Introduction

Investigation of the association between socio-economic status (SES) and mental disorders has been a research focus since the 1930s [1–6]. Results have confirmed a relationship between SES and various mental disorders (e.g., depression) in Western societies, leading to hypotheses about “social stress” or “social causation” [7–9].

Depression commonly has a dynamic relationship with SES. There is a negative association at an individual level when surveys of SES have primarily used income as a variable [10–12]; the lower the income, the greater the risk of depression [13–16]. The mechanism may be explained absolutely by one’s ability to secure basic needs (e.g., in purchasing goods and services), including health, or relatively by implying social position and emotional satisfaction in terms of SES producing happiness [11,17,18]. It may also imply the modality of higher income (compared with other modes of SES) having a substantial impact on psychological problems such as depression, mainly at low SES levels [5,11,12,19,20].

However, SES factors such as education and employment have frequently been found associated with depression [9,11,21,22], as have other SES factors [23,24], as well as sociodemographic factors [11,25–29]. Those socio-economic-demographic (SED) factors could be independently [5,30,31] or dependently [3,13,20,22,32–34]—directly or indirectly [12]—associated with depression. There may also be negative societal perception or adverse socio-economic impacts if a person consults with a mental health professional or is hospitalized for a mental disorder in a society that stigmatizes such disorders. Additionally, those with low SES in developing countries may not be able afford medical treatment fees.

Depression is the seventh most common cause of disability in Indonesia, with a high prevalence of 6.1% [35–37]. To our knowledge, while some studies have investigated SED factors related to depression (commonly these are community-based studies) [20,26], there are no reviews of those factors with regard to depression among hospital outpatients (hospital-based study) in Indonesia. Since the implementation of universal health insurance in Indonesia (named Badan Penyelenggara Jaminan Sosial), access to medical facilities has improved for people of low socioeconomic status. However, there may be risks of lost opportunity cost from undergoing medical treatment (e.g., because of time spent for hospitalization or because of prejudice toward patients with depression). It is therefore important to explore characteristics of patients at hospitals.

Materials and methods

Data analysis

Data analyses comprised (1) multiple imputation methods for handling missing values, (2) univariate and bivariate analyses of the variables, and (3) multivariable model analyses.

Missing data and refusals to answer are unavoidable; however, analyses without replacing missing values (i.e., excluding participants with missing values) yield misleading conclusions. Several statistical approaches have been proposed to counteract this [38]. Table 1 shows the proportions of missing values for the SED variables. The greatest amount of missing values (39.06%) was observed for individual cash income. Initially, we performed logical replacement by replacing missing cash income with 0 if the participant, for occupation, answered “unemployed” or answered “housewife” and worked <1 hour per day. This slightly reduced the proportion to 30.94%.

Table 1

Variable and missing values.

Variable	% Missing Values
Income (annual)a	30.94b
Occupation	7.19
Hours worked per day	6.56
Age (years)	3.12
Cohabitating family members	0.62
Education	0.31

^a The high missing value percentage for income related to occupation, as many who reported being unemployed or a housewife did not want to provide their income: 94.4% and 58.9%, respectively. Missing values indicated missing at random.

^b This figure is after inputting values for missing data with logical consideration (i.e., missing income replaced with 0 for unemployed and for housewives who worked <1 hour per day). Before inputting values, this was 39.06%.

We then conducted multiple imputation analysis as follows. We regarded all missing values as neither caused by complete randomness (missing completely at random) nor by systematic bias (missing not at random), though they can be incorporated using partially observed variables (missing at random [MAR]) [39]. The multiple imputation analyses for MAR were based on a Bayesian approach; the overall association among all SED variables was simulated from the observed values through multiple imputation processes by using a fully conditional specification method [40,41]. For example, we replaced missing cash income values for each occupation with random values that were generated with the same probabilistic distribution of the observed cash income for the respective occupation at each imputation. We replaced the missing values for occupation with random values in accordance with the same probabilistic distribution of the observed occupation. We used fraction of missing information (FMI) and relative efficiency (RE) score to indicate the output of five imputations was appropriate. Scores closest to 0 for FMI and 1 for RE indicated the best output. In this study, we mainly analyzed the average of five imputations as a full set of data, while for the purpose of transparency, we also showed the original data (with missing values) and the result of each imputation (Table 2).

Table 2

Variables, including original data with missing data and after multiple imputations.

Variable	Original Data	Imputations					Fraction of Missing Information	Relative Efficiency
		1	2	3	4	5
Income (annual, IDR)
Mean	30,056,108	30,435,411	30,448,317	30,257,215	30,111,008	30,025,365	0.019	0.996
Standard error of the mean	1,939,228	1,532,283	1,504,881	1,480,188	1,479,880	1,461,166
Occupation (%)
Housewife	30.3	29.7	30.3	30.0	30.0	29.4	0.012	0.998
Retired	4.4	4.7	4.1	4.7	4.1	4.4
Civil servant	11.4	11.6	11.6	11.3	10.9	11.6
Private sector	47.8	47.8	47.8	48.1	48.8	48.4
Unemployed	6.1	6.3	6.3	5.9	6.6	6.3
Hours worked per day (%)
<1	8.7	8.1	9.4	8.4	9.1	9.4	0.042	0.992
1–2	4.7	5.6	4.4	6.3	5.6	5.0
2–4	7.4	8.4	7.8	7.8	7.2	7.5
4–6	20.4	20.0	20.3	20.3	20.6	21.3
6–8	27.8	26.6	28.4	26.6	26.6	27.2
8–10	20.4	20.3	19.4	20.3	19.7	19.4
>10	10.7	10.9	10.3	10.3	11.3	10.3
Age (years)
Mean	42.78	42.58	42.81	42.69	42.79	42.78	0.020	0.996
Standard error of the mean	0.756	0.739	0.737	0.750	0.735	0.736
Cohabitating family members
Mean	4.57	4.58	4.58	4.57	4.54	4.55	0.012	0.998
Standard error of mean	0.146	0.145	0.145	0.145	0.146	0.146
Education (%)
Elementary school or lower	15.7	15.6	15.6	15.6	15.6	15.6	0.003	0.999
Junior high school	15.4	15.3	15.3	15.3	15.3	15.3
Senior high school	27.3	27.2	27.2	27.2	27.5	27.2
Diploma	7.2	7.5	7.2	7.5	7.2	7.2
Graduate	30.4	30.3	30.3	30.3	30.3	30.3
Postgraduate	4.1	4.1	4.4	4.1	4.1	4.4

IDR, Indonesian rupiah.

This output showed FMI and RE score closest to 0 and 1, respectively, for each variable

A maximum of 15 iterations was found to be best after running at a range of 10–50 iterations.

For classifying ethnicities, the major ethnic groups in South Sulawesi Province (Bugis, Makassar, and Toraja) were first defined, and mixed ethnicity included any combination of the three primary groups. Other ethnic groups comprised an additional, “others,” group. We converted all ordinal and continuous data (e.g., education, income) to nominal data. Age and hours worked per day were classified for each quartile.

We used a chi-square (χ²) test for detecting raw associations between the case or control and each variable; continuity correction was applied in the case of 2×2 cross-comparison tables. We also measured the strength of association symmetrically using Phi (φ) and Cramer’s (ν) test, and directionally using Goodman and Kruskal’s lambda (λ) or Goodman and Kruskal’s tau (τ) test as an alternative for proportional reduction in error (PRE) detection.

In further investigation, we used the Mantel–Haenszel method to measure the estimated odds ratio (OR) toward the magnetic SED variable. We compared pooled crude OR with original and complete case analysis (CCA) for greater precision [42]. Finally, we performed multiple logistic regression analysis including all variables in one model to confirm a variable’s effect after adjusting for all other variables.

All data analysis was conducted using IBM SPSS for Windows, Version 26.0 (IBM Corp., Armonk, NY, USA).

Results

First, we investigated the sociodemographic characteristics of the case group (patients with depression) compared with those of the control group (Table 3) and (S1 Appendix). No significant between-groups difference was found in age composition or sex. The ethnic composition differed between the groups especially in that the case group had fewer people of mixed ethnicity and more “others.” Among the socio-economic characteristics, only the income level differed between the groups, showing more with “high” income group and fewer with “middle” income for the case group. The strength of symmetric association for ethnicity or income for depression was >0.20. Directionally, income contributed to PRE for depression at 18.1%.

Table 3

Socio-economic-demographic characteristics of participants.

Item	Casea (n = 160)	Controla (n = 160)	χ2b	φ/νc	λ/τd
Socio-demographic characteristic
Age (years)
<32	34.6 (21.6)	43 (26.9)	0.089	-	-
32–42	36.8 (23.0)	51 (31.9)
43–52	44 (27.5)	33 (20.6)
>52	44.6 (27.9)	33 (20.6)
Sex
Male	65 (40.6)	54 (33.8)	0.247	-	-
Female	95 (59.4)	106 (66.3)
Ethnicity
Bugis	57 (35.6)	58 (36.3)	<0.001	0.25	0.063
Makassar	37 (23.1)	47 (29.4)
Toraja	13 (8.1)	8 (5.0)
Mixed	12 (7.5)	30 (18.8)
Others	41 (25.6)	17 (10.6)
Socio-economic characteristic
Income (annual)
High	63 (39.4)	43.8 (27.4)	0.002	0.20	0.181
Middle	39 (24.4)	68 (42.5)
Low	58 (36.3)	48.2 (30.1)
Education levele
High	64.8 (40.5)	69 (43.1)	0.164	-	-
Middle	75.2 (47.0)	61 (38.1)
Low	20 (12.5)	30 (18.8)
Occupation
Housewife	12.6 (7.9)	7.4 (4.6)	0.156	-	-
Civil servant	46.8 (29.3)	48.6 (30.4)
Retired	11 (6.9)	3 (1.9)
Private sector	17.4 (10.9)	19 (11.9)
Unemployed	72.2 (45.1)	82 (51.2)
Hours worked per day
<6	70 (43.8)	66 (41.3)	0.734	-	-
≥6	90 (56.2)	94 (58.8)
Cohabitating family members
0 (alone)	7 (4.4)	12 (7.5)	0.096	-	-
1–3	72.8 (45.6)	86 (53.8)
≥4	80 (50.1)	62 (38.8)

^a N (%).

^b Continuity correction was applied in case of 2×2 cross-comparison.

^c Phi, Cramer’s V test.

^d Goodman and Kruskal’s lambda, Goodman and Kruskal’s tau test (%).

^e High (S0, S1, S2, or S3), middle (junior or senior high school), and low (elementary school or lower).

We did not put the value for non-significant tests on φ/ν or λ/τ, α = 0.05. All values that appeared on the χ² test were p-values.

All variables with missing values are presented using pooled frequency after imputations.

Table 4 (and S2 Appendix) shows an association between income and patients with depression. Pooled data (OR = 1.723, [95%CI = 1.076–2.758]) and original and CCA data showed a significant association in high probability of depression for the high-income group.

Table 4

Income and patients with depression (High income vs. lower or middle income).

Income	Depression		p-valuea	Odds Ratio (95% Confidence Interval)
	Case N (%)	Control N (%)
Pooled (n = 320)
High	63 (39.4)	43.8 (27.4)	0.023	1.723 (1.076–2.758)
Middle or low	97 (60.6)	116.2 (72.6)
Original (n = 221)
High	34 (38.2)	33 (25.0)	0.037	1.855 (1.037–3.317)
Middle or low	55 (61.8)	99 (75.0)
Complete case analysis (n = 171)
High	26 (46.4)	32 (27.8)	0.017	2.248 (1.156–4.371)
Middle or low	30 (53.6)	83 (72.2)

^aMantel–Haenszel common odds-ratio estimate.

Table 5 (and S3 Appendix) shows the association between high income and the case group, and other SED variables. The greater proportions of high-income patients with depression (case) were >52 years old, female, of Makassar ethnicity, had middle-level education, were private sector workers, and were cohabitating with four or more family.

Table 5

Proportion of high-income participants in different socio-economic-demographic classes, by case and control groups.

Socio-economic Demographic	Income	Casea	Controla
Age (years)
<32	High	15.2 (43.9)	15 (34.9)
	Middle or low	19.4 (56.1)	28 (65.1)
32–42	High	13.4 (36.4)	14.4 (28.2)
	Middle or low	23.4 (63.6)	36.6 (71.8)
43–52	High	15.4 (35.0)	8.2 (24.8)
	Middle or low	28.6 (65.0)	24.8 (75.2)
>52b	High	19 (42.2)	6.2 (18.8)
	Middle or low	25.6 (57.4)	26.8 (81.2)
Sex
Femaleb	High	34 (35.8)	24 (22.6)
	Middle or low	61 (64.2)	82 (77.4)
Male	High	29 (44.6)	20 (37.0)
	Middle or low	36 (55.4)	34 (63.0)
Ethnicity
Bugis	High	21 (36.8)	17 (29.3)
	Middle or low	36 (63.2)	41 (70.7)
Makassarb	High	15 (40.5)	9 (19.1)
	Middle or low	22 (59.5)	38 (80.9)
Toraja	High	4 (30.8)	5 (62.5)
	Middle or low	9 (69.2)	3 (37.5)
Mixed	High	7 (58.3)	9 (30.0)
	Middle or low	5 (41.7)	21 (70.0)
Others	High	17 (41.5)	4 (23.5)
	Middle or low	24 (58.5)	13 (76.5)
Education level
High	High	35.2 (54.3)	34.4 (49.9)
	Middle or low	29.6 (45.7)	34.6 (50.1)
Middleb	High	22.2 (29.5)	6.6 (10.8)
	Middle or low	53 (70.5)	54.4 (89.2)
Low	High	5.6 (28.0)	2.8 (10.8)
	Middle or low	14.4 (72.0)	27.2 (90.7)
Occupation
Housewife	High	8.8 (18.8)	3.6 (7.4)
	Middle or low	38 (81.2)	45 (92.6)
Civil Servant	High	10 (57.5)	9 (47.4)
	Middle or low	7.4 (42.5)	10 (52.6)
Retired	High	7.4 (67.3)	3 (100)
	Middle or low	3.6 (32.7)	0 (0)
Private sectorb	High	36.2 (50.1)	28 (34.1)
	Middle or low	36 (49.9)	54 (65.9)
Unemployed	High	0.6 (4.8)	0.2 (2.7)
	Middle or low	12 (95.2)	7.2 (97.3)
Hours worked per day
<6	High	22 (31.4)	13 (19.7)
	Middle or low	48 (68.6)	53 (80.3)
≥6	High	41 (45.6)	30.8 (32.8)
	Middle or low	49 (54.4)	63.2 (67.2)
Cohabitating family members
0	High	2.4 (34.3)	3.2 (26.7)
	Middle or low	4.6 (65.7)	8.8 (73.3)
1–3	High	29.6 (40.7)	27 (31.4)
	Middle or low	43.2 (59.3)	59 (68.6)
≥4b	High	30.8 (38.5)	13.6 (21.9)
	Middle or low	49.2 (61.5)	48.4 (78.1)

^a N (%)

^b p<0.05 chi-square test

All variables with missing data are presented using pooled frequency after imputations.

Table 6 (and S4 Appendix) shows the results of the multiple logistic regression analyses. Among all SED variables, high income (OR = 2.088 [95%CI = 1.178–3.700]), middle-level education (1.688 [1.042–2.734]), and cohabitating with four or more family members (1.632 [1.025–2.597]) were significant predictors of depression.

Table 6

Multiple logistic regression analysis of socio-economic-demographic determinants related to depression.

Socio-economic Demographic	Coefficient	Standard Error	p-value	Odds Ratio	95% Confidence Interval
Age (>52 years)	0.381	0.289	0.187	1.464	0.831–2.579
Sex (female)	−0.243	0.255	0.340	0.784	0.476–1.293
Ethnicity (Makassar)	−0.331	0.273	0.225	0.718	0.421–1.226
Education (middle-level)	0.524	0.246	0.033	1.688	1.042–2.734
Income (high)	0.736	0.289	0.012	2.088	1.178–3.700
Occupation (private sector)	−0.289	0.268	0.281	0.749	0.443–1.267
Hours worked per day (≥6)	−0.086	0.266	0.747	0.918	0.545–1.547
Cohabitating family members (≥4)	0.490	0.237	0.039	1.632	1.025–2.597

All variables with missing data are presented using pooled frequency after imputations.

Discussion

The results suggest that income has a substantial impact on depression [5,10,12]. High-income status was related to a greater risk of depression then was lower or middle income, even after adjusting with other SEDs. This is despite many studies having suggested that high income does not correlate with a high risk for depression, such as in it facilitating the ability to manage and cope with financial stress and to foster individual and social well-being [5,10–12,18–20,22,43–46].

This study had limitations in the frequency of missing values for cash income. Income has typically been a limitation to competing questionnaires, but exclusion either of participants with the missing values in the data or the income variable may cause substantial bias [27,47]; accordingly, multiple imputation methods for missing values have been recommended [38]. The results in the present study were consistent among the original data and each imputation; therefore, we can reasonably deem that the results from the pooled data of the five imputations were reliable. There are various reasons participants were reluctant and even embarrassed to inform of their income [27,47]. We sought to avert that, though we did not entirely succeed. Some studies have shown similar experiences and it was validated by Mossakowski [9].

Another limitation was the difficulty sampling control participants to match the case participants. We initially attempted to recruit control participants with identical neighborhood characteristics to the case sample. However, most individuals declined to participate because they lived too far away from the hospital. As an alternative approach, we recruited participants by visiting communities close to the hospital. Thus, the control participants may have had different neighborhood characteristics to the case participants. Although there were no significant differences in age or sex between the groups, the difference in the ethnicity may owe to sampling bias. Ethnicity only differed for mixed ethnicity or “others,” but not among the major ethnic groups (Bugis, Makassar, and Toraja). Our analyses regarded ethnicity as a confounding factor in the final model (Table 6), rather than the explanatory variable. Despite the study’s potential limitations, we deemed the final result showing effects of income, education, and the number of cohabitating family members on patients with depression to be reliable.

Since implementing a national social security system (Sistem Jaminan Sosial Nasional for health care in 2014 in Indonesia, all citizens acquired de facto no-cost primary care consultation at hospitals. However, accessing hospitals involves other costs, such as transportation [48–51]. There is also an opportunity cost that citizens lose by visiting a hospital; e.g., day laborers will lose 1 day of income for every day they visit a hospital [52] and ordinary employees with lower earned income hesitated to take leave for mental disorders [53].

All the above factors might have led to the indication that patients with depression were more likely to have higher income. These findings raise a critical issue about how lower income people experience and express depression and common care-seeking pathways. Although few studies have examined this issue, symptoms of depression are often not recognized as a treatable medical condition among individuals with lower socioeconomic status and/or lower levels of education in Indonesia (Byron Good, personal communication). Presumably, some individuals in this group experience depression symptoms, but their condition may not be recognized as a treatable medical disease [52]. In addition, depression may be expressed in local cultural and religious terms [54]. Hence, people in such conditions may prefer to receive treatment from traditional or spiritual healers rather than going to the hospital [55]. In some cases, restraints are still used for treating severe depression in this group [56]. The bivariate analyses (Table 5) and multiple regression analyses (Table 6) also revealed that patients with depression most commonly had achieved a middle level of education and/or cohabitated with a larger number of family members. This may indicate that education could increase expectations [33] toward achieving a certain income threshold in the community (a perception of economic position) posing a more substantial goal for patients with depression to achieve, thus creating a complex situation, as ten Kate et al. noted in the context of cultural entitlement [17], while also requiring more family members to bear the expenses and care for these patients. These can be regarded as exclusive characteristics regarding depression among hospital outpatients.

Another possible interpretation involves social causation [9], in that high income itself is related to higher risk of depression [25,57,58]. It should be noted that our income analysis used personal income, which reflects personal social and economic status. Unexpectedly, the current findings revealed that high personal income did not protect individuals from depression. Instead, high personal income may have caused depression independently [59,60], in accord with a phenomenon known as the Easterlin Paradox [61,62]. Previous studies have reported that individuals may have a high income but still experience financial pressure and even hold high levels of debt [9,25] because of loss-aversion, which causes them to experience hardship and lead to depression [63]. Individuals’ inability to reach a certain income threshold within the social groups in which they live may fail to meet exceptionally high standards, which may be a particular problem when compared with what exists in other groups [5,18]. This is also called “relative deprivation” [14]. Hence, the implications of relative income related to the “fear of losing status” may be correlated with the occurrence of depression.

The phenomenon described above occurs more commonly in Western societies [18,25,64]. We argue this may be the effect of an “evolutionary mismatch” [65] or “evolutionary cleft stick” [66] phenomenon via modernity like that which takes place in Western societies (e.g., competition) [67,68] and is already being mirrored in Indonesian society (e.g., Makassar society) [69]. In Indonesia, therefore, there is an association between income and depression, in which high relative income may impact the high risk of depression because of the effects of socio-cultural transformation.

In the discussion above, we critically assessed possible explanations for the correlation between high income and depression, either as an exclusive determinant at the hospital due to the lack of access of lower income people to the hospital, or through relative social causation related to socio-cultural transition. Thus, this study proposes that the government of Indonesia reevaluate the effectiveness of hospital-based service in terms of (more) coverage of service and access for low- and middle-income households (e.g., cost, health insurance, procedures). Future research should examine hospital-related paradigms in society (i.e., lower or middle society) in depth and investigate why socio-cultural transformation causes some high income individuals to be at greater risk of depression.

Conclusions

Cash income appears to be a determinant of depression in outpatients in hospitals in Indonesia. This study offers health policy implications toward enabling better hospital access and service for people with depression in low- or middle-income households, and recommendations for considering high income (and SED) as a determinant for high risk of depression occurrence, due to socio-culture transition.