Investigating sleep disturbance and its correlates among formerly homeless adults in permanent supportive housing

Abstract

Background:

Adults experiencing homelessness have a high burden of sleep disturbance, which may be reduced by accessing permanent supportive housing.

Objectives:

To assess sleep disturbances and their correlates, including demographics, activity level, health status, age-related health issues (e.g., functionality and cognitive impairment), substance use, and homelessness history in a sample of PSH tenants.

Research Design:

Cross-sectional survey design.

Subjects:

237 formerly homeless adults between 45 and 80 years old.

Measures:

The Patient-Reported Outcomes Measurement Information System (PROMIS) Sleep Disturbance short form was used to measure sleep disturbance.

Results:

28% of our sample had PROMIS scores indicative of a moderate or severe sleep disturbance. Functional impairment, pain, and mental health comorbidities were associated with increased sleep disturbance in multivariable linear regression analyses. The number of years a person experienced homelessness was inversely associated with sleep disturbance.

Conclusions:

This study supports the need to screen for sleep disturbances among PSH tenants. The findings suggest that supportive services in PSH may need to include integrated physical and behavioral health care, pain management, and interventions designed to address ADLs to improve tenant sleep. They also suggest that improved sleep may help reduce PSH tenant pain, impairment, and mental health symptoms among PSH tenants.

Introduction

Homelessness is associated with poor sleep,^1–6 insomnia, and daytime fatigue² due to the daily struggle to obtain quality sleep in emergency shelter or unsheltered environments that are typically noisy, overcrowded, uncomfortable and perceived as unsafe.⁶ This chronic sleep deprivation can contribute to disproportionately high rates of disease (e.g., obesity, cancer, depression)^7–10 and decades-early mortality^11–13 among homeless adults. Evidence-based solutions to homelessness—namely, permanent supportive housing (PSH) using a housing first approach, which refers to providing low-barrier affordable housing paired with flexible health and social services¹⁴—provide a stabilizing setting^15,16 that can alter a person’s sleep–wake environment and contribute to sufficient sleep. In the only known study that has examined sleep health in PSH, Henwood et al.³ found that rates of insomnia dropped considerably upon entering PSH for chronically homeless adults and remained lower throughout a 1-year follow-up period. Nevertheless, 22% of the sample consistently screened positive for insomnia throughout the first year in housing and more than 42% screened positive for insomnia at any given point after being housed. This compares to an overall age-adjusted prevalence of insomnia in the past year of 19% in a population-based sample of adults in the United States.¹⁷

Although poor sleep constitutes a health risk factor, it is also the case that disproportionately high rates of chronic physical and behavioral health conditions among PSH tenants^10,18 can contribute to disturbed sleep in a bidirectional relationship. Serious mental illness¹⁹ and substance use disorder,²⁰ along with other conditions such as cardiovascular,²¹ neurological,²² autoimmune,²³ and respiratory^24,25 (e.g., asthma and COPD) diseases, have all been associated with sleep disturbances. Further, chronically homeless adults, who are the target population for PSH, experience early onset of geriatric syndrome including functional and cognitive impairment, pain, and limited mobility,^7,8,26–29 which can negatively affect sleep.^30,31 Cumulative adversity including past traumas experienced by this population³² can also have a lasting impact on sleep.^33,34 To date, it is unclear if PSH can provide a stabilized sleep environment that can both improve daily sleep quality and quantity but also mitigate other correlated health and behavioral issues.

The current study sought to expand a limited body of research by investigating sleep disturbance in a sample of 237 formerly homeless adults living in PSH. This is the first study of PSH tenants to use the Patient-Reported Outcomes Measurement Information System (PROMIS) sleep disturbance measure, though the PROMIS measure has been utilized to assess sleep disturbance of adults who remain homeless.⁶ PROMIS measures were developed and validated to be psychometrically sound and standardized, allowing for assessment of many patient-reported outcome domains based on common metrics that allow for comparisons across domains, across chronic diseases, and with the general population.³⁵ The PROMIS sleep disturbance measure has been calibrated based on a reference sample that is a 50% mix from the general population and 50% from a clinical population with increased levels of chronic illness.³⁶ This allows for some comparison of sleep disturbances to a reference population that has a higher disease burden than the general population. In addition, we explored correlates of sleep issues, including demographics, activity level, health status, age-related health issues (e.g., functionality and cognitive impairment), substance use, and homelessness history to help elucidate other conditions that may contribute to sleep disturbance.

Methods

Participants and Procedures

Residents aged 45 or older in two PSH programs located in the Skid Row area of Los Angeles were recruited for participation as part of a larger study exploring the early onset of geriatric conditions of adults with prior experiences of homelessness.³⁷ Less than 10% of residents in either program was younger than 45 reflecting that the target population for PSH skews to older adults. Residents were invited to participate in the study through a mailed letter explaining the purpose of the study and inviting them to contact the research team. The letter also indicated that residents could enroll in the study when interviewers were on site at their respective buildings. For the larger PSH program, which had more than 1,000 residents, determining who received invitations was done by sampling individuals aged 45 or older using a random number generator. For the smaller PSH program, all residents aged 45 and older were invited to join the study since there were only 100 residents that qualified. Participants older than 45 were eligible if they spoke English and could pass a screener for delirium.

During recruitment, 506 residents from both sites were mailed letters inviting them to participate. 275 people were screened in person for eligibility by the research team. 32 (11%) potential participants were screened as ineligible for various reasons (not part of randomized sample, too young, not living in PSH). Of those who were eligible (n=243; 48% response rate), three participants scheduled interviews, but did not show up and could not be rescheduled. Three more participants were disqualified based on the delirium screener administered during the informed consent process. This resulted in a final sample of 237 participants who were consented and asked to complete a standardized in-person interview querying basic demographics, physical and mental health, cognitive impairments and other geriatric-specific constructs, diet and physical activity, and socioeconomic domains (e.g., number of years spent homeless). Each interview took about 90 minutes to complete, and participants received $25 as compensation. The study was approved by the Institutional Review Board at the University of Southern California. Further details of the overall study can be found elsewhere.³⁷

Measures

Participants were queried regarding the number of years they were homeless prior to being housed (i.e., “If you add all the times in your life, how many months (or years) have you been homeless?”) and the number of years they had lived in permanent supportive housing (“How long have you been living in your apartment”). Age, race, gender and education were obtained with standard questions. Race was coded as white, African American, Latinx, and multiracial or another race or ethnicity, and gender was coded as male vs. female (one transgender participant was excluded from these analyses, as it was not possible to run regression analyses with a 3-category gender variable where one category has only a single respondent and we did not want to misrepresent this person’s gender by assigning them to a male or female category). Diagnoses of current physical and mental health conditions were queried using a list of conditions based on the National Health Interview Survey³⁸ and recoded as indicators of specific types of physical health conditions. Participants were presented with a list of conditions and asked if they had been diagnosed with each. Indicators were created for diabetes (“Diabetes (high blood sugar”), autoimmune disorders (“Lupus (systemic lupus erythematosus)”; “HIV or AIDS”; “Polymyalgia rheumatica”; “A thyroid problem”), cardiovascular disease (“Heart attack, coronary heart disease, clogged arteries in your legs”; “A stroke, cerebrovascular accident, blood clot or bleeding in the brain, or transient ischemic attack (TIA)”; “Heart failure (you may have been short of breath and the doctor may have told you that you had fluid in your lungs or that your heart was not pumping well”) , neurological disorders (“Movement disorders, such as Parkinson’s Disease, ALS, or Lou Gehrig’s disease”; “Epilepsy or seizures”; “Alzheimer’s disease, or another form of dementia”), and respiratory conditions (“Emphysema, chronic bronchitis, or chronic obstructive pulmonary disease”; “Asthma”). An indicator of serious mental illness was created if participants self-reported that they had been diagnosed with depression, bipolar disorder, or schizophrenia. Participants reported whether they were taking any medications, including specific medications for diabetes, rheumatoid arthritis, asthma, or lung disease; an indicator variable was created for any medication use.

An activities of daily living (ADL) screener assessed reported difficulty with five activities (bathing, dressing, eating, transferring, toileting)³⁹; based on existing literature, those who have difficulty performing one or more basic ADLs were coded as ADL impaired.⁸ The Minnesota Leisure Time Questionnaire⁴⁰ was used to assess level of physical activity; physical activities were converted in kilocalories (kcals) per week (M = 1,737), and a calculated variable divided kcals/week by 100 to facilitate regression interpretation (i.e., without standardization the regression beta coefficient is exceedingly small, <0.0001). Participants were asked the extent to which pain had interfered with general activity in the past week (0 = did not interfere to 10 = completely interfered). Participants were asked what substances they had used in the past 30 days, and indicators were created for any use of illicit substances (heroin, cocaine, or methamphetamines) and alcohol to intoxication (3 or more drinks in a sitting or 5 or more drinks in one day).

Sleep disturbance was measured using the PROMIS sleep disturbance 4-item short form,³⁶ a reliable measure to assess the presence of sleep disturbance among adults without focusing on the presence or absence of specific sleep disorders. The PROMIS short form includes four 5-point Likert scales that respondents use to rank their general sleep quality, the extent that their “sleep was refreshing,” and if they had “a problem with [their] sleep” or “difficulty falling asleep” during the past 7 days, with higher scores corresponding to greater sleep disturbance. PROMIS T-scores were derived by uploading participant responses to these four items to the Health Measures Scoring Service,⁴¹ a free web-based platform that uses response pattern scoring to calculate normalized T-scores for PROMIS measures. The PROMIS sleep calibration sample was a mixed clinical and community-based adult sample, which generally has a greater burden of chronic illness than the general population.³⁶ These normalized T-scores have a mean of 50 and standard deviation of 10, meaning a T-score of 60 indicates sleep disturbance that is one standard deviation worse than the average for the calibration sample. T-scores can be converted to cutoff points indicating mild, moderate, and severe sleep disturbance (severe: ≥70 or 2 SDs; moderate: ≥60 or 1 SD; mild: ≥55 or 0.5 SD; or within normal limits: <55) to aid in interpretation of scores; however, for analytic purposes we treat this as a continuous measure, given that a primary benefit of this measure is its ability to provide precise estimates of sleep disturbance for individuals.^36,42,43

Data Analyses

Analysis started with descriptive statistics for variables found to be associated with sleep in existing research findings, as well as demographic control variables. We then estimated a series of bivariate regressions between each of these variables and the sleep disturbance T-score, including demographics, health (medication use, physical health conditions, and serious mental illness), ADL impairment, physical activity level, pain, and substance use. Because all variables are either motivated by existing research findings or demographic controls, all were included in the subsequent multivariable linear regression model examining correlates of sleep disturbance in PSH tenants. There is very little missing data in this sample and so we utilized a complete case analysis approach with an analytic n of 235 in the regression model (the two excluded participants are one who declined to answer the race/ethnicity item and the transgender respondent excluded as described above). All analyses were conducted in Stata 16.

Results

As noted in Table 1, about 60% of the 237 participants in this study identified as male and more than 60% identified as African American, with a mean age of 58 years (range: 45–80 years old). Two thirds of the sample had a high school diploma or more education. Participants reported being homeless in the past for an average of 8 years (range: < 1–45 years) and living in PSH for an average of nearly 5 years (range: <1–22 years). Nearly 90% reported that they currently took medication and 75% reported a serious mental illness diagnosis. Nearly a quarter (23%) had been diagnosed with diabetes, 22% with an autoimmune disorder, 69% with a cardiovascular disease, 18% with a neurological disorder, and 23% with respiratory disease. The average number of kcals associated with physical activity in the past week was 1,734 (table presents 17.37, the mean of the standardized variable used in regression). Around 40% of respondents were classified as ADL impaired. On the pain interference scale (0 = no interference and 10 = complete interference), participants reported an average score of 4.4. Illicit substance use in the past month was reported by 16% and alcohol to intoxication by 22%. Participants had a mean sleep disturbance T-score of 53.3; converting the T-score to cutoff points of sleep disturbance severity, 55% could be categorized as having normal sleep, 17% as mild sleep disturbance, 23% as moderate sleep disturbance, and 5% as severe sleep disturbance.

Table 1.

Sample characteristics and relationships with sleep disturbance score, formerly homeless adults (N = 237)

			Bivariate Regression
	M (n)	SD (%)	b	SE
Age (years)	57.72	10.79	−0.37	0.11**
Race
White	43	18.2	referent
African American	144	61.0	0.92	1.87
Latinx	17	7.2	3.77	3.08
Multiracial, another race or ethnicity	32	13.6	2.47	2.51
Gender
Female	87	36.9	−0.96	1.45
Male	149	63.1	referent
Education
Less than high school	81	34.2	referent
High school or more	156	65.8	0.42	1.48
Time homeless before housing (years)	7.94	8.12	−0.14	0.09
Time in PSH (years)	4.75	3.68	−0.48	0.19*
Taking any medication	213	89.9	2.69	2.32
Physical health conditions
Diabetes	55	23.2	0.93	1.66
Autoimmune	51	21.5	0.92	1.71
Cardiovascular	164	69.2	0.41	1.52
Neurological	42	17.7	2.72	1.83
Respiratory	60	23.3	3.16	1.60
Serious mental illness	178	75.1	6.88	1.56**
ADL impairment	99	41.8	6.61	1.36**
Average daily physical activity	17.37	21.45	−0.04	0.03
Pain interference	4.40	4.45	0.74	0.15**
Illicit substance use (30 days)	38	16.0	0.12	1.91
Alcohol to intoxication (30 days)	53	22.4	0.05	0.11
Sleep disturbance
T-score	53.26	10.79	---
Severity
Normal	131	55.3	---
Mild	40	16.7	---
Moderate	54	22.8	---
Severe	12	5.1	---

^*p < .05.

^**p < .01.

Also shown in Table 1, statistically significant bivariate relationships were found between sleep and age (b = −0.37, p = .001), years living in PSH (b = −0.48, p = .01), serious mental illness diagnosis (b = 6.88, p = <.001), ADL impairment (b = 6.61, p = <.001), and level of pain interference with daily life (b = 0.74, p = <.001). Results from the multivariable regression are detailed in Table 2. Bivariate relationships between sleep and age or years in PSH did not persist in the multivariable model; all other bivariate relationships remained statistically significant in the multivariable model, though the strength of the relationship diminished slightly. After adjusting for all other covariates, there was a statistically significant negative association between years of homelessness prior to housing and sleep disturbance (b = −0.23, p = <.01). Women had statistically significantly lower levels of sleep disturbance than men (b = −3.48, p = .02). Results revealed statistically significant positive associations between increased sleep disturbance scores and ADL impairment (b = 4.59, p < .01), the extent to which pain interferes with activity (b = 0.47, p < .01), and diagnoses of serious mental illness (b = 4.33, p = .01). There were no statistically significant associations between sleep disturbance and other demographics, physical health diagnoses, taking medications, physical activity, or substance use after adjusting for all other covariates (ps > .05).

Table 2.

Results from multivariable linear regression examining the relationship between sleep disturbance, health, mental health, and substance use, with demographic controls, in a sample of formerly homeless older adults (n=235)

	b	SE
Age (years)	−0.14	0.12
Race
White	referent
African American	0.51	1.80
Latinx	2.22	2.94
Multiracial, another race or ethnicity	−0.58	2.39
Gender
Male	referent
Female	−3.48	1.44*
Education
Less than high school	referent
High school or more	0.43	1.41
Time homeless before housing (years)	−0.23	0.08**
Time in PSH (years)	−0.24	0.20
Taking any medication	−1.81	2.31
Physical health conditions
Diabetes	−0.56	1.64
Autoimmune	0.65	1.69
Cardiovascular	0.13	1.46
Neurological	0.59	1.89
Respiratory	1.66	1.58
Serious mental illness	4.33	1.66*
ADL impairment	4.59	1.56**
Average daily physical activity	−0.01	0.03
Pain interference	0.47	0.16**
Illicit substance use (30 days)	−1.61	1.88
Alcohol to intoxication (30 days)	−0.02	0.10
N	235
R2	.23
AIC	1,758.54

^*p < .05.

^**p < .01.

Discussion

Findings from this study show that 28% of our sample had PROMIS scores indicative of a moderate or severe sleep disturbance. This compares to only 20% in a reference population that has a higher disease burden than the general population,³⁶ which is consistent with a previous published study that showed higher rates of insomnia among PSH tenants compared to a population-based sample of adults in the United States.³ The continuation of moderate to severe sleep disturbances in this population after acquiring stable housing in the form of PSH further supports the need to screen for sleep disorders and disturbances in PSH programs and test interventions to improve sleep quality for this population. PSH may provide an ideal setting to explore evidence-based treatment approaches for insomnia, for example, that also consider comorbid physical and mental conditions such as Cognitive Behavioral Therapy.⁴⁴

Our results also indicate that disease burden—specifically ADL impairment, pain, and mental health conditions—was positively correlated with sleep disturbance in this population, similar to past research among homeless adults.⁶ History of homelessness was inversely associated with sleep disturbance, such that individuals who had experienced homelessness for longer periods had lower sleep disturbance. Although the effect was small, the mechanism behind this relationship is currently unclear though it may speak to a degree of distress tolerance accumulated with years of sleeping in chaotic, unstable environments.⁴⁵

Surprisingly, this study did not find substance use or taking prescribed medications to be associated with sleep disturbances, though the use of substances has been cited as a sleep aid in qualitative inquiries exploring barriers to sleep among homeless individuals.^6,20 Furthermore, it was expected that engaging in physical activity would be associated with sufficient sleep but no significant relationship was found, which is contrary to other research on physical activity and sleep problems among homeless adults.⁴⁶ Future research involving accelerometry could provide a more complete understanding of the relationship between physical activity and sleep in this population.^46–48

This study is primarily limited by the cross-sectional and post-housing-only design that prevented inference of causality. Findings may not be generalizable to PSH programs in other localities. In addition, the limited sample size may account for why some expected relationship were not detected (e.g., sleep disturbance and alcohol use). The PROMIS measure relies on self-reported data and has not been validated specifically in this population. Future research should consider leveraging the use of wearable devices to objectively measure sleep patterns.⁴⁷ As noted, some measures used in the study are not congruent temporally. For example, it may be that the assessment of substance use including alcohol, which was assessed during the past 30 days, needs to be better aligned to the 7-day sleep disturbance measure time period to detect a relationship. Alternately, the use of intensive longitudinal methods such as ecological momentary assessment may be needed to understand the more immediate relationship or interaction between substance use and sleep.⁴⁹ Further, we did not have detailed information about the specific medications each person was taking to uncouple disease-related and medication-related effects on sleep disturbance, or even if some were using medication to address sleep disturbances.⁵⁰

This study further supports the need to address sleep disturbance in formerly homeless adults who now reside in PSH. Though these individuals have been provided with an evidence-based approach to end homelessness, the findings of this study suggest that supportive services in PSH may need to include integrated physical and behavioral health care, pain management, and interventions designed to address ADLs to improve tenant sleep. Quality sleep is an essential component of healthy functioning and disease prevention (9) and improved sleep may help reduce PSH tenant pain, impairment, and mental health symptoms. Cost-effective approaches may to address sleep quality may include training existing service providers in Brief Behavioral Treatments for Insomnia.⁵¹ Future research should address the varying impact of sleep environment characteristics, such as bedding, noise, and light, or sleep hygiene (e.g., screen time prior to sleep) in PSH which may account for variance in sleep disturbances and provide insights for similar co-living situations,^52,53 which could impact policies on the design and development of PSH.