Improving Sleep Using Mentored Behavioral and Environmental Restructuring (SLUMBER)

Abstract

Objectives:

To evaluate the impact of a mentoring program to encourage staff-delivered sleep-promoting strategies on sleep, function, depression, and anxiety among skilled nursing facility (SNF) residents.

Setting and Participants:

Seventy-two residents (mean age 75±15 years; 55.9% female, 41% non-Hispanic white, 35% black, 20% Hispanic, 3% Asian) of two New York City urban SNFs.

Methods:

Using a modified stepped-wedge unit-level intervention, expert mentors provided SNF staff webinars, in-person workshops and weekly sleep pearls via text messaging. Resident data were collected at baseline, post-intervention (V₁), and 3-month follow-up (V₂), including wrist actigraphy, resident behavioral observations, Pittsburgh Sleep Quality Index (PSQI), Patient Health Questionnaire-9 (PHQ-9) depression scale, Brief Anxiety and Depression Scale (BADS), Brief Cognitive Assessment Tool (BCAT), and select Minimum Data Set 3.0 (MDS 3.0) measures. Linear mixed models were fit for continuous outcomes and mixed effects logistic models for binary outcomes. Outcomes were modeled as a function of time. Planned contrasts compared baseline to V₁ and V₂.

Results:

There was significant improvement in PSQI scores from baseline to V₁ (p=.010), and from baseline to V₂ (p=.007). Other significant changes between baseline and V₁ included decreased depression (PHQ-9) (p=.027), increased daytime observed out of bed (p= <0.001), and increased daytime observed being awake (p <0.001). At V₂ (versus baseline) being observed out of bed decreased (p < 0.001). Daytime sleeping by actigraphy increased from baseline to V₁ (p=.005), but not V₂. MDS 3.0 Activities of Daily Living (ADLs) and pain showed improvements by the second quarter following implementation of SLUMBER (p’s ≤ 0.022). There were no significant changes in BADS or BCAT between baseline and V₁ or V₂.

Conclusions and Implications:

SNF residents had improvements in sleep quality and depression with intervention, but improvements were not sustained at 3-months follow-up. The COVID-19 pandemic led to premature study termination, so full impacts remain unknown.

Brief summary:

The SLUMBER staff mentoring sleep intervention program led to short-term improvements in nursing home residents’ sleep quality and depression.

INTRODUCTION

Among the 1.4 million Americans who reside in skilled nursing facilities (SNFs),¹ over two-thirds experience sleep problems.^2,3 Residents seldom sleep for a full hour at night and are seldom awake for a full hour during the day,^4,5 and those with dementia have even poorer sleep quality.^6,7 Poor sleep is strongly correlated with multiple negative conditions and symptoms including circadian dysregulation, depressed mood,⁸ anxiety,⁹ pain,¹⁰ cognitive impairment,¹¹ physical disability, and decreased engagement in activities.³ Resident sleep difficulties are also a source of stress for nursing staff.^12,13

Residents who are sleepy during the day spend more time in bed, engage in fewer social and physical activities, and require more assistance with ADLs than residents who are more alert,³ and sleep disruption is related to less daytime activity participation among SNF residents.¹⁴ The relationship between poor sleep and functional impairment is likely bidirectional. Residents with poor nighttime sleep are likely to sleep more during the day and are more depressed, both of which limit participation in activities to enhance cognitive and other domains of function. Sleep disturbance may also increase social isolation, with greater risk for morbidity and mortality.¹⁵ Use of sedating medications for sleep may be ineffective and contribute to declining health.¹⁶

Dependencies in ADLs^17,18 require frequent interactions with staff to meet ongoing care needs;¹⁹ however, nursing care practices (which impact time spent in bed, bright light exposure, and nighttime environmental noise) and institutional environments may contribute to impaired sleep.²⁰ Prior research shows that 93% of nursing home staff received no training about sleep or sleep improvement strategies after completing their nursing education.²¹ Interventions focused on the environment and staff interactions with residents are essential to achieving better sleep.

The current study evaluated a mentored staff-directed approach to improving sleep in nursing home residents called Improving Sleep Using Mentored Behavioral and Environmental Restructuring (SLUMBER). We hypothesized that SLUMBER would improve residents’ subjective and objective sleep quality, decrease depression and anxiety, and improve function (including cognitive function, daytime wakefulness and time out of bed).

METHODS

Study Design

We designed SLUMBER as a four-year hybrid type I effectiveness/implementation randomized modified stepped wedge trial (SWT) in which all clusters (SNF units) receive the intervention but at different time points²² to test the effects of this sleep-focused education and behavioral program for staff on resident outcomes (NCT03327324). The study protocol was previously published (see Chodosh, et al.²³). The main analytic approach was a pre-versus-post intervention evaluation with each resident unit as the unit of randomization. Due to COVID-related restrictions implemented in March 2020, New York City SNFs limited non-staff entry, leading to suspension of in-person study activities 1 year early. Therefore, study recruitment and data collection were conducted from March 2018 through March 2020.

We collected baseline data prior to intervention implementation on each unit. Assessments were repeated immediately after the intervention (V₁) and again at 3-months follow-up (V₂). All study procedures were reviewed and approved by the New York University Grossman School of Medicine Institutional Review Board.

Setting and Participants

We initiated SLUMBER at three not-for-profit SNFs in New York City (bed size: 362 to 500 residents). Facility units averaged 40 beds/unit and residents were ethnically diverse (40–45% other than non-Hispanic White). We randomized the order of facility engagement and units received the intervention within each facility in random order. Prior to unit-level study initiation, letters were sent to residents’ families/next-of-kin informing them of the planned research and the opportunity to opt out of participation. Due to premature termination of study activities because of the COVID-19 pandemic, only data from the first two facilities are available for these analyses.

We conducted the intervention with staff at the unit level and study assessments were performed among individual residents living on those participating units. All unit residents were eligible, except those that staff indicated should not be approached due to inability to communicate, significant behavioral concerns (e.g., risk for physical aggression), severe illness, or family preferences to opt out.

Inclusion criteria were: (1) living on the participating unit; (2) ability to communicate and follow simple commands in English or Spanish; and (3) ability to provide verbal informed consent or with their surrogate (using IRB-approved procedures).

Procedures

At each facility, leadership identified two units for participation based on unit size and likely resident eligibility. Data collection during baseline included use of resident-worn sleep monitoring devices (wrist actigraphs), behavioral observation by research staff (9am-5pm, every 15 minutes for 3 days), questionnaires (administered to residents in interview format by research staff), and specific MDS 3.0²⁴ survey items collected quarterly by facility staff, including the surveys before and after baseline assessments. Actigraphy, 3-day behavioral observations, and resident questionnaires were repeated immediately post-intervention (V₁; 3–4 months after baseline) and again 3-months later (V₂). We used MDS 3.0 data measures closest to these assessment periods for analysis.

SLUMBER Intervention

The intervention with SNF staff took place over a 3-to-4-month period on each unit and included three main components: (1) three webinars; (2) four in-person workshops; and (3) short message service (SMS) text messages reinforcing SLUMBER content, sleep facts, and resident tips. Across all units, 293 staff members participated in one or more activities across all three facilities (245 participated in the two fully engaged facilities). This included 193 CNAs, 62 licensed professional nurses and/or registered nurses, 10 social workers and 28 other facility staff members (e.g., activities director, psychologist). Details of the intervention components are described in Supplementary Table S1.

A sleep psychologist remotely conducted three 20-minute webinars with each shift (day, evening, and night) based on content developed by the intervention team. These live sessions were recorded and made available to all unit staff who wished to provide their contact number via a texted link using a mass messaging platform. A URL was provided, and the recording could be viewed on a smartphone, tablet, or computer. On average 5.4 staff members attended the live presentation of the webinar per shift on each unit.

The four workshops were conducted with staff on each shift by an interdisciplinary team of sleep/geriatrics experts including a geriatrician, a clinical psychologist and one or more nurse practitioners. Workshops reinforced webinar content and focused on unit-specific and resident-specific sleep-related issues and challenges. Using a collaborative “mentoring” style approach, proposed solutions were identified and a plan to evaluate these approaches was developed during the workshop. On average, 4.8 staff members attended the workshop sessions per shift on each unit.

Reminders about the webinar and workshop schedule, core webinar information, and tips and strategies to improve sleep were shared in a weekly text message during and after the intervention phase.

Study Measures and Assessments

Study measures have been previously described (see Chodosh et al. 2023),²³ and are summarized below. Demographic data, including age, gender and race/ethnicity were extracted from MDS 3.0. Length of stay at the facility was calculated as the days from first facility admission to the baseline assessment date.

Sleep measures

For objective sleep assessment, participants wore a wrist actigraph (Micro Motionlogger, Ambulatory Monitoring Inc., Ardsley, NY) on the dominant wrist (common in nursing home research due to low activity levels in this setting) for at least three consecutive days and nights. The actigraph is a small watch-sized device that is useful in longitudinal, naturalistic assessments of sleep-wake patterns. Since daily sleep diaries may not capture a resident’s “sleep period” in nursing homes,²⁵ we used pre-specified time for daytime (8AM-4PM) and nighttime (10PM-6AM). Actigraphy measures included: 1) percent of nighttime asleep (the main actigraphy outcome and an estimate of sleep efficiency), 2) number of nighttime awakenings, and 3) nighttime total sleep time.

We assessed resident-reported sleep quality using the Pittsburgh Sleep Quality Index (PSQI), which consists of 18-items evaluating sleep quality and sleep disturbances over the previous month. The PSQI is reported as a total score (range 0–21) and three subscales: sleep efficiency, perceived sleep quality, and daily disturbances.²⁶

Psychological symptom measures

Depression was assessed via questionnaire using the Patient Health Questionnaire (PHQ-9), which includes 9 items to assesses the frequency (0, not at all, to 3, nearly every day) of depressive symptoms over the previous 2 weeks. In the MDS 3.0 development study in 71 SNFs,²⁷ 86% of residents completed the instrument with very good agreement between PHQ-9 and a gold standard (weighted κ=.69, 95% CI=.61-.76). Anxiety and depression were assessed with the Brief Anxiety and Depression Scale (BADS), an 8-item measure of depression and generalized anxiety.²⁸

Functional Status and Pain measures

Cognitive function was assessed using the Brief Cognitive Assessment Tool (BCAT),²⁹ a 21-item instrument. Total scores range from 0–50 and a score <38 is indicative of dementia. The BCAT measures contextual memory, executive control, and attentional capacity. Physical function was calculated from MDS 3.0 resident assessments using a total score from seven ADL items (i.e., bed mobility, transfer, locomotion on unit, dressing, eating, toilet use, personal hygiene)³⁰ with a range 0–28 where higher scores indicate more impairment. Pain intensity (MDS 3.0) was measured over the preceding 5 days on a scale of 0 to 10, where higher scores indicate more pain.

Observed behaviors

We measured daytime activity using behavioral observations from 9am-5pm over a 3-day period, where research staff observed participants every 15 minutes noting whether the resident was in or out of bed, and whether they were observed to be awake or asleep.³

Data Analysis

Power analysis:

While a priori power calculations were done for the study, we also performed a power analysis using available data at the point when data collection was halted due to the COVID pandemic, focusing on two primary outcomes: sleep efficiency (by actigraphy) and sleep quality (PSQI total score). We computed Pearson correlations to estimate correlations of the outcomes across the three time points. The Variance Inflation Factor (VIF) was computed as a function of cluster size and intraclass correlation to estimate the variance accounting for the clustering of observations within facilities and units.³¹ We calculated the detectable effect size assuming power=80%, alpha=0.05, and two-tailed tests, using the calculated correlations and VIF values. There was sufficient power to detect a 10.9% change in actigraphy sleep efficiency from baseline to post-intervention (V₁; d=0.68), and an 11% change from baseline to 3-months follow-up (V₂; d=0.69). Computations for PSQI found sufficient power to detect a 2.3 unit change in total PSQI from baseline to V₁ (d=0.55), and sufficient power to detect a 2.9 unit change in total PSQI from baseline to V₂ (d=0.70). These power calculations were similar to the a priori analysis with anticipated power to detect effect sizes of d≥=0.60.

Data analysis:

We used mixed effects models, with a fixed effect for facility, to assess changes in sleep outcomes due to intervention. When applied to repeated measure designs, mixed-effects models accommodate incomplete data across time points (e.g., due to resident death) and can permit specification of a wide variety of residual covariance structures.³² We analyzed continuous outcomes using a three-level linear mixed-effects model with time nested within residents, and residents nested within SNF units. Time is represented as three levels: baseline, V₁, and V₂. Non-independence of residuals across time were modeled using an unstructured residual covariance matrix. Hypothesis testing compared baseline versus V1, and baseline versus V2. Marginal means were computed for each outcome as a function of the three levels of time.

For the two daytime observed behaviors, a two-level mixed effects logistic regression with random intercepts was estimated with time (baseline, V₁ and V₂) nested within person. Two contrasts were performed with respect to time; (1) V₁ versus baseline; and (2) V₂ versus baseline. Estimated marginal probabilities were estimated at each time point.

For analyses of MDS 3.0 data, we created 90-day bins (Quarters), beginning from the start of the intervention as day 0. The baseline quarter was the 90 days before the intervention start. Planned contrasts included: (1) baseline quarter versus the first intervention quarter and (2) baseline quarter versus the post-intervention quarter. Tests of the stability of the measures prior to the intervention were tested via planned contrasts comparing all available pre-intervention quarters to the baseline quarter. Analyses were conducted using Stata version 15³⁶.

RESULTS

Participant flow and characteristics:

Through March 2020, 241 residents had been approached for screening, of whom 92 (38%) enrolled in the study and completed the pre-intervention baseline assessments. Data from 72 (78%) residents with at least one outcome measure completed at baseline from Facilities 1 and 2 were included in the outcome analyses (Figure 1). Data from Facility 3 were dropped as outcome data were not collected (due to COVID-19). There were significant differences across facilities in terms of age, race/ethnicity, pain, cognitive function and some sleep variables at baseline (see Table 1).

Figure 1.

CONSORT-style diagram showing participant flow through the study.

Table 1:

Sample demographics for two participating facilities

	Total (𝑁=72)	Facility 1 (𝑁=38)	Facility 2 (𝑁=34)	𝑝-value
Age (years)2, Mean±SD [Min-Max]	74.6±15.2 [21–96]	64.5±13.6 [21–95]	85.6±7.0 [70–96]	<0.001
Gender2, n (%)
Male	25 (38.5%)	15 (44.1%)	10 (32.3%)	0.445
Female	40 (61.5%)	19 (55.9%)	21 (67.7%)
Race/Ethnicity1,2
White	27 (41.5%)	3 (8.8%)	24 (77.4%)	<0.001
Asian	2 (3.1%)	0 (0.0%)	2 (6.5%)	0.224
Black/African American	23 (35.4%)	22 (64.7%)	1 (3.2%)	<0.001
Hispanic	13 (20.0%)	9 (26.5%)	4 (12.9%)	0.222
Native Hawiian/Pacific Islander	0 (0%)	0 (0%)	0 (0%)	N/A
Length of stay (days),2 Mean±SD [Min-Max]	877±788 [15–3704]	879±729 [15–2695]	876±860 [31–3704]	0.987
Pain in last 5 days,2 n (%)
No	49 (75.4%)	20 (58.8%)	29 (93.5%)
Yes	16 (24.6%)	14 (41.2%)	2 (6.5%)	0.001
Pain Intensity2, Mean±SD [Min-Max]	1.0±1.9 [0–6]	1.6±2.2 [0–6]	0.2±1.0 [0–5]	0.001
BCAT total score2, Mean±SD [Min-Max]	27.6±10.1 [7–50]	28.8±10.0 [7–50]	26.1±10.1 [7–42]	0.273
ADL tot A-J, excluding C D F,2 Mean±SD [Min-Max]	16.6±5.5 [0–24]	16.3±5.8 [0–24]	16.9±5.2 [0–23]	0.691
PHQ-9 Total,3 Mean±SD	3.6±4.3	4.5±4.5	2.5±3.8	0.052
PSQI Total,3 Mean±SD	8.0±4.2	9.1±3.9	6.7±4.2	0.018
PSQI F1, Sleep Efficiency,3 Mean±SD	3.4±2.0	4.0±1.8	2.7±2.1	0.008
PSQI F2, Sleep Quality,3 Mean±SD	3.1±2.3	3.3±2.3	2.9±2.4	0.493
PSQI F3, Daily Disturbances,3 Mean±SD	1.6±1.2	1.9±1.3	1.2±1.1	0.024
Sleep Efficiency (Acti),4 Mean±SD [Min-Max]	75.2±16.4 [34–99]	67.1±15.9 [37–94]	83.0±12.7 [34–99]	<0.001
Wake Time (Minutes, Acti),4 Mean±SD [Min-Max]	119±795 [6–317]	158±77 [30–302]	82±61 [6–317]	<0.001

Results shown (% of non-missing) or Mean (SD), unless otherwise noted. Abbreviations: BCAT=Brief Cognitive Assessment Tool; ADL=Activities of Daily Living; PHQ-9=Patient Health Questionaire-9; PSQI=Pittsburgh Sleep Quality Inventory; Acti=Actigraphy.

¹Results shown as n(% of non-missing) for each response, multiple responses were allowed.

²𝑁=[65, 34, 31].

³𝑁=[69, 38, 31].

⁴𝑁=[63, 31, 32].

Effects of the intervention on outcomes

The intervention demonstrated a significant reduction (improvement) in PSQI total score from baseline to V₁, and from baseline to V₂. Although the PSQI subscale findings varied by time point, there was significant improvement on the sleep efficiency subscale at V₁, and on the daytime disturbances subscale at V₂. There was a significant increase in minutes of daytime sleep (by actigraphy) from baseline to V₁ (p=.005), which was not significant at V₂. There were no significant changes in the other actigraphy variables (Tables 2 and 3).

Table 2.

Contrast Estimates with Sample Size, Standard Error, Confidence Intervals and P-Values for Each Outcome.

		V1 vs baseline				V2 vs baseline
	N 3	Estimate	(SE)	[95% CI]	p-value	Estimate	(SE)	[95% CI]	p-value
Actigraphy % Sleep Efficiency (nighttime)1	67	0.44	(1.84)	[−3.17, 4.04]	0.813	−0.26	(1.96)	[−4.10, 3.57]	0.894
# Of Awakenings (nighttime)1	67	−0.58	(0.92)	[−2.39, 1.23]	0.531	0.34	(1.18)	[−1.98, 2.66]	0.772
Minutes Sleep (daytime napping)1	65	56.67	(19.67)	[18.11, 95.23]	0.004	35.75	(21.25)	[−5.90, 77.39]	0.092
PSQI Total Score	69	−1.22	(0.47)	[−2.14, −0.30]	0.009	−1.53	(0.57)	[−2.66, −0.41]	0.008
Factor 1: Sleep Efficiency1	69	−0.56	(0.28)	[−1.11, −0.01]	0.048	−0.59	(0.31)	[−1.20, 0.02]	0.060
Factor 2: Sleep Quality1	69	−0.42	(0.27)	[−0.95, 0.12]	0.127	−0.51	(0.28)	[−1.05, 0.03]	0.062
Factor 3: Daily Disturbances1	69	−0.24	(0.14)	[−0.52, 0.04]	0.092	−0.40	(0.17)	[−0.74, −0.06]	0.023
PHQ-9 1	69	−1.18	(0.54)	[−2.23, −0.12]	0.028	−0.99	(0.67)	[−2.30, 0.32]	0.140
BADS
Anxiety1	69	−0.33	(0.31)	[−0.94, 0.27]	0.279	−0.12	(0.39)	[−0.89, 0.65]	0.756
Depression1	69	−0.09	(0.23)	[−0.54, 0.36]	0.695	−0.17	(0.37)	[−0.89, 0.55]	0.641
Total Score1	69	−0.42	(0.41)	[−1.22, 0.38]	0.302	−0.24	(0.66)	[−1.54, 1.05]	0.711
BCAT	69	−0.52	(0.60)	[−1.68, 0.65]	0.387	−0.23	(1.03)	[−2.25, 1.78]	0.820
Behavioral Observations Out of Bed2	70	0.0361	(0.0067)	[0.0230, 0.0492]	<0.001	−0.0393	0.0075	[−0.0541, −0.0245]	<0.001
Awake2	70	0.0300	(0.0071)	[.0161, .0438]	<0.001	−.0124	.0083	[−.0287, .0039]	0.137

Note: V₁ = post-intervention; V₂ = 3-months post-intervention. Actigraphy = wristwatch device worn by subjects for three consecutive days. Abbreviations: PSQI= Pittsburgh Sleep Quality Index; PHQ-9 = Patient Health Questionnaire-9; BCAT = Brief Cognitive Assessment Tool; BADS = Brief Anxiety and Depression Scale.

¹Linear mixed model.

²Mixed effects logistic model.

³Number of subjects (residents)

Table 3.

Estimated Margins from Mixed Model (see Table 2), with standard error SE and 95% confidence interval by Outcome and Time-Point

	Baseline			V1			V2
	Estimate	(SE)	[95% CI]	Estimate	(SE)	[95% CI]	Estimate	(SE)	[95% CI]
Actigraphy % Sleep Efficiency (nighttime)1	74.48	(1.82)	[70.92, 78.05]	74.92	(2.17)	[70.67, 79.17]	74.22	(2.41)	[69.50, 78.95]
# Of Awakenings (nighttime) 1	11.71	(0.74)	[10.27, 13.15]	11.13	(0.88)	[9.40, 12.86]	12.05	(1.14)	[9.82, 14.29]
Minutes Sleep (daytime napping) 1	310.04	(18.63)	[273.52,346.55]	366.71	(24.90)	[317.90,415.52]	345.79	(23.91)	[298.93,392.64]
PSQI Total Score 1	8.06	(0.49)	[7.11, 9.01]	6.84	(0.45)	[5.95, 7.72]	6.53	(0.53)	[5.50, 7.56]
Factor 1: Sleep Efficiency 1	3.40	(0.23)	[2.94, 3.86]	2.84	(0.24)	[2.37, 3.31]	2.81	(0.22)	[2.37, 3.24]
Factor 2: Sleep Quality 1	3.09	(0.28)	[2.54, 3.65]	2.68	(0.27)	[2.15, 3.20]	2.58	(0.32)	[1.96, 3.20]
Factor 3: Daily Disturbances 1	1.57	(0.14)	[1.28, 1.85]	1.33	(0.13)	[1.08, 1.58]	1.17	(0.15)	[0.87, 1.47]
PHQ-9 1	3.64	(0.50)	[2.65, 4.63]	2.46	(0.45)	[1.57, 3.35]	2.65	(0.59)	[1.50, 3.80]
BADS
Anxiety 1	3.10	(0.35)	[2.42, 3.78]	2.76	(0.37)	[2.04, 3.49]	2.98	(0.47)	[2.05, 3.90]
Depression 1	2.29	(0.19)	[1.91, 2.67]	2.20	(0.26)	[1.69, 2.71]	2.12	(0.34)	[1.45, 2.79]
Total Score 1	5.39	(0.48)	[4.45, 6.34]	4.97	(0.54)	[3.91, 6.04]	5.15	(0.72)	[3.74, 6.56]
BCAT 1	28.56	(1.24)	[26.12, 30.99]	28.04	(1.31)	[25.47, 30.61]	28.32	(1.49)	[25.39, 31.25]
Behavioral Observations Out of Bed2	0.4892	(0.0409)	[.4091, .5694]	.5253	(.0409)	[.4451,.6055]	.4499	(.0404)	[.3707, .5694]
Awake2	0.7457	(0.0244)	[.6979, .7935]	.7759	.0231	[.7304, .8210]	.7333	(.0254)	[.6834, .7832]

Note: V₁ = immediately post-intervention; V₂ = 3-months post-intervention. Actigraphy = wrist watch device worn by subjects for three consecutive days. Abbreviations: PSQI= Pittsburgh Sleep Quality Index; PHQ-9 = Patient Health Questionnaire-9; BCAT = Brief Cognitive Assessment Tool; BADS = Brief Anxiety and Depression Scale.

¹Marginal means from linear mixed model.

²Marginal probabilities from mixed effects logistic model.

PHQ-9 total scores were lower at V₁ compared to baseline, but not at V₂. Anxiety/depression as assessed with BADS did not change from baseline to V₁ or V₂ (Tables 2 and 3).

Cognitive function (BCAT total score) showed no significant changes from baseline to V₁ or V₂ (Tables 2 and 3). MDS 3.0 assessments of ADLs and pain showed no significant changes across the quarters prior to intervention. ADLs and pain were not significantly different comparing the first intervention quarter to the baseline quarter; however, both were significantly improved comparing the post-intervention quarter to baseline. ADLs improved (increased) by 0.74 (p=0.022, 95% CI=[0.10, 1.38]) and pain improved (decreased) by −0.55 (p=0.034, 95% CI=[−1.06, −0.04]) (Supplemental Tables S2 and S3 and Figure S1).

Daytime behavioral observations out of bed and awake both showed significant improvements (increases) from baseline to V₁; however, observations out of bed decreased from baseline to V₂, and there was no change from baseline to V₂ in observations awake (Tables 2 and 3)_.

DISCUSSION

This study examined the impact of SLUMBER, a mentored staff-directed approach to improving sleep in nursing home residents, on resident outcomes. The intervention led to improved self-reported sleep quality between baseline and post-intervention assessment, and improvement was sustained at 3-month follow-up. Sustained improvement in subjective sleep over 6 months from a staff mentoring program like SLUMBER has not been found in prior studies. One prior study of a staff education intervention found improvement in PSQI-measured sleep quality after the initial 3-month intervention but no change from baseline to 6-months.³³ Of note, we found that the PSQI subscale of sleep efficiency improved at V₁ but was not sustained at V₂, while the PSQI daytime dysfunction scale showed no change at V₁ but improved at V₂, suggesting that daytime symptoms may take longer to improve than nighttime sleep_.

We found no change in nighttime sleep measured by actigraphy. Daytime sleep (by actigraphy) increased (worsened) from baseline to V₁ but showed no overall change from baseline to V₂. A prior staff education study that included actigraphy showed less daytime and more nighttime sleep in the intervention versus control groups.³⁴ In that study participants had severe dementia and the intervention was implemented by research staff. Results in prior research involving research staff implementation of interventions have had mixed results, with some reporting improved sleep by actigraphy,^35–37 and another reporting no change.³⁸

Using direct observations of enrolled residents, we found that, with intervention, residents were more awake and out of bed during the daytime at post-intervention, but this was not sustained (it was reversed) at V_2. The reason for this is unclear; however, in the absence of a control group, it is difficult to know whether this trend represents a natural worsening of sleepiness among residents over time. In addition, we were not able to directly observe staff behavior due to the pragmatic nature of this trial (i.e., observation itself can change behavior in unnatural ways) and privacy of non-enrolled residents.

PHQ-9 depression scores improved from baseline to V₁ but were not sustained at V₂, and there was no change in measures of anxiety. These findings of improvement in depression but not anxiety are consistent with results from a study in Turkey in which research staff provided CBT to cognitively intact residents in NHs.³⁹

We found no immediate changes in ADL function or cognitive function following the intervention, but some improvement in ADLs at the post-intervention quarter. This is consistent with the PSQI findings showing delayed improvements in daytime symptoms after the intervention began.

While this study had multiple strengths including its pragmatic approach, diversity of patients and staff and interdisciplinary mentoring team, there are notable limitations. First, we do not know the frequency and intensity of sleep-promoting strategies used by facility staff in their direct work with individual residents. Staff were responsible for all residents on each unit (not only enrolled residents), and non-consented residents may have benefited from the intervention in ways that could not be captured within the study. Also, some factors contributing to poor sleep may not have been addressed in the intervention and the etiology of sleep disturbance in SNFs is multifactorial. The COVID-19 pandemic impacted data collection and prevented sufficient data collection at the third facility resulting in a small sample size that prevented subgroup analyses.

Aside from data on general attendance at webinars and workshops, we lack individual staff data and it is not clear what specific strategies they implemented during their work shifts. Also, the relationship between participating staff characteristics (e.g., level of training) and resident-level outcomes is unknown. Informal comments at workshops suggested that facilities use staff working on non-engaged units and staff from outside agency on the intervention units may have diluted the impact of SLUMBER at some points during the study. As recently demonstrated,⁴⁰ staffing shortages and care provided by less familiar agency staff may independently contribute to decreased total nighttime sleep among residents.

Although our intervention provided staff with access to an interdisciplinary team of experts and allowed the ability to build on intervention approaches in real time, this is unlikely to be feasible in other SNF settings. However, SLUMBER focus on mentoring and multicomponent aspects of behavioral and environmental restructuring may be useful in other settings.

In conclusion, the SLUMBER intervention improved resident self-reported sleep quality and depression symptoms although no changes in actigraphy measures of nighttime sleep, ADLs or cognitive function were found. Improved subjective sleep quality and depression may be early indicators of benefit while outcomes related to daytime functioning may emerge later. The COVID-19 pandemic forced early termination of the study, so the full impact of the intervention is unknown. Content, strategies, and technology used in this study may provide direction for future models of sleep intervention among SNF residents.