Impact of obstructive sleep apnea on cardiometabolic health in a random sample of older adults in rural South Africa: building the case for the treatment of sleep disorders in underresourced settings

Abstract

Study Objectives:

The association between obstructive sleep apnea (OSA) and increased cardiometabolic risk (CMR) has been well documented in higher-income countries. However, OSA and its association with CMR have not yet been investigated, based on objective measures, in southern Africa. We measured polysomnography-derived sleep characteristics, OSA prevalence, and its association with cardiometabolic diseases in a rural, low-income, African-ancestry sample of older adult participants in South Africa.

Methods:

Seventy-five participants completed the study. Body mass index, hypertension, diabetes, dyslipidemia, and HIV status were determined. A continuous CMR score was calculated using waist circumference, random glucose, high-density-lipoprotein cholesterol, triglycerides, and mean arterial blood pressure. Sleep architecture, arousal index, and apnea-hypopnea index for detection of the OSA (apnea-hypopnea index ≥ 15 events/h) were assessed by home-based polysomnography. Associations between CMR score and age, sex, socioeconomic status, apnea-hypopnea index, and total sleep time were investigated by multivariable analysis.

Results:

In our sample (53 women, age 66.1 ± 10.7 years, 12 HIV+), 60.7% of participants were overweight/obese, 61.3% were hypertensive, and 29.3% had undiagnosed OSA. Being older (P = .02) and having a higher body mass index (P = .02) and higher waist circumference (P < .01) were associated with OSA. Apnea-hypopnea index severity (β = 0.011; P = .01) and being a woman (β = 0.369; P = .01) were independently associated with a higher CMR score in socioeconomic status– and age-adjusted analyses.

Conclusions:

In this South African community with older adults with obesity and hypertension, OSA prevalence is alarming and associated with CMR. We show the feasibility of detecting OSA in a rural setting using polysomnography. Our results highlight the necessity for actively promoting health education and systematic screening and treatment of OSA in this population to prevent future cardiovascular morbidity, especially among women.

Citation:

Roche J, Rae DE, Redman KN, et al. Impact of obstructive sleep apnea on cardiometabolic health in a random sample of older adults in rural South Africa: Building the case for the treatment of sleep disorders in underrresourced settings. J Clin Sleep Med. 2021;17(7):1423–1434.

BRIEF SUMMARY

Current Knowledge/Study Rationale: Although the association of obstructive sleep apnea (OSA) with higher cardiometabolic risk is known, it has never been investigated in sub-Saharan Africa. We performed home-based polysomnography in a sample of rural, older South Africans of African ancestry.

Study Impact: Along with an alarming prevalence of obesity, dyslipidemia, and hypertension, 29% of participants had OSA; in addition, being a woman and having severe obstructive sleep apnea were associated with increased cardiometabolic risk independently of age and socioeconomic status. Because OSA is a potentially modifiable factor, these results call for including the treatment of sleep disorders in the South African public health care system and raising awareness of its health care stakeholders to limit future cardiovascular mortality and assert that it is urgent to promote health education in rural communities, especially among women.

INTRODUCTION

Obstructive sleep apnea (OSA) is known as a proinflammatory factor¹ associated with increased risk for cardiometabolic diseases.² It is one of the most prevalent respiratory sleep disorders worldwide, with a prevalence ranging from 6% to 17% among the general adult population, according to data from high-income countries.³

In low- to middle-income countries such as those in sub-Saharan Africa, objective measures of sleep and OSA are scarce, partly because of restricted access to health care. Most of the limited number of studies on sleep performed in the region have assessed self-reported sleep quality using questionnaires such as the Pittsburgh Sleep Quality Index,⁴ and those studies measured the risk of OSA using the Berlin⁵ or STOP-BANG^6,7 questionnaires. Consequently, knowledge regarding the prevalence and consequences of sleep disorders in African populations remains limited. There is strong evidence for an increased risk of cardiometabolic disease associated with sleep disorders in African Americans compared to other ethnicities in the United States.^8,9 However, significant differences in genetic (including admixture), environmental, and socioeconomic circumstances¹⁰ precludes extrapolating those findings to the situation of Africans living in Africa. Sub-Saharan Africa, and particularly South Africa, is experiencing a profound health transition, characterized by the emergence of noncommunicable diseases and the aging of the population in both rural and urban areas.¹¹ For instance, the absolute number of cardiovascular deaths in sub-Saharan Africa increased by 81% between 1990 and 2013.¹² A study performed in a rural South African population of African-ancestry adults aged ≥30 years reported an alarming prevalence of obesity, dyslipidemia, and hypertension and concluded that 18.9% of women and 32.1% of men had a 20% or higher chance of having a cardiovascular event in the next 10 years, according to their Framingham risk score.¹³

In South Africa, the current prevalence of HIV infection (13%) is one of the highest in the world.¹⁴ HIV, based on what we know from studies in other populations, constitutes an additional risk factor for sleep disorders¹⁵ and cardiometabolic comorbidities.¹⁶ Moreover, rural communities represent a group of interest given the limited access to health services and the high prevalence of noncommunicable diseases.^4,14 To date, no community-based study assessing the association between objective sleep and cardiometabolic risk has been conducted in sub-Saharan Africa. Such studies are warranted to improve the understanding, detection, and treatment of cardiometabolic disorders in this population.

The aims of this cross-sectional study were as follows: first, to measure polysomnography (PSG)-derived sleep characteristics and the prevalence of OSA in an African-ancestry random sample of older adults with low socioeconomic status (SES) in a rural community in South Africa; second, to compare the clinical, cardiometabolic, and sleep parameters between participants with and without OSA; and third, to assess sleep parameters associated with cardiometabolic risk (CMR) in this sample in age-, SES-, and sex- adjusted analyses.

METHODS

Participants

Since 1992, the MRC/Wits Rural Public Health and Health Transitions Research Unit in South Africa has run the Agincourt Health and Demographic Surveillance System.¹⁷ In 2015, the study area of approximately 450 km² included 31 villages with a population of approximately 11,600 people. From the Agincourt Health and Demographic Surveillance System, the “Health and Ageing in Africa: a Longitudinal Study of an INDEPTH Community in South Africa” (HAALSI) was implemented in 2015 and randomly selected 5,890 eligible women and men aged ≥40 years living in the rural Agincourt subdistrict in Mpumalanga Province, South Africa.¹⁸ Among these eligible participants, 5,059 persons took part in the HAALSI study. In 2018, a sleep study on a random subset of the HAALSI cohort was conducted, for which a total of 1,000 participants had been randomly selected and randomly divided into 4 groups: group 1 (target n = 500, sleep questionnaires), group 2 (target n = 250, sleep questionnaires + actimetry), group 3 (target n = 200, sleep questionnaires + actimetry + home-based PSG), and group 4 (target n = 50, sleep questionnaires + actimetry + home-based PSG + dim light melatonin onset (DLMO)). For the present study, our analysis focused on participants from groups 3 and 4 who completed home-based PSG. In total, 182 participants were randomized in either group 3 or 4. Among them, 130 agreed to take part in the study and 92 individuals completed home-based PSG (37 from group 3 and 55 from group 4) before the study was suspended because of the COVID-19 pandemic and of the limitations imposed by the human research committee for human research of the University of the Witwatersrand (Johannesburg, South Africa) during the national lockdown.

We excluded participants with uninterpretable PSG data (n = 11) and uninterpretable respiratory parameters (n = 6). The final sample included 75 participants (53 women, 22 men). All participants belong to the Shangaan ethnic group. The flow chart of the study is presented in Figure 1.

Figure 1. Flowchart of the study population.

DLMO = dim light melatonin onset, PSG = polysomnography.

Study design

This is a cross-sectional, observational study in which participants were interviewed in their home by a trained fieldworker. As part of the HAALSI study, demographic information including SES and relevant personal history of cardiovascular and metabolic disease was collected and anthropometric parameters, blood pressure, and random glucose, lipids, and HIV status were assessed. A trained nurse and a locally trained senior fieldworker performed the home-based PSG.

This investigation conformed to the tenets of the Declaration of Helsinki and was approved by the University of the Witwatersrand Human Research Ethics Committee (#M180667) and by the Mpumalanga Province Research Ethics Committee in South Africa. Written informed consent was obtained from all participants. For participants who were not literate, a witness not linked to the study was present during the consenting process.

Experimental procedures

Demographics

Age and sex were recorded. Household SES level of the participants was determined using a validated SES index¹⁹ from the Agincourt Health and Demographic Surveillance System, including data on household asset indicators in a typical rural South African setting, over the period 2001–2013. Full methodology can be found in Kabudula et al.¹⁹ This index is presented in raw values and spans from 0 to 5, a lower value indicating lower SES level.

History of cardiometabolic diseases

Participants were asked whether they had had a diagnosis or were/had been under treatment for the following diseases: angina, stroke, heart attack, diabetes, or hypertension. They were classified as having a history of a cardiovascular event if they reported at least 1 of the following events: diagnosis of angina, stroke, or heart attack. Classifications of diabetes, dyslipidemia, and hypertension were established according to the blood glucose, lipid, and blood pressure measures and the participants’ declaration of treatment (see “Blood pressure measurement” and “Biochemical measures” sections).

Anthropometric evaluation

Body weight and height were measured to calculate body mass index (BMI; kg/m²). According to the World Health Organization recommendations, body weight status was categorized as follows: underweight was defined as BMI < 18.5 kg/m²; normal weight as BMI ≥ 18.5 kg/m² and < 25 kg/m²; overweight as BMI ≥ 25 kg/m² and < 30 kg/m²; and obese as BMI ≥ 30 kg/m². Waist circumference (WC) was measured in a standing position with a standard nonelastic tape that was applied horizontally midway between the last rib and the superior iliac crest.

Blood pressure measurement

Systolic blood pressure (SBP) and diastolic blood pressure (DBP) were measured in a seated position after 20 minutes of rest using a blood pressure cuff adapted to the arm circumference (Omron M6W automated cuff, Kyoto, Japan). Measurement was repeated 3 times at 2-minute intervals, and the average of the second and third measurements was used to calculate SBP and DBP. Hypertension stage 2 was defined as an SBP ≥ 140 mm Hg, or a DBP ≥ 90 mm Hg, or when participants reported current use of medication to treat hypertension.²⁰ Mean arterial pressure (MAP) was calculated using the following formula: [(DBP*2) + SBP]/3.

Biochemical measures

Blood was collected using a finger prick for dried blood spots for HIV. Point-of-care measurements were taken for glucose (CareSense, De Puy Synthes, Rainham, MA) and lipids (Cardiochek, PTS Diagnostics, Indianapolis, IN). Participants were classified as having diabetes if their nonfasting glucose concentration was ≥ 11.1 mmol/L or if they reported using medication to control diabetes.²¹ Participants were classified as having dyslipidemia if they had elevated total cholesterol (≥ 6.21 mmol/L), low high-density lipoprotein cholesterol (HDL-C; < 1.19 mmol/L), elevated low-density lipoprotein-C (> 4.1 mmol/L), elevated triglycerides levels (> 2.25 mmol/L), or reported current use of dyslipidemia treatment at the time of the interview.⁴ HIV status was determined by testing dried blood spots for HIV antibodies with enzyme-linked immunosorbent assays (Vironostika Uniform 11; BioMérieux, Marcy-l’Étoile, France) and confirmed using the Elecsys HIV combi PT assay (Roche Diagnostics, Basel, Switzerland) and the ADVIA Centaur HIV Ag/Ab Combo assay (Siemens Healthcare Diagnostics, Norwood, MA).

Home-based PSG

All participants underwent a single home-based PSG study on a weekday (Track-It Mk2, Nihon Kohden, Tokyo, Japan). Sleep was assessed using the 10-20 system,²² and the following electrode derivations were measured and recorded: Fz, Cz, F4-M1, C4-M1, O2-M1, F3-M2, C3-M2, and O1-M2; left and right electrooculogram; and chin electromyogram. Respiratory efforts were measured using thoracic and abdominal inductance plethysmography. Airflow was measured with a nasal pressure cannula. Peripheral oxygen saturation (SpO₂) was recorded by pulse oximetry (Nonin Medical, Plymouth, MN).

The electroencephalogram recordings were visually scored in 30-second periods by the first author, using the American Academy of Sleep Medicine’s standard rules to obtain the overnight pattern of sleep stages.²³ We extracted sleep latency (time from lights out to sleep defined as the first epoch of any sleep stage, minutes), total sleep time (TST, hours), lights out and lights on (hh:mm, lights out determined as alpha rhythm appearance and lights on as being awake and active), time in bed (lights out to lights on, hours), sleep efficiency (TST/time in bed, %), wake after sleep onset (WASO, minutes), arousal index (events/h of TST), arousal with respiratory events (events), awakening index (events/h of TST), percentage of nonrapid eye movement stage 1 sleep in TST (N1, %), percentage of nonrapid eye movement stage 2 sleep in TST (N2, %), percentage of nonrapid eye movement stage 3 sleep in TST (N3, %) and percentage of rapid eye movement (REM) sleep stage in TST (%) and REM sleep latency (minutes).

Respiratory events were scored in 3-minute periods for airflow according to the adult criteria of the AASM²³; an apnea event was defined as a ≥ 90% reduction in airflow for at least 10 seconds associated with the presence of respiratory effort for obstructive apnea, or associated with absent respiratory effort during 1 portion of the event and the presence of inspiratory effort in another portion for mixed apnea. Central apnea was defined as a ≥ 90% reduction in airflow for at least 10 seconds with absent inspiratory effort throughout the entire event. A hypopnea event was defined as a ≥ 30% reduction in airflow for at least 10 seconds associated with a ≥ 3% fall in oxygen saturation and/or arousal.

The obstructive apnea, central apnea, mixed apnea, and hypopnea indexes and the oxygen desaturation ≥ 3% index (ODI; events/h of TST) were determined. The apnea-hypopnea index (AHI) and ODI were calculated in TST, REM sleep, and non-REM sleep. Mean SpO₂, minimum SpO₂, time spent between 71% and 80% of SpO₂, time spent between 81% and 90% of SpO₂, and time spent between 91% and 100% of SpO₂ (%) were reported. OSA was defined by the presence of an AHI ≥ 15 events/h of TST.

CMR score and metabolic syndrome

A continuous CMR score was calculated in the whole sample of participants (n = 92), as described previously.²⁴ Z scores of glucose, triglycerides, HDL-C, WC, and MAP were calculated. Each z score was obtained by subtracting the sample mean from the individual value divided by the standard deviation of the sample mean: z score = (individual value – sample mean)/standard deviation. Then fasting glucose, triglycerides, WC, and MAP z scores were summed and the HDL-C z score was deducted because of its decreased health risk with higher values. The final value was then divided by 5 to create the continuous CMR score.

The presence of metabolic syndrome (MetS) was determined when a participant presented at least 3 of the following criteria²⁵: triglycerides ≥ 1.7 mmol/L or when the participant reported current use of medication to treat hypertriglyceridemia; HDL-C < 1.03 mmol/L for men and < 1.3 mmol/L for women; nonfasting glucose ≥ 11.1 mmol/L; WC ≥ 102 cm for men and ≥ 88 cm for women; and SBP ≥ 130 mmHg or DBP ≥ 85 mmHg or when the participant reported current use of medication to treat hypertension.

Statistical analyses

Statistical analyses were performed using GraphPad Prism 8.4.3 (GraphPad Software, San Diego, CA). The Kolmogorov-Smirnov test was used to test the assumption of distribution normality for quantitative continuous parameters with a level of significance set at P < .05 to reject normality. Data are presented as mean ± standard deviation (standard deviation) for parametric data, median (25%–75% interquartile range [IQR]) for nonparametric data, or n (%) for qualitative data. We first compared the OSA and non-OSA groups, then investigated the association between cardiometabolic variables and OSA severity markers in sex-, SES- and age-adjusted analyses. Because we found a main effect of sex on several cardiometabolic variables, we also compared men and women for their main demographic, sleep, and cardiometabolic characteristics.

Comparisons of clinical, cardiometabolic, and sleep variables between participants with and without OSA and between women and men were performed using unpaired t tests (parametric data) or Mann-Whitney U tests (nonparametric data). Fisher exact tests were used for comparison between categorical variables.

To investigate the association between cardiometabolic variables and OSA severity markers, we first performed linear univariate analyses between the cardiometabolic dependent variables (ie, CMR score, glucose, HDL-C, low-density lipoprotein-C, triglycerides, MAP, WC, and BMI) and the following independent variables: age, sex, SES, AHI, ODI, arousal index, TST, and HIV status. We chose these 5 specific variables (AHI, ODI, arousal index, TST, and HIV status) because they are traditionally associated with a higher CMR.^1,26–28 We did not pursue multivariable analysis if we found no significant association between the independent variables and the cardiometabolic dependent variable of interest. We included in our multivariable model analysis independent variables whose univariate P was < 0.2 (to allow for possible confounding effects) and further adjusted for SES, sex, and age.

RESULTS

Out of 92 participants, 17 were excluded from analysis because of either uninterpretable polysomnographic data (n = 11) or uninterpretable respiratory parameters during PSG (n = 6). These 17 excluded participants (7 women, 10 men) were aged 67.4 ± 7.5 years, had a mean BMI of 28.7 ± 5.9 kg/m², and had a median SES of 2.81 (IQR, 2.53–3.05).

Our analysis included 75 participants (53 women, 22 men) with a mean age of 66.1 ± 10.7 years and a median SES of 2.83 (IQR, 2.46–3.03). Table 1 shows the general and cardiometabolic characteristics of all participants, and PSG-derived sleep parameters are shown in Table 2. Average TST was 6.57 ± 1.03 hours, with 14.0% (IQR, 12%–18%) of TST classified as stage N1 sleep, 45.2 ± 8.4% as stage N2 sleep, 19.3 ± 8.4% as stage N3 sleep, and 19.6 ± 5.5% as REM sleep. Median AHI was 7.6 events/h (IQR, 4.3–16.7). Twenty-two participants had an AHI ≥ 15 events/h, constituting the OSA group.

Table 1.

Demographics and CMR characteristics of all participants and comparison between participants with and without OSA.

	All (n = 75)	Non-OSA (n = 53)	OSA (n = 22)	P
Demographics
Age, y	66.1 ± 10.7	64.3 ± 10.9	70.4 ± 8.9	.02
Women/men, n (% women)	53/22 (70.7)	38/15 (71.7)	15/7 (68.2)	.76
SES index	2.83 (2.49–3.03)	2.79 (2.42–3.02)	2.90 (2.71–3.10)	.18
Cardiometabolic parameters
BMI, kg/m2	28.9 ± 7.4	27.6 ± 6.7	31.8 ± 8.3	.03
WC, cm	96.8 ± 18.8	93.0 ± 16.2	105.4 ± 21.3	.01
Body weight status				.20
Underweight, n (%)	5 (6.8)	5 (9.6)	0 (0.0)
Normal weight, n (%)	24 (32.4)	18 (34.6)	6 (27.3)
Overweight, n (%)	17 (22.9)	11 (21.2)	6 (27.3)
Obese, n (%)	28 (37.8)	18 (34.6)	10 (45.4)
SBP, mm Hg	139.1 ± 23.6	136.1 ± 23.2	146.4 ± 23.6	.09
DBP, mm Hg	81.2 ± 11.1	80.1 ± 10.7	83.7 ± 11.8	.21
MAP, mm Hg	100.5 ± 13.6	98.8 ± 13.5	104.6 ± 13.4	.09
HTN stage 2, n (%)	46 (61.3)	31 (58.5)	15 (68.1)	.60
Controlled, n (%)	13 (30.2)	10 (34.5)	3 (21.4)	—
Resistant, n (%)	8 (18.6)	4 (13.8)	4 (28.6)	—
Untreated, n (%)	22 (51.2)	15 (51.7)	7 (50.0)	—
History of CV event, n (%)	8 (10.7)	8 (15.1)	0 (0.0)	.10
HDL-C, mmol/L	1.61 ± 0.47	1.72 ± 0.45	1.35 ± 0.44	< .01
LDL-C, mmol/L	1.97 ± 0.86	1.95 ± 0.89	2.02 ± 0.83	.76
Triglycerides, mmol/L	1.88 ± 1.02	1.81 ± 1.00	2.04 ± 1.07	.37
Dyslipidemia, n (%)	30 (42.9)	17 (34.7)	13 (61.9)	.06
Nonfasting glucose, mmol/L	5.80 (5.15–7.25)	5.80 (5.00–7.60)	6.00 (5.38–6.73)	.92
Diabetes, n (%)	7 (9.6)	5 (9.8)	2 (9.1)	> .99
CMR score	0.03 ± 0.56	–0.09 ± 0.57	0.29 ± 0.44	< .01
MetS, n (%)	29 (41.4)	15 (30.6)	14 (66.7)	< .01
HIV+, n (%)	12 (16.7)	9 (18.0)	3 (13.6)	.74

Data are presented as mean ± SD, median (25%–75% IQR), or n (%). P values represent between-group comparisons using Fisher exact tests, Student t tests, or Mann-Whitney U tests. For a Fisher exact analysis of body weight status between participants with and without OSA, underweight and normal-weight participants and overweight and obese participants were grouped. Missing data included the following: body weight status (n = 1, non-OSA group); information regarding HTN treatment (non-OSA group: n = 2; OSA group: n = 1); diabetes (non-OSA group: n = 2); dyslipidemia (non-OSA group: n = 4; OSA group: n = 1); MetS (non-OSA group: n = 4; OSA group: n = 1); and HIV status (non-OSA group: n = 3). BMI = body mass index, CMR = cardiometabolic risk, CV = cardiovascular, DBP = diastolic blood pressure, HDL-C = high-density lipoprotein cholesterol; HTN = hypertension, IQR = interquartile range, LDL-C = low-density lipoprotein cholesterol, MAP = mean arterial pressure, MetS = metabolic syndrome, OSA = obstructive sleep apnea, SBP = systolic blood pressure, SD = standard deviation, SES = socioeconomic status, WC = waist circumference.

Table 2.

Polysomnographic-derived sleep and respiratory parameters of all participants and comparison between participants with and without OSA.

	All (n = 75)	Non-OSA (n = 53)	OSA (n = 22)	P
Sleep parameters
Lights out, hh:mm	20:48 ± 02:33	20:43 ± 03:00	20:59 ± 00:39	.69
Lights on, hh:mm	05:40 ± 00:56	05:40 ± 00:54	05:40 ± 01:01	.94
Time in bed, h	8.5 ± 1.1	8.4 ± 1.2	8.8 ± 0.9	.15
Sleep latency, min	10 (5.0–19.8)	9.0 (5.0–16.0)	11.0 (5.6–26.4)	.39
TST, h	6.57 ± 1.03	6.63 ± 1.08	6.42 ± 0.88	.43
Sleep efficiency, %	81.0 (72.0–85.0)	82.0 (73.0–87.0)	75.0 (67.3–82.0)	.04
WASO, min	80.0 (47.3–126.8)	74.0 (40.0–108.5)	84.5 (68.3–135.1)	.04
Stage N1 sleep, % of TST	14.0 (12.0–18.0)	13.0 (11.0–17.0)	15.5 (14.0–20.0)	< .01
Stage N2 sleep, % of TST	45.2 ± 8.4	46.5 ± 8.5	42.0 ± 7.6	.03
Stage N3 sleep, % of TST	19.3 ± 8.4	19.2 ± 9.1	19.5 ± 6.8	.89
REM sleep, % of TST	19.6 ± 5.5	19.7 ± 6.0	19.5 ± 4.1	.89
REM sleep latency, min	93.8 ± 45.0	95.4 ± 44.8	89.9 ± 46.3	.63
Total arousals, n	81.0 (50.0–123.5)	63.0 (48.0–99.0)	134.0 (92.3–157.0)	< .0001
Arousal index, events /h	14.0 ± 7.3	11.4 ± 5.7	20.3 ± 7.2	< .0001
Arousal w/ respiratory event, n	18.0 (8.0–39.5)	13.0 (6.0–19.0)	51.5 (39.3–69.5)	< .0001
Awakening index, events/h	4.6 (3.4–5.6)	4.6 (3.3–5.5)	4.9 (3.6–5.6)	.42
Respiratory parameters
AHI < 5 events/h, n (%)	27 (36)	27 (50.9)	—	—
5 ≤ AHI < 15 events/h, n (%)	26 (34.7)	26 (49.1)	—	—
15 ≤ AHI < 30 events/h, n (%)	14 (18.7)	—	14 (63.6)	—
30 ≤ AHI < 45 events/h, n (%)	4 (5.3)	—	4 (18.2)	—
AHI ≥ 45 events/h, n (%)	4 (5.3)	—	4 (18.2)	—
AHI, events/h	7.6 (4.3–16.7)	4.9 (2.9–7.7)	23.2 (18.7–38.9)	< .0001
AHI, REM sleep, events/h	10.5 (4.1–33.2)	6.4 (1.9–13.6)	42.6 (30.3–60.8)	< .0001
AHI, NREM sleep, events/h	5.9 (3.0–10.9)	4.0 (1.8–6.3)	21.2 (14.9–35.5)	< .0001
OAI, events/h	2.9 (0.6–6.6)	1.2 (0.3–4.0)	11.0 (5.4–21.8)	< .0001
MAI, events/h	0.0 (0.0–0.0)	0.0 (0.0–0.0)	0.0 (0.0–0.2)	< .01
CAI, events/h	0.0 (0.0–0.0)	0.0 (0.0–0.0)	0.0 (0.0–0.7)	< .01
HI, events/h	3.8 (2.0–1.8)	2.7 (1.8–4.7)	12.8 (11.5–16.4)	< .0001
Mean SpO2, %	95.5 ± 2.6	967 ± 1.5	92.9 ± 2.6	< .0001
Min SpO2, %	89.5 (82.8–93.0)	91.0 (89.0–94.0)	81.0 (76.0–84.5)	< .0001
Total desaturation ≥ 3%, n	18.2 (3.7–69.0)	8.7 (1.6–20.1)	170.2 (71.4–225.3)	< .0001
ODI, TST 3%, events/h	2.7 (0.6–11.6)	1.3 (0.3–3.0)	25.3 (11.8–31.0)	< .0001
ODI, REM sleep, events/h	6.6 (0.0–29.3)	2.6 (0.0–6.8)	38.6 (26.6–68.8)	< .0001
ODI, NREM sleep, events/h	1.8 (0.4–7.6)	0.7 (0.2–2.0)	21.4 (7.9–26.6)	< .0001
Time spent between 71%–80% SpO2, %	0.0 (0.0–0.0)	0.0 (0.0–0.0)	0.0 (0.0–0.8)	< .001
Time spent between 81%–90% SpO2, %	0.1 (0.0–2.5)	0.0 (0.0–0.1)	7.5 (2.4–24.2)	< .0001
Time spent between 91%–100% SpO2, %	99.5 (93.8–99.9)	99.9 (98.8–100.0)	91.9 (58.7–97.5)	< .0001

Data are presented as mean ± SD or median (25%–75% IQR). P values represent between-group comparisons using Student t tests, Mann-Whitney U tests, or Fisher exact tests. AHI = apnea-hypopnea index, CAI = central apnea index, HI = hypopnea index, MAI = mixed apnea index, NREM = non-rapid eye movement, OAI = obstructive apnea index, ODI = oxygen desaturation index, OSA = obstructive sleep apnea, REM = rapid eye movement, SD = standard deviation, SpO₂ = peripheral oxygen saturation, TST = total sleep time, WASO = wake after sleep onset.

Comparisons between the OSA and non-OSA groups

SES status and HIV prevalence were similar between the non-OSA and OSA groups. A comparison of cardiometabolic characteristics between the 2 groups is presented in Table 1. OSA participants were older (P = .02), had a higher BMI (P = .02), a higher WC (P < .01), a higher prevalence of MetS (P < .01), and a greater CMR score (P < .01) than the non-OSA participants.

PSG-derived sleep and respiratory parameters of the 2 groups are presented in Table 2. Participants with OSA had lower sleep efficiency (P = .04), a lower proportion of stage N2 sleep (P = .03), a higher proportion of stage N1 sleep (P < .01), a higher arousal index (P < .001), and more WASO (P = .04), arousals (P < .001), and arousals with respiratory events (P < .001) than participants in the non-OSA group. All respiratory parameters differed between the 2 groups (all P < .01).

Univariate and multivariable linear regressions

In univariate analyses including all participants (n = 75), the CMR score was associated with a higher AHI (P = .004) and ODI (P = .028) and with being a woman (P = .043; Table 3). We also tested for an interaction between sex and AHI, which was not significant (P = .30). Therefore, our final multivariable model of the CMR score included AHI, SES, age, and sex. CMR was associated with AHI (P = .011) and with sex (P = .011) independently of age (P = .483) and SES (P = .548; Table 3), whereby a higher AHI and being a woman were associated with a higher CMR score.

Table 3.

Predictors of CMR (outcome variable: CMR score) in univariate and multivariable analyses (n = 75).

	Univariate			Multivariable
	β	SE	P	β	SE	P
Age, y	0.005	0.007	.450	0.004	0.006	.483
Sex (female)	0.307	0.148	.043	0.369	0.143	.011
SES index	0.201	0.199	.318	0.114	0.188	.548
AHI, events/h	0.012	0.005	.004	0.011	0.004	.011
ODI 3%, events/h	0.013	0.006	.028	—	—	—
Arousal index, events/h	0.004	0.009	.684	—	—	—
TST, h	–0.000	0.001	.700	—	—	—
HIV+	–0.097	0.185	.601	—	—	—

Univariate and multivariable analyses were performed using linear regression. AHI = apnea-hypopnea index, CMR = cardiometabolic risk, ODI = oxygen desaturation index, SE = standard error, SES = socioeconomic status, TST= total sleep time.

Table S1 ^{(65.3KB, pdf)} in the supplemental material shows the results of the univariate and final multivariable analyses performed on the other cardiometabolic variables. In univariate analyses, glucose and MAP were not associated with any of the independent variables considered (all P > .2). Therefore, no multivariable analysis was pursued. Triglycerides were associated with BMI only (P = .036), but this association was no longer significant after adjustment for age, sex, and SES in a multivariable model. Low-density lipoprotein-C was associated with age only (P = .019) in univariate analysis. In the multivariable model, none of the included variables (age, sex, SES, AHI) were associated with low-density lipoprotein-C. HDL-C was inversely associated with AHI (P = .001), ODI (P < .01), WC (P < .0001), and BMI (P < .001) in unadjusted analyses. In the final multivariable model, HDL-C was negatively associated with AHI (P = .015) independently of BMI, age, sex, and SES (all P > .05). A higher BMI was associated with a higher AHI (P = .004), a higher ODI (P < .001), being a woman (P = .003), and a higher SES (P = .012). In the final multivariable model, BMI was associated with AHI (P = .004), sex (P < .0001), and SES (P = .033) independent of age (P = .923). Finally, a larger WC was associated with a higher AHI (P < .0001), ODI (P < .001), and SES (P = .023) in univariate analyses. In the final multivariate model, WC was associated with AHI (P < .0001) and being a woman (P = .004) independently of age (P = .722) and SES (P = .116).

Comparisons between women and men

Because being a woman was associated with several of our cardiometabolic outcome variables, we further compared the demographic, clinical, PSG, and cardiometabolic characteristics of women and men. This comparison is presented in Table 4. Women and men had a similar age and SES. The women had a higher BMI (P < .001), more overweight/obesity (P = .02), and a higher CMR score (P = .04) than the men but a similar prevalence of MetS. Similar proportions of OSA were found among the women (28.3%) and men (31.8%; P = .76), but men had more stage N1 sleep (< 0.01) and a higher awakening index (P < .01) compared to women.

Table 4.

Cardiometabolic and polysomnographic comparisons between women and men.

	Women (n = 53)	Men (n = 22)	P
Demographics
Age, y	65.5 ± 10.8	67.5 ± 10.6	.45
SES index	2.79 (2.42–3.01)	2.90 (2.72–3.09)	.13
Cardiometabolic parameters
BMI, kg/m2	30.4 ± 7.6	24.9 ± 5.3	< .001
WC, cm	99.5 ± 19.8	90.9 ± 14.8	.08
Body weight status			.02
Underweight, n (%)	2 (3.8)	3 (14.3)	—
Normal weight, n (%)	14 (26.4)	10 (47.6)	—
Overweight, n (%)	13 (24.5)	4 (19.05)	—
Obese, n (%)	24 (45.3)	4 (19.05)	—
SBP, mm Hg	139.2 ± 23.3	138.8 ± 24.9	.94
DBP, mm Hg	82.0 ± 10.8	79.2 ± 11.8	.33
MAP, mm Hg	101.1 ± 13.6	99.1 ± 13.8	.57
HTN stage 2, n (%)	35 (66.0)	11 (50.0)	.21
Controlled, n (%)	12 (34.3)	1 (9.1)	—
Resistant, n (%)	5 (14.3)	3 (27.3)	—
Untreated, n (%)	18 (51.4)	7 (63.6)	—
History of CV event, n (%)	7 (13.2)	1 (0.5)	.42
HDL-C, mmol/L	1.56 ± 0.49	1.72 ± 0.43	.21
LDL-C, mmol/L	1.86 ± 0.83	2.28 ± 0.90	.09
Triglycerides, mmol/L	1.91 ± 0.99	1.78 ± 1.10	.63
Dyslipidemia, n (%)	27 (52.9)	5 (26.3)	.06
Nonfasting glucose, mmol/L	6.0 (5.6–7.4)	5.6 (4.7–6.8)	.06
Diabetes, n (%)	5 (9.4)	2 (10.0)	> .99
CMR score	0.12 ± 0.55	–0.18 ± 0.54	.04
MetS, n (%)	24 (47.1)	5 (26.3)	.17
HIV+, n (%)	8 (15.7)	4 (19.0)	.73
Polysomnographic parameters
Lights out, hh:mm	20:50 ± 03:00	20:41 ± 00:42	.82
Lights on, hh:mm	05:40 ± 00:53	05:40 ± 01:02	.99
Time in bed, h	8.3 ± 1.1	8.9 ± 1.0	.05
Sleep latency, min	9.5 (4.5–16.0)	13.8 (6.1–26.5)	.25
TST, h	6.6 ± 1.0	6.6 ± 1.0	.88
Sleep efficiency, %	82.0 (73.0–85.0)	75.5 (65.5–84.8)	.17
WASO, min	74.5 (47.0–118.5)	84.3 (51.0–149.5)	.26
Stage N1 sleep, % of TST	14.0 (11.0–16.0)	18.0 (13.0–21.5)	< .01
Stage N2 sleep, % of TST	46.2 ± 7.3	42.8 ± 10.4	.11
Stage N3 sleep, % of TST	20.4 ± 8.5	16.7 ± 7.8	.08
REM sleep, % of TST	19.4 ± 5.8	20.3 ± 4.7	.49
Arousal index, events/h	13.1 ± 6.9	16.2 ± 8.0	.09
Arousal with respiratory event, n	15.0 (7.0–30.0)	24.5 (12.0–46.5)	.08
Awakening index, events/h	4.1 (3.0–5.1)	5.8 (4.5–7.4)	< .01
OSA, n (%)	15 (28.3)	7 (31.8)	.76
AHI, events/h	7.5 (3.9–16.5)	7.9 (4.6–19.9)	.48
Mean SpO2, %	95.6 ± 2.4	95.3 ± 3.3	.73
Min SpO2, %	90.0 (82.5–93.0)	86.0 (83.0–92.0)	.87
ODI, TST 3%, events/h	2.7 (0.6–11.1)	3.4 (1.5–11.6)	.69

Data are presented as mean ± SD, median (25%–75% IQR), or n (%). P values represent between-group comparisons made using Fisher exact tests, Student t tests, or Mann-Whitney U tests. For the Fisher exact analysis of body weight status between women and men, underweight and normal-weight participants and overweight and obese participants were grouped. Missing data: body weight status (n = 1, men); diabetes (n = 2, men); dyslipidemia (n = 2, women; n = 3, men); MetS (n = 2, women; n = 3, men); and HIV status (n = 2, women; n = 1, men). AHI = apnea-hypopnea index, BMI = body mass index, CMR = cardiometabolic risk, CV = cardiovascular, DBP = diastolic blood pressure, HDL-C = high-density lipoprotein cholesterol, HTN = hypertension, LDL-C = low-density lipoprotein cholesterol, MAP = mean arterial pressure, MetS = metabolic syndrome, ODI = oxygen desaturation index, OSA = obstructive sleep apnea, REM = rapid eye movement, SBP = systolic blood pressure, SD = standard deviation, SES = socioeconomic status, SpO₂ = peripheral oxygen saturation, TST = total sleep time, WASO = wake after sleep onset, WC = waist circumference.

DISCUSSION

In this first objective sleep study in a random sample of African-ancestry adults living in rural sub-Saharan Africa, we reported an OSA prevalence of 29.3%. Moreover, OSA severity and being a woman were independently associated with a greater CMR score in SES- and age-adjusted analyses. Overall, our rural sample was predominantly female, on average aged 66 years and of low SES. The sex distribution is representative of this region, where the low employment opportunities lead to men seeking employment in the bigger cities, and therefore the adults remaining are mainly women and/or retired.¹⁸

We present some of the first sleep architecture data on sub-Saharan Africans of African ancestry. As a whole group, our participants exhibited similar sleep latency, sleep efficiency, proportions of stage N3 and REM sleep, and arousal index compared to other age-matched populations.²⁹ Time spent in stage N2 sleep (∼45%) was lower than expected (∼53%), and time spent in stage N1 sleep and WASO were greater than normative values for this age group.²⁹ This finding may be partly explained by the elevated prevalence of OSA in our sample. Indeed, our OSA group had lower sleep efficiency, more WASO and stage N1 sleep, less stage N2 sleep, and more arousals than the non-OSA group, indicating a more fragmented sleep. To date, most of the literature on people of African ancestry has focused on the comparison between African Americans, whose genetic background is typically admixed, and European Americans.^30,31 Those studies found that sleep quality is impaired among African Americans, with reduced sleep duration and reduced sleep efficiency, and in contrast to our findings, they also found a reduced stage N3 sleep proportion and longer sleep latency when adjusting for BMI, age, sex, and SES.^30,31 Perceived discrimination or habitual living conditions have been incriminated in altered sleep quality among African Americans, suggesting that poor sleep may not be attributable only to genetics.^30,31 In the present study, our participants underwent PSG in their home. Thus, it seems that sleep quality is not negatively affected by the environment in older adults living in rural communities. However, given the modest living conditions in rural communities of South Africa (single-room houses, no temperature control indoors, few electric installations¹⁷), further studies should investigate the possible effects of these parameters on sleep.

Using a standard AHI cutoff of ≥ 15 events/h, we found an alarming prevalence of OSA of 29.3%. To date, only 2 studies have reported OSA and the risk of OSA prevalence in sub-Saharan Africa.^7,32 Ozoh et al⁷ recruited 1,100 patients attending a tertiary health facility in Nigeria with a mean age of 43.9 years. The risk of OSA, as assessed by the STOP-BANG questionnaire, was found in 36.3% of the sample. More recently, Poka-Mayap and colleagues³² assessed OSA prevalence using home sleep apnea testing among patients admitted to a tertiary hospital in Cameroon. Among the 111 participants, with a mean age of 58 years, 31.5% had moderate-to-severe OSA as defined using an AHI > 15 events/h. Even though our results are consistent with these 2 studies, ours is the first to use the gold standard for OSA detection—full PSG with a recording of respiratory events—in a sub-Saharan African setting. In contrast with these 2 studies, we did not recruit participants in health facilities but used a population-based random sample from an existing older-age cohort (the HAALSI study¹⁸). Therefore, our study provides a more accurate estimate of the actual prevalence of OSA in this rural community with older adults.

As expected, we found a greater sleep fragmentation and intermittent hypoxia in participants with OSA, expressed by a longer WASO, a greater proportion of stage N1 sleep and arousals, a lower sleep efficiency, and elevated desaturation indices. Both sleep fragmentation and intermittent hypoxia have been involved in the establishment of systemic inflammation, through sympathetic overactivity³³ and the overproduction of radical oxygen species pathways,²⁷ all factors associated with an increased CMR.

Chronic inflammation has been reported in treated and untreated patients with HIV and has been hypothesized to contribute to the significantly higher risk of cardiovascular disease observed in people living with HIV.³⁴ A recent study also showed that reporting highly bothersome insomnia was associated with a higher risk of cardiovascular events in people living with HIV over a follow-up period of 10.8 years.³⁵ In this study, we did not measure markers of inflammation, and the HIV prevalence (16.7%) was slightly lower than that seen in other studies in the region (23%¹⁸), although it was still higher than the national level of 13%. Because of this low representation, we did not have the power to further investigate any interaction between OSA, HIV status, and CMR. However, it is likely that the increased activation of inflammatory pathways observed in OSA may have a more deleterious impact on people living with HIV, hence contributing more to CMR in this population.

In the general adult population of higher-income countries, the main identified risk factors for OSA are central obesity, sex, age, heredity, and being of African ancestry.³⁶ In our sample, 73% of the participants with OSA were overweight/obese and exhibited a higher BMI and WC than their counterparts. In addition, BMI and WC were positively associated with AHI in adjusted multivariate analyses. Ozoh et al⁷ also reported that abdominal obesity, as assessed by WC, was a determinant of nearly double the risk of OSA in a comparable population. Taken together, these results seem to confirm the implication of adiposity, especially central obesity, in the pathogenesis of OSA in this sample. In addition, and in accordance with the literature,^7,36 older age seems to increase the likelihood of exhibiting OSA. Interestingly, in our sample, men and women were equally affected by OSA, whereas the literature reports a higher risk of OSA in men.³⁷ However, women exhibited a greater BMI than men, and this factor may explain why OSA is as prevalent in women as in men.

Unexpectedly, we did not find significant cardiometabolic differences between the OSA and non-OSA groups, apart from the higher CMR score and greater prevalence of MetS that was more than twice as high among participants with OSA compared to participants without OSA (66.7% vs 30.6%). For instance, the high prevalence of hypertension and dyslipidemia was similar in the 2 groups. Again, these results may be explained, at least in part, by the extremely high prevalence of overweight/obesity found in our entire sample. South Africa is still experiencing the emergence of noncommunicable diseases in both urban and rural areas,¹¹ and previous surveys conducted in the 2000s already documented the rise of obesity and cardiovascular diseases in rural populations of African ancestry in South Africa.^13,38 Regarding components of MetS, the CMR score was associated with increased AHI and with being female, independent of age and SES. As highlighted earlier, OSA is a well-established proinflammatory factor promoting the emergence of cardiometabolic disorders and, ultimately, endothelial dysfunction.^1,39,40 Previous studies performed among either European,^41,42 or Asian populations⁴³ have already reported such a relationship between OSA severity and CMR, assessed with MetS, in women⁴² and men.⁴³ In this regard, this study is the first to confirm that the findings reported among other groups apply to aging adults with African ancestry in a rural area from South Africa.

Regarding the female sex being associated with the CMR score, this result echoes previous studies conducted in Southern Africa that raised some concerns regarding cardiovascular health in women.^4,12,13 For instance, a study with a systematic analysis of data from the 2013 Global Burden of Disease Study reported that the cardiovascular disease mortality rate in sub-Saharan African women was higher than in developing countries, whereas it had been lower in 1990.¹² In line with another previous study¹³ and with the South African National Health and Nutrition Examination Survey,⁴⁴ Gómez-Olivé et al⁴ recently reported that aging women of African ancestry and living in rural areas of South Africa exhibited a greater incidence of either obesity or hypertension than men. Although it was not possible to determine causality in this study, the authors highlighted that this higher burden of diseases among women coincided with less formal education, larger households, and lower employment than men.⁴ In the present study, we observed a greater overweight/obesity prevalence among women compared to men (69.8% vs 38.1%; P = .02).

Accordingly, the present findings call for better health prevention among rural communities of South Africa where the high prevalence of HIV infection represents an additional CMR factor.²⁸ However, overweight/obesity, which promotes the development of OSA, is a modifiable lifestyle factor. A recent multicentric study investigated OSA awareness among primary care physicians in Kenya, Nigeria, and South Africa.⁴⁵ Although physicians from South Africa reported a satisfactory knowledge regarding OSA, as investigated by the OSA Knowledge and Attitudes questionnaire, overall respondents showed less confidence in their ability to identify patients at high risk of OSA along with lower confidence in their ability to manage the disorder.⁴⁵ Given the fact that based on our findings, OSA may involve almost one-third of the aging population living in rural South Africa, an effort considering education on sleep and OSA among physicians seems warranted to prevent cardiovascular morbidity/mortality.

This study has some limitations that deserve to be highlighted. Eleven PSG recordings out of the original 92 were uninterpretable. Moreover, an adaptation night for PSG was not feasible with our experimental setup. We were also unable to report on any measures of daytime sleepiness or dysfunction, known traits of OSA.

In conclusion, we have provided the first PSG-derived sleep data from a sub-Saharan African cohort of African-ancestry adults and observed an alarming prevalence of undiagnosed and untreated OSA associated with higher CMR. OSA is a modifiable risk factor for cardiometabolic disorders that is currently not screened or offered treatment for in the public health care sector in South Africa, which caters to approximately 80% of the general population. In a population already at high CMR because of traditional risk factors, coupled with a high prevalence of HIV infection, we show that sleep disorders and their treatment may be pivotal aspects of CMR prevention in low- to middle-income countries such as South Africa. We highlight the feasibility of performing objective sleep measurements in an underresourced rural community using home-based PSG, which can then be used to detect OSA so that treatment can be initiated. Accordingly, it would be worth investing in the introduction of continuous positive airway pressure treatment in participants with OSA in this region and monitoring cardiovascular risk markers during treatment. In the meantime, health and sleep education should be implemented among rural communities, especially among women and rural health practitioners to reduce future cardiovascular morbidity.

DISCLOSURE STATEMENT

All authors have seen and approved the manuscript. The study was supported by the Academy of Medical Sciences Newton Advance Fellowship to M.v.S. and F.X.G.-O., and by a postdoctoral fellowship from the University of the Witwatersrand’s University Research Council to J.R. The authors report no conflicts of interest.

ABBREVIATIONS

AHI

apnea-hypopnea index

BMI

body mass index

CMR

cardiometabolic risk

DBP

diastolic blood pressure

DLMO

dim light melatonin onset

HAALSI

Health and Ageing in Africa: a Longitudinal Study of and INDEPTH Community in South Africa

HDL-C

high-density lipoprotein cholesterol

25%-75% IQR

25%-75% interquartile range

MAP

mean arterial pressure

MetS

metabolic syndrome

ODI

oxygen desaturation ≥3% index

OSA

obstructive sleep apnea

PSG

polysomnography

REM

rapid eye movement

SBP

systolic blood pressure

SES

socioeconomic status

SpO₂

peripheral oxygen saturation

TST

total sleep time

WASO

wake after sleep onset

WC

waist circumference