Diabetes causes 4.6 million deaths per year, accounting for 8.2% of global all-cause mortality, and it is estimated that 366 million adults have diabetes 1 . The global mortality burden of diabetes is not evenly distributed, with low and middle income countries carrying a disproportionate burden. It is projected that by 2030 around 82.5% of people with diabetes will live in developing countries 1 . The age distribution of adults with diabetes differs by country.

The occurrence of depression appears to be linked with the occurrence of diabetes. In 1684, Thomas Willis, the physician who first identified glycosuria as a sign of diabetes, suggested that diabetes resulted from ‘sadness or long sorrow and other depressions or disorders’ 2 . Further studies have demonstrated that a comorbid state of depression incrementally worsens health compared with depression alone 3 . According to the latest global burden of disease estimates unipolar depressive disorder are third in the ranking (65.5 mil DALY worldwide of which 26.5 in LICs). Unipolar depressive disorders are set to become the leading disease in 2030 with 6.3% of the overall burden and Diabetes the 10^th place with 2.3% as a percentage of the overall DALYs 4 .

Comorbidity has various definitions and previous literature has highlighted the difficulty of defining it but in general, in medicine, it is usually considered as the presence of one or more disorders (in addition to a primary disease or disorder), or also the effect of such additional disorders or diseases 5 . In this study we look at the co-presence of diabetes and depression regardless of whether diabetes or depression is the primary disorder. The identification of co-morbidities is fundamental in order to understand whether the primary disorder or disease might either cause or affect the secondary one but also to understand any association between the two.

Studies have scrutinized the association of diabetes with depression and the bidirectional nature of this relationship; considering that depression may occur as a consequence of having diabetes, but may also be a risk factor for the onset of type 2 diabetes ^a 6 7 8 . One study showed how there is a higher risk of mood and anxiety disorders among individuals with diabetes relative to those without, with an odds ratio for depression of 1.38 (95% CI 1.14-1.66) after adjusting for age and gender 9 . A meta-analysis concluded that the presence of diabetes doubles the odds of comorbid depression and the prevalence of comorbid depression among people with diabetes was 11% 10 . Estimates of depression prevalence among people with diabetes appear to vary by diabetes type and between lower and higher income countries, although the evidence base for lower income countries is much smaller than that for HICs 11 . A study conducted in 2007 which looked at depression worldwide using the WHO World Health Survey (WHS) found that 9.3% of people with depression were also with diabetes 3 .

Two hypotheses attempt to explain the causal pathway between diabetes and depression. One hypothesis asserts that depression precedes type 2 diabetes, with depression occurring as a result of increased counter regulatory hormone release and action, alterations in glucose transport function and increased immuno-inflammatory activation. These physiologic alterations are thought to contribute to insulin resistance and beta islet cell dysfunction, leading to development of type 2 diabetes 12 . The second hypothesis is that depression in patients with both type 1 and type 2 diabetes results from chronic psychosocial stressors of having a chronic medical condition 13 .

Evidence from HICs suggests that depression among people with diabetes is associated with socio-economic status 14 , marital status 15 , and physical activity and chronic somatic diseases 16 . Psychosocial factors may mediate the relation between SES and depression in people with diabetes, including social isolation or social support, coping styles, behaviour and job stress or strain 17 18 . Most studies show inverse social gradients, meaning that the risk is higher for people with lower SES 14 15 16 19 . However, the relationship may vary depending on the social and economic context of the country. In LICs, higher SES may be associated with higher levels of chronic disease risk factors in general 20 while the poor experience a double burden of infectious and chronic diseases according to the protracted polarised model of epidemiological transition 21 .

In addition, the burden of risk factors for depression among people with diabetes in particular has been found to shift towards the less affluent in countries undergoing the epidemiologic transition where the cause of deaths shifts from infectious to non-infectious causes 22 .

The aim of this systematic mapping is to identify the socio-economic factors associated with diabetes and depression as a comorbid condition exclusively in low and middle income countries.

Abstracts were screened and items included if they addressed diabetes (type 2 or type 1 and 2 together to keep the disease type homogeneous), and depression as a comorbidity with diabetes, and SES. Studies were excluded if: they were conducted exclusively in HICs; did not address SES as a risk factor or consequence; the full-text was not in English; the study population was aged below 16 years; and, if there was no abstract.

Our search strategy was deliberately inclusive in order to map the available evidence as widely as possible. Reflecting the different approaches to conceptualising socio-economic status are the indicators used to measure it. Rather than considering just one term such as “poverty”, we considered socio-economic status (SES) in general, including both individual-level (e.g.: education, occupation, income, household assets, place of residence, age, marital status, family type and social support) and household-, family- and community-level characteristics. Debates about conceptualising, defining and operationalizing socio-economic status are well-established and beyond the scope of this systematic mapping 29 30 31 32 33 34 .

Characteristics of communities or neighbourhoods, such as the availability and accessibility of health services, infrastructure deprivation, prevailing attitudes towards health, levels of stress and social support, and environmental conditions, may influence general health outcomes 35 . The socio-economic status of a community may determine the educational, employment, and income opportunities of individuals and may also directly influence the social environment, although it is subject to the ‘ecological fallacy’ of assuming that all individuals in an area have similar characteristics 36 .

Measurement of depression usually relies on structured interviews conducted by a professionally trained clinician or nursing staff using established criteria to identify a cluster of symptoms that may accompany depression (e.g.: loss of interest or pleasure in everyday activities, lack of appetite, fatigue, sleep disturbances, suicidal ideation). Most tools used to identify and rate the severity of depression rely on a multiple choice questionnaire, for example, The Hamilton Rating Scale for Depression (HRSD, HAM-D) the MINI questionnaire, the Beck Depression Inventory (BDI) 37 . It should be noted, however, that in many LICs lack the resources – both human and financial – to detect depression 38 .

Similar problems of detection and diagnosis affect the valid measurement of diabetes in LICs Diabetes can be identified by either clinical blood glucose measurements, although some studies use “self-reported diagnosis” associated with diabetes. Self reports of diagnosis tend to be used in settings where glucose data are unavailable, and cannot distinguish between Type 1 and Type 2 diabetes. It is important to note that self-report might underestimate type 2 diabetes due to undiagnosed cases 11 .

All the 14 included studies were published post-2007, reflecting the nascent interest in depression-diabetes comorbidity in LICs. All of the included studies were cross sectional in design, and we did not identify any longitudinal or intervention studies, meaning that causal inference was not a possibility in our mapping. Just five studies 39 40 41 42 43 used a control case design to compare diabetic patients with and without depression. Three studies were community-based 43 44 45 while the rest where facility-based. It is important to separate facility- and community-based studies in order to take account of bias (Berkson’s bias), as barriers to accessing health care might bias results from facility based studies because they are more likely to include patients: from higher socio-economic strata; with more advanced disease; and, more likely to have another comorbid disorder than those in the general population 46 .

Facility-based studies tended to have relatively small (at most n = 400) sample sizes and were carried out at tertiary hospitals, which in LICs might be more likely to cater for patients from higher socio-economic strata and with more advanced disease. This difference needs to be taken into account when making statements about true population differences, which might account for inconsistencies in association between socio-economic status and diabetes-depression comorbidity across studies. Studies which used control groups for comparison were not always clear about the characteristics of the control groups which could have potentially affected the effect of sample sizes on the overall results.

Of the facility-based (n = 11) studies, 4 studies had a control group, although they differed in control group selection 39 40 41 43 47 . A study from Nigeria recruited diabetic patients as cases and apparently healthy controls without a history of diabetes mellitus from local government staff of three local government areas 40 . A similar approach was used in a study from Iraq, which compared diabetic patients (case) with healthy controls drawn from hospital staff 41 . A study conducted in Turkey recruited diabetic patients and assessed them for presence of depression 39 . Finally, 42 assessed the prevalence of depression in Hispanics of Mexican origin with type 2 diabetes living on both sides of the Texas-Mexico border, recruiting people with type 2 diabetes from clinical settings which included hospitals and physicians’ offices on both sides of the border.

The operationalization and definition of SES in studies included in our mapping are heterogeneous. There is little or no discussion about the validity or reliability of the many difference measures of, and proxies for, SES. Studies that cautiously and robustly identify the presence of diabetes and depression comorbidity tend not to apply the same rigour to SES and its measurement. SES indicators in studies included in our mapping include indicators at a variety of scales, including individual and household.

Employment and education were the most frequently used variables to assess SES. Most studies included education as a proxy of SES 39 40 41 42 48 49 50 51 52 53 . Categorisations varied from literate-illiterate dichotomy 48 52 to years of education 40 50 51 . Employment was considered as a dichotomy (employed vs. unemployed) 40 48 . Three studies used income 40 41 52 , and just one study used place of residence 52 to represent SES.

Finally, three studies used composite indicators of SES 44 53 54 55 . For example, a study from Syria assigned a score for SES based on work status, number of earning members within a household, household income, education level, item ownership and household density (number of individuals living in the household divided by the number of rooms) 44 .

No study sought a causal relationship between SES and diabetes-depression comorbidity. The majority of studies considered the risk of, and risk factors for, depression in diabetic patients. Two community-based studies addressed diabetes and depression as a comorbidity – hereafter referred to as “direct diagnosis of comorbidity” 44 45 . The Kilzieh study 44 assessed the comorbidity of depression with other chronic diseases in a single Syrian city, using two stage, stratified cluster sampling, with a sample size of 2038. The second study, from China, was community based and conducted among people with type 2 diabetes (n = 148) and assessed the association between diabetes and depression comorbidity with SES 45 .

The remaining studies looked at depression risk in patients with diabetes, hereafter referred to as “indirect diagnosis of comorbidity” because the comorbidity was assessed indirectly by considering the patient’s risk of depression. A notable finding, which helps to explain the lack of studies in the area of diabetes and depression comorbidity, is that in studies conducted at geriatric or diabetic clinics where patients came for treatment of chronic medical conditions, patients were often diagnosed with psychiatric comorbidity only as a result of going to the clinic. This suggests that there is a substantial burden of undiagnosed psychiatric disorders, including depression 48 50 51 53 54 56 .

The study by Kilzieh 44 in Syria showed that depression comorbidity with any chronic disease decreased with higher SES (middle vs. low: OR = 0.41, 95% CI:0.22-0.78; high vs. low: OR = 0.52, 95% CI:0). An increase in comorbid depression in women with lower SES underlines the higher vulnerability of women to adverse mental health effects of lower SES. This relationship was not, however, confirmed in the relationship with education where a significant increase in depression comorbidity was reported in those with 1–9 years of education, which, according to the authors, may reflect ascertainment bias. That is, more educated individuals are more likely to seek medical care and consequently to be diagnosed with depression and chronic disease. This study also considered other proxies for SES, including the community-level proxy of place of residence, and found depression to be associated with disadvantaged neighbourhoods or “informal zones” (OR = 0.22, 95% CI:0.06-0.80) in the Kilzieh study 44 . Informal zones are areas in which houses were built without government approval, reflecting disadvantaged status.

Unemployment was significantly associated with depression in diabetic patients in the study by Yang 45 . At household levels, those with low income, less wealthy or those with fewer household assets were more likely to be depressed 44 . Finally, lower levels of social support were significantly associated with depression in the study by Yang 45 using a multidimensional scale of perceived social support.

Socio-economic indicators at the individual level (e.g.: unemployment, education) were associated with depression in these studies that indirectly diagnosed depression-diabetes comorbidity 22 45 57 .

A study from China found no significant difference in depressive symptoms between rural and urban dwellers (p = 0.129) 49 . This study was conducted in one rural county and two urban districts in two geographical locations of Beijing and Shanghai, which might account for the lack of an observed statistical difference because of the predominance of an urban population. However, this study did note a statistically significant association for women (but not men) between depressive symptoms and insulin resistance (OR 1.58, CI 1.14-2.18; P = 0.006) after adjusting for geographic location, residential region, age, educational level, smoking and drinking status, physical activity level, BMI category and comorbidity. By contrast, no significant association between depression comorbidity with place of residence was found in studies from Nigeria (p = 0.80) 48 and India (OR 0.76, CI 0.44-1.34, p = 0.35) 52 . Both studies were carried out in tertiary health care facilities, meaning that their samples tended to involve complicated cases, not necessarily representative of a true population difference.

Mansour et al.’s 41 Iraq study derived an indicator for “social class” based on an aggregate score of education, occupation and income. The control group had a higher social class than patients with diabetes, which could be explained by the recruitment of controls from the medical staff of the hospital.

Monthly income for diabetic patients was significantly and negatively correlated with depression scores in a study from Nigeria 40 (Pearson coefficient(r) = −0.207, p = 0.003). Similar findings are found from research in Iran which reported that depressed patients were poorer (64.1% vs. 52.4% had a low income level, p < 0.05) 53 . A decline in economic condition was significantly associated with depression among people with diabetes in a study from China using multiple regression analysis with adjustment for sex, age, marital status, educational level, income, employment, years since diagnosis of disease, and presence or absence of diabetes complications. (Beta 0.482, t value 2.059, p = 0.041 and partial correlation 0.132.) 54 55 . By contrast, in India depression comorbidity was significantly associated with high monthly income (OR 1.22, CI 1.03-1.41, P < 0.001) 52 . Finally, no significant association with monthly income and depression comorbidity was found in the Agbir study (P = 0.110) from Nigeria 48 .

Drawing conclusions about the relationship between education and depression-diabetes comorbidity is difficult because of the highly heterogeneous ways in which education was conceptualised across the different studies, in part reflecting different education systems between countries. The majority of studies found no significant association between the depression comorbidity and education level for a range of countries including Nigeria (Chi-square 1.229, df = 1, P = 0.268) 48 ; India 52 (literate vs. illiterate, OR 1.12; CI 0.93-1.46, P = 0.07). Studies that compared depressive and non-depressive groups also showed no significant difference in Nigeria (Chi-square = 0.705, P = 0.343) 40 and Turkey (t = 1.31, P > 0.05) 39 and (r = −0.07, P = 0.49) 43 .

The remaining four studies all suggest that lower education is associated with depression among people with diabetes, including: education up to secondary was significantly associated with depression among people with diabetes (OR 2.39; CI 1.09-5.21, P = 0.029) 42 ; and, people with diabetes who had <5 years of education were more likely to be depressed (OR 3.26, CI 1.57-6.80, p = 0.0004) 50 . Diabetic patients in Thailand with less than 12 years education were significantly more likely to be depressed (OR 2.33, 1.28-4.29, p < 0.01) 51 . Finally, depressed patients were less educated than non-depressed patients in an Iranian study (OR 4.20 CI 1.10-5.60; p < 0.0001) 53 .

Considering the relationship between employment, as a proxy for SES, and the comorbidity the findings are equally mixed. Three out of five studies found no significant association between depression-diabetes comorbidity and employment, including studies from Malaysia (Chi-square = 0.429, p = 0.512) 47 and Nigeria (Chi square = 0.04, df = 1, P = 0.84) 48 and (Chi-square = 0.087, P = 0.445) 40 . Of the two studies, both from Mexico, which did find a relationship between employment status and depression-diabetes comorbidity both report the same direction: lower employment status was significantly associated with depression among people with diabetes 42 50 .

Poorer levels of social (including family) support were significantly associated with depression among people with diabetes studies from Thailand (OR 4.10, CI 1.78-9.53, p < 0.01) 51 and Mexico (OR 2.79, CI 1.02-7.82, p = 0.02) 50 . Depressive symptoms were negatively correlated with subjective social support in China (Beta −0.162, t = −3.635, p = <0.000 and partial correlation −0.228) 54 55 .

There are limitations of our search strategy that have implications for the scope of included evidence. Firstly, we only included items with English abstracts, meaning that we are likely to have excluded from the mapping substantial research evidence which may be of relevance for this topic. We did, however, review the type and content of these non-English items on the basis of their title and abstract only. Among these studies, only three studies, all from Latin America, appeared to be relevant to our study. A study from Brazil concludes that among people with diabetes, higher education, low family and individual income predispose to symptoms of depression 58 . A study in Mexico concludes that among people with type 2 diabetes, significant differences between depressed and non-depressed participants were found in schooling, marriage type and occupation 59 . A study assessing trends in social and demographic inequalities in the prevalence of chronic diseases including diabetes and depression in Brazil 60 revealed a higher presence of chronic diseases in low socio-economic strata. The remaining non-English studies did not provide sufficient evidence in their abstract for us to describe them here 61 62 63 64 65 66 . Secondly, we excluded studies that consider diabetes and depression in low and middle income countries that did not explicitly include reference to SES or one of its proxies. Therefore, there are themes that are potentially linked with the pathways between SES and diabetes and depression that we have not explored in this mapping, which may further our understanding of the relationship. A third limitation is methodological. As a systematic mapping, rather than a systematic review, we have not assessed the quality of the included studies. This means that the evidence base that we have identified is not necessarily all of high quality. However, as a systematic mapping we set out to describe the available research in order to show the gaps in the literature and, by taking an inclusive approach to our search, we have identified studies of research and policy relevance. Fourth, studies that failed to find any significant relationship between depression and diabetes as a comorbidity and SES, might not be published, introducing the possibility of publication bias. However, this possibility is diminished by the fact that we did find, but did not include studies in which diabetes and depression comorbidity was not the principle focus of interest of many of the included studies and that depression was reported as the commonest psychiatric disorder while diabetes was one of the many chronic disorders in the populations under study 56 . Fifth, assessment of SES is heterogeneous, limiting statistical comparability. Sixth, we included studies that used self-reports of diabetes, meaning there is no differentiation between type 1 and type 2 diabetes. There are further limitations linked to our analysis which are due to the quality and quantity of the papers found. Given the heterogeneity of the SES indicators and the small number of studies found we could not perform either a meta-analysis or a causal chain analysis. Finally, our inability to electronically retrieve 37 full text items, identified on the basis of our abstract search, means that we were unable to review some potentially relevant items. The majority of these items were non-peer-reviewed items such as unpublished working papers. The inability to retrieve some items means that we have been unable to include some potentially relevant material in our mapping, limiting its breadth.