Quality of Life Scores for Nursing Home Residents are Stable Over Time: Evidence from Minnesota

Abstract

Quality of life (QoL) is important to nursing home (NH) residents, yet QoL is only publicly reported in a few states, in part because of concerns regarding measure stability. This study used QoL data from Minnesota, one of the few states that collects the measures, to test the stability of QoL over time. To do so, we assessed responses from two resident cohorts who were surveyed in subsequent years (2012–2013 and 2014–2015). Stability was measured using intra-class correlation (ICC) obtained from hierarchical linear models. Overall QoL had ICCs of 0.604 and 0.614, respectively. Our findings show that person-reported QoL has adequate stability over a period of one year. Findings have implications for higher adoption of person-reported QoL measure in long-term care.

Introduction

Nursing homes (NHs) are a major component of the US long-term care system, providing care to about 1.6 million older adults and people with disabilities (National Center for Health Statistics, 2019). While NHs are highly regulated, NHs also continue to have quality gaps. Attention from researchers and policy makers has focused on these gaps, with key metrics being quality of care (QoC) measures. QoC measures include clinical aspects of care, such as preventive care, use of antipsychotics, maintenance of physical function, and prevention of readmission to hospital (Castle & Ferguson, 2010). One reason for this focus on QoC is that the data used to construct these measures, namely the Minimum Data Set (MDS) and Medicare claims, are available for all NHs nationally. Many QoC indicators have been the subject of validation studies to determine their reliability and validity (Hutchinson et al., 2010; Mor et al., 2003). In addition, validation work has been conducted on scales that measure cognitive performance, physical function pain, depression (Saliba & Buchanan, 2008), and disease diagnoses (Mor et al., 2011). These items form the building blocks of QoC indicators. Thus, QoC indicators have achieved widespread acceptance among researchers.

However, QoC is just one aspect of NH quality (Castle & Ferguson, 2010). A growing body of literature (Kane et al., 2003; Shippee et al., 2015, 2020; Shippee, Ng, Duan et al., 2020) and the Centers for Medicare and Medicaid Services have recognized the need to have more comprehensive measures of NH quality that include quality of life (QoL). QoL includes person-reported measures about the residents’ overall well-being and experience of receiving NH care. Studies have identified that QoL is a multidimensional construct spanning person-environment fit (e.g., ability to get around), attention from staff, meaningful relationships and activities, meal enjoyment, and positive and negative mood (Kane, 2003; Shippee et al., 2015). Good QoC can be a building block of QoL, for example, by preserving physical function and other aspects of health. However, QoL is conceptually distinct from QoC, and it should be measured separately (Kane, 2001). Yet, there is relatively little research on QoL. Possible reasons for this gap include the lack of any QoL measures in the current MDS, and the lack of a nationally accepted measure of QoL. We are aware of only two U.S. states, Minnesota and Ohio, that regularly use validated measures of QoL or satisfaction with care to NH residents (Shippee, Ng, Roberts et al., 2020).

Whereas a number of QoC measures have been shown to be reliable, the same work is needed to establish that QoL is sufficiently reliable to measure NH quality. This study aims to fill this gap by utilizing Minnesota’s QoL measure to assess its longitudinal stability. The QoL score is constructed from a 31-item in-person survey, administered annually to a random sample of NH residents at each facility by an independent survey firm. The measure has been validated (Kane et al., 2004a, 2003; Shippee et al., 2015) and has been used for quality improvement efforts by the state since 2005. This presents a unique opportunity to assess the stability of the QoL measure over time (Mor et al., 2011). Using longitudinal cohort analyses, the study has two objectives. First, the primary objective is to determine the level of stability in QoL responses for a group of residents who were surveyed in two consecutive years. The second objective is to determine if key resident and facility-level covariates affect this stability.

Background on quality of life in nursing homes

QoL in NHs is a multidimensional construct. It covers the preservation of individuality, dignity, and autonomy, adapting one’s room to one’s level of physical function, enjoyment of food, meaningful relationships with both staff and other residents, access to meaningful activities, and good mood and emotional health (Kane, 2001, 2003; Kane et al., 2003). There is a growing emphasis on having NHs focus on supporting resident QoL. Previous work has shown that QoL is impacted by resident and facility factors, including resident age, ADL race/ethnicity and facility racial composition (Shippee et al., 2015, 2016, 2020; Shippee, Ng, Duan et al., 2020). However, testing the reliability of QoL measures is one area that remains to be explored.

Background on the stability of existing measures of quality of life for nursing homes

One aspect of validation is test-retest reliability, which determines if a measure is reliable or stable over time. Each NH resident is assumed to have a true, underlying value of QoL. Because this is a subjective construct, the true score cannot be directly, “objectively” measured. However, it can be estimated: an individual survey does produce an observed measure of QoL for each resident, containing the person’s true QoL plus some noise or error. To measure the short-term test-retest reliability of the QoL instrument, the same residents are given the survey instrument twice within a short time, usually one or two weeks, and the consistency of the survey results is compared. Within a short time, the resident’s true QoL should have changed minimally. If the observed measures are stable, this provides evidence that the instrument accurately measures what is was designed to measure, or that it has a high ratio of signal to noise (Devellis, 2012; J. P. Weir, 2005; Streiner & Norman, 2008)

Two widely accepted estimates of test-retest reliability are the Pearson correlation between the two tests or the intra-class correlation (ICC), both of which are scaled from 0 to 1. One guideline suggests that the test-retest reliability should be at least 0.70 when the retest is completed within two weeks.(Nunnally, 1994) Two studies of QoL measures for community residents with dementia or users of long-term services and supports have found one- to three-week test-retest reliabilities to generally be over 0.7 (Bouman et al., 2011; Rand et al., 2017), with one study showing reliability over 0.6 (Thorgrimsen et al., 2003). In the NH QoL measure used in this paper, two- to five-day test-retest reliability of 11 different domains ranged from 0.554 to 0.748 (Kane et al., 2004b).

While short-term reliability is important, when used as a measure to compare quality across facilities, a key concern is that surveying a resident on a particularly “bad day” or “good day” may affect the reliability of QoL measures when making comparisons across NHs. Therefore, long-term stability is also a desired characteristic. A recent systematic review of the literature, focused on QoL measures for persons living in care homes or nursing homes, found that long-term stability for QoL measures in NHs had not generally been assessed (Aspden et al., 2014). We are aware of only two studies that assessed the longterm stability of QoL measures in the non-institutionalized population. Both studies estimated a 12-month test-retest reliability of between 0.55 and 0.635 for the Dementia Quality of Life scale (Adler & Resnick, 2010; Carpenter et al., 2007). Additionally, previous work showed that QoL was sensitive to depressive symptoms, declines in physical function, pressure ulcers, and was relatively stable over a 6-month period (Degenholtz et al., 2008).

Therefore, this paper makes an important contribution by assessing the stability of a person-reported, multi-domain measure of QoL for a period of one year and determining whether including resident and facility-level controls materially alters stability.

Method

Data and sample

The Minnesota QoL survey selects a random sample of residents in all Medicaid-certified nursing homes each year. While it was not designed as a cohort study, several thousand residents are sampled in consecutive years by chance. Taking the 2012 to 2015 NH QoL surveys, we created two 2-year cohorts of residents who remained in the same facility in each pair of years. We linked each observation (i.e. each respondent had two observations) to the MDS assessment temporally closest to the QoL survey data. We also linked each observation to facility characteristics from the state of Minnesota’s cost report data. The 2012–2013 cohort had 4,448 respondents in 365 NHs, and the 2014–2015 cohort had 4,644 respondents in 354 NHs.

These respondents’ characteristics are somewhat different from the overall random sample. First, they tend to be in smaller NHs, because this gives them a higher probability of being sampled. Second, whereas mortality in NHs can be high, the members of these cohorts survived at least one year beyond their initial survey. Compared to residents not sampled for either cohort, our analytic sample has, among other things, slightly better physical function, and lower rates of cognitive impairment and dementia. Yet, the sample is similar in mean age to the residents who were not sampled. The sample’s average facility occupancy rate and all-staff retention is also similar to the general population. A list of characteristics that shows the difference between residents in the cohorts compared to those who are not in the cohorts is included in Appendix Table A1.

This study obtained approval from the University of Minnesota IRB for a HIPAA waiver, and we have a Data Use Agreement from the Centers for Medicare and Medicaid Services for the use of all data described above.

Construction of QoL summary and domains

We calculated an overall summary QoL score, constructed from 31 items. The items cover six domains: 1) attention: staff are polite and respectful, 2) environmental adaptations: are the resident’s room and belongings accessible, 3) food enjoyment, 4) engagement: are there meaningful activities and meaningful relationships with staff and other residents, 5) lack of negative mood, and 6) positive mood. All domains were scaled 0 to 100, i.e. each point represents a percentage point, with higher numbers indicating better QoL (e.g., higher scores on the mood domains indicate less negative mood and more positive mood; Shippee et al.).

The overall QoL summary score is the average of the domains in which domain scores could be calculated. For residents who failed to answer all questions in a domain, we treated the domain as missing. Otherwise, we calculated a domain score based only on the questions answered. While many residents miss or refuse to respond to some questions, the QoL score data are generally complete. In 2012, out of 31 questions, 95.4% of the sample had 6 or fewer missing questions, and 98.3% had no missing domain scores.

Approach

To test the long-term reliability of the Minnesota NH resident QoL measure, we utilize the test-retest reliability framework, but we are instead examining the stability of measures taken for persons in the same NH approximately one year apart. We first examined the unadjusted and adjusted average change in QoL scores between the two years. Both unadjusted and adjusted changes were determined from a linear mixed model that excluded and included control variables (x_it). With controls, for resident i in period t, the linear mixed model has the following functional form:

$${\mathrm{y}}_{\text{it}}={\mathrm{x}}_{\text{it}}\beta +\tau +{\alpha }_{\mathrm{i}}+{\epsilon }_{\text{it}}$$

where y_it is a QoL measure, τ is an indicator variable for time, α_i is a resident random effect, and ε_it is an error term. Without covariates, the model above is equivalent to a repeated measures ANOVA. We estimate test-retest stability at one year using the intra-class correlation of the above model. The ICC is the ratio of the variance of the random intercept to the combined variances of the random intercept and error term. In other words,

$$\text{ICC}=\text{Var}({\alpha }_{\mathrm{i}})∕[\text{Var}({\alpha }_{\mathrm{i}})+\text{Var}({\epsilon }_{\text{it}})]$$

We used the ICC to measure reliability because it is sensitive to systematic variation in scores between the measurements, whereas the Pearson correlation is not (Baumgartner, 2000; Bédard et al., 2000; J. Weir, 2005). Additionally, using the ICC permits us to include covariates which we believe could affect a resident’s QoL response over time.

In estimating the above equation, a few details are worth noting. First, when resident and facility-level control variables are included in the model, all covariates are treated as time-varying. Second, we explored an additional random effect for the nursing home. However, it explained a minimal additional proportion of the variance. For the 2014–2015 survey year, the NH-level ICC was 0.025, and domain-level ICCs ranged from 0.015 for positive mood to 0.032 for food. Thus, we chose to omit the NH-level random effects for simplicity. Finally, we chose to run two separate analyses for each cohort, mainly to ensure that our estimates would be unaffected by any unusual effects in one cohort and to validate our results in two samples.

Control variables

Based on previous literature, we included key resident-level and facility-level covariates, which can vary with time. We included the following covariates from the MDS: age group at first survey (21 to 64, 65 to 84, and 85 years or older, treated as categorical because we believed that age might have a non-linear effect), Black, Indigenous, or other People of Color (BIPOC race/ethnicity), living in a high-BIPOC facility defined as over 10.1 percentage points (90^th percentile in 2011) BIPOC residents at census, physical function (0–28 point Activities of Daily Living scale), length of stay in years at the first survey, being married, count of 5 chronic conditions (congestive heart failure, diabetes, hip fracture, paralysis, and stroke), diagnoses of depression, anxiety, serious mental illness, and dementia, and moderate or severe cognitive impairment. We included flags for hospital admission or any other significant change of status during the second survey year (i.e., we excluded any such events in the first year). The following facility-level characteristics were derived from Minnesota facility cost reports: location (Twin Cities metropolitan area, other metropolitan area, micropolitan area, or rural), ownership (nonprofit, for-profit, government), chain ownership, Minnesota facility acuity index, proportion of rooms occupied, retention proportion for all staff, hours per resident-day for registered nurses, licensed practical nurses, certified nursing assistants, activities staff, and social workers, and high use of temporary/agency staff (i.e. proportion of direct care staff hours per resident day greater than 10%).

Results

Table 1 provides descriptive statistics of each cohort. Both cohorts had mean ages around 81 years, with 90% being over the age of 65. About 4% were BIPOC residents, and approximately 10% lived in high-BIPOC facilities. Both cohorts were similar in terms of most variables of interest, though some differences exist in terms of number of chronic conditions, length of stay, and registered nurse staffing levels.

Table 1.

Descriptive Statistics.

Variable	2012 to 2013	2014 to 2015
N Respondents	4448	4644
N Facilities	363	354
Dependent Variables
QoL Summary Score, mean (SD)	81.382 (14.748)	82.025 (14.224)
Environment Score, mean (SD)	85.513 (27.219)	86.624 (25.370)
Attention Score, mean (SD)	93.310 (16.699)	93.887 (15.176)
Food Enjoyment Score, mean (SD)	82.671 (30.311)	83.134 (29.965)
Engagement Score, mean (SD)	81.396 (22.343)	82.736 (21.871)
Positive Mood Score, mean (SD)	78.036 (22.288)	78.192 (22.123)
Resident Characteristics
Hospitalization (in second year)	7.756%	6.524%
Other Significant Change of Status (in second year)	12.972%	14.538%
Black, Indigenous, or Other People of Color (BIPOC)	4.024%	4.177%
Age 21–64	9.015%	9.690%
Age 65–84	39.523%	38.264%
Age 85+	51.461%	52.046%
ADL Long-Form Scale (0–28), mean (SD)	13.119 (7.307)	13.381 (7.102)
Length of Stay (Years), mean (SD)	2.046 (2.709)	2.924 (3.368)
Married	19.874%	19.251%
Count of Chronic Conditions (0– 5), mean (SD)	0.758 (0.790)	0.332 (0.477)
Depression Diagnosis	53.395%	51.464%
Anxiety Diagnosis	24.438%	25.345%
Moderate or Severe Cognitive Impairment	23.539%	22.976%
Dementia Diagnosis	40.917%	40.676%
Serious Mental Illness Diagnosis	15.018%	14.470%
Facility Characteristics
High-BIPOC Facility	10.881%	9.755%
Twin Cities	34.442%	34.345%
Other Metropolitan Area	19.020%	18.928%
Micropolitan Area	20.594%	20.564%
Rural	25.944%	26.163%
Nonprofit	60.274%	60.013%
For-Profit	28.777%	29.328%
Government Owned	10.949%	10.659%
Chain Ownership	51.911%	52.627%
Minnesota Acuity Index, mean (SD)	1.003 (0.135)	1.005 (0.127)
Occupancy, mean (SD)	0.899 (0.071)	0.891 (0.078)
Retention, All Staff, mean (SD)	0.698 (0.118)	0.691 (0.118)
Registered Nurse Hours/Resident-Day, mean (SD)	0.462 (0.215)	0.515 (0.238)
Licensed Practical Nurse Hours/Resident-Day, mean (SD)	0.732 (0.231)	0.708 (0.232)
Certified Nursing Assistant Hours/Resident-Day, mean (SD)	2.166 (0.481)	2.158 (0.502)
Activities Hours/Resident-Day, mean (SD)	0.249 (0.097)	0.246 (0.098)
Social Work or Mental Health Hours/Resident-Day, mean (SD)	0.112 (0.077)	0.113 (0.083)
High Pooled Staff	0.922%	1.120%

To measure test-retest reliability, the unadjusted and adjusted ICCs are reported in Table 2. A few interesting patterns emerge. First, the ICCs for each cohort are similar, suggesting that our results are not due to cohort effects. Second, the ICCs for the summary score are the highest. Both cohorts have unadjusted ICCs over 0.60, and the ICCs are over 0.57 in the adjusted models. Third, the adjusted models tend to have slightly lower ICCs than the unadjusted models. While the domains have lower ICCs than the summary score, they are mostly close to 0.5. The only exceptions are environmental adaptations (unadjusted ICCs 0.42–0.45, adjusted 0.35–0.39) and positive mood (unadjusted ICCs 0.37–0.40, adjusted 0.34–0.38).

Table 2.

Intra-Class Correlations for Quality-of-Life Summary and Domain Scores.

	Unadjusted		Partly Adjusted		Fully Adjusted
Intra-Class Correlations	2012 to 2013	2014 to 2015	2012 to 2013	2014 to 2015	2012 to 2013	2014 to 2015
Summary	0.602	0.614	0.572	0.581	0.565	0.573
Attention	0.496	0.439	0.477	0.418	0.475	0.414
Environmental Adaptations	0.423	0.447	0.352	0.386	0.353	0.383
Food	0.504	0.517	0.485	0.496	0.483	0.49
Engagement	0.466	0.49	0.454	0.478	0.45	0.472
Lack of Negative Mood	0.571	0.574	0.544	0.543	0.543	0.539
Positive Mood	0.374	0.401	0.346	0.378	0.344	0.377

Partly adjusted models include individual age, physical function, sex, BIPOC race/ethnicity, living in a high-BIPOC NH, length of stay, being married, count of 5 chronic conditions (congestive heart failure, diabetes, hip fracture, paralysis, and stroke), diagnoses of depression, anxiety, serious mental illness, and dementia, moderate or severe cognitive impairment, and hospital admission or any other significant change of status during the second survey year. Fully adjusted models add facility location, ownership, occupancy, retention, staffing, Minnesota acuity index, and high use of pooled staff.

Table 3 reports the unadjusted and adjusted changes in the QoL summary score and domain scores. Focusing on the unadjusted models, we observed slight declines in the summary score and in most domain scores. The unadjusted decline in summary score was 0.7 to 1.1 percentage points depending on the cohort. Among the domains, the largest declines were for positive mood (1.9 to 2.0 percentage point decline). Engagement was an exception, as it improved in both cohorts. The increase was only statistically significant in the 2012–2013 cohort. In contrast, both sets of adjusted models find no statistically significant change in the 2014–2015 cohort for the summary score or any of the six domains. The 2012–2013 cohort has two domains (engagement and positive mood) that have statistically significant changes in both the partly and fully adjusted models.

Table 3.

Annual Change in Quality of Life Summary and Domain Scores.

	Unadjusted		Partly Adjusted		Fully Adjusted
Annual Change	2012 to 2013	2014 to 2015	2012 to 2013	2014 to 2015	2012 to 2013	2014 to 2015
Summary	−0.772***	−1.087***	−0.187	−0.317	−0.211	−0.317
Attention	−0.771**	−1.239***	−0.439	−0.5	−0.461	−0.544
Environmental Adaptations	−1.341**	−1.353***	0.054	−0.141	0.06	−0.114
Food	−0.72	−1.217*	−0.759	−0.835	−0.804	−0.976
Engagement	1.316***	0.12	1.534***	0.669	1.547***	0.621
Lack of Negative Mood	−1.081**	−0.700*	−0.561	0.039	−0.594	0.14
Positive Mood	−1.976***	−1.937***	−0.912*	−0.711	−0.948*	−0.611

Partly adjusted models include individual age, physical function, sex, BIPOC race/ethnicity, living in a high-BIPOC NH, length of stay, being married, count of 5 chronic conditions (congestive heart failure, diabetes, hip fracture, paralysis, and stroke), diagnoses of depression, anxiety, serious mental illness, and dementia, moderate or severe cognitive impairment, and hospital admission or any other significant change of status during the second survey year. Fully adjusted models add facility location, ownership, occupancy, retention, staffing, Minnesota acuity index, and high use of pooled staff.

While the full regression results are reported in Appendix Table A2, key independent that are associated with changes in QoL scores in the fully adjusted models included: having a hospitalization or other significant change of status in the second year, BIPOC race/ethnicity, living in a high-BIPOC facility, physical function, diagnoses of depression or anxiety, number of other chronic conditions, cognitive impairment, dementia, younger age, length of stay, facility location, ownership, occupancy rates, and staff retention. These associations are similar to those reported in prior work (Shippee et al., 2015, 2016, 2020; Shippee, Ng, Duan et al., 2020). The coefficients for the partially adjusted models are similar, and are available on request.

Discussion

In this study, we assessed the stability of a validated NH QoL tool over a one-year period. Our findings show that the QoL overall scores are stable, with somewhat less stability for individual domains. While mean QoL scores generally decline slightly with time, changes in resident characteristics like physical function account for most of this decline. Our results are consistent with earlier work on the QoL measures, which showed high test-retest reliability for the domain scores over 2–5 days (Kane et al., 2004b).

Our results and their consistency with the other studies mentioned give us confidence that the overall summary score has sufficient power to detect intraresident changes over a longer time interval such as one year. This also bodes well for reporting NH mean scores: if the QoL score were not stable, then sampling variation could mean that a facility’s scores might jump from year to year depending on the residents interviewed. Because the QoL score is sufficiently stable, it should also be a useful publicly-reported measure for NH quality. For states other than Minnesota which would want to adopt a similar measure of QoL, our findings should increase their confidence that a facility’s publicly-reported summary score in one year reflects real performance, consistent with previous work on ranking facilities by QoL scores (R. L. Kane et al., 2004).

Given these findings, policymakers should consider publicly reporting QoL scores, as they can provide valuable information to NH consumers. In particular, our research supports reporting the summary score and each domain, but only using the summary score if policymakers plan on ranking NHs. This is due to the summary score being relatively stable, while the domain scores have more year-to-year variability potentially because of the lower number of questions per domain. This recommendation is consistent with existing public reporting programs ran by CMS, specifically, how CMS’ Care Compare website reports multiple quality metrics for the quality measure domain but only uses a subset of quality measures to calculate the domain’s star rating.

The summary score and most domains declined slightly each year. However, most of this decline was accounted for by resident and facility characteristics. Engagement in activities was an outlier for the early cohort, as the domain with improved scores over time, even after adjusting for resident characteristics. There is not enough evidence to explain why the mean score for engagement, which covers the presence of meaningful activities and meaningful relationships with staff and other residents, exhibited a positive average change, contrary to all the other domains, in the earlier cohort. This remained true after adjusting for resident and facility characteristics related to QoL. This domain could simply be an outlier or could be a result of institutionalization trajectory, experienced by long-stay residents who end up making nursing home their “home” by virtue of higher satisfaction with engagement. (Shippee, 2009) This trend could also reflect increasing emphasis by policymakers and providers on nursing home culture change and honoring resident preferences, and it is possible that these actions improved engagement. However, we did not have data on culture change or similar initiatives, and therefore we could not analyze their impact.

Limitations

This article is not without limitations. First, our sample differs from the overall NH population as described earlier. It over-represents residents in smaller facilities, and it somewhat under-represents residents with diagnoses of Alzheimer’s and related dementia or cognitive impairment. Nevertheless, because a significant proportion of the cohort members do still have dementia and cognitive impairment, we argue that our findings are generalizable beyond this cohort. Further, publicly reporting a quality measure that NHs have little ability to improve has little value. Because we lack data on culture change or similar initiatives, we did not test whether our QoL measure is responsive to interventions. Yet prior work on other QoL measures have found them to be responsive to interventions (Degenholtz et al., 2008; Van Haitsma et al., 2013), and we would expect that would be true of the measure used in this study. Finally, our data are limited to Minnesota, which has markedly different demographics and policy environments, and may not be generalizable to other states.

Conclusions

QoL is substantively important to residents, and it is conceptually distinct from QoC. Yet most research on quality in NHs focuses on QoC measures. Our analyses of Minnesota’s measure of QoL in NHs show that these measures are stable over a period of about one year. We acknowledge that the time and expense of interviewing residents to measure QoL, which is a potential barrier to adoption. The expense barrier can be minimized if resident QOL is included as part of the MDS assessment, although this approach suffers from potential self-reporting bias. However, our findings lend weight to the argument that Minnesota’s or a similar validated measure of QoL should be tested and implemented nationally. This would provide invaluable information to consumers selecting facilities and researchers looking to compare NH quality. Our results should increase other researchers’ confidence that QoL as a person-reported measure is stable enough to detect intra-individual change in QoL and is meaningful for inclusion in report cards and other public reporting. Other QoL measures have been shown to be responsive to changes in resident health or to external interventions (Degenholtz et al., 2008; Van Haitsma et al., 2013), and an important next step should be to show that this QoL measure is similarly responsive.

Key points.

• Quality of life is important for nursing home (NH) residents, but it is not measured nationally.

• One reason for the lack of a national measure is concern about lack of stability of person-reported QoL over time.

• We find that Minnesota’s quality of life measure has an intra-class correlation of over 0.60 over a one-year period.

• Our results show that QoL is a construct that is stable enough to measure and to report on broadly. NH QoL measures should be implemented nationally.

Appendix

Table A1:

Comparing the characteristics between those in study cohorts vs those not in the analytic sample.

Variables	Not In Study Cohort	In Study Cohort	p-value
N	16285	9092
Summary Score, mean (SD)	78.872 (16.145)	81.710 (14.485)	<0.001
Age, mean (SD)	82.037 (12.452)	82.161 (11.611)	0.44
ADL Long-Form Scale (0–28), mean (SD)	14.896 (6.693)	13.253 (7.204)	<0.001
Length of Stay (Years), mean (SD)	1.899 (2.760)	2.494 (3.094)	<0.001
Married	23.706%	19.226%	<0.001
Count of Chronic Conditions (0– 5), mean (SD)	0.564 (0.716)	0.540 (0.683)	0.009
Depression Diagnosis	50.573%	52.409%	0.005
Anxiety Diagnosis	24.104%	24.901%	0.16
Moderate or Severe Cognitive Impairment	32.711%	23.251%	<0.001
Dementia Diagnosis	43.615%	40.794%	<0.001
Serious Mental Illness Diagnosis	13.202%	14.738%	<0.001
Behavioral Symptoms	17.298%	14.430%	<0.001
Location			<0.001
Twin Cities	43.605%	34.635%
Other Metro	17.255%	18.973%
Micropolitan	20.178%	20.579%
Rural	18.962%	25.814%
Ownership			<0.001
Government	64.802%	61.659%
For-Profit	27.848%	27.706%
Nonprofit	7.350%	10.636%
Chain Affiliation	55.302%	50.506%	<0.001
Minnesota Acuity Index, mean (SD)	1.025 (0.139)	1.004 (0.131)	<0.001
Occupancy, mean (SD)	0.894 (0.087)	0.895 (0.075)	0.52
Retention, All Staff, mean (SD)	0.693 (0.120)	0.695 (0.118)	0.49
Registered Nurse Hours/Resident-Day, mean (SD)	0.551 (0.346)	0.489 (0.229)	<0.001
Licensed Practical Nurse Hours/Resident-Day, mean (SD)	0.717 (0.232)	0.719 (0.232)	0.35
Certified Nursing Assistant Hours/Resident-Day, mean (SD)	2.193 (0.453)	2.162 (0.492)	<0.001
Activities Hours/Resident-Day, mean (SD)	0.234 (0.097)	0.248 (0.098)	<0.001
Social Work or Mental Health Hours/Resident-Day, mean (SD)	0.123 (0.097)	0.112 (0.080)	<0.001