Socioeconomic predictors and consequences of depression among primary care attenders with non-communicable diseases in the Western Cape, South Africa: cohort study within a randomised trial

Depression is a common mental disorder, causing a high level of disease burden. There were an estimated 298 million cases of major depressive disorder worldwide in 2010 [1] and this disorder was ranked the second leading cause of years lived with disability (YLD) [1, 2].

Mental disorders are also an important cause of disease burden in South Africa. The South African Stress and Health (SASH) study indicated a lifetime prevalence of major depression of 9.7 % and a 12 month prevalence of 4.9 % [3].

Approximately 80 % of South Africans are estimated to be dependent on public health sector services [4], which are inadequately equipped to address the high prevalence of mental disorders. There is marked under-treatment of mental disorders in primary care in South Africa. Three quarters of adults with a mental disorder in the SASH study received no treatment in the year of the interview [5]. This treatment gap is consistent with evidence from many low and middle income countries (LMICs) [6].

South Africa is one of the most unequal countries in the world [7], with wide disparities in wealth and health [8]. Income inequality has been shown to be positively associated with mental illness [9–11]. A study among older Americans found those living in counties with higher income inequality were more depressed, independent of their demographic characteristics, socioeconomic status, and physical health [10]. In South Africa, the burden of ill-health has been demonstrated to be greater among lower socio-economic groups [12].

A number of studies in LMICs have shown an association between indicators of poverty and mental disorders [11, 13]. A systematic review of the relationship between poverty and common mental disorders in LMICs found a relatively consistent and strong association between common mental disorders and education, food insecurity, housing, social class, socio-economic status and financial stress; whereas income, employment and consumption were found to be more equivocal [11]. A second systematic review of poverty and common mental disorders in developing countries found most studies showed an association between risk of common mental disorders and low levels of education, and many studies also showed a relationship with other indicators of poverty such as poor housing or low income [13].

Associations have been found between depression and non-communicable diseases. High depression scores have been found to be an independent risk factor for hypertension, and there is evidence for an association between mental disorder and diabetes. In addition, depression has been shown to be associated with poor glycaemic control [14].

Depression has been shown to be associated with more frequent exacerbations in patients with chronic obstructive pulmonary disease (COPD), worse short-term survival, and higher rates of post-exacerbation readmission to hospital. An interaction effect has also been reported between symptoms of depression and death among patients with COPD [15].

The majority of past studies looking at socio-economic associations with depression in LMICs have been cross-sectional, making it difficult to draw conclusions on causality [11]. Two explanations have been proposed for the inverse relationship between psychiatric disorders and socioeconomic status. Social causation postulates that adversity and stress due to conditions of poverty increase the risk of mental illness, whereas social selection/drift postulates that people with mental illness are at increased risk of drifting into or remaining in poverty due to factors such as loss of employment, reduced productivity, stigma and increased health expenditure [16]. Although these causal pathways are complex, evidence suggests social causation may be more important for common mental disorders such as depression, particularly in women, while social selection/drift processes may be more important for schizophrenia [17].

Few studies have addressed predictors of change in depression symptoms, and predictors of change in treatment for depression over time in LMICs. In addition, few studies have explored the health and economic impact of depression over time.

The aims of this cohort study were to investigate the extent to which socioeconomic position and physical illness (hypertension, diabetes, chronic respiratory disease) predict depression symptoms over time among primary care attenders, and the extent to which these factors and health service characteristics predict treatment of depression over time.

A total of 4393 participants were enrolled at baseline, of whom 90.5 % were followed up. Prescription records were available for 4364 (99.3 %) participants at baseline and 4284 (97.5 %) participants at follow up.

Baseline CESD-10 scores had a mean value of 10.8 units (standard deviation (SD) 6.4, median 11, interquartile range 6 to 15). Change in CESD-10 scores had a mean value of 3.1 units (SD 7.0, median 7, interquartile range −1 to 8). Both baseline and change in CESD-10 score had symmetrical bell-shaped distributions, except that the baseline score was truncated at zero. However both distributions were significantly different from Normal according to Stata’s combined skewness and kurtosis tests for Normality. At baseline, CESD-10 scores were positively associated with female sex, chronic respiratory disease, antidepressant use and housing density, and were inversely associated with age, hypertension, diabetes and income.

The secondary analyses mostly confirmed the robustness of the results reported in Table 2, as follows. An equivalent logistic regression model with higher (greater than or equal to ten) versus lower baseline CESD-10 scores as binary outcome found the same variables as in Table 2 to be significant predictors, except that language was not significant (p = 0.360), and employment was (p = 0.013). Housing density was not independently associated with CESD-10 score as a continuous outcome variable (p = 0.148), but was associated with higher CESD-10 score modelled as a binary outcome variable (p = 0.05). Logistic regression with higher CESD-10 score at follow-up as a binary outcome variable, adjusted for baseline CESD-10 score, found age, sex, chronic respiratory disease, education and welfare grant, but not hypertension, to be significant predictors. Greater housing density was independently associated with increasing CESD-10 scores in linear and in logistic regression models.

Baseline CESD-10 score, whether coded as a continuous or binary variable, was not associated with changes in blood pressure control, glycaemic control, respiratory symptom score or ten year risk of death from cardiovascular disease.

Psychiatric referral between baseline and follow-up was more likely in participants with tertiary education or higher income and was less likely in participants who were older, male, Xhosa-speaking or in intervention clinics.

Counselling between baseline and follow-up was more likely in participants with more education or receiving welfare grants, and in clinics that supplied drugs away from the clinics, and was less likely in participants who were older or had hypertension, and in intervention clinics or clinics with a psychiatric nurse.

This study shows that depression symptoms in adults attending primary care clinics in two districts of South Africa, most of whom had common chronic conditions, were strongly and independently associated with several indicators of disadvantaged socioeconomic position. Depression symptoms, as indicated by higher CESD-10 scores at baseline, were independently associated with being less educated and having lower income. CESD-10 scores at follow-up had increased since baseline in participants who were less educated or receiving welfare grants. Level of education was however not associated with baseline CESD-10 score in the crude analysis, being confounded by the other socioeconomic indicators. This is consistent with findings from several other LMICs, where education was less frequently associated with common mental disorders in bivariate analyses than in multivariate analyses [11].

Previous studies, the majority of which have been community based, have similarly demonstrated associations between common mental disorders and socioeconomic factors, including less education [11, 13, 33, 34], low socio-economic status [11] and low income [13, 34].

Our study population comprised patients already using primary care facilities, and therefore relatively easy to reach for diagnosis and treatment of depression. It showed that, at baseline, participants were less likely to have received treatment with antidepressants if they were socially disadvantaged, in particular if they had lower income or less education. However, participants were more likely to receive treatment if they were unemployed. This may be because it is easier for unemployed participants to attend clinics for treatment. In contrast, the SASH study found no significant associations between receiving treatment for mental disorders and income or level of education [35]. At follow-up, clinic characteristics were more important than socio-economic factors in predicting depression treatment, with participants less likely to have received antidepressant medication if they attended less resourced clinics, without a pharmacist or off-site drug delivery, or if they had lower income. Primary care clinics should be adequately staffed and have pharmacists on site but also enable patients to collect their repeated medicines at more convenient locations. Strategies to deal with the shortage of doctors and nurses in the South African public sector, especially in rural areas, have included community service for doctors, monetary incentives, introducing a cadre of mid-level workers such as pharmacists, contracting non-professional health workers to take on various responsibilities such as counselling and adherence support, and introducing innovative clinical guidelines to enable nurses to manage patients who would otherwise be seen by doctors [36]. Nevertheless, our results reflect the effects of variation in patient:staff ratios within a resource-constrained system, and suggest the need to equalise workloads between clinics, with existing resources. Our finding that patients in clinics with a psychiatric nurse were less likely to receive counselling at follow-up is counter-intuitive. It may be that psychiatric nurses are managing patients with more severe psychiatric disease, that is, psychoses mostly treated with drugs, and do not have the time or skills to provide counselling for depression.

Patients who were more depressed at baseline were more likely to receive antidepressant medication subsequently. Causal inference about the cross-sectional association between depression symptoms at baseline and treatment at baseline is not as clear, but it is more plausible that depression led to treatment rather than that treatment led to depression. Women were more likely than men to have higher CESD-10 scores and to receive treatment with antidepressants at baseline and follow-up. These findings are consistent with work from HIV cohorts in Southern Africa which have shown that proportionally more women than men are on antiretroviral therapy [37], and highlights the need to identify barriers to men accessing healthcare [38, 39]. The role of gender in the causation, experience, reporting and care of depression is however an enormous subject which was beyond the scope of this study.

Depression could potentially have affected participants’ physical health through biological mechanisms, or through their health care use, treatment adherence or interpretation of physical symptoms. However, we found that depression symptoms at baseline were not associated with changes in blood pressure control, glycaemic control, respiratory symptom score or ten year risk of death from cardiovascular disease. This differs from studies which have shown a positive association between depression and poor glycaemic control in diabetic patients [14].

Our findings suggest that the association between depression symptoms and socio-economic position is bidirectional. That is, in addition to disadvantaged social position predicting worse depression symptoms at follow-up, participants who had depression symptoms at baseline were more socially disadvantaged at follow-up, showing 25 % higher odds of being unemployed. The bidirectional link between depression symptoms and social disadvantage therefore supports both the social causation and social selection theories. Our findings suggest that, in this study setting, socioeconomic disadvantage is both a cause and a consequence of depression, and may also be a barrier to treatment, with participants less likely to receive treatment if they had a lower income (baseline and follow-up) or less education (baseline).

The study had a number of strengths. The sample size was large, high rates of follow-up were achieved, and a wide range of socio-economic variables were investigated. A key strength of the study was the longitudinal design, which allowed potential causal relationships to be identified, demonstrating that the relationship between socioeconomic position and depression worked in both directions.

There were a number of limitations to the study. The CESD-10 questionnaire was used to identify participants with depression symptoms, but not to confirm the clinical diagnosis of depression. It was originally derived and validated on an older adult population [22] but has subsequently been validated in a younger population on antiretroviral therapy [25]. Participants were only enrolled into the study if they had hypertension, diabetes, chronic respiratory disease or depression symptoms, so the results may not be generalisable to primary care attenders without these conditions. Thirty-four percent of participants in the depression group did not answer the question at baseline on whether they had received counselling in the past year. This was due to an error in the electronic questionnaire that resulted in this question being skipped during several weeks of fieldwork before it was detected and corrected. Socioeconomic factors that could influence depression symptoms that were not measured include food insecurity, poor housing, lack of social support, and disability.

Further research is needed to investigate the relative contributions of both social causation and social selection/drift mechanisms to the well documented association between socio-economic disadvantage and depression in LMICs; to identify what specific intervention strategies are needed to reach vulnerable low socioeconomic populations living with depression; and to evaluate the effectiveness of interventions that are designed to target each of the above mechanisms. Feasible examples might include brief psychological interventions with financial risk protection as part of universal health coverage.

This study provides new evidence from South Africa in support of the bidirectional relationship between poverty and depression. Mental health interventions have been shown to be associated with improved economic outcomes in LMICs [16]. This study reinforces arguments for the expansion of mental health services and improving the prevention, detection and treatment of depression in primary health care settings in South Africa and other LMICs, for clinical and economic reasons. While there is currently an emphasis on integrating communicable and non-communicable chronic disease care in South Africa, we must not lose sight of the importance of ensuring better management and access to mental health care.