Depressive Symptoms and Their Relation to Age and Chronic Diseases Among Middle-Aged and Older Adults in Rural South Africa

Abstract

Background

Understanding how depression is associated with chronic conditions and sociodemographic characteristics can inform the design and effective targeting of depression screening and care interventions. In this study, we present some of the first evidence from sub-Saharan Africa on the association between depressive symptoms and a range of chronic conditions (diabetes, HIV, hypertension, and obesity) as well as sociodemographic characteristics.

Methods

A questionnaire was administered to a population-based simple random sample of 5,059 adults aged 40 years and older in Agincourt, South Africa. Depressive symptoms were measured using a modified version of the eight-item Center for Epidemiological Studies—Depression screening tool. Diabetes was assessed using a capillary blood glucose measurement and HIV using a dried blood spot.

Results

17.0% (95% confidence interval: 15.9%–18.1%) of participants had at least three depressive symptoms. None of the chronic conditions were significantly associated with depressive symptoms in multivariable regressions. Older age was the strongest correlate of depressive symptoms with those aged 80 years and older having on average 0.63 (95% confidence interval: 0.40–0.86; p < .001) more depressive symptoms than those aged 40–49 years. Household wealth quintile and education were not significant correlates.

Conclusions

This study provides some evidence that the positive associations of depression with diabetes, HIV, hypertension, and obesity that are commonly reported in high-income settings might not exist in rural South Africa. Our finding that increasing age is strongly associated with depressive symptoms suggests that there is a particularly high need for depression screening and treatment among the elderly adults in rural South Africa.

The proportion of the total disease burden attributable to major depressive disorder is thought to be increasing in sub-Saharan Africa (SSA) (1). Understanding whether and how strongly depression is associated with other chronic conditions in this region is important from both a health system perspective and an etiological perspective. Using a health system lens, knowing what patient groups are more likely to suffer from depression can inform planning and targeting of screening and care. Routine screening for, and treatment of depression may be particularly beneficial in chronic disease care settings given the strong associations between depression and nonadherence to treatment for diabetes, hypertension, and HIV (2–4). More broadly, evidence on comorbidities of depression with noncommunicable diseases and HIV could provide an impetus for more integrated care of these conditions. From an etiological perspective, associations of depression with diabetes, hypertension, obesity, or HIV could provide evidence suggestive of causal links (whether biological or behavioral) between these conditions.

Supplementary Table S1 details the systematic reviews and meta-analyses that have examined the correlation of depression with diabetes, HIV, hypertension, and obesity. All these reviews (5–10), which mostly identified studies from high-income countries, found positive associations of depression with these chronic conditions. An exception is the systematic review by Atlantis and colleagues (8), which found no association between depression and obesity in cross-sectional studies among non-U.S. populations (but did find positive associations in cross-sectional studies among U.S. populations and in prospective cohort studies).

This article reports on data collected in a large population-based sample of middle-aged and older adults in rural South Africa. To our knowledge, this is the first population-based study in SSA to examine the relationship between depressive symptoms and chronic diseases defined by their associated biomarkers. We aimed to ascertain (i) sociodemographic correlates of depressive symptoms and (ii) whether associations exist in this population between depressive symptoms and diabetes, HIV, hypertension, or obesity.

Methods

Study Setting

This study was conducted in the Medical Research Council/University of Witwatersrand Agincourt Health and Socio-Demographic Surveillance System (HDSS) site. The Agincourt HDSS site is located in a rural area of the Mpumalanga province of South Africa. It covers an area of 475 km² and includes 31 contiguous villages with a population of approximately 110,000 individuals residing in 21,000 households.

Study Population

Only individuals who were aged 40 years or older as of July 1, 2014 and continuously living in the area during the 12 months prior to study enrollment were eligible to participate in the study. We identified 12,484 (70% female) eligible individuals. Of those, 5,890 individuals (50% female) were selected through simple random sampling (stratified by sex). Participants were visited at home to seek informed consent and to be interviewed. Five thousand and fifty-nine (86%) individuals consented to the survey and were included in the final sample. The study measured several health outcomes and was not powered to any specific outcome or exposure. Instead, the sample size was chosen to be able to detect effect sizes of clinical relevance for nonrare exposures (eg, a prevalence of at least 10%).

Data Collection

Local fieldworkers administered a questionnaire to all participants using computer-assisted personal interviews between November 2014 and November 2015. Participants were asked a series of questions about their cardiometabolic health, physical and cognitive functioning, economic well-being, and sexual health. In addition, fieldworkers measured blood pressure (BP; Omron M6W automated cuff), weight (Genesis Growth Management Electronic Scale), and height (using a height sensor with infrared measurement) and took a dried blood spot from the participant. The dried blood spot samples were analyzed for HIV status (testing for HIV antibodies) and, among those who tested HIV positive, nucleoside reverse transcriptase inhibitors. Last, the field workers took finger prick blood samples to measure glucose with a point-of-care glucose meter (Caresens N Monitor). BP was measured on the arm three times with each measurement taken 2 minutes apart. All interviews were conducted in Shangaan.

Outcome Variable

Depressive symptoms were measured using the eight-item version of the Center for Epidemiological Studies—Depression (CES-D) screening tool, which is an abbreviated version of the 20-item CES-D tool. The tool was translated into Shangaan by staff from the Agincourt MRC/Wits Rural Public Health and Health Transitions Research Unit’s Public Engagement Office and then translated back into English (by a person different to the one doing the translation into Shangaan) to verify accuracy of the translation. Extensive piloting of the tool prior to the study led to further minor edits. To our knowledge, the Shangaan version of the eight-item tool has not been validated or used previously. The tool comprises a set of eight “yes/no” questions on whether a respondent has experienced symptoms indicative of depression in the prior week, including feeling depressed, feeling “like everything is an effort” (because of difficulties in translating this phrase into Shangaan, this was translated to “did you feel that everything you did was easy” and reverse coded), restless sleep, feeling happy, feeling lonely, enjoying life, feeling sad, and being “unable to get going.” Each individual’s set of responses (coded as 1 for “yes” and 0 for “no”) was translated into a summed score, with equal weighting given to each question (11). We refer to this score (ranging from 0 to 8) as the “depression score” and used this score as the dependent variable in all regression analyses.

Explanatory Variables

Socioeconomic and demographic variables used as independent variables were sex, age group, education (none, primary, secondary, or more than secondary), employment status (employed or unemployed), marital status (married or unmarried), and household wealth. Household wealth was measured using a series of questions on the physical characteristics of the dwelling (eg, material of the floor, roof, and walls) and whether the household owned each of 40 different durable goods (eg, a car or a television). As per the methodology developed by Filmer and Pritchett (12), we summarized the answers to these questions using a principal component analysis from which we extracted the first principal component, which was then categorized into quintiles.

Chronic conditions used as independent variables were HIV infection, obesity, hypertension, and diabetes. Obesity was defined as a body mass index (BMI) of 30 kg/m² or greater. Mean BP was calculated by averaging the second and the third reading (13). Hypertension was defined as a participant reporting taking hypertensive medication or having a mean systolic BP ≥ 140 mmHg or a mean diastolic BP ≥ 90 mmHg (14). Diabetes was defined as reporting to be on treatment for diabetes (insulin or an oral antidiabetic drug) or having a fasting blood glucose level ≥ 7.0 mmol/L or a random blood glucose level ≥ 11.1 mmol/L (15). A participant was considered to be fasting if he/she reported no caloric intake for at least 8 hours prior to blood sample collection—this was the case for 24% of participants. We defined participants to have been diagnosed with diabetes or hypertension if they reported that a health worker had told them that they have a raised BP or blood sugar. Similarly, participants were considered to be on treatment for diabetes or hypertension if they answered “yes” to the question “Are you currently receiving any treatment for [diabetes/high BP] prescribed by a doctor, nurse, or other healthcare worker?”. Being on antiretroviral therapy (ART) was defined as having a positive blood test result for the presence of Emtricitabine (FTC) and Lamivudine (3TC). Either FTC or 3TC (or both) was part of all first- and second-line ART regimens ever recommended in South Africa (16).

Statistical Analysis

The sample is self-weighting by sex, and thus, no sampling weights were necessary to calculate the prevalence of depression. We ran univariable and multivariable ordinary least squares regressions with the depression score as the dependent variable. The multivariable model included sociodemographic characteristics and an indicator for each chronic condition (BMI group, diabetes, HIV, and hypertension) as defined in the Explanatory variables section. In addition, we fitted both univariable and multivariable ordinary least squares regressions of the depression score on an indicator for whether the participant reported to have been diagnosed with the condition (diabetes, HIV, and hypertension) or not and whether the participant reported to be currently on treatment for the condition or not. Standard errors for all prevalence estimates and regressions were adjusted for clustering at the household level. In sensitivity analyses, we also fitted generalized ordered logit models. All statistical analyses were conducted in Stata 14.0 (College Station, TX).

Ethics

This study received ethics approval from the University of Witwatersrand (#M141159), the Harvard T.H. Chan School of Public Health (#13-1608), and the Mpumalanga Provincial Research and Ethics Committee.

Results

Ninety-seven percent (4,929/5,059) of participants answered all eight questions in the depression tool, 93% (4,689/5,059) had both their height and weight measured, and 97% (4,904/5,059) had a BP measurement. An HIV test result was available for 91% (4,582/5,059) of respondents and a blood glucose measurement for 92% (4,648/5,059).

Sample Characteristics

Women comprised 53.7% of the study sample (Table 1). More than half (54.2%) of participants were aged between 50 and 70 years (mean age was 62 years). 45.7% of respondents reported no formal education, and employment was low with only 26.8% of respondents reporting active employment. 29.8% of respondents were obese (BMI ≥ 30.0 kg/m²) whereby women were more likely to be obese than men (41.5% vs. 15.9%, respectively). 10.5% of participants had diabetes of whom approximately half were diagnosed and currently on treatment (54.6% and 49.8%, respectively). 23.2% were HIV positive of whom 76.3% were aware of their status and 63.9% on ART. More than half (58.5%) of participants had hypertension—64.6% of these adults reported a prior diagnosis of hypertension and 49.9% were currently on antihypertensive treatment. Although diagnosis and treatment proportions were similar between men and women for diabetes and HIV, hypertensive men were less likely than women to report a diagnosis of hypertension or to currently be on treatment.

Table 1.

Sample Characteristics

Sample Characteristic	Total	Female	Male
	N = 4,929	n = 2,649	n = 2,280
Age group (y)
40–49	902 (18.3%)	498 (18.8%)	404 (17.8%)
50–59	1,391 (28.3%)	779 (29.4%)	612 (26.9%)
60–69	1,282 (26.0%)	649 (24.5%)	633 (27.8%)
70–79	846 (17.2%)	418 (15.8%)	428 (18.8%)
≥80	502 (10.2%)	303 (11.5%)	199 (8.7%)
Education
None	2,211 (45.7%)	1,294 (49.0%)	917 (40.3%)
Primary schoolinga	1,695 (34.5%)	876 (33.2%)	819 (36.0%)
Secondary schoolingb	570 (11.6%)	259 (9.8%)	311 (13.7%)
>Secondary schooling	441 (9.0%)	212 (8.0%)	229 (10.1%)
Employedc	1,318 (26.8%)	693 (26.2%)	625 (27.5%)
Currently married	2,548 (51.7%)	970 (36.7%)	1,578 (69.2%)
Body mass index
<18.5 kg/m2	247 (5.5%)	65 (2.6%)	182 (8.6%)
18.5–24.9 kg/m2	1,685 (36.5%)	688 (27.6%)	997 (47.1%)
25.0–29.9 kg/m2	1,308 (28.4%)	706 (28.3%)	602 (28.4%)
≥30.0 kg/m2	1,372 (29.8%)	1,035 (41.5%)	337 (15.9%)
Diabetes	480 (10.5%)	282 (11.5%)	198 (9.5%)
Diagnosed with diabetes	262 (54.6%d)	152 (53.9%d)	110 (55.6%d)
On treatment for diabetes	239 (49.8%d)	139 (49.3%d)	100 (50.5%d)
HIV positive	1,037 (23.2%)	560 (23.2%)	477 (23.2%)
Diagnosed with HIV	790 (76.3%d)	428 (76.4%d)	362 (76.1%d)
On ART	654 (63.9%d)	342 (61.6%d)	312 (66.5%d)
Hypertension	2,821 (58.5%)	1,611 (61.8%)	1,210 (54.6%)
Diagnosed with hypertension	1,822 (64.6%d)	1,140 (70.8%d)	682 (56.4%d)
On treatment for hypertension	1,407 (49.9%d)	905 (56.2%d)	502 (41.5%d)

Note: ART = antiretroviral therapy; HIV = human immunodeficiency virus.

^aHad at least some primary schooling. ^bHad at least some secondary schooling. ^cEmployment was defined as being engaged in full-time or part-time work. ^dThe denominator for the percentage was all those who had the condition (diabetes, HIV, or hypertension).

Prevalence of Depressive Symptoms

17.0% (95% confidence interval: 15.9%–18.1%) of participants had a depression score of 3 or more. Supplementary Figure S1 shows the distribution of scores on the eight-item CES-D tool. A third (33.1%) of respondents had a depression score of 0.

Association of Depressive Symptoms With Chronic Conditions

Table 2 displays the coefficients from ordinary least squares regressions of the depression score on sociodemographic variables and chronic conditions. These coefficients should be interpreted as the mean absolute change in the number of depressive symptoms (on a scale from zero to eight symptoms) associated with a one unit increase in the explanatory variable. Compared to being of normal weight, both underweight and overweight were positively correlated with the depression score, whereas the associations for obesity were insignificant. Diabetes and hypertension were both positively associated with the depression score in univariable regressions (coefficient of 0.27 [0.10–0.44] and 0.16 [0.07–0.25], respectively), whereas those with HIV had on average 0.21 (0.11–0.32) less points on the eight-point depression score. The coefficients on these chronic conditions maintained their direction but became insignificant and their absolute value smaller when adjusted for BMI group and sociodemographic characteristics. Regarding participants’ sociodemographic characteristics, older age was positively associated with the depression score in both the univariable and multivariable regression, whereas being married and employment were negatively associated. Male sex, household wealth, and education were only significantly (negatively) correlated with the depression score in univariable regressions.

Table 2.

OLS Regressions of the Depression Score (Ranging From Zero to Eight) on Sociodemographic Variables and Chronic Disease Indicators

	Univariable		Multivariablea
	Coefficient (95% CI)	p Value	Coefficient (95% CI)	p Value
Body mass index
<18.5 kg/m2	0.46 (0.20 to 0.72)	.001	0.45 (0.19 to 0.71)	.001
18.5–24.9 kg/m2	Reference		Reference
25.0–29.9 kg/m2	0.12 (0.01 to 0.24)	.031	0.17 (0.06 to 0.29)	.004
≥30.0 kg/m2	0.01 (−0.10 to 0.12)	.873	0.07 (−0.06 to 0.19)	.276
Diabetes	0.27 (0.10 to 0.44)	.002	0.13 (−0.04 to 0.29)	.135
HIV+	−0.21 (−0.32 to −0.11)	<.001	−0.08 (−0.19 to 0.03)	.173
Hypertension	0.16 (0.07 to 0.25)	.001	0.06 (−0.04 to 0.15)	.260
Female	0.11 (0.03 to 0.20)	.012	0.02 (−0.08 to 0.13)	.660
Age group (y)
40–49	Reference	<.001b	Reference	<.001b
50–59	0.24 (0.12 to 0.36)		0.15 (0.02 to 0.28)
60–69	0.35 (0.23 to 0.48)		0.19 (0.05 to 0.34)
70–79	0.56 (0.42 to 0.70)		0.30 (0.13 to 0.46)
≥80	1.04 (0.85 to 1.24)		0.63 (0.40 to 0.86)
Currently married	−0.37 (−0.46 to −0.28)	<.001	−0.24 (−0.35 to −0.14)	<.001
Employedc	−0.59 (−0.69 to −0.50)	<.001	−0.42 (−0.53 to −0.32)	<.001
Wealth quintile
1 (poorest)	Reference	<.001b	Reference	.075b
2	0.09 (−0.06 to 0.24)		0.09 (−0.07 to 0.25)
3	0.04 (−0.11 to 0.19)		0.06 (−0.10 to 0.22)
4	−0.16 (−0.30 to −0.02)		−0.09 (−0.24 to 0.07)
5 (wealthiest)	−0.25 (−0.40 to −0.11)		−0.08 (−0.25 to 0.08)
Education
None	Reference	<.001b	Reference	.286b
Primary	−0.13 (−0.23 to −0.02)		0.03 (−0.08 to 0.14)
Secondary	−0.39 (−0.53 to −0.24)		−0.07 (−0.23 to 0.09)
>Secondary	−0.58 (−0.72 to −0.44)		−0.07 (−0.25 to 0.10)

Note: CI = confidence interval; HIV+ = HIV positive; OLS = ordinary least squares.

^aThe multivariable regression included all variables listed in this table as independent variables. ^bp Value for linear trend. ^cEmployment was defined as being engaged in full-time or part-time work.

The results were similar when stratifying the regressions by age group (results displayed in Supplementary Table S2). Of note, however, is that diabetes was most strongly associated with the depression score among those aged 70 years and older (univariable: 0.49 [0.15–0.83], p = .005; multivariable: 0.33 [−0.01 to 0.67], p = .058). Similarly, interacting age—as a continuous variable in years—separately with each of the chronic disease indicators (results shown in Supplementary Table S3) identified a significant positive interaction with diabetes (0.014 [0.000–0.028], p = .046). However, all other age–chronic disease interactions were not significant.

Neither did the multivariable regression results change substantially when each chronic condition was adjusted separately for all sociodemographic characteristics (results not shown), nor when the regressions were run separately for men and women. An exception was overweight (25.0 ≤ BMI < 30.0 kg/m²), which was significantly associated with depressive symptoms among women (coefficient of 0.35 [0.17 to 0.53], p < .001) but not men (coefficient of 0.03 [−0.12 to 0.18], p = .732). The interaction term between sex and overweight was also the only significant interaction between sex and the chronic disease indicators (Supplementary Table S4).

The results did not differ substantially when using generalized ordered logit models instead of ordinary least squares regressions (Figure 1 and Supplementary Table S5). Figure 1 shows the predicted probability of each possible score on the eight-point depression scale by whether a participant had each chronic condition (HIV, diabetes, hypertension, and obesity) or not.

Figure 1.

The predicted probability of having each possible depression score, by chronic condition. The predicted probabilities were obtained from multivariable generalized ordered logit models adjusting for participants’ sociodemographic characteristics (sex, age, marital status, employment status, household wealth quintile, and education). The predicted probabilities shown held participants’ sociodemographic characteristics at their observed values. Vertical lines are 95% confidence intervals. The corresponding table for this figure is shown in Supplementary Table S2. HIV− = HIV negative; HIV+ = HIV positive.

Association of Depressive Symptoms With Diagnosis and Treatment of a Chronic Condition

Neither diagnosis nor treatment of HIV was associated with depressive symptoms when compared with being uninfected (Supplementary Table S6). However, reporting to have been diagnosed with or to be currently on treatment for diabetes and hypertension was positively associated (when compared with not having the condition) with depressive symptoms among females, but not among males. This impression was confirmed when including interaction terms between sex and indicators of chronic disease diagnosis and treatment (Supplementary Table S7).

Discussion

In contrast to studies from high-income countries, we did not identify significant associations of depressive symptoms with diabetes, HIV status, hypertension, or obesity once adjusted for sociodemographic variables. Although a larger sample size may have led to statistically significant findings, the coefficients on each chronic condition were small (not exceeding an absolute value of 0.13 in the multivariable regression) and, in fact, negative for HIV. In addition, the sample size of this survey (~5,000 participants) was comparatively large and sufficient to detect highly significant associations between the depression score and sociodemographic characteristics. Interestingly, reporting to have been diagnosed with, or to be currently on treatment for diabetes and hypertension was positively associated with depressive symptoms among females, though not among males. Although the cross-sectional design of this study limits interpretation of temporal relationships and causal inference, this finding nonetheless provides suggestive evidence that hypertension and diabetes diagnosis causes or worsens depressive symptoms among women and/or that depressed women are less likely to seek a diagnosis and to adhere to treatment for these conditions.

Even though the association was not significant in multivariable regressions, those with HIV had on average less depressive symptoms than those without HIV. In a population-based cohort in Uganda, Manne-Goehler and colleagues have recently reported less depressive symptoms among HIV-infected adults on ART compared with age- and sex-matched HIV-uninfected adults (17). In addition, in a population-based sample of 36,000 adults in Zambia and South Africa (participants of the “PopART” trial), Thomas and colleagues found that HIV-infected and HIV-uninfected adults had a similar health-related quality of life (18). It is, therefore, possible that our study is part of an emerging body of evidence from SSA that mental health and health-related quality of life among people with HIV is not worse than among those without HIV. The fact that ART is available widely and free of charge in South Africa may reduce depressive symptoms among those with HIV (16), which is supported by our observation that being on treatment for HIV increased the negative association between HIV and depressive symptoms. In addition, it is also possible that HIV care services provide an avenue to better health care for conditions other than HIV, which may conceivably lead to improved mental health. In initial support of this hypothesis, we have shown in this same sample of middle-aged and older adults that those with HIV were more likely to have received preventive care for diabetes and hypertension than those without HIV (19). However, we also recognize that selection effects may be responsible for the negative (or null) association between HIV and depressive symptoms. For instance, those who are at a higher risk of being infected with HIV may be less likely to have (or subsequently develop) depression—one potential pathway is the reduced interest in sexual activity that often accompanies depression (20).

The South African population is aging rapidly and is projected to continue doing so over the next decades (21). This population aging is partly due to the massive scale-up of ART, which greatly extended the life expectancy of people living with HIV in the country. In fact, in a similar community in rural South Africa, Bor and colleagues showed that life expectancy had increased by 11.3 years between 2003 (the year prior to ART becoming available in the public sector) and 2011 (22). In this study, we found highly significant positive correlations between older age and depressive symptoms in both univariable and multivariable regressions. In addition, the magnitude of the coefficients on age group was generally the highest of any of the variables included in the regression models. Thus, our results suggest that the aging of many societies in the region might be an important driver of the increasing burden of depression in SSA (1). In addition, although depression and psychological well-being are not equivalent, our findings imply that the relationship between psychological well-being and depression with age may be different in rural SSA than in high-income countries. Despite remaining uncertainty on this topic (23), in high-income settings, depression and poor psychological well-being are generally thought to have the highest prevalence in middle-aged adults (23–25). However, similar to our analysis, studies from rural settings in Ethiopia, Malawi, and South Africa all found monotonically increasing odds of depression with older age (26–28).

This study has several limitations. First, the cross-sectional design of this study does not allow for a causal interpretation of the findings. Second, to our knowledge, the CES-D 8 tool has not been validated for depression screening by comparing its performance to a “gold standard” for diagnosing depression. Furthermore, the factor structure and psychometric properties of the CES-D 8 tool for measuring depression have only been assessed in European adults (29,30). However, another abbreviated version of the CES-D tool—the 10-item “CES-D 10”—has been used extensively in SSA [including in South Africa (31), Kenya (32), and Ghana (33)] and has been found to be both valid and reliable for Zulu, Xhosa, and Afrikaans populations in South Africa (34). We chose the 8-item rather than the 10-item (or the even more widely used and validated 20-item) version of the CES-D tool to allow for direct comparability with the U.S. Health and Retirement Survey (11), on which the entire study and questionnaire were modeled. Nonetheless, to explicitly acknowledge this limitation, we refer to the outcome studied in this analysis as “depressive symptoms” rather than “depression.” Last, the findings of this study do not necessarily extend to settings other than rural South Africa nor to younger adults, who were not part of this sample.

In conclusion, we found that diabetes, HIV, hypertension, and obesity were not associated with depressive symptoms in this cross-sectional study of 5,000 middle-aged and older adults in rural South Africa. This study, therefore, provides some indication that the positive correlation of depression with diabetes, HIV, hypertension, and obesity, which has been widely reported in studies from high-income countries, might not (or to a lesser degree) hold true for rural settings in SSA. An implication of our findings for health systems in the region is that targeting depression screening at those with chronic diseases could be less effective than targeting population groups based on their sociodemographic characteristics, particularly age.

Funding

The HAALSI study, funded by the National Institute on Aging (P01 AG041710), is nested within the Agincourt Health and Socio-Demographic Surveillance System site, supported by the University of the Witwatersrand and Medical Research Council, South Africa, and the Wellcome Trust, UK (grants 058893/Z/99/A; 069683/Z/02/Z; 085477/Z/08/Z; 085477/B/08/Z). None of the funders had a role in the design, method, subject recruitment, data collection, analysis, and preparation of the article.

Conflict of Interest

None reported.