Detecting Depression in People Living with HIV in South Africa: The Factor Structure and Convergent Validity of the South African Depression Scale (SADS)

Abstract

Screening measures for depression developed in high-income countries have not always demonstrated strong psychometric properties in South Africa and with people living with HIV (PLWH). The present study explored the psychometric properties of the 16-item South African Depression Scale (SADS) comprised of idioms of distress specific to isiXhosa culture in PLWH. The SADS was administered to 137 Xhosa-speaking PLWH who met diagnostic criteria for major depressive disorder (MDD) together with the Hamilton Depression Scale (HAM-D) and the Center for Epidemiological Studies Depression Scale (CES-D). We conducted exploratory factor analysis, correlation, and reliability statistics. Four factors of the SADS emerged: Sadness, lethargy/burdened, anhedonia/withdrawal, and cognitive/somatic. All factors correlated significantly with the HAM-D and CES-D. Internal consistency of the overall measure was high (α = .89). The SADS promises to be a robust measure of depression in isiXhosa-speaking PLWH in South Africa likely due to the inclusion of local idioms of distress.

Introduction

Depression is a global epidemic and the leading cause of disability worldwide [1]. Low- and middle-income countries (LMICs) carry the highest burden with the greatest number of people with depression residing in these settings [2]. The first and only national psychiatric epidemiological study conducted in South Africa found a 9.8% 12-month prevalence rate for major depressive disorder [3]. The prevalence rate increases significantly with at-risk sub-populations, such as in people living with HIV (PLWH) due to the many challenges they face including stigma, loss of social support, grief, and HIV disclosure concerns [4]. Furthermore, the underlying biological processes of HIV, such as inflammation [5] and immune activation [6], may increase the risk for depression.

South Africa has the greatest HIV/AIDS burden in the world and rates of mild to major depression have been found to be as high as 41% in PLWH [4]. Not only does depression affect well-being and functioning [7], but the double disease burden of depression and HIV can have serious health implications, including sub-optimal adherence to antiretroviral therapy (ART) [8–10], accelerated disease progression [11], and mortality [12]. Identifying major depressive disorder (MDD) in PLWH by means of culturally sensitive assessments, and treating MDD in South Africa informed by such assessments should therefore be a public health priority.

To date, studies that have assessed clinical depression in PLWH in sub-Saharan Africa have taken an etic approach and have used translated versions of scales developed in high-income countries or by the World Health Organization. Certain measures have not been subjected to thorough psychometric examination in South Africa, or indeed in PLWH in South Africa, while others have demonstrated sub-optimal psychometric properties with particularly high false positive rates [13, 14]. For example, the Centre for Epidemiological Studies Depression Scale (CES-D) [15] and the Kessler Psychological Distress Scale (K10) [16] have shown adequate sensitivity and specificity when administered to PLWH in South Africa; however, the positive predictive values for both scales were well below 50% in these samples [15, 16]. In resource-constrained settings like South Africa, screening measures should ideally identify true cases of clinical depression while yielding low false positives rates. This is to minimize diagnostic burden and to ensure that the scarce mental health resources that are available in public health are allocated to patients with true mental disorders. If clinical depression screening is to be feasibly integrated into primary HIV care in South Africa, a measure is needed that not only manages to identify those who are depressed, but also succeeds in accurately identifying those who are not depressed.

While some global mental health studies suggest cross-cultural similarities in clinical depression symptomatology (e.g. sadness, anhedonia, lethargy, etc.), distinct cultural descriptions and expressions of clinical depression have also been identified in some sub-Saharan African settings [17]. In South Africa, descriptions of the symptoms of clinical depression include “thinking too much”, “not being myself”, “having pain in my heart and head”, and “carrying a lot of weight on my shoulders” [18]. These cultural distinctions in the experience of depression could account for the sub-optimal predictive properties evident in studies of translated measures. Developing a local, culture-specific screener for clinical depression in PLWH could provide a more robust measure of this prevalent psychiatric condition.

Identifying the various aspects of clinical depression experienced by PLWH in South Africa is not only important diagnostically and for screening purposes, it is necessary for the development of appropriate and effective psychological interventions. Based on the culture-specific factors identified, evidence-based treatments such as cognitive-behavioral therapy could be appropriate treatment approaches, informed by culturally sensitive assessments. For example, behavioral activation could be appropriate for targeting the symptoms of thinking too much and behavioral withdrawal, while relaxation training may be particularly pertinent for addressing stress and sleep difficulties. Ultimately the experience of clinical depression in isiXhosa-speaking PLWH will dictate the most appropriate treatment modalities to comprise a psychotherapeutic intervention.

The South African Depression Scale (SADS) is a 16-item measure of clinical depression developed from an emic approach to reflect the culture-specific characteristics of depression experienced by isi-Xhosa speaking PLWH in South Africa. In a previous study, we conducted semi-structured qualitative interviews on the experience of clinical depression with isiXhosa-speaking PLWH who met diagnostic criteria for MDD [18]. The culture-specific and traditional symptoms of depression identified were combined to create the SADS. This study sought to examine the factor structure, convergent validity and internal reliability of the SADS in isi-Xhosa speaking PLWH in South Africa, as a first step to ascertaining the psychometric properties of this scale.

Methods

Participants

The sample of patients who met diagnostic criteria for MDD were drawn from 137 participants who were enrolled in the Ziphamandla study, an ongoing randomized controlled trial of the effectiveness of cognitive-behavioral therapy for adherence and depression (CBT-AD) among PLWH with poor antiretroviral therapy (ART) adherence in South Africa. Adult patients who failed first-line ART, were virally unsuppressed (VL ≥ 400 g/mL), spoke English or isiXhosa, and met diagnostic criteria for MDD as determined by the clinician-administered Mini International Psychiatric Interview (MINI) [19] were eligible. Participants were recruited from six Infectious Disease clinics in Khayelitsha run by the City of Cape Town Department of Health. Khayelitsha is a large peri-urban area situated east of Cape Town. isiXhosa is the most commonly spoken language in this area. Baseline data obtained between July 2016 and July 2018 were included in the analyses.

Procedure

As part of the baseline evaluation for the parent trial, all participants completed a clinician-administered assessment of their clinical depression, including the MINI and the Hamilton Depression Scale (HAM-D) [20]. They also completed a battery of self-report measures administered by a bilingual research assistant in their preferred language, either English or isiXhosa.

Ethical approval was obtained from the University of Cape Town’s Human Research Ethics Committee (HREC), South Africa and the University of Miami’s Institutional Review Board (IRB), U.S.A. All participants provided written informed consent. Clinic access was granted by the City of Cape Town’s Health Department.

Measures

South African Depression Scale (SADS)

Semi-structured qualitative interviews on the manifestation and experience of clinical depression were previously conducted with 14 isi-Xhosa speaking participants who were HIV-positive and met diagnostic criteria for major depressive disorder on the MINI [18]. Local idioms of distress were identified (9 items) and combined with traditional symptoms of depression (6 items) as well as a final item, isingqala (deep sorrow) added from consultation with local, isiXhosa speaking mental health providers. These mental health providers included a psychiatrist, two psychiatric nurses, and two counsellors working in primary HIV care. The idioms of distress were combined in the 16-item SADS self-report scale. An isiXhosa and English version of the SADS was created. Items were translated from English to isiXhosa and from isiXhosa to English using Brislin’s back-translation procedure [21].

On the scale, participants are asked how many days in the past week they have been experiencing each symptom. Each item is scored on a 4 -point Likert scale (0 days, 1–2 days, 3–4 days, 5 or more days). The scores range from 0 to 48 with higher scores indicating greater severity. Participants are asked at the end to indicate how long they have been feeling this way (the response option is open ended). The measure is designed to be administered via either self-report or interview. In the current study, the measure was administered via interview by a trained, bilingual research assistant in isiXhosa or English. The measure took approximately 12 min to administer in the participant’s preferred language.

Center for Epidemiological Studies Depression Scale (CES‑D)

The CES-D [15] is a 20-item measure of depression where participants are asked to rate each item on a 4-point Likert scale (rarely or none of the time, some or a little of the time, occassionally or a moderate amount of time, most or all of the time. The four positive items on the scale are reverse coded. The scores range from 0 to 60 with higher scores indicating greater severity of depression. The CES-D has been validated in South Africa in PLWH [22]. The measure was administered by interview via a trained bilingual research assistant.

HAM‑D

The HAM-D is a widely-used 17-item clinician-administered scale for assessing the severity of depressive symptoms [23]. It has been used in several antidepressant medication trials in South Africa [24–26]. The scores range from 0 to 50 with higher scores indicating greater severity of depression [20]. The measure was administered by interview via trained bilingual psychiatric nurses.

Data Analysis

The Statistical Package for the Social Sciences (SPSS) [27] version 25, as well as the psych [27] and stats packages of R version 3.5.1 [28], were used to conduct the analyses. Descriptive statistics were performed to describe the sociodemographic characteristics of the sample. An exploratory factor analysis of the SADS was conducted. Exploratory factor analysis is appropriate when there are no a priori expectations of the composition of the sub-scales based on previous research [29]. The structure of the SADS was examined using a principle component structure. Principle axis factoring was used as the extraction method. An oblique rotation was specified because the factors were thought to be related. Model fit was assessed using the root mean square of the residuals (RMSR), the root mean square error of approximation (RMSEA) and the Tucker Lewis Index (TLI). In accordance with procedures specified by Floyd and Widaman [30] items with factor loadings greater than or equal to 0.3 were retained. Pearson’s correlations were used to determine convergent validity, a subtype of construct validity, which indicates how similar constructs are by assessing the associations between the SADS and the CES-D and HAM-D. Cronbach’s alpha coefficients were calculated to determine the internal consistency of the scale and its factors.

Results

Sociodemographic Characteristics

Table 1 details the sociodemographic characteristics of the sample. Participants ranged in age from 22 to 69, with a mean age of 39.2. The majority of participants were female (69.7%) with one participant identifying as transmale. isiXhosa was the most common language spoken at home (81.7%). Most of the participants had at least one child (90.1%), although only 39.4% of participants were currently receiving a child support grant (despite parents/guardians below a certain income level being eligible for a child support grant in South Africa). With respect to employment status, 81% of participants were unemployed, 5.6% were employed full-time, and 7% had casual employment. Only 8% of participants were receiving a disability grant. Household income ranged from ZAR 0 to 15,000 per month, with a mean monthly income of ZAR 1,950.63 (approximately USD136).

Table 1.

Sociodemographic characteristics of participants

Variable	N	%
Sex
Female	99	69.7
Male	42	29.6
Transmale	1	0.7
Home language
isiXhosa	116	81.7
English	3	2.1
Other	23	16.2
Children	128	90.1
Employment status
Unemployed	115	81.0
Full-time	8	5.6
Casual	10	7.0
Grant
Child	56	39.4
Disability	11	7.7
	Mean	S.D
Age	39.2	9.2
Monthly household income (ZAR)	1951	2404
HIV characteristics
Viral load (copies/mL)	49,045	132,124
CD4 count	254	196
Years since diagnosis	8	5

Characteristics of the SADS

Total SADS scores ranged from 5 to 43, with a mean score of 31.14 (SD = 9.64). Table 2 lists the frequencies of endorsed items. The two most frequently endorsed items were ‘I was carrying a lot of weight on my shoulders’ and ‘I had stress’. The least endorsed item was ‘I cried’.

Table 2.

Frequency of SADS items endorsed

	Item	N (of 146)	%
1	Things were not going well in my life	126	89.8
2	I had difficulties sleeping	134	94.9
3	I was thinking too much	132	94.2
4	I did not want to be around people	122	85.4
5	I was feeling tired or without energy	119	85.4
6	I was not my usual self	119	87.6
7	I was carrying a lot of weight on my shoulders	122	94.9
8	I had stress	134	88.3
9	I did not feel like talking	123	89.8
10	I had pain in my heart	126	88.7
11	I cried	94	67.9
12	I was irritated easily	128	91.2
13	I was sad	128	91.2
14	I had no appetite or I ate more than usual	117	83.9
15	I was not interested in the things I used to like doing	122	87.6
16	I had isingqala (deep sorrow)	103	72.3

Exploratory Factor Analysis

Exploratory factor analysis of the SADS yielded four factors having an eigenvalue greater than 1, explaining 62.93% of the variance in the data (see Table 3). An examination of the unrotated scree plots of eigenvalues indicated the potential of a two-factor solution. However, the absolute fit indices (RMSR, TLI, RMSEA) demonstrated that the two-factor solution did not have adequate fit, 0.09, 0.84, and 0.07 respectively. The four-factor solution, however, seemed a better theoretical fit and had adequate absolute fit. The RMSR was 0.05. The TLI was 0.95, and the RMSEA was 0.05 with a 90% confidence interval of 0.07.

Table 3.

Factors with eigenvalues greater than 1.0 and degrees of variance accounted for by each factor

Factor	Eigenvalue	% of variance	Cumulative % of variance
1	6.11	38.19	38.19
2	1.49	9.32	47.51
3	1.20	7.47	54.98
4	1.12	7.01	61.99

Table 4 depicts the factors, sorted by factor loading score. The four factors were labelled, ‘sadness’, ‘anhedonia/withdrawal’, ‘burdened/lethargy’, and ‘cognitive/somatic’. The highest factor loading was for the item “I had pain in my heart” (0.79), which is a culture-specific idiom of distress, with sadness having the second highest loading (0.68). Two items loaded onto more than one factor (“I was feeling tired or without energy” and “I was thinking too much”) and were included with the factor on which the item had the highest loading. No items were dropped after oblique rotation as all factor loadings were higher than 0.30.

Table 4.

Rotated factor matrix

Item number and wording	Factor 1: Sadness	Factor 2: Anhedonia/withdrawal	Factor 3: Burdened/lethargy	Factor 4: Cognitive/somatic
I had pain in my heart	.79	.16	.18	.11
I was sad	.68	.29	.22	.19
I cried	.60	.12	.11	.11
I had isingqala (deep sorrow)	.52	.08	.26	.12
Things were not going well in my life	.51	.24	.03	.08
I had stress	.42	.10	.35	.25
I did not feel like talking	.18	.74	.20	.11
I was not interested in the things I used to like doing	.22	.61	.21	.23
I did not want to be around people	.29	.57	.18	.27
I was irritated easily	.30	.33	.28	.18
I was feeling tired or without energy	.13	.44	.74	.02
I was not my usual self	.26	.32	.66	.23
I was carrying a lot of weight on my shoulders	.19	.07	.63	.19
I was thinking too much	.37	.28	.07	.59
I had difficulties sleeping	.28	.04	.17	.55
I had no appetite or I ate more than usual	− .02	.23	.15	.55

The items comprising each factor are indicated by factor loadings in bold

Convergent Validity: Pearson Correlations

Convergent validity was assessed by examining the relationship of the SADS with measures of the same construct: the CES-D and the HAM-D. The total SADS and each of the factors were positively correlated with the CES-D (p < 0.001) and the HAM-D (p < 0.001). The burdened/lethargy factor correlated most strongly with the CES-D, while the sadness factor correlated most strongly with the HAM-D. The CES-D and the HAM-D were also correlated with each other (r = 0.49). Table 5 shows the correlations between these measures and the SADS and each of the four factors.

Table 5.

Relationship of the CESD-D and HAM-D with the SADS and Factors

	Total SADS	Factor 1 Sadness	Factor 2 Burdened/lethargy	Factor 3 Anhedonia/withdrawal	Factor 4 Cognitive/somatic
CES-D	.68	.52	.61	.52	.52
HAM-D	.51	.55	.51	.45	.30

Reliability: Internal Consistency

Cronbach’s alpha coefficients were calculated for the SADS, CES-D, and HAM-D. The internal consistency of the SADS was excellent (α = 0.89), and comparable to yet slightly higher than the CES-D (α = 0.88) and the HAM-D (α = 0.75).

Discussion

This study presents strong initial psychometric properties of the SADS, the only locally developed measure for clinical depression in HIV in South Africa that we are aware of. The exploratory factor analysis of the SADS yielded a four-factor solution: sadness, burdened/lethargy, anhedonia/withdrawal, and cognitive/somatic. Across studies characterizing the factor structure of depression measures developed in high-income countries, i.e. the HADS, BDIII [31] and CES-D [32] in South Africa and the HSCLD in Uganda [33] three similar factors have emerged: depressed affect, somatic complaints, and behavioral withdrawal. These are consistent with the factors ‘sadness’, ‘cognitive/somatic’, and ‘anhedonia/withdrawal’ identified in the factor structure of the SADS. However, the burdened/lethargy factor identified in the SADS factor structure appears to be unique to this scale. This factor consists of three items, two of which (“I am carrying a lot of weight on my shoulders” and “I am not my usual self”) are culture-specific descriptions of clinical depression provided by isi-Xhosa speaking PLWH in South Africa identified in our prior work [18]. This factor may highlight a distinct aspect of the culture-specific presentation of depression that could be missed using customary scales developed in high-income countries.

In high income countries, the construct of clinical depression requires sadness or anhedonia followed by at least four other symptoms for diagnosis, with sadness being a key symptom. In our study, we found that the item ‘I was sad’ only had the second highest factor loading, while the item ‘I have pain in my heart’ had the highest factor loading. This seems to indicate that in the South African HIV care context the factor ‘I have pain in my heart’ is a more robust representation of a core symptom of clinical depression. This finding is consistent with other qualitative studies conducted in sub-Saharan Africa where “pain in my heart” was used by youth in Uganda [34] and by mothers in DR Congo [35] to describe psychological distress. The inclusion of this symptom on the SADS depression scale may improve its sensitivity and positive predictive value.

The SADS demonstrated excellent internal consistency with an alpha of 0.89, comparable to, or slightly better than both the CES-D and HAM-D in this sample. This should be explored in future research. The convergent validity, a subtype of construct validity, was supported by the SADS correlating positively with both the CES-D and the HAM-D. The cognitive/somatic factor of the SADS correlated the weakest with the HAM-D. This finding could be due to the somatic features of the HAM-D being constructed as somatic anxiety and tension rather than the vegetative features of depression included in the SADS.

Differing views exist in the literature on whether depression scales that include somatic symptoms should be used to screen for depression in PLWH. Kalichman et al. [36] went so far as to conclude that patients who have experienced symptoms of HIV-related disease should only be screened for depression using measures that do not include somatic symptoms. Others have emphasized the importance of including somatic symptoms when assessing depression in sub-Saharan Africa as they found somatic complaints to feature highly in patients’ experiences of depression [33, 37]. Our own experience, as described previously [18] is that participants in South Africa often report somatic complaints, rather than cognitive, affective, and behavioral complaints, as they perceive these as medically relevant. However, when probed about their experiences of clinical depression, they describe a variety of cognitive, affective, and behavioral idioms of distress. Only two somatic items were included in the SADS, difficulties with sleep and changes in appetite; both these items were highly endorsed by participants, 94% and 82% respectively. While we cannot say these symptoms are exclusively attributed to depression in this sample, it does indicate the pervasiveness of these symptoms in PLWH with clinical depression in South Africa.

There are limitations to the current study. Of note, it is possible that solutions with a larger number of factors may fit the data very well. However, the 4-factor solution was chosen to balance empirical fit, theoretical fit, and generalizability. Because the measures were only administered to depressed participants, we were unable to examine the ability of the scale to differentiate between depressed and non-depressed participants. Receiver Operating Characteristic (ROC) analyses are needed to determine whether the SADS is able to accurately discriminate between depressed and non-depressed individuals, using a diagnostic interview as a gold standard. Research is recommended with depressed and non-depressed participants to identify cut-off scores on the SADS that indicate clinically significant depression. Confirmatory factor analyses would also be useful to further support the factor structure identified in this study.

Conclusions

This study examined the factor structure, convergent validity and internal reliability of the SADS, a locally developed measure of clinical depression in PLWH, as a first step in establishing the psychometric properties of this scale. The SADS appears to be a promising measure of clinical depression in this context. By including local idioms of distress, the scale may be more sensitive to the identification of clinical depression in this population. If found to have high sensitivity, specificity and positive predictive value, the SADS could be a feasible option for use in HIV care in South Africa where routine screening for depression has yet to be implemented.