Associations between sleep parameters, non-communicable diseases, HIV status and medications in older, rural South Africans

As part of the Health and Aging in Africa: A Longitudinal Study of an INDEPTH Community in South Africa (HAALSI), we investigated sleep habits and their interactions with HIV or non-communicable diseases (NCDs) in 5059 participants (median age: 61, interquartile range: 52—71, 54% females). Self-reported sleep duration was 8.2 ± 1.6h, and bed and rise times were 20:48 ± 1:15 and 05:31 ± 1:05 respectively. Ratings of insufficient sleep were associated with older age, lack of formal education, unemployment, and obesity (p < 0.05). Ratings of restless sleep were associated with being older, female, having more education, being unemployed, and single. Hypertension was associated with shorter self-reported sleep duration, poor sleep quality, restless sleep, and periods of stopping breathing during the night (p < 0.05). HIV positive individuals not on antiretroviral treatment (ART) reported more nocturnal awakenings than those on ART (p = 0.029) and HIV negative individuals (p = 0.024), suggesting a negative net effect of untreated infection, but not of ART, on sleep quality. In this cohort, shorter, poor-quality sleep was associated with hypertension, but average self-reported sleep duration was longer than reported in other regions globally. It remains to be determined whether this is particular to this cohort, South Africa in general, or low- to middle-income countries undergoing transition.

Sleep parameter associations. Table 2 displays associations between demographic and sleep parameters. Self-reported sleep duration was associated with age, sex, employment, marital status, and wealth quintile. Participants aged 50-59 y (p = 0.030) and 60-69 y (p < 0.001) had shorter self-reported sleep durations than 40-49-year-olds; females reported shorter sleep durations than males (p < 0.001); employed individuals and homemakers reported shorter sleep durations than those not working (p < 0.001); married individuals reported shorter sleep durations than those divorced (p = 0.037) or widowed (p = 0.003), and middle, upper-middle and upper income individuals reported shorter sleep durations than individuals in the lowest wealth quintile (p = 0.024, p = 0.001, and p = 0.005, respectively).  Table 1. Descriptive, demographic, health and sleep characteristics of the cohort. Data are presented as mean ± standard deviation, median (interquartile range) or count (%). BMI: body mass index; ART: antiretroviral therapy. *Self-reported sleep duration. a "Very good" or "Fairly good"; b Once a week or more; c "Often" or "Very often"; d "Yes". The p-values represent differences between the two gender groups analyzed using an independent t-test, a Mann-Whitney U test or a Fisher's Exact test. Self-reported bedtime was associated with age, BMI, education level, marital status and wealth quintile. Individuals aged 70-79 y (p = 0.058) and ≥80 y (p < 0.001) had earlier bedtimes than those aged 40-49 y; underweight participants had earlier bedtimes than normal weight participants (p = 0.006); individuals with 1-7 y (p = 0.005) or >12 y (p = 0.002) of education had later bedtimes than those with no formal education; widowed individuals had earlier bedtimes than married individuals (p = 0.001); middle, upper-middle and upper income individuals had later bedtimes than those in the lowest wealth quintile (p = 0.002, p < 0.001 and p < 0.001 respectively). Waking time was associated with age, sex, employment status and marital status. Participants aged 50-69 y woke earlier than 40-49-year-olds (p < 0.001 females woke earlier than males (p < 0.001), employed individuals woke earlier than others (p < 0.001), and currently married individuals woke earlier than those not married (p < 0.001), separated/divorced (p < 0.001), or widowed (p = 0.009).
Ratings of insufficient sleep were associated with age, BMI, level of education, and employment status. Participants older ≥80 y were more likely to rate their sleep duration as being insufficient compared to 40-49-year-olds (p = 0.025); obese individuals were more likely to rate their sleep as being insufficient compared to normal weight individuals (p = 0.001); those with 8 to 11 years of education (p = 0.001) and those with 12 +years of education (p = 0.017) were less likely to rate their sleep as being insufficient compared to those with no formal education; and homemakers were less likely to rate their sleep as being insufficient compared to those not working (p < 0.001).
Ratings of restless sleep were associated with age, sex, education level, employment and marital status. Individuals ≥50 y were more likely to report restless sleep compared to those aged 40-49 y (p < 0.001); females were more likely to report restless sleep than males (p = 0.001); those with ≥ 12 years of education were less likely to report restless sleep compared to those with no formal education (p < 0.001); employed individuals were less likely to report restless sleep compared to those not working (p = 0.011); and separated/divorced (p = 0.002) and widowed (p < 0.001) participants were more likely to report restless sleep compared to married participants.
Self-reported snoring was associated with age, sex, BMI, education level and marital status. Participants ≥ 50 y were more likely to report snoring than 40-49-year-olds (p = 0.044 for 50-59 y, p < 0.001 for 60-69 y, p = 0.004 for 70-79 y, p < 0.001 for ≥ 80 y); overweight (p = 0.009) and obese (p < 0.001) individuals were more likely to report snoring compared to normal weight individuals; males were more likely to report snoring than females (p < 0.001); those with 1-7 y of education were more likely to report snoring compared to those with no formal education (p = 0.005); and separated/divorced (p = 0.031) and widowed (p < 0.001) participants were less likely to report snoring compared to those currently married. Sleep and health outcomes. Associations between self-reported sleep parameters and health outcomes are presented in Table 3. Obesity was associated with self-reported bedtime, wake time, insufficient sleep, snoring, gasping and periods of stopping breathing during sleep. Those reporting bedtimes between 21:00-21:30 (p = 0.012) and wake times from 05:00-06:00 (p = 0.019) were more likely to be obese than those going to bed before 21:00 and waking up before 05:00 respectively. In addition, those reporting not getting enough sleep (p = 0.008), snoring (p < 0.001), gasping (p < 0.001) and have breathing pauses (p < 0.001) during the night were also more likely to be obese. Hypertension was associated with self-reported sleep duration, bedtime, wake time, sleep quality, restless sleep, and periods of stopping breathing during the night. In this cohort, those with earlier bedtimes (<21:00, p = 0.028), early waking times (<05:00, p = 0.030), better sleep quality (p = 0.010), less restless sleep (p = 0.011) and those not reporting periods of stopping breathing (p = 0.009) were less likely to have hypertension. Diabetes was associated with bedtime, restless sleep, and snoring. Participants reporting the latest bedtimes (>21:30) were less likely to have diabetes compared to those with the earliest bedtimes (<20:00, p = 0.047); those reporting restless sleep were more likely to have diabetes than those not reporting restless sleep (p = 0.015); and those reporting snoring were more likely to have diabetes compared to those not reporting snoring (p = 0.003). Dyslipidaemia was only associated with sleep sufficiency such that those reporting insufficient sleep were more likely to have dyslipidemia compared to those who felt they were obtaining sufficient sleep (p = 0.032). Anaemia was associated with self-reported sleep duration such that individuals in the shortest (4-7 h, p = 0.041) and longest (>= 10 h, p = 0.007) sleep quartiles were more likely to have anaemia compared to those in the second sleep quartile (8 h). HIV infection was associated with bedtime only: participants with moderately early bedtimes (20:00 to <21:00) were more likely to be HIV positive compared to those with the earliest bedtimes (<20:00, p = 0.025).

ART and viral load associations with sleep parameters. While HIV positive individuals not on
ART reported more awakenings during the night compared to those on ART (p = 0.024; Table 4), viral load amongst the HIV positive participants was not associated with any of the self-reported sleep parameters measured (Supplementary Table 1).

Discussion
This cohort of older, rural, black South African adults reported a longer sleep duration (8.2 ± 1.6 h per night) than observed in other populations globally, despite the fact that one third described their sleep duration as being insufficient. A study of self-reported sleep duration and sleep problems in older adults found that participants from South Africa reported sleep durations of 8.6 h (SD ± 2.1), and a greater proportion of rural dwelling participants than urban dwellers reported sleep durations >10 h 25 . A review of data from 168 studies around the world (n = 6052) between 2010 and 2015 found total sleep time (actigraphy-or PSG-derived) to range between 5.8 and 7.8 h per night 26 . Other large-scale epidemiological studies have observed self-reported median sleep durations of 7-8 h [27][28][29][30] . Although self-reported sleep duration is typically longer than objectively measured total sleep time, possible explanations for the longer self-reported sleep duration observed in this study may relate to differences in ethnicity (or tribal origin), geographical region, culture, proportion of females, age, employment status, level of education, or a rural living environment.
There may be a strong argument relating to ethnicity, or culture, since three other South African cohorts have also found longer self-reported sleep durations (8.5-10.5 h per night) in black South African adults 19,20,31 . Since a previous study 20 reported longer sleep durations in black Africans than in the other ethnic groups living in South Africa, it is tempting to surmise that a longer self-reported sleep duration is particular to ethnicity or culture in this setting. Interestingly, data from the 2005-2012 National Health and Nutrition Examination Survey (NHANES) in the USA indicated that non-Hispanic black Americans reported shorter sleep durations than other ethnic groups 32 . Furthermore, a meta-analysis comparing the sleep of African-Americans to European-Americans using polysomnography, actigraphy, diaries and questionnaires found that African-Americans had shorter total sleep time (objectively and subjectively measured) and poorer quality sleep, as measured by worse sleep efficiency, longer sleep onset latency and less slow wave sleep 33 . Therefore, it is important to try to disentangle the factors underlying these variations between ethnic groups in different countries with disparate socio-economic histories which appear to reoccur in spite of variability in general lifestyle and culture.
Longer self-reported sleep duration has been shown to be more prevalent in adults younger than 25 years or older than 60 years in the United States 34 . The longer self-reported sleep duration observed in the present study may be, at least in part, due to the older age of the cohort. Counter to this argument, however, is that those between 50 and 70 years old in this study reported shorter sleep durations than the 40-year-olds, and the sleep durations reported in previous South African studies, 8.75 h 19 and 10.35 h 31 , were both longer than this study, despite the younger ages of their participants (39.9 years and 33.4 years respectively). There is evidence to suggest that males typically have longer, better quality sleep than females (reviewed in 35 ). Since our sample was mildly biased towards females (54%), whose average sleep duration was 8.15 h, sex seems unlikely to explain the longer self-reported sleep duration observed in this cohort.
Another point to consider relates to employment. A study of adults in the USA found that being employed had a strong effect on reducing self-reported sleep duration 34 . Since only 16% of the individuals in the present study were employed, the longer sleep durations are not unexpected, and indeed employed individuals from this cohort did report shorter sleep durations. At odds with our observation is that the Soweto cohort, in whom the employment rate was nearly double (30%), reported longer sleep durations 19 . Of interest, too, is that despite the lower proportion of females in paid employment in this study, they reported earlier rise times and shorter sleep One might also argue that the longer sleep durations reported in this study relate to a rural living environment. Norwegian 36 and South African studies 25 have observed that individuals living in rural areas reported longer sleep durations than those in urban areas. However, the fact that an urban, albeit younger, population in Soweto, South Africa reported a longer sleep duration (8.8 ± 1.7 h) 19 , counters this argument. Generally, self-reported sleep durations from native Africans living in South Africa -those reported in the Soweto cohort 19 , the black African participants from the SAGE cohort 25 and the black South Africans from the Modelling the Epidemiological Transition Study (METS) cohort 31 -are all similar to those reported in the present study. We note that the sleep durations reported in this study are also similar to the objectively measured sleep periods (i.e. time-in-bed) in rural (9.23 ± 0.13 h) and semi-rural (8.78 ± 0.18 h) African communities in Mozambique 18 but longer than those measured in southern African hunter-gathers (6.9-8.5 h) (Yetish, 2015, #3241}. We interpret these comparisons, with caution, however, owing to the different data collection methods used.
Thus, these findings raise interesting questions. Do South Africans, and in particular those of black African ancestry, have different sleep habits or even different sleep needs compared to other populations? Are these self-reported findings truly indicative of longer sleep, or could self-reported sleep in this cohort be subject to interpretation bias? Could this longer sleep period indicate sleep quality issues? What does the sleep period and NCD risk profile association look like in this group? To truly address these questions, future studies making use of objective sleep measures are required. The regression analyses from the present study do, however, provide some insight to the latter questions.
Data from the WHO indicate that the two strongest contributing risk factors to NCD-related deaths in South Africa are elevated blood pressure (33.7%) and obesity (31.3%) 38 . This sample of rural, older, longer sleeping adults was on average overweight, and had a high prevalence of obesity (30%) and hypertension (58%). Similar to the SA-NHANES report 1 , the incidence of obesity and hypertension was higher in the females in this study than in the males. The females also had higher incidences of depression and anaemia than the males. This higher burden of disease in the females was coincident with less formal education, larger households and lower employment. Since they also spent fewer hours in bed, had earlier wake times and reported more restless sleep and nocturnal awakenings than the males, one might speculate that their shorter, poorer quality sleep places them at higher risk for cardiometabolic diseases.
Prospective cohort studies indicate that both long-and short-total sleep times are associated with higher mortality rates and risk for cardiometabolic diseases [10][11][12][13]39 . In a recent large-scale study in the USA, 7 h of sleep was associated with the lowest risk for cardiometabolic disease in all ethnic groups except non-Hispanic blacks, in whom 8 h of sleep was associated with the lowest risk score for cardiometabolic disease 32 . As in previous studies 40,41 , the long sleepers (>10 h) in the present study had a lower incidence of hypertension than the shorter sleepers (<7 h). Sleep timing and ratings of sufficiency also appear to be important contributors to obesity in this cohort, since we observed an association between later bedtimes, ratings of insufficient sleep and obesity. A  previous study in a cohort of black South Africans living in Soweto reported that a longer total sleep time and naps appeared to be protective against a high BMI in older females and high blood pressure, BMI and abdominal obesity in males 19 . Collectively, these data suggest that optimal sleep need may well differ by ethnicity in the context of minimising risk for NCDs. It is also worth noting that although we did not assess sleep apnoea, there are indications that this cohort may suffer from the condition, which may be contributing to poorer health since those reporting snoring, gasping or having breathing pauses during the night were more likely to be obese. The relationship between HIV infection, and its treatment, is complex. Sleep disturbances are a known effect of HIV infection 3 . This occurs already at an early stage of the infection, and has been suggested to be associated with chronic activation of the immune system 4,42 . HIV infection in itself has neurotoxic effects, quite apart from the effects of any opportunistic pathogens (reviewed in 43 ). It is also well known that patients taking efavirenz, one of the common ART for HIV, experience insomnia and other sleep disturbances such as longer sleep onset latency and shorter sleep duration, in particular in the earlier stages of treatment 5,6 . Electroencephalogram monitoring has shown longer sleep latency and shorter duration of deep sleep in patients taking efavirenz 5 , with higher plasma concentrations of efavirenz associated with insomnia 5,7 . Although it has been reported that some patients stop taking efavirenz as they find the sleep disturbances too debilitating 7,44 , long-term studies find that for most patients these and other neurological and psychiatric side-effects are transient and do not last beyond the first four weeks of treatment 45,46 . Further, the immunoprotective effects of ART have a paradoxical effect when successful treatment commences at an advanced stage of the infection. It has been shown that a high increase in CD4 counts, although a sign of restored immune function, also associates with lower sleep quality (as well as higher incidence of pain and depression) 42 . Our finding that participants with untreated HIV infection and a high viral load reported more nocturnal awakenings than HIV negative participants (p = 0.024) is consistent with previous reports of an effect of the infection on sleep quality. Our finding of less self-reported sleep disturbance in treated than in untreated HIV positive participants are not suggestive of a negative net effect on sleep of ART treatment, either via the effects of the ART drugs or via an asssociated CD4 rebound, in this older population. Although based on self-reported data on sleep quality and quantity, these data do not support any suggestion of replacing efavirenz as first-line treatment specifically because of its effects on sleep, at least in older patients. As with other aspects of these findings, objective measures of sleep parameters, and cellular markers of immune function status would be required in order to determine whether there are effects that are not captured by questionnaires and categorical data on treatment and viral load. Additionally, the present study did not examine other neuropsychiatric symptoms, or younger patients.
In instances where a longer than expected total sleep time is reported, it is reasonable to investigate quality since long sleep may be an indicator for disrupted or inefficient sleep, or an underlying medical condition 47 . In the present study, although the vast majority of participants considered their sleep to be of good quality. There were indications that quality was suboptimal, since one-third took longer than 30 minutes to fall asleep at night and one-third experienced restless sleep. Those with restless sleep were more likely to have hypertension and diabetes, suggesting that sleep quality warrants further investigation in this population.
A limitation of this study is that data collection took place throughout the year across all seasons, since participants were part of the parent HAALSI study. Thus, seasonal variation in photoperiod, with daylight hours in Agincourt ranging between 10h37min and 13h39min, may have influenced self-reported total sleep time. Although this is a rural community, most (but not all) participants have access to electricity and thus light at night. In general, this population rose at, or before, sunrise and remained awake for a few hours after sunset, similar to what has been observed in other rural African populations 16,18 . We were unable to obtain data on exposure to electricity at night, which would have been a helpful covariate in interpretation of these results. Further studies distinguishing between participants with and without access to electricity would be of interest, and would need to be performed before electrification of the region has been completed.
In conclusion, this cohort has the potential to address a number of important questions of current high interest. What is not known is whether the longer total sleep time, and higher prevalence of long sleep (9 h or more) seen in this and other South African studies (i) reflects a greater sleep need, (ii) is a consequence of higher levels of unemployment, (iii) is protective against NCDs, and/or (iv) is underpinned by society-specific beliefs or value placed on sleep in South Africans of African origin. The data presented here provide a basis for further objective studies of the relationship between sleep quality and quantity to other risk factors in South African populations, a complex and underexplored area of research 48 . On a more general note, it also provides valuable data about sleep habits in Africa, with the potential for future analysis of how it is affected by urbanisation and industrialisation. Finally, it offers a unique opportunity for a population-based comparison of the effects of treated and untreated HIV infection, which is increasingly unavailable elsewhere.

Methods
Setting. The HAALSI study 23  Study design. This is a cross-sectional, observational study in which each participant was interviewed at their home by a trained fieldworker. Interviews were performed using a computer-assisted personal interview (CAPI) on a laptop. The individual interview included an extensive set of questions on socio-demographic variables, self-reported health, and wellbeing, as well as anthropometric, blood pressure, and point of care measurements for glucose, lipids, haemoglobin, and HIV. • Over the past 4 weeks, what time did you usually get out of bed? (Hours, minutes)

Outcome measures.
• Over the past 4 weeks, how many hours do you think you actually slept each day? (Hours) • During the past 4 weeks, how often did you wake up in the middle of the night or early morning? ("Never"/"Less than once per week"/"Once or twice a week"/"Three of more times a week") • During the past month, have you snored, or ever been told that you were snoring? ("Yes"/"No") • During the past month, have you snored loudly, or ever been told that you were snoring loudly? ("Yes"/"No") • During the last month, have you had, or ever been told that your breathing stops or you struggle for breath? ("Yes"/"No") • During the last month, have you had, or ever been told that you were snorting or gasping? ("Yes"/"No") • On a scale of 0 to 10, where 0 is "Does not interfere" and 10 is "Completely interferes", select the one number that describes how, during the past 24 hours, pain has interfered with your sleep.
• Over the past 4 weeks, how would you rate your sleep quality overall? ("Very good"/"Fairly good"/"Fairly bad"/"Very bad") • During the past 4 weeks, how often could you not get to sleep within 30 minutes? ("Never"/"Less than once per week"/"Once or twice a week"/"Three of more times a week") • In your life, have you ever had any experience that was so frightening, horrible, or upsetting that, in the past 30 days you had more trouble than usual falling asleep or staying asleep?
• Much of the time in the past week, your sleep was restless. ("Yes"/"No") • How often during the past 4 weeks did you get enough sleep to feel rested upon waking up? ("Never"/"Rarely"/"Sometimes"/"Often"/"Very often") of a household asset score, based on five asset categories -'modern assets' , 'power supply' , 'water and sanitation' , 'quality of housing' , and 'livestock assets' -with increasing values corresponding to greater SES. This score was then divided in quintiles by household. Anthropometric measurements included height (m), weight (kg), waist and hip circumferences (cm). BMI was calculated and participants classified as obese for BMI ≥ 30 kg/m 2 . Resting blood pressure (BP) was measured (Omron M6W ® automated cuff) and participants classified as hypertensive for diastolic BP ≥ 90 mmHg or systolic BP ≥ 140 mmHg or participants reported current use of medication to treat hypertension 53 . Blood was collected using the finger prick technique for dried blood spots (DBS) and point of care measurements were taken for glucose (CareSense, De Puy Synthes, Rainham, MA), lipids (Cardiochek, PTS Diagnostics, Indianapolis, IN), and haemoglobin (HemoCue, Ängelholm, Sweden). Participants were classified as diabetic if their fasting glucose concentration was >7 mmol/L, or non-fasting concentration was ≥11.1 mmol/L or they reported using medication to control diabetes (American Diabetes Association. 2013). Participants were classified as having dyslipidaemia if they had elevated total cholesterol (≥6.21 mmol/L), or low HDL cholesterol (<1.19 mmol/L), or elevated LDL cholesterol (>4.1 mmol/L), or elevated triglycerides (>2.25 mmol/L) levels, or reported current use of cholesterolaemia treatment at time of interview. Anaemia was diagnosed if haemoglobin concentration was <11 g/dl) 1 49.
HIV status was determined by testing DBS for HIV antibodies with enzyme-linked immunosorbent assays [ELISA; 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). Among those who tested HIV positive, assays were performed for viral load and to detect presence of ART. The Division of Clinical Pharmacology Laboratory of the University of Cape Town performed the ART assays, which detected metabolites of emtricitabine (FTC) or lamivudine (3TC) at a concentration as low as 0.02 µg/ml 50 . Either FTC or 3TC or both are included in all standard first-and second-line ART regimens in South Africa, so presence of metabolites of either drug was used to define those currently on ART. A subset of the major individual components of the Pittsburgh Sleep Quality index (PSQI) 24 were analysed in this study, in addition to a number of additional questions. All questions were translated from English to Shangaan and then back translated to English for consistency. The questionnaire in English is listed in Table 5 and the Shangaan version in Annex A. For the purposes of this study the question "Over the past 4 weeks, how many hours do you think you actually slept each day?" was used to determine self-reported sleep duration.
Data and statistical analyses. Data collected in the field were checked regularly for quality and completeness following HAALSI standard operating procedures. Descriptive data are presented as mean ± standard deviation, median (interquartile range) or count (%). Sex comparisons were carried out using an independent t-test, a Mann-Whitney U test or a Fisher's Exact test. The association of demographic variables and sleep parameters were measured using logistic regression. Models for continuous variables were estimated using ordered logistic regression, with sleep duration, bedtime and wake time quartiles as the dependent variables. Quartiles of sleep duration were 4-7 h (quartile 1, n = 1,542), 8 h (quartile 2, n = 1,271), 9 h (quartile 3, n = 1,007), and 10-14 h (quartile 4, n = 853). The quartiles for bed time were earlier than 20:00 (quartile 1, n = 667), 20:00 to <21:00 (quartile 2, n = 1,602), 21:00-21:30 (quartile 3, n = 1,622), and later than 21:30 (quartile 4, n = 1,109). The quartiles for wake time were earlier than 05:00 (quartile 1, n = 887), 05:00 (quartile 2, n = 1,626), >05:00 to 06:00 (quartile 3, n = 1,537), and later than 06:00 (quartile 4, n = 893). We then modelled the association of sleep parameters and health conditions (hypertension, dyslipidaemia, diabetes, anaemia and HIV) also using logistic regression. In these analyses, each sleep parameter was evaluated in a separate model, adjusted for BMI and demographic characteristics (age group, sex, education group, employment status, marital status, wealth index quintile). For those participants who were HIV positive, we studied the association of being on ART and viral load on the main sleep variables (duration, quality, quantity, restless, awaking at night, snoring, gasping, nocturnal apnea). These models were adjusted for BMI and demographic characteristics (age group, sex, education group, employment status, marital status, wealth index quintile). Data were analysed using SAS software v9.3, and significance was accepted for p < 0.05.