074614
       Jul-190.0582820.0065598.89<.001.0454270.071138
       Aug-190.0382360.0055846.85<.001.0272910.04918
      Major RUG GroupRUG CodeCoefficientRobust SEzP > z95% CI
      Behavioral Symptoms and Cognitive Performanceba10.0065470.00053112.33<.0010.0055060.007588
      ba20.0041660.0016412.54.0110.000950.007381
      bb10.0085050.00053116<.0010.0074640.009547
      bb20.0036710.0015432.38.0170.0006460.006695
      Clinically Complexca10.0102290.0008811.62<.0010.0085040.011954
      ca20.0117790.0017036.92<.0010.0084410.015117
      cb10.0118810.00086413.75<.0010.0101870.013575
      cb20.0163740.0034384.76<.0010.0096350.023113
      cc10.0101640.00056218.08<.0010.0090620.011266
      cc20.0103320.0020085.15<.0010.0063960.014267
      cd10.0102930.00059317.37<.0010.0091310.011455
      cd20.0101850.0019865.13<.0010.0062930.014077
      ce10.0080070.0009378.55<.0010.0061710.009844
      ce20.0088220.0038262.31.0210.0013240.01632
      Extensive Serviceses10.0118220.0013988.45<.0010.0090810.014562
      es20.0109380.0015976.85<.0010.0078070.014068
      es30.0087120.0019964.37<.0010.0048010.012624
      Special Care

      High
      hb10.0109220.00097411.22<.0010.0090140.01283
      hb20.0083330.0017274.82<.0010.0049480.011718
      hc10.0094310.0008781.74<.0010.007710.011152
      hc20.0068850.0016264.23<.0010.0036980.010072
      hd10.0099220.00077612.79<.0010.0084010.011443
      hd20.0062780.0016193.88<.0010.0031050.00945
      he10.0103290.0010799.58<.0010.0082140.012443
      he20.0069030.0020763.33.0010.0028350.010971
      Special Care

      Low
      lb10.0150330.00105814.21<.0010.012960.017106
      lb20.0141140.0037023.81<.0010.0068590.021369
      lc10.0114520.00057719.87<.0010.0103220.012582
      lc20.0117640.0021695.42<.0010.0075130.016014
      ld10.0103390.00057617.94<.0010.009210.011468
      ld20.0112360.0016926.64<.0010.0079190.014553
      le10.0098840.00076412.93<.0010.0083860.011382
      le20.0078560.0022063.56<.0010.0035320.01218
      Reduced Physical Functioningpa10.007770.0005214.95<.0010.0067510.008788
      pa20.0080260.0014315.61<.0010.0052220.010831
      pb10.0076340.00053914.15<.0010.0065770.008691
      pb20.0087970.0011057.96<.0010.0066310.010963
      pc10.0083810.0004518.64<.0010.00750.009263
      pc20.006760.0007479.05<.0010.0052970.008224
      pd10.00840.0004132.32<.0010.0075890.00921
      pd20.005530.0008746.33<.0010.0038170.007243
      pe10.0066930.0006719.98<.0010.0053780.008008
      pe20.0046360.0013253.5<.0010.0020390.007233
      Rehab and Rehab Plus Extensiverha0.0084480.00060413.98<.0010.0072640.009633
      rhb0.0115130.00063218.22<.0010.0102740.012751
      rhc0.0115040.0005682.26<.0010.0103910.012617
      rhl0.0167350.0031255.35<.0010.0106090.022861
      rhx0.0174190.0023937.28<.0010.0127290.022108
      rla0.0073490.0007211.19<.0010.0059350.008762
      rlb0.0080290.0009868.14<.0010.0060970.009961
      rlx0.0152180.0046223.29.0010.0061590.024276
      rma0.0076890.00046616.49<.0010.0067750.008603
      rmb0.0091950.0004321.38<.0010.0083520.010038
      rmc0.0092790.00043321.43<.0010.0084310.010128
      rml0.0136840.003933.48<.0010.0059820.021386
      rmx0.0072140.0032172.24.0250.0009090.013519
      rua0.0103690.00044323.42<.0010.0095010.011236
      rub0.0092720.00042421.86<.0010.0084410.010103
      ruc0.0111120.00047623.35<.0010.010180.012045
      rul0.0155710.0020037.78<.0010.0116460.019496
      rux0.0125840.0027954.5<.0010.0071060.018063
      rva0.011640.0005919.73<.0010.0104840.012796
      rvb0.0120820.00060819.86<.0010.010890.013274
      rvc0.0135940.00060922.32<.0010.0124010.014788
      rvl0.0255550.0024931.25<.0010.0206680.030441
      rvx0.0217250.00188311.54<.0010.0180340.025416
      Unknown0.0329390.00276811.90<0.0010.038364
      Supplementary Table 2Sensitivity Analysis Summary of Findings: Percent Change in Daily Hospitalizations Due to an Increase of a 1 Full-Time Equivalent Staff Position
      Each line reports findings from a separate regression model that also controls for age, gender, case mix, seasonality, and facility fixed effects. Models differ by either staffing variables (eg, salaried vs contract), or facility characteristics (eg, low quality vs high quality) and sample (full and stratified). Bolded entries are those significant at P < .001.
      Same DayPrevious Day
      RNsLPNsCNAsRNsLPNsCNAs
      Models Estimated on Stratified Samples
       Midweek days only
      1.0% [0.9, 1.1] (<.001)1.0% [0.9, 1.1] (<.001)0.0% [−0.0, 0.1] (0.698)−0.5% [−0.6, −0.4] (<.001)−0.4% [−0.5, −0.3] (<.001)−0.0% [−0.1, 0.0] (0.148)
       NH Quality measured by 5 Star QMs
      High Average Quality

      5-Star Rating (70th percentile)
      2.5% [2.2, 2.7] (<.001)2.2% [1.9, 2.5] (<.001)0.5% [0.3, 0.7] (<.001)−0.9% [−1.1, −0.7] (<.001)−0.8% [−1.0, −0.6] (<.001)−0.3% [−0.4, −0.1] (<.001)
      Low Average Quality

      5-Star Rating (30th percentile)
      2.1% [1.9, 2.3] (<.001)2.1% [1.9, 2.3] (<.001)0.9% [0.8, 1.0] (<.001)−0.8% [−1.0, −0.6] (<.001)−0.9% [−1.1, −0.8] (<.001)−0.4% [−0.5, −0.3] (<.001)
       NH Quality measured by 5-Star Survey
      High Average Survey

      5-Star Rating (70th percentile)
      2.4% [2.1, 2.8] (<.001)2.2% [1.9, 2.5] (<.001)0.5% [0.3, 0.8] (<.001)−0.8% [−1.0, −0.6] (<.001)−0.8% [−1.1, −0.6] (<.001)−0.3% [−0.5, −0.2] (<.001)
      Low Average Survey

      5-Star Rating (30th percentile)
      2.3% [2.1, 2.6] (<.001)2.2% [2.0, 2.4] (<.001)0.7% [0.6, 0.8] (<.001)−0.9% [−1.1, −0.8] (<.001)−0.9% [−1.0, −0.8] (<.001)−0.4% [−0.5, −0.3] (<.001)
       Total nursing level variations
      High Average Nursing Level (85th percentile)2.2% [1.9, 2.5] (<.001)1.8% [1.6, 2.1] (<.001)0.3% [0.1, 0.5] (<.001)−0.8% [−0.9, −0.6] (<.001)−0.7% [−0.9, −0.5] (<.001)−0.2% [−0.3, −0.1] (0.004)
      Low Average Nursing Level (15th percentile)2.1% [1.7, 2.4] (<.001)2.3% [1.9, 2.7] (<.001)1.0% [0.7, 1.2] (<.001)−0.9% [−1.3, −0.6] (<.001)−0.9% [−1.2, −0.6] (<.001)−0.7% [−0.9, −0.5] (<.001)
       Hospitalization-level variations
      High Hospitalization Rate (85th percentile)2.5% [2.2, 2.7] (<.001)2.4% [2.2, 2.6] (<.001)0.7% [0.5, 0.8] (<.001)−1.0% [−1.2, −0.8] (<.001)−1.0% [−1.2, −0.9] (<.001)−0.5% [−0.6, −0.4] (<.001)
      Low Hospitalization Rate (15th percentile)1.9% [1.3, 2.5] (<.001)1.6% [1.2, 2.0] (<.001)0.5% [0.3, 0.6] (<.001)−0.3% [−0.6, −0.1] (0.013)−0.2% [−0.4, 0.1] (0.226)−0.3% [−0.5, −0.1] (0.001)
       High vs low Medicare census nursing homes
      High Medicare facilities (85th percentile) only2.5% [2.2, 2.8] (<.001)2.3% [2.0, 2.6] (<.001)0.5% [0.1, 0.8] (0.004)−0.9% [−1.1, −0.7] (<.001)−0.9% [−1.1, −0.7] (<.001)−0.4% [−0.7, −0.2] (<.001)
      Low Medicare facilities (15th percentile) only1.9% [1.5, 2.3] (<.001)1.7% [1.4, 2.1] (<.001)0.6% [0.5, 0.8] (<.001)−0.5% [−0.8, −0.4] (<.001)−0.6% [−0.8, −0.4] (<.001)−0.3% [−0.5, −0.2] (<.001)
       High vs low Medicaid census nursing homes
      High Medicaid facilities (70th percentile) only2.1% [1.9, 2.4] (<.001)2.2% [1.9, 2.4] (<.001)0.8% [0.7, 0.9] (<.001)−0.7% [−0.9, −0.5] (<.001)−0.8% [−0.9, −0.6] (<.001)−0.4% [−0.5, −0.3] (<.001)
      Low Medicaid facilities (30th percentile) only2.3% [2.1, 2.5] (<.001)2.1% [1.9, 2.3] (<.001)0.4% [0.3, 0.6] (<.001)−0.8% [−1.0, −0.6] (<.001)−0.7% [−0.9, −0.6] (<.001)−0.3% [−0.4, −0.2] (<.001)
      High Medicaid facilities (85th percentile) only1.9% [1.6, 2.3] (<.001)2.0% [1.7, 2.3] (<.001)0.8% [0.6, 0.9] (<.001)−0.8% [−1.0, −0.5] (<.001)−0.8% [−1.0, −0.6] (<.001)−0.3% [−0.4, −0.1] (<.001)
      Low Medicaid facilities (15th percentile) only2.2% [1.8, 2.6] (<.001)2.00% [1.7, 2.3] (<.001)0.3% [0.1, 0.6] (0.013)−0.9% [−1.2, −0.6] (<.001)−0.9% [−1.1, −0.7] (<.001)−0.2% [−0.4, −0.0] (0.014)
       Nursing homes with high and low percentage of patients with ADRD and cognitive impairment
      High ADRD (85th percentile)2.1% [1.9, 2.4] (<.001)2.3% [2.0, 2.6] (<.001)0.6% [0.4, 0.7] (<.001)−0.6% [−0.8, −0.4] (<.001)−0.7% [−0.9, −0.5] (<.001)−0.4% [−0.5, −0.2] (<.001)
      Low ADRD (15th percentile)2.8% [2.6, 3.1] (<.001)2.5% [2.2, 2.8] (<.001)0.9% [0.7, 1.0] (<.001)−1.1% [−1.3, −0.9] (<.001)−1.0% [−1.2, −0.9] (<.001)−0.6% [−0.8, −0.5] (<.001)
      Each line reports findings from a separate regression model that also controls for age, gender, case mix, seasonality, and facility fixed effects. Models differ by either staffing variables (eg, salaried vs contract), or facility characteristics (eg, low quality vs high quality) and sample (full and stratified). Bolded entries are those significant at P < .001.

      References

        • Creditor M.C.
        Hazards of hospitalization of the elderly.
        Ann Intern Med. 1993; 118: 219-223
        • Dwyer R.
        • Gabbe B.
        • Stoelwinder J.U.
        • Lowthian J.
        A systematic review of outcomes following emergency transfer to hospital for residents of aged care facilities.
        Age Ageing. 2014; 43: 759-766
        • Ouslander J.G.
        • Lamb G.
        • Perloe M.
        • et al.
        Potentially avoidable hospitalizations of nursing home residents: frequency, causes, and costs: [see editorial comments by Drs. Jean F. Wyman and William R. Hazzard, pp 760–761].
        J Am Geriatr Soc. 2010; 58: 627-635
        • CMS.gov
        Nursing home value-based purchasing demonstration.
        • CMS.gov
        Nursing homes - quality of resident care.
        • Saliba D.
        • Weimer D.L.
        • Shi Y.
        • Mukamel D.B.
        Examination of the new short-stay nursing home quality measures: rehospitalizations, emergency department visits, and successful returns to the community.
        Inquiry. 2018; 55 (46958018786816)
        • Laging B.
        • Ford R.
        • Bauer M.
        • Nay R.
        A meta-synthesis of factors influencing nursing home staff decisions to transfer residents to hospital.
        J Adv Nurs. 2015; 71: 2224-2236
        • Backhaus R.
        • Verbeek H.
        • van Rossum E.
        • Capezuti E.
        • Hamers J.P.
        Nurse staffing impact on quality of care in nursing homes: a systematic review of longitudinal studies.
        J Am Med Dir Assoc. 2014; 15: 383-393
        • Spilsbury K.
        • Hewitt C.
        • Stirk L.
        • Bowman C.
        The relationship between nurse staffing and quality of care in nursing homes: a systematic review.
        Int J Nurs Stud. 2011; 48: 732-750
        • Grabowski D.C.
        • Stewart K.A.
        • Broderick S.M.
        • Coots L.A.
        Predictors of nursing home hospitalization: a review of the literature.
        Med Care Res Rev. 2008; 65: 3-39
        • Castle N.G.
        Nursing home caregiver staffing levels and quality of care: a literature review.
        J Appl Gerontol. 2008; 27: 375-405
        • Bostick J.E.
        • Rantz M.J.
        • Flesner M.K.
        • Riggs C.J.
        Systematic review of studies of staffing and quality in nursing homes.
        J Am Med Dir Assoc. 2006; 7: 366-376
        • Kash B.A.
        • Hawes C.
        • Phillips C.D.
        Comparing Staffing Levels in the Online Survey Certification and Reporting (OSCAR) system with the Medicaid cost report data: are differences systematic?.
        Gerontologist. 2007; 47: 480-489
        • Feng Z.
        • Katz P.R.
        • Intrator O.
        • Karuza J.
        • Mor V.
        Physician and nurse staffing in nursing homes: the role and limitations of the Online Survey Certification and Reporting (OSCAR) system.
        J Am Med Dir Assoc. 2005; 6: 27-33
        • LTC Focus, Brown University
        Facts on Care in the US - Create Custom Reports on Long-Term Care.
        https://ltcfocus.org/
        Date accessed: May 11, 2022
        • Centers for Medicare and Medicaid Services
        Design for care compare nursing home five-star quality rating system: technical users’ guide-January 2022.