Prevalence and correlates of obstructive sleep apnea in urban-dwelling, low-income, predominantly African-American Women

Abstract

Study objectives:

The current study examined the prevalence and correlates of obstructive sleep apnea in a sample of low-income, predominantly African-American women using two waves of data.

Methods:

Participants were adults from two urban neighborhoods who enrolled in the PHRESH Zzz Study (N = 828; Pittsburgh Hill/Homewood Research on Neighborhoods, Sleep, and Health). A subsample who reported never receiving OSA diagnosis completed home sleep apnea testing in 2016 (n = 269, mean age 55.0 years, 79.6% female) and again in 2018 (n = 135). Correlates of OSA tested included demographic and anthropometric variables, health behavior/conditions, psychological distress and general health, smoking status, actigraphy-measured sleep, and neighborhood factors measured at baseline.

Results:

18.0% of all 2016 participants reported receiving physician diagnoses of OSA. Among those who completed in-home assessment, 19.3% had AHI ≥ 15 and 33.8% had AHI ≥ 5 plus one or more sleep symptoms. Estimates of the prevalence of OSA in all 2016 participants were 33.8% to 45.7% based on physician diagnoses and AHI results, depending on the criteria used. Age, gender, BMI, blood pressure, habitual snoring, neighborhood walkability, actigraphy-measured sleep characteristics, and smoking were concurrently associated with OSA in 2016. Changes in AHI categories from 2016 to 2018 were documented.

Conclusions:

Low-income African Americans, including women, are a high-risk group for OSA, but remain under-diagnosed and under-treated. The current findings show a high prevalence of OSA in African-American women and are among the first to demonstrate that both individual and neighborhood factors are implicated in OSA prevalence.

Introduction

African Americans are disproportionally affected by, and undertreated for, obstructive sleep apnea (OSA).^1–3 Blacks not only exhibit a higher prevalence of OSA than non-Hispanic Whites,^4,5 they are also more likely to develop OSA at a younger age⁴ and have more severe sleep-disordered breathing,⁶ and are less likely to receive screening and be adherent to treatment for OSA.¹ The high prevalence and inadequate treatment of OSA in African Americans may contribute to the alarming health disparities among African Americans, as they experience worse health outcomes than any other racial/ethnic groups in the U.S.^7,8 Several risk factors and correlates for OSA are well-documented across racial/ethnic groups, including male gender, older age, obesity as indexed by body mass index (BMI), and snoring in non-Hispanic Whites,^9–11 Hispanic/Latino, ¹² and Blacks. ¹³ However, several risk factors and correlates of OSA that were previously reported from primarily white samples, including waist circumference, hypertension, diabetes, sleepiness, insomnia, and actigraphy-assessed sleep duration and efficiency, were not supported in a large sample of African-American adults after controlling for age, sex, and BMI.¹³ Furthermore, evidence of gender differences in OSA prevalence and risk factors also calls for further examination of OSA in women.^14,15 Sex affects the symptomatology and presentation of OSA,¹⁶ such that women with OSA report lower scores on sleepiness,¹⁷ and have lower AHI¹⁸ and relatively more hypopneas,¹⁹ compared to men. Certain risk factors, such as waist-hip ratio, may have a stronger association with OSA severity in men than women.²⁰ Hence, more research is needed to further examine OSA prevalence and risk factors in racial/ethnic minorities and in women.

Few studies have examined neighborhood characteristics in relation to OSA risk in socio-economically disadvantaged minority communities. Neighborhood characteristics such as the walkability, perceived safety, and overall aesthetics of the community, may be implicated in OSA risk by influencing physical activity levels, sedentary behavior, and exposures to environmental pollutants.^21,22 In a large, multi-ethnic U.S. sample of adults (N = 1,896), lower perceived walking environment quality (e.g., endorsing “it is a pleasure to walk in my neighborhood”) was associated with greater severity of sleep apnea, especially in male and obese individuals.²¹ Furthermore, race/ethnicity-stratified analyses suggested that the associations between poorer walking environment and activity level and risk of OSA were strongest in African-American participants, compared to other racial/ethnic groups.²¹ In a clinical sample of patients with probable sleep-disordered breathing (N = 1,789, including 43% African-American patients), neighborhood-level crowding, a proxy for poor housing environment, was associated with severity of sleep-disordered breathing, and BMI partially mediated this association.²³ Among African-American participants only, the association between neighborhood-level crowding and sleep-disordered breathing was attenuated and no longer statistically significant after adjusting for sex, age, and BMI.²³ In contrast, in a predominantly white sample (N = 1,298), there was no association between perceived neighborhood quality (e.g., perceptions of crime, litter, and pleasantness in the neighborhood) and self-reported physician diagnosis of OSA.²⁴ Given these results and the limited extant literature, it is critical to examine both individual and neighborhood level factors in OSA risk in African Americans.

The goal of the current study is to examine the prevalence and correlates of OSA in urban-dwelling, predominantly African American women residents from two socioeconomically disadvantaged neighborhoods. We expected that there would be a high prevalence of OSA, and that male gender, older age, higher BMI, presence of comorbid conditions (hypertension and diabetes), presence of sleep concerns, and greater neighborhood disadvantages would be associated with OSA. We also tested for sex differences in the prevalence and correlates of OSA. In addition, using two waves of data, we documented change in AHI over a 2-year period and explored correlates of change in AHI.

Methods

Study population and participants

Participants in the current study were enrolled in the Pittsburgh Hill/Homewood Research on Neighborhood Change and Sleep study (PHRESH Zzz),²⁵ which is part of a series of studies that build upon the Pittsburgh Hill/Homewood Eating, Shopping, and Health (PHRESH) study.^26,27 About 32%–35% of the households were living below poverty line based on 2006–2009 American Community Survey estimates.²⁸ Started in 2011, the original PHRESH study recruited a random sample of households from two low-income, predominantly African-American neighborhoods in Pittsburgh, Pennsylvania. The primary goal of the PHRESH studies was to examine how changes in the neighborhood environment affect a range of health behaviors and outcomes among residents, including sleep. The current study used data collected in 2016 and 2018. The total sample size for 2016 survey participants was 828. Figure 1 is a flow chart of study participation.

Figure 1.

Study Flowchart.

Data collection included individual-level data (e.g., sleep, sociodemographic variables) collected during an in-home assessment with participants (originally recruited as the primary food shopper for the household) and neighborhood-level characteristics. All participants provided informed consent. The study protocol was approved by RAND Institutional Review Board.

Physician Diagnoses of OSA

At baseline in 2016, all survey participants were asked whether they had ever received a diagnosis of OSA from a physician. The analytic sample for this variable was 824, with 4 missing values.

Objective assessment of OSA

In 2016, a subsample (described below) of PHRESH Zzz participants (N = 291) were invited to participate in an in-home sleep apnea assessment using the ApneaLink Plus, an FDA-approved, a Type 3 home sleep apnea testing (HSAT) device.^29,30 Based on budget and equipment limitations, the in-home assessment was limited to the first 291 eligible participants who agreed to participate in the supplemental assessment. To be eligible for the in-home apnea assessment, participants were excluded from a previous diagnosis of OSA (148 out of 828 total participants in the 2016 survey reported a previous diagnosis with apnea and were ineligible for the study).

Eligible participants who completed the HSAT were significantly younger than eligible participants who did not complete the home sleep apnea test, but the two groups were similar in other demographic characteristics (sex, BMI) and whether they screened positive for OSA based on the Berlin Questionnaire that was part of the larger survey. Participants’ data were excluded from analysis if there were less than 3.5 hours of evaluation duration¹ (n = 22 excluded), resulting in the analytic sample of 269 in 2016. In 2018, participants who completed the HSAT in 2016 and did not initiate OSA treatment between 2016 and 2018 were invited to participate in the same in-home apnea assessment. Of the 269 participants with available data in 2016, 49 were lost to follow-up (i.e., did not complete any of the 2018 PHRESH assessments), 6 were excluded because they had initiated OSA treatment, 65 refused re-assessment, and 14 were screened but their data could not be used due to < 3.5 hours of evaluation duration. Hence, the analytic sample for the change in AHI included 135 participants with data from both 2016 and 2018.

AHI was calculated as the sum of apneas and hypopneas divided by total evaluation time in hours. 4% desaturation threshold was used for identifying hypopneas.³¹ Hypopnea was scored with reduction of 30% and then edited by technician to ensure accuracy. OSA severity was classified according to the AHI categories: AHI < 5 (normal), 5 ≤ AHI < 15 (mild OSA), 15 ≤ AHI < 30 (moderate OSA), and AHI ≥ 30 (severe OSA). All participants received feedback about their AHI results and were provided information and recommendations for further clinical evaluation for elevated AHI or if they had concerns about their sleep at 2016 or 2018.

Objective measures of sleep

Sleep was measured using the ActiGraph GT3x+, which has been validated against polysomnography and Actiwatch for measuring sleep-wake characteristics.^32–35 Participants were asked to wear the accelerometer on their wrist continuously for 7 consecutive days. Participants with < 4 nights of data were excluded from the analyses concerning actigraphy (n = 24 excluded, of whom 17 were missing all actigraphy data and 7 had actigraphy data for < 4 nights). The average number of nights of actigraphy for the analytic sample (n = 245) was 6.80 nights (SD = 0.60, range = 4 – 7). Identification of sleep intervals was based on a combination of reported bedtimes and waketimes from the sleep diary and visual inspection of actigraphy data. The Cole-Kripke algorithm³⁶ was used to determine the sleep and wake periods and to derive the sleep variables, including sleep duration, wake after sleep onset (WASO), and sleep efficiency.

Sleep variables were averaged across all nights with available actigraphy data. Sleep duration was calculated as the total sleep time in minutes, i.e., the total amount of time slept during the time in bed. Sleep efficiency was calculated as sleep duration divided by the total time in bed (x100). WASO was calculated as the total number of minutes scored as awake after initial sleep onset. All three sleep variables were analyzed as continuous measures in the primary analyses, with higher scores indicating greater sleep duration, higher sleep efficiency, or longer WASO.

Self-reported sleep symptoms

The Berlin Questionnaire was administered to all participants. A widely-used self-report screener for OSA,³⁷ it includes questions assessing three categories of risk factors: 1) habitual snoring, 2) sleepiness, and 3) hypertension or BMI > 30. We used habitual snoring and sleepiness in the current paper. Both categories were scored as present or positive if the total score ≥ 2, according to the scoring instruction of the Berlin Questionnaire.

Neighborhood factors

Four sets of variables were included as neighborhood factors: 1) an objective measure of walkability, 2) crime exposure derived using police department data, 3) measures of perceived neighborhood characteristics (i.e., safety and satisfaction) and 4) perceived housing conditions.

Walkability index³⁸ was derived from observations of neighborhood street segments in both neighborhoods that addressed the following items: traffic signs at the intersection (4 points) pedestrian crossings (2 points), sidewalks (10 points), lighting (2 points), transit (2 points), and mixed use (2 points). Items were summed up for each street segment, and a weighted average across street segments was derived from the sampled segments within a quarter mile of a participant’s home; weights were proportional to the length of each segment (Cronbach’s α = 0.55 for the total 2016 sample). The scale score ranges from 0 to 22, with higher scores indicating greater walkability.³⁸ A detailed description of the rating procedures and derivation of the walkability index was described in Troxel et al.²⁵

Total crime was derived using 2016 and 2018 incident-level crime data provided by the City of Pittsburgh police department. Using the ArcGIS 10.2 software, street network distances from each household to each approximate crime location was calculated. 95 percent of the incident-level crime data were geocoded using the address information from the raw data. We derived the total number of crimes that occurred within a 0.1-mile network distance for each household for 2016 and 2018 separately, following prior neighborhood research.^39,40

Perceived neighborhood safety was measured using a 4-item scale validated in prior literature.⁴¹ Participants rated on a 5-point Likert scale, ranging from 1 (strongly agree) to 5 (strongly disagree), the following items: “You feel safe walking in your neighborhood during the day,” “You feel safe walking in your neighborhood at night,” “Your neighborhood is safe from crime,” “Violence is a problem in your neighborhood (reverse coded)” (Cronbach’s α = 0.70). A summary score of the 4 items was derived by averaging responses, with higher scores indicating greater perceived neighborhood safety.

Perceived neighborhood satisfaction was measured using a single-item, rated on a 5-point Likert scale ranging from 1 (very dissatisfied) to 5 (very satisfied): “All things considered, would you say you are very satisfied, satisfied, dissatisfied, very dissatisfied or neutral - neither satisfied nor dissatisfied- with your neighborhood as a place to live?” This single-item measure has been used in prior research.⁴² Higher scores indicate greater perceived neighborhood satisfaction.

Perceived housing conditions were measured using 7 items drawn from a prior study examining the impact of neighborhoods changes (via housing vouchers) on participants’ health and well-being.⁴³ Participants were asked to rate “a big problem,” “a small problem,” or “no problem at all” in their homes or apartments on each of the following issues: (1) peeling paint or broken plaster, (2) plumbing that does not work, (3) rats or mice, (4) cockroaches, (5) broken locks or no locks on door to unit, (6) broken windows or windows without screens, and (7) a heating system that does not work. Similar to a prior study,⁴⁴ due to low base rates for some of the individual items, we derived a binary variable of any housing distress, where 1 indicates one or more of these seven issues being rated as a small or big problem, and 0 indicates a rating of no problem at all for each item.

Individual-level factors

Demographic variables.

Participants’ age (continuous), sex (female or male), educational achievement, employment status (employed or unemployed), and annual family income (continuous) were examined. Education was categorized into less than high school, high school or GED, some college or training, and college degree or higher.

Anthropometric variables.

Height was measured to the nearest eighth inch using a carpenter’s square (triangle) and an eight-foot folding wooden ruler marked in inches. Weight was measured to the nearest tenth pound using a Seca Robusta 813 digital scale. BMI (kg/m²) was calculated from participants’ measured height and weight. BMI was analyzed as a continuous variable.

Health behaviors/conditions.

Smoking status was assessed with a question asking whether the participant currently smokes during the 2014 survey. We classified someone as a smoker if they reported smoking in 2014. For hypertension, we included measurements of systolic (SBP) and diastolic blood pressure (DBP) as well as self-reported hypertension and hypertensive medication use. Two blood pressure measurements (taken 60 seconds apart) were obtained during an in-home assessment using a Micro Life automated blood pressure monitor after the participant had been seated for five minutes. The average of the two measurements was used to calculate the average SBP and DBP, respectively. For diabetes, we included hemoglobin A1c (HbA1c) level and a binary variable of HbA1c > 6.5% or prior diagnosis for diabetes. HbA1c was measured via collection of non-fasting blood samples in the research clinic or in the participant’s home. Blood samples were obtained from the antecubital vein by a trained phlebotomist while the participant was seated. Assays for HbA1c were performed at the University of Pittsburgh Heinz Nutrition Laboratory at the Graduate School of Public Health.

In addition, we assessed general psychological distress and general health. General psychological distress was measured using the 6-item Kessler 6 (K6).⁴⁵ The K6 is a well-validated self-report measure that assesses the frequency of experiencing general psychological distress symptoms (e.g., “feel nervous,” “feel hopeless”) during the last 30 days (Cronbach’s alpha= .86).⁴⁶ General health was measured using a single-item measure, asking the participant to rate their general health on a Likert scale of 1 (excellent) to 5 (poor).

Statistical analyses

Data analyses were completed in SAS Version 9.4. We reported the AHI categories in the subsample (n = 269) that completed the HSAT. To estimate the overall prevalence of OSA in the total 2016 sample (N = 828), we combined prevalence estimates across two distinct sub-groups of participants: (i) those self-reporting physician diagnoses of OSA (and ineligible for HSAT) (18%) (ii) and those not reporting a physician diagnosis of OSA and eligible to receive the HSAT (82%). The overall prevalence is a weighted combination of prevalence estimates in the two groups, with the weights representing sample proportions. Everyone in the first group is assumed to have OSA. In the second group, prevalence is estimated using AHI results from the HSAT conducted with a sub-sample. Therefore, this calculation assumes that 1) participants report of physician diagnosis is accurate, and 2) participants who completed the HSAT are representative of those who did not complete the HSAT. A gender-stratified bootstrap approach, with 1000 draws of the original sample, was used to obtain approximate 95% confidence intervals for the combined estimate of apnea prevalence.⁴⁷ We report two types of OSA prevalence, based on ICSD-3 criteria:⁴⁸ 1) OSA defined by AHI ≥ 15, and 2) clinically relevant OSA, defined by AHI ≥ 5 plus at least one sleep symptom (snoring or sleepiness). In addition, we reported the prevalence of self-reported physician diagnoses of OSA. Notably, for the second definition, we did not have a validated measure of insomnia, which is a deviation from the ICSD-3 criteria; thus we may underestimate the actual prevalence of clinically relevant OSA, particularly in women.

Analyses considering the correlates of AHI were conducted in two steps. Due to the skewness of the AHI variable, for these analyses, a binary AHI variable was used as the outcome variable for the analysis and was coded as: 0 = AHI < 15 (normal to mild apnea), 1 = AHI ≥ 15 (moderate to severe apnea). In step one, we examined whether participants with AHI < 15 and AHI ≥ 15 were significantly different on demographics, anthropometry, health behaviors and conditions, neighborhood factors, and sleep characteristics using Wilcoxon two-sample tests (for continuous variables) and Fisher’s exact tests (for categorical variables). Only the variables that differed between the two AHI categories were further evaluated in the next step. For this selection, we used p < 0.25 as the cutoff. ⁴⁹ In step two, logistic regression was used. Model 1 included the “standard” risk factors (i.e., sex, age, and BMI), and Model 2 included a series of expanded risk factors, examining the impact of each correlate on AHI after controlling for the standard risk factors. We also tested for possible sex differences in the correlates of OSA by adding the interaction term between sex and each correlate of interest for each of the expanded models (i.e., Model 2). In addition, we cross-tabulated AHI categories from 2016 and 2018 to provide descriptive statistics.

Results

AHI Results and Estimated Prevalence of OSA

Table 1 shows the AHI results from the in-home assessment at baseline in 2016 (n = 269) for the total subsample who completed the in-home assessment, overall and by sex. In the overall sample, 43.5% had normal AHI scores (< 5), 37.2% had mild sleep apnea (AHI: 5 - < 15), 12.3% had moderate sleep apnea (AHI: 15 - < 30), and 7.1% had severe sleep apnea (AHI ≥ 30). In total, 67.3% of men and 53.7% of women had AHI ≥ 5.

Table 1.

AHI Categories and Prevalence Estimates at Baseline (2016).

	Total Estimate (95% CI)	Male Estimate (95% CI)	Female Estimate (95% CI)
OSA prevalence in the subsample with AHI data	n = 269	n = 55	n = 214
AHI categories, %
Normal, < 5	43.5% (37.5%, 49.7%)	32.7% (20.7%, 46.7%)	46.3% (39.4%, 53.2%)
Mild, 5- <15	37.2% (31.4%, 43.3%)	36.4% (23.8%, 50.4%)	37.4% (30.9%, 44.2%)
Moderate, 15- <30	12.3% (8.6%, 16.8%)	16.4% (7.8%, 28.8%)	11.2% (7.3%, 16.2%)
Severe, ≥ 30	7.1% (4.3%, 10.8%)	14.5% (6.5%, 26.7%)	5.1% (2.6%, 9.0%)
Moderate to severe OSA (AHI ≥ 15)	19.3% (14.8%, 24.6%)	30.9% (19.1%, 44.8%)	16.4% (11.7%, 22.0%)
Clinically relevant OSA (AHI ≥ 5 plus sleepiness or snoring)	33.8% (28.2%, 39.8%)	49.1% (35.4%, 62.9%)	29.9% (23.9%, 36.5%)
OSA prevalence in the total 2016 sample	N = 828	n = 172	n = 656
Self-reported physician diagnosis a	18.0% (15.4%, 20.8%)	19.9% (14.2%, 26.7%)	17.5% (14.6%, 20.6%)
Physician diagnosis or AHI ≥ 15 b	33.8% (29.3%, 38.2%)	44.6% (34.2%, 55.9%)	31.0% (26.2%, 35.8%)
Physician diagnosis or AHI ≥ 5 plus sleepiness or snoring b	45.7% (40.4%, 50.5%)	59.2% (47.1%, 69.8%)	42.1% (36.1%, 47.1%)

Note.

^aMissing = 4 for this variable (N = 824).

^bThese are estimated prevalence of OSA combining the AHI data and physician diagnosis of OSA (see Statistical Analysis for details of this calculation).

In the subsample with AHI data (n = 269), the prevalence of moderate to severe OSA based on AHI ≥ 15 was 19.3%, with 30.9% in males and 16.4% in females. The prevalence of clinically relevant OSA based on AHI 5 - < 15 plus a sleep symptom (i.e., sleepiness or snoring) was 33.8%, including 49.1% in males and 29.9% in females.

In the total 2016 sample (N = 828), 18.0% of the participants (19.9% of males; 17.5% of females) reported a physician diagnosis of OSA. Among the 82.0% of study participants not reporting a physician diagnosis, we assumed that 19.3% would have OSA defined by AHI ≥ 15 if screened, and extrapolated that 82.0%*19.3% of the total 2016 sample would have OSA defined by AHI ≥ 15. The estimated prevalence of OSA in the total 2016 sample based on either physician diagnoses or AHI ≥ 15 was 18% + (82.0%*19.3%) = 33.8% (44.6% in males; 31.0% in females). Similarly, the estimated prevalence of OSA in the total 2016 sample based on either physician diagnoses or AHI ≥ 5 plus a sleep symptom was 18% + (82.0%*33.8%) = 45.7% (59.2% in males; 42.1% in females).

Descriptive Statistics

Table 2 presents the descriptive statistics of demographic variables and all study variables in 2016 for the total sample and by AHI categories. In 2016, the mean age of the total sample was 55.0 years (SD = 14.4 years). The sample included predominantly women (79.6%) and had low-income (mean annual family income $21,700), limited education attainment (13.4% having a college degree) and high unemployment (58.0%). Compared to those with AHI < 15, those with moderate or severe apnea (AHI ≥ 15) were significantly older, more likely to be men, had higher mean BMI, had higher mean measured blood pressure (both SBP and DBP), had lower mean neighborhood walkability, were more likely to have habitual snoring, had lower mean actigraphy-measured sleep efficiency and higher mean WASO, and were more likely to be smokers.

Table 2.

Demographic variables and sample characteristics in 2016 by AHI categories (n = 269)

Characteristics	n	Total (n = 269)	By AHI categories
			AHI < 15 (n = 217)	AHI ≥ 15 (n = 52)	p
		Mean (SD) or n (%)	Mean (SD) or n (%)	Mean (SD) or n (%)
Demographics
Age (years)	269	55.0 (14.4)	54.0 (15.1)	58.9 (10.6)	0.04
Female, n (%)	269	214 (79.6)	179 (82.5)	35 (67.3)	0.02
Education, n (%)	269				0.46
< High school		30 (11.2)	21 (9.7)	9 (17.3)
High school or GED		103 (38.3)	84 (38.7)	19 (36.5)
Some college/training		100 (37.2)	83 (38.2)	17 (32.7)
College degree or higher		36 (13.4)	29 (13.4)	7 (13.5)
Employed, n (%)	269	113 (42.0)	90 (41.5)	23 (44.2)	0.76
Annual family income	269	21,700	21,800	21,400
		(17,500)	(17,800)	(16,700)	0.96
Anthropometry
BMI (kg/m2)	269	30.7 (7.3)	30.1 (7.0)	33.4 (7.7)	0.002
Comorbiditv/conditions
Hypertension, self-report, n (%)	268	159 (59.3)	123 (56.9)	36 (69.2)	0.11
Systolic Blood Pressure	257	128.7 (19.7)	126.2 (18.9)	139.0 (19.8)	<.0001
Diastolic Blood Pressure	257	80.0 (11.7)	78.9 (11.2)	84.6 (12.6)	0.002
Hypertension or taking medication,* n (%)	257	166 (64.6)	131 (63.0)	35 (71.4)	0.32
HbAlc	l85	6.1 (1.3)	6.1 (1.3)	6.4 (1.4)	0.10
HbAlc ≥ 6.5% or prior diagnosis of diabetes, n (%)	185	57 (30.8)	42 (28.2)	15 (41.7)	0.16
Psychological distress (k6)	268	4.4 (4.3)	4.5 (4.2)	4.0 (4.5)	0.31
Mental distress (k6 ≥ 8), n (%)	268	54 (20.1)	47 (21.8)	7 (13.5)	0.25
Mental illness (k6 ≥ 13), n (%)	268	12 (4.5)	10 (4.6)	2 (3.8)	1.00
General self-rated health	269	3.0 (1.0)	3.0 (1.0)	3.0 (0.9)	0.83
Rated “fair” or “poor”	269	82 (30.5)	69 (31.8)	13 (25.0)	0.40
Any report of smoking in 2014, n (%)	231	89 (38.5)	67 (35.8)	22 (50.0)	0.09
Neighborhood factors
Walkability index	245	8.8 (1.9)	8.9 (2.0)	8.2 (1.6)	0.01
Total crime	253	22.5 (16.1)	22.3 (15.9)	23.3 (17.2)	0.70
Perceived neighborhood safety	269	3.0 (0.7)	3.0 (0.7)	3.0 (0.8)	0.60
Perceived neighborhood satisfaction	269	3.6 (1.1)	3.6 (1.1)	3.4 (1.2)	0.28
Perceived housing distress, n (%)	269	124 (46.1)	104 (47.9)	20 (38.5)	0.28
Sleep characteristics
Berlin questionnaire
Habitual snoring, n (%)	269	86 (32.0)	62 (28.6)	24 (46.2)	0.02
Sleepiness, n (%)	269	36 (13.4)	31 (14.3)	5 (9.6)	0.50
Actigraphv-measured sleep
Sleep duration (min)	245	341.4 (73.6)	346.6 (71.3)	320.2 (79.4)	0.07
Sleep duration < 7 hours, n (%)		209 (85.3)	165 (83.8)	44 (91.7)	0.25
Sleep efficiency (%)	245	75.0 (11.3)	76.3 (10.8)	69.7 (12.0)	0.001
Sleep efficiency < 85%, n (%)		197 (80.4)	152 (77.2)	45 (93.8)	0.01
WASO (min)	245	100.7 (55.4)	95.7 (52.1)	121.3 (63.8)	0.01

Note. AHI = apnea hypopnea index; GED = General Education Diploma; BMI = Body Mass Index; kg/m² = kilogram/meter squared; HbA1c = Hemoglobin A1c; WASO = Wake After Sleep Onset.

^*Hypertension defined as High BP (SBP ≥ 140 or DBP ≥ 90 or currently taking BP Medication).

We tested sex differences on all study variables at baseline in 2016. Women had higher mean BMI (M = 31.6 in women, M = 27.1 in men, p < 0.001), higher mean perceived neighborhood satisfaction (M = 3.7 in women, M = 3.2 in men, p < 0.01), and longer mean sleep duration (M = 5.8 hours in women, M = 5.3 hours in men, p < 0.05), relative to men. No sex differences were found for the other variables listed in Table 2.

Correlates of OSA

Table 3 presents the results of logistic regression predicting the odds of having AHI ≥ 15 in 2016 (reference group: AHI < 15) after adjusting for age, sex, and BMI (Model 1). Variables that distinguish individuals with AHI ≥ vs. < 15 in Table 2 (p < 0.25) were entered in the logistic regression analyses. Results shows that older age, male gender, higher BMI, higher SBP, higher DBP, presence of habitual snoring, smoking, and longer actigraphy-measured WASO were associated with increased odds of having moderate or severe OSA defined by AHI ≥ 15. Living in more walkable neighborhood and higher sleep efficiency were associated with decreased odds of having moderate or severe sleep apnea. We conducted a sensitivity analysis using clinically relevant OSA (AHI ≥ 5 plus a sleep symptom) as the outcome variable and found that the results were consistent with using AHI ≥ 15 as the outcome variable. We only present the results using OSA defined by AHI ≥ 15 as the outcome variable because of limitations in our definition of clinically relevant OSA (i.e., we did not have a validated measure of insomnia). We also conducted sensitivity analyses for the logistic regression combining AHI ≥ 15 or self-reported prior physician diagnosis of OSA, with AHI < 15 as the reference group. The results are presented in Supplemental Tables 1 and 2. These results were also consistent with those using AHI ≥ 15 (with reference group being AHI < 15) as the outcome variables, with the exception that walkability was no longer a significant correlate.

Table 3.

Logistic regression results showing odds ratios of having moderate or severe apnea (AHI ≥ 15) at baseline.

		Model 1	Model 2
		(age, sex, and BMI)	(Model 1 + each correlate of interest entered separately)
	n	OR (95% CI)	Adj. OR (95% CI)
Demographics
Age	269	1.45 (1.13, 1.86) **
Male	269	3.89 (1.81, 8.36) ***
Anthropometry
BMI (kg/m2)	269	2.67 (1.65, 4.31) ***
Comorbiditv/conditions
Hypertension, self-report	268		1.17 (0.56, 2.45)
Systolic Blood Pressure	257		1.37 (1.15, 1.63) ***
Diastolic Blood Pressure	257		1.71 (1.25, 2.34) ***
HbA1c	185		1.09 (0.82, 1.44)
HbA1c ≥ 6.5% or prior diagnosis of diabetes	185		1.31 (0.58, 2.95)
Mental distress (k6 ≥ 8)			0.66 (0.26, 1.66)
Smoking (yes/no, in 2014)	231		2.70 (1.25, 5.86) *
Neighborhood factors
Walkability index	245		0.83 (0.70, 0.99) *
Sleep characteristics
Presence of habitual snoring	269		2.23 (1.14, 4.37) *
Actigraphv-measured sleep
Sleep duration	245		0.75 (0.57, 0.98) *
Sleep efficiency	245		0.64 (0.48, 0.85) **
WASO	245		1.48 (1.06, 2.07) *

Note. OR = Odds Ratio. Adj. OR = Adjusted Odds Ratio. AHI = apnea hypopnea index; BMI = Body Mass Index; WASO = wake after sleep onset. Age was coded as a 10-year increase; BMI, systolic and diastolic blood pressure, and sleep efficiency were coded as a 10-point increase; sleep duration and WASO were coded as a 1-hour increase. Smoking was assessed as any report of current smoking in 2014 (this question was not asked in 2016).

^*p < .05

^**p < .01

^***p < .001

We also tested for effect modification by sex. Significant sex differences were found on age (p = 0.03) and the presence of habitual snoring (p = 0.002). Specifically, older age was associated with OSA in males (OR = 3.23 per 10-year increase, 95% CI: [1.35, 7.76]) but not in females (OR = 1.28, 95% CI: [0.98, 1.68]), adjusting for BMI. Habitual snoring was associated with OSA in females (OR = 4.14, 95% CI: [1.85, 9.24]) but not in males (OR = 0.36, 95% CI: [0.08, 1.68]), adjusting for age and BMI. No sex differences were found for the other variables listed in Table 3.

Change of AHI Categories from 2016 to 2018

Table 4 presents descriptive statistics related to the changes in objectively measured AHI categories from 2016 to 2018. The majority of participants (90%) who had AHI < 15 in 2016 remained in this category in 2018; only 10% worsened and moved to AHI > 15 in 2018. Among those who were in the AHI ≥ 15 category in 2016, 56% remained in this category and 44% moved down to the AHI < 15 category. In a set of exploratory analyses, we used baseline variables measured in 2016 to predict the increasing or stably high AHI from 2016 to 2018. Higher baseline levels of BMI, HbA1C, habitual snoring score, and being a smoker were associated with increasing AHI (from < 15 in 2016 to ≥ 15 in 2018) or having AHI ≥ 15 in both years, relative to decreasing AHI (from ≥ 15 in 2016 to < 15 in 2018) or having AHI < 15 in both years.

Table 4.

Cross-tabulation of AHI Categories from 2016 and 2018 (n = 135)

	AHI categories 2018
	AHI < 15	AHI ≥ 15	Total
AHI categories 2016	n (%)	n (%)	n (%)
AHI < 15	99 (90%)	11 (10%)	110
AHI ≥ 15	11 (44%)	14 (56%)	25
Total n	110	25	135

Note. AHI = apnea hypopnea index. No apnea (AHI < 5) was collapsed with mild apnea (AHI 5-<15), and moderate apnea (AHI 15- <30) was collapsed with severe apnea (AHI ≥ 30) due to small sample sizes in the majority of cells.

Discussion

Consistent with prior studies,^4–6,13,50 our results show a high prevalence of OSA in African American adults. A previous study of a large African-American cohort reported that 23.6% of the sample screened positive for OSA (AHI ≥ 15), out of whom only 5% reported a physician diagnosis of OSA.¹³ Our percentage (19.3%) of AHI ≥ 15 was slightly lower, possibly because we excluded participants previously diagnosed with OSA and our sample was predominantly women. The high prevalence of OSA, especially undiagnosed OSA (19.3% AHI ≥ 15) and AHI ≥ 5 (56.6%) in the subsample, suggests a critical need to increase clinical screening of OSA in minority groups and women. The ICSD-3 definition of OSA, based on a combination of symptoms and findings, has been criticized as being too inclusive,^51,52 resulting in high prevalence estimates of OSA. We acknowledge that the prevalence of OSA events or OSA syndrome are sensitive to the definitions being used, which depends on the purpose of the investigation (e.g., epidemiological, clinical case-finding, treatment). Our purpose in this paper was to provide an epidemiological study of OSA in an understudied, underserved population.

Our results regarding the correlates of OSA are generally consistent with prior studies in African-American samples ^6,13 and other racial groups.^9–12 In particular, we confirmed that habitual snoring but not sleepiness was associated with OSA in African-American adults.¹³ However, unlike the Johnson et al.¹³ study that did not find associations between actigraphy-measured sleep characteristics and OSA in African-American adults,¹³ we found that shorter actigraphy-measured sleep duration, poorer sleep efficiency, and greater WASO were associated with OSA. These differences may relate to the extremely poor sleep health in our sample; mean actigraphy-measured sleep duration in the current sample was only 5.68 hours (85.3% of the total sample had sleep duration < 7 hours) and mean sleep efficiency was 75% (80.4% of the total sample had sleep efficiency < 85%), both far lower than previous reports in African-American samples.^5,13 For instance, in the Johnson et al.¹³ study using the Jackson Heart Study cohort (with primarily college-educated African Americans in their early 60s and median family income around $43,000), mean actigraphy-measured sleep duration was 6.7 hours and 30% of the sample had sleep efficiency below 85%. These discrepant results suggest that sleep health is poorer among residents from vulnerable African-American neighborhoods compared to other African-American samples, and that the association between poor sleep and OSA is present among socio-economically disadvantaged African-American adults. In addition, our results are the first to show a relationship between an objectively measured neighborhood factor (i.e., walkability index) and OSA in African-American adults, suggesting that neighborhood level interventions may also be an important target for reducing sleep and other health disparities.

Although the two repeated measurements were only two years apart, we documented changes in AHI categories within this short period of time and reported basic correlates of the changes. All participants were given feedback of their apnea screening results from the in-home assessment in 201,6 including recommendations to consult with their provider if they had concerns about their sleep or had elevated AHI; however, only 6 participants had initiated treatment between 2016 to 2018. Though we made a research recommendation, this result is consistent with prior evidence that African Americans have low adherence to the physician’s recommendation for a sleep evaluation ⁵⁰ and to OSA treatment,^53,54 or may reflect lack of access to care or failure of medical providers to recommend further evaluation. Future study with a large sample size could examine associations between individual- and block or neighborhood-level characteristics and OSA using a multilevel modeling approach.

Several limitations should be noted. First, the sample size is sufficient to detect direct associations, but it is still relatively small. Therefore, the test of gender differences may have been under-powered due to small number of male participants. The changes in AHI from 2016 to 2018 might be due to measurement error at either time point. Future studies with large sample sizes could also examine individual- and block- or neighborhood-level characteristics that might be associated with OSA using a multilevel modeling approach. Secondly, our OSA prevalence estimates may be biased given limitations associated with self-reported physician diagnoses of OSA in the total sample and that the sample completing HSAT was significantly younger than the rest of the eligible participants. While prior evidence showed a high agreement between self-reported physician diagnoses and objective measures of OSA in a sample of nurses and health professionals,⁵⁵ in a sample of African-American adults only 5% of those with an objectively-measured OSA reported a physician diagnosis of OSA.¹³ Thus, both self-reported physician diagnoses and objective-measures of OSA obtained from a younger subsample may have resulted in underestimation of the actual prevalence. Our use of 4% (vs. 3%) desaturation used for hypopneas may also contribute to lower levels of OSA in our study.⁵⁶ We used 4% desaturation because AASM scoring manual version 2.6 states it is acceptable to use ≥ 4% oxygen desaturation from pre-event baseline ⁵⁷ and it is recommended by Centers for Medicare & Medicaid Services.⁵⁸

In addition, the ApneaLink device used in the current study does not provide sleep staging information. Therefore, we were unable to examine whether OSA during REM sleep in particular might be driving the associations between OSA and the outcomes for women. Future studies should examine this possibility, especially given the recent evidence of sex differences in AHI in NREM sleep (but not in REM sleep).¹⁵ Also, we did not formally assess insomnia, hence the clinically relevant OSA only included AHI ≥ 5 plus snoring or sleepiness, which, as mentioned, may underestimate actual prevalence, particularly in women who are at greater risk for insomnia. Women and men with OSA may have different clinical presentations (e.g., symptoms, PSG characteristics). Including a requirement of daytime sleepiness (rather than a report of tiredness) may further underestimate the prevalence in women compared to men.¹⁷ Furthermore, risk factors and correlates of OSA were examined using cross-sectional data collected in 2016, which precludes causal inferences on the directionality of these relationships. Finally, generalizability of these results to other disadvantaged communities needs to be assessed in future studies as the current sample was predominantly African-American females in their 50s. Nevertheless, African Americans and women are underrepresented and understudied populations in the extant literature concerning OSA. These results make a unique contribution by documenting correlates of OSA in socio-economically disadvantaged minority women.

To summarize, the current study reports a high prevalence of OSA (19.3% – 45.7%, depending on the methods and criteria used) based on self-reported physician diagnoses and in-home sleep assessment in a sample of socio-economically-disadvantaged, predominantly African-American women. The current study is the first to report an association between objectively measured neighborhood walkability and OSA in urban African-American adults. We are also the first to document changes in AHI over time. Future research should replicate the current findings using larger samples with more racial/ethnic and socioeconomic diversity, examine the mechanisms through which neighborhood characteristics influence OSA risk and how individual- and neighborhood-factors interact to influence OSA risk, and develop intervention strategies accordingly to reduce OSA risk among residents in vulnerable minority communities.

Highlights.

• Racial/ethnic minorities and women are understudied in the study of obstructive sleep apnea

• High prevalence of OSA was reported in low-income, predominantly African American women

• Objectively measured neighborhood walkability was associated with OSA

• Considerable changes were documented in Apnea–Hypopnea Index over a 2-year period

• Interventions targeting both individual- and neighborhood-level factors might be needed