Study design and study population
Data from the Korean Social Life, Health, and Aging Project (KSHAP) were used. KSHAP data have been collected across five waves: wave 1 (2011), wave 2 (2012), wave 3 (2014-2015), wave 4 (2015-2016), and wave 5 (2018-2019). Initiated in 2011, the KSHAP was designed to examine the entire population of adults 60 years old or older, and their spouses in Township K in South Korea [38]. Township K is a typical rural community in Korea. KSHAP completed a face-to-face survey with 814 adults at baseline, 710 adults in the second wave, 591 adults in the third wave, 572 adults in the fourth wave, and 506 adults in the fifth wave. Twenty-five participants joined after wave 1, making it a total of 839 people examined during all the study periods.
Our research targeted adults identified in the first, third, fourth, and fifth waves of KSHAP panel data in which the CES-D questionnaire was administered. There was some missing data concerning spousal panel data (n=14), marital relationship (n=4), annual household income (n=4), and network members (none were indicated; n=3). For analysis, 291 couples out of a total of 316 couples were included.
Depressive symptoms
Depressive symptoms were measured with the Center for Epidemiologic Studies Depression (CES-D) Scale [39]. The Korean version of CES-D from Cho and Kim (1993) was utilized. The CES-D consisted of 20 items with a four-point scale ranging from 0 to 3 for each item (0 = rarely or none of the time, 1 = some or little of the time, 2 = moderately or much of the time, 3 = most or almost all the time). Scores ranged from 0 to 60, with high scores indicating greater depressive symptoms. A respondent's depressive symptoms were measured with continuous CES-D scores, and their spouses' depressive symptoms were dichotomized using a cut-off point to screen for an indication of mild depressive symptomatology. The CES-D cut-off score of 16 was used as it was validated by the Diagnostic and Statistical Manual of Mental Disorders [40].
Spousal relationships
Spousal support
Spousal support was identified using the questions from Schuster, Kessler, and Aseltine Jr [41] and assessed by two subscales, a supportive relation and a negative relation. The first item asked, "How often can you open up to your spouse/partner if you need to talk about your worries?" and "How often can you rely on your spouse/partner for help if you have a problem?" The second item was, "How often does your spouse/partner make too many demands on you?" and "How often does your spouse/partner criticize you?" Each question was answered by selecting one of four response options in a scale: never, hardly ever, sometimes, and often. A supportive or negative spousal relation index was respectively standardized in the scale to a mean of 0 and a standard deviation of 1.
Network overlap proportion
To collect social network data, the KSHAP adopted an approach developed by social network researchers, which involve using questions called name generators to permit the respondent to enumerate relevant social network members [42, 43]. The KSHAP asked respondents to list people with whom they had discussed important matters over the last 12 months (up to five) and a spouse, if any (up to six members in total) [38, 44]. This name generator collects tends to extract names of close, frequently met, and long-term contacts who are thought particularly important by the older adults [35]. In addition to the relationships between a respondent and their social network members, the KSHAP surveyed the relationships among their social network members. Respondents indicated the communication frequencies between social network members, including with a spouse. The frequency was reported on a 9-point scale ranging from 0 (have never spoken to each other) to 8 (every day).
To construct a complete Township K network based on the respondents, KSHAP identified the same social network members appearing in more than one social network of different respondents (i.e., duplicates). Based on the respondent's report, the KSHAP collected detailed information about social network members, including their names, genders, ages, and addresses in the smallest administrative unit, the Ri. The KSHAP assumed that two social network members were the same person if they satisfied the following four criteria: 1) at least two out of three Korean characters in their names matched, 2) their gender was the same, 3) their age difference was less than five years, and 4) their addresses were in the same Ri (see Youm et al. [38]). In a global network, the presence of the same network members between respondents can be identified.
Spousal network overlap was measured by two indicators, the self-response social network (ego-centric network) data and the complete network data. The self-response social network was comprised of the communication frequencies between a spouse and each social network member. It was assumed that a spouse and one of the social network members were connected if a respondent reported social network member communication with a spouse at least once a week, and the number of all existing overlapping network members with a spouse was counted. The complete network was constructed by matching a respondent's and spouse's social network members. The number of duplicate network members appearing in both networks was counted as spousal network overlap. The spousal network overlap variable ranged from 0 to 5. To control for the respondent’s network size effect, the proportion of spousal overlap network was used. The number of spousal overlap networks was divided by the social network size. For analysis, the network overlap proportion was categorized as less than 0.5 and 0.5 or more.
Covariates
Socio-demographic covariates included age (in years), education level (below middle school and middle school or higher), and annual household income (below median, and above median). Annual household income was recorded in Korean won then dichotomized at the median. Alcohol consumption was categorized into never, rarely, or once a week or more; three categories were collapsed into two (never or rarely vs. once a week or more). The total number of comorbidities was calculated using hypertension, hyperlipidemia, diabetes, osteoporosis, and cancer; this was a continuous variable ranging from 0 to 5. Cognitive impairment was assessed using the Korean version of the Mini-Mental State Examination for Dementia Screening, and the score ranged from 0 to 30. Cognitive impairment was dichotomized using a cut-off point to screen dementia. This was risk-adjusted following a previous standardization study in Korea [45]. The number of social activity scales was applied as a continuous variable ranging from 0 to 3 (higher values for less social activity), based on the types of activity in which the participants mainly engaged: (1) gathering with relatives or friends, (2) participating in senior center programs, or (3) attending church or religious services.
Data analysis
We used a t-test or chi-square test to compare differences in socio-demographics, depressive symptoms, comorbidities, and marital relationships between husband and wife. Further analysis was processed using a panel data seemingly unrelated regression (SUR) model [46]. The SUR model simultaneously estimates the impact of spousal depressive symptoms on the respondent's depressive symptoms, together with the effects of the respondent's depressive symptoms on their spouse's symptoms. These estimates consider unmeasured and unobserved factors, common to husband and wife, which are likely to affect depressive symptoms. In other words, SUR takes into account the contemporaneous correlation between the residuals of the regression equations for husband and wife's depression symptoms [47]. This improves the efficiency of the regression estimates. The model was developed initially by Biørn [48], and the equation models are summarized in the supplementary material.
This approach allowed for correlation between the unobservable components of spouses' depressive symptoms. All analyses were conducted separately by gender. The main effects of spouses' depressive symptoms and three marital relationships were investigated, and the interaction effects between a spouse's depressive symptoms and three marital relationships were performed, respectively. Statistical analyses were carried out using Stata 15.0.
Ethical approval
Yonsei University’s institutional review board approved this study (YUIRB-2011-012-01 in 2011; 1040917-201505-SB-152-05 in 2014; 7001988-201806-HRBR-244-04 in 2016; 7001988-201812-HR-505-02 in 2018).