The search yielded 1,747 items, of which 1647 were excluded after abstract screening. Of the remaining 100 items, the full text via institutional (London School of Economics) access was only available for 63 studies, of which a further 11 were excluded because the full-text was not in English. Where full-text institutional access was not available, we used secondary databases (e.g.: Google Scholar) to try to retrieve full text of the remaining 37 items but none were available through this route. Most non-retrievable items were unpublished working papers with abstracts that were identified by the search, but not available electronically. The remaining 52 items were screened for inclusion on the basis of a review of their full text, after which a total of just 14 studies were selected for inclusion in the mapping. The main reason for exclusion at this stage was that an association between depression-diabetes comorbidity and SES was not sought or diabetes and depression cases were considered as separate diseases in two different populations rather than as a comorbidity (e.g.: diabetic patients with depression or viceversa) in a specific group of people.
Description of included studies
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‐
43] used a control case design to compare diabetic patients with and without depression. Three studies were community-based [
43‐
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‐
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.
Assessing socio-economic status
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‐
42,
48‐
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‐
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].
Studying comorbidities: diabetes and depression
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].
Direct diagnosis of comorbidities and their relationship with SES
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.
Indirect diagnosis of comorbidity and its relationship with SES
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].
Study limitations
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‐
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.