Association of Staffing Instability With Quality of Nursing Home Care

Key Points Question Is day-to-day staffing instability of registered nurses, licensed practical nurses, and certified nurse aides, measured as the percentage of days substantially below average levels, associated with nursing home quality? Findings In this quality improvement study of 14 717 nursing homes, after controlling for average staffing levels, staffing instability of licensed practical nurses and certified nurse aides was associated with lower quality across standard quality measures. Meaning This study suggests that day-to-day staffing instability is an important marker of nursing home quality, limiting measurement, reporting, or policy to average hours per resident-day likely will miss important opportunities to improve care.


Introduction
The association between staffing levels and quality of nursing home care has been a topic of research, policy debates, and federal and state regulations for decades. Recently, the White House announced that, within a year the federal government will implement a new minimum staffing policy for nursing homes. 1 The evidence linking staffing levels to quality of care is substantial. [2][3][4][5][6][7][8][9] Several studies found that higher staffing levels (more registered nurses [RNs] and certified nurse aides [CNAs], and, in fewer studies, also more licensed practical nurses [LPNs]) led to fewer deficiency citations for deviations from state and federal quality standards, 10  the first to introduce the concept that quality of care may depend not only on the average level of staffing but also on staffing stability over time, a concept common in industrial quality improvement, 13 but infrequently applied when reporting on factors associated with health care outcomes and quality. The study demonstrated the importance and relevance of this concept in producing high-quality care in nursing homes. 12 In the present study we add to this foundational work that established a working definition of staffing instability by rigorously testing the hypothesis that staffing instability adds information about quality of care above and beyond the traditional average staffing measure. We estimate 12 regression models, 1 for each of 12 standard QMs, with all models including both the instability and the average staffing measures for RNs, LPNs, and CNAs. These models allow us to test the hypothesis that instability in any of these of staffing measures is associated with lower quality of care as measured by these QMs, while controlling for average staffing levels as well as other nursing home characteristics.

Data and Sample
Of the 15 790 nursing homes submitting staffing data to the CMS Payroll Based Journal during fiscal years 2017-2019, we excluded those missing independent or dependent variables. Because we estimated separate models for each of the dependent variables, sample sizes varied from 10 037 to 14 183. A total of 14 717 facilities (93% of the Payroll Based Journal sample) appeared in at least 1 model. This study followed the Standards for Quality Improvement Reporting Excellence (SQUIRE) reporting guideline. 14 The study was reviewed by the University of California Irvine's institutional review board with consent waived due to the large number of individuals whose data were included in the Minimum Data Set, making it infeasible to obtain consent. The data were obtained deidentified under a data use agreement from CMS.
We created an analytical data set by linking data sets at the nursing home level using the CMS Provider Number of the nursing home. Daily hours worked by direct-care nurses by type (RNs, LPNs, and CNAs) as well as daily resident census were obtained from the Payroll Based Journal, which is the source of staffing data for NHCC. Quality measures, obtained from NHCC, 15   past. Our analyses assume that deficiency citations reflect care during the 6 months preceding the survey. Therefore, the date of the 180-day period over which we defined this measure is specific to each nursing home.

Independent Variables
We calculated daily staffing HPRDs for RNs, LPNs, and CNAs and for each day and each staffing type.
If HPRDs for the day were below 20% of the facility's average HPRD level for the period (see periodicity definitions), the day counted toward the percentage of below-average days for that staffing type for that period. This percentage measured the instability of staffing by type during the period.
Minimum Data Set data were used to calculate average facility daily case mix of residents in terms of age, sex, and Resource Utilization Group (RUGs) IV score for each nursing home. These daily case mixes of residents were based on residents' admission and discharge dates, and assumed that a resident's RUGs score does not change until the next Minimum Data Set assessment, which for short-stay residents could be as short as 11 days and for long-stay residents was typically 90 days.
Annual values for ownership, payer mix, occupancy, hospital-based affiliation, and chain affiliation were obtained from the Long-Term-Care Focus website. Thus, these variables provided values that were updated only annually.

Aligning Variables and Outcomes Periodicity Definitions
Because outcomes (dependent variables) were measured over different time periods (quarters, semi-annual, annual, and facility-specific), independent variables associated with the observed outcomes had to be defined over the same time periods. We, therefore, averaged all the daily independent variables over time periods corresponding to the time period defined for each outcome.
For example, for the long-stay outcomes that were measured over 1 quarter (eg, activities of daily living), all the daily measures of staffing (HPRD and instability), average daily age of the residents, and average daily RUGs score were averaged separately for each of the 11 quarters between 2017 and quarter 3 of 2019, 19 resulting in 11 observations for each nursing home. Similarly for the short-stay measures, which were measured over 6 months, we averaged the same independent daily variables for each 6-month period, resulting in 6 observations per nursing home.

Statistical Analysis Estimated Models
Statistical analysis was performed from February 8 to November 14, 2022. We estimated 12 separate models, 1 for each QM as a dependent variable. The unit of observation was the nursing home, with repeated observations for nursing homes and number of repetitions depending on the outcome (eg, 11 for those measured quarterly). The independent variables included the 3 instability staffing measures (ie, the below-average staffing percentages) and the 3 average staffing measures in terms of HPRD, for RNs, LPNs, and CNAs. In addition, all models controlled for mean resident sex and age, payer, RUGs score, facility characteristics, and time trend. The models were estimated as ordinary least-squares regressions with random facility effects, fixed state effects, and robust SEs.
We tested the hypothesis that the instability measures for RNs, LPNs, and CNAs offer independent information about quality, above and beyond the information offered by the HPRD staffing measures in 12 separate regressions, by assessing whether the P values were below the .05 threshold for significance. To account for multiple comparisons we applied the Benjamini-Hochberg adjustment. 20 We present the Benjamini-Hochberg-adjusted P values.

Sensitivity Analyses
We tested an alternative specification for instability, defining it as total days exceeding a 20% band above and below the average staffing level, and repeated the analysis with this measure.

Estimated Association With Outcomes
We estimated the marginal association of decreasing instability of each of the staff types-RNs, LPNs, and CNAs-by 1 SD of the instability measure for each separately, on each of the 12 outcomes. We then calculated it as a percentage of the mean of the outcome measure. We calculated similarly the same percentage for a 1-SD increase in the average staffing measures for comparison.

Results
Of the 14 717 nursing homes participating in the study, 70.3% were for-profit facilities, 59.0% were affiliated with a chain, and 4.1% were hospital based (    quality, positive coefficients indicate that an increase of 1 unit in the independent variable is associated with higher quality. For CNAs, we found support for the hypothesis that staffing instability (percentage of days of below-average staffing levels) is significantly associated with worse quality in 9 of the 12 tests. The largest association was with the QM of functioning failing to improve by discharge among short-stay residents (regression coefficient, 0.030; P = .01), followed by the QM of independent mobility worsening among long-stay residents (regression coefficient, 0.017; P = .006).
For LPNs, we found support for the hypothesis in 10 of 12 tests. The largest associations were with ED visits among short-stay residents (regression coefficient, 0.020; P < .001) and decline in activities of daily living among long-stay residents (regression coefficient, 0.020; P < .001). For RNs, we found no significant associations with any of the measures we tested.
Sensitivity analyses led to findings similar to those of the main analysis with respect to CNAs and LPNs, showing that total days' outliers were also associated with lower quality of care, although the effect sizes tended to be smaller. The full regression models (eTable in Supplement 1) show similar findings to those found in prior studies with respect to average HPRDs for RNs, LPNs, and CNAs.
Higher RN and CNA average staffing hours were significantly associated with better quality in most cases, while LPN staffing hours were either not significantly associated with the QMs or were associated with worse quality. [2][3][4][5][6][7][8]   Abbreviations: ED, emergency department; HPRD, hours per resident-day.
a As the outcomes are all negative, a negative percentage change indicates improvement in the outcome. A larger negative percentage indicates larger improvement relative to the mean. For example, a 1-SD decrease in certified nurse aide and licensed practical nurse staffing instability was associated with similar improvements in deficiency citations of 0.05% and 0.04%, respectively. b Adjusted for multiple comparisons using the Benjamini-Hochberg adjustment. change in staffing. For RNs, only HPRD increases were associated with significant improvements, with 8 outcomes showing significant improvements, ranging from 0.18% (P < .001) for the deficiencies score within the past 6 months to 6.96% (P < .001) for ED visits per 1000 long-stay residents. For CNAs, 6 instability measures were associated with more improvement compared with HPRD, ranging from 0.05% (P < .001) for the deficiencies score within the past 6 months to 2.12% (P < .001) for ED visits per 1000 long-stay residents. The most associations of instability with outcomes were found for LPNs, with 10 outcomes associated with improvement relative to 1 for HPRD, ranging from 0.04% (P = .006) for the deficiencies score within the past 6 months to 2.66% (P = .004) for percentage of short-stay residents receiving antipsychotic drugs for the first time.

Discussion
In this study, we tested the hypothesis that staff instability, measured as percentage of belowaverage staffing days, is an indicator of nursing home care quality, above and beyond the traditional measure of average HPRD staffing, a measure that has been used in numerous prior studies. The SD of the instability measure compared with an increase of 1 SD in HPRD is for LPNs at 10 of the outcomes compared with 1 outcome for an increase in HPRD. For CNAs, instability is associated with more improvement in 6 of the outcomes compared with 4 of the outcomes for an increase in HPRD. Table 3 demonstrates that a 1-SD increase in HPRD also improved many of the outcomes, particularly for RNs and CNAs, a finding reported in many prior studies. [2][3][4][5][6][7][8][9] These findings highlight the fact that both measures are important and offer different perspectives on quality of care. In particular, it seems that having enough RN hours is essential to nursing home quality, but stability of those hours does not matter as much, perhaps because nursing home managers find ways to compensate when an RN cannot show up, possibly by delegating some of those tasks to LPNs or to administrative RNs, or postponing those tasks. On the other hand, instability of LPN and CNA staffing seems to be a red flag for quality, perhaps one that consumers should know about. The ability of a nursing home to avoid days with low LPN and CNA staffing, perhaps by building more flexibility into staffing availability or better planning and anticipation of changes in staff availability or resident census, appears to offer a new pathway to quality improvement. Limitations This study has some limitations. Nursing homes were excluded from our study if data on staffing or outcomes were not available, which resulted in samples of different sizes for different outcomes, primarily with missingness associated with small nursing homes, which are less likely to meet the threshold for minimum sample size for inclusion in the NHCC measures. These nursing homes tend to have higher HPRD but also higher instability. Nevertheless, the study captured 93% of nursing homes nationally in at least 1 of the estimated models. Because staffing data were available only at the facility level, all adjustments for resident risks, such as age and case mix, could be made only at the facility level.

Conclusions
After President Biden's recent call for setting federal minimum staffing standards for nursing homes, 1 the CMS has announced a new study to determine the optimal level and type of nursing home staffing needs and is currently soliciting input from the public. It intends to issue proposed rules on minimum staffing requirement for nursing homes within 1 year. 21 The information provided here, about both the HPRD and below-average staffing days, is relevant and may inform these efforts.
Most important, it is clear that average staffing levels do not tell the whole story; now that the data are available to examine more nuanced aspects of staffing, such as the frequency with which a facility has below-average staffing, a more comprehensive view of nursing home staffing should be pursued.
Funding/Support: Research reported in this publication was supported by grant R01AG066742 from the National Institute on Aging of the National Institutes of Health.

Role of the Funder/Sponsor:
The funding source had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Disclaimer:
The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. The views presented here do not represent those of the US Department of Veterans Affairs. The sponsor had no role except for funding. All information and materials in the manuscript are original.
Data Sharing Statement: See Supplement 2.