Study sample and design
Data were obtained from the Korean Longitudinal Study of Aging (KLoSA), a nationwide survey of community-dwelling people at least 45 years of age (in 2006), analyzed using multistage stratified cluster sampling. Those results collected between the first and fourth wave of data collection were used. The KLoSA is conducted by the Korea Labor Institute, which aims to devise and implement effective social and economic policies addressing emerging trends in population aging. Additional waves of data collection are undertaken every even-numbered year. The original KLoSA study population comprised South Korean adults, at least 45 years of age, living in 15 large administrative areas.
In the first baseline survey, conducted in 2006, 10,254 individuals among 6,171 households (1.7 per household) were interviewed using the Computer-Assisted Personal Interviewing method. A follow-up survey in 2008 comprised 8,688 subjects, who represented 86.6% of the original panel. The third survey in 2010 used 7,920 subjects, who represented 80.3% of the original panel. Finally, the fourth survey in 2012 involved 7,486 subjects, who represented 76.2% of the original panel.
We restricted our population to married couples experiencing no change in marital status during the previous 8 years and of these participants, we excluded 4,209 subjects that one of them (couples) did not respond to survey in 2006.
We additionally excluded 75 subjects who did not provide the required information.
Our final analysis included 2,881 couples (i.e., 2,881 households) in whom both partners were at least 45 years of age during the 2006 baseline survey; 3,033 couples in 2008; 2,772 couples in 2010; and 2,711 couples in 2012. The KLoSA represents a national public database (http://www.kli.re.kr/klosa/en/about/introduce.jsp), and open data that not included human material or human data. Therefore we don’t need to approve Institutional Review Board.
Control variables
Four age group categories were used as follows: ≤ 55, 56–65 66–75, and ≥ 75 years. Two employment status categories, “yes” or “no”, were employed; the number of occasions on which couples socialized with friends during the past week was also recorded, according to the following three categories: everyday, sometimes, and never. Economic activity status was divided into two categories: employed and unemployed. Self-rated health was assessed with the question: “How do you usually perceive your health?” The response “very bad” or “bad” indicated “Bad”, and the response “normal”, “good”, or “very good” indicated “Good”, thus dichotomizing the response. Marital satisfaction was assessed with the question: “Are you satisfied with the relationship with your spouse”? The response was ranked from lowest to highest and grouped into three groups (High, Middle, and Low) using the SAS Rank function and the depression status of the partner (spouse effect) were included as covariates in the analyses.
Depressive symptoms – CESD 10
Several depressive symptoms measurement tools exist, such as the Beck Depression Inventory and the Zung Self-Rating Depression Scale. The 20-item Center for Epidemiological Studies Depression Scale (CES-D), first used in the late 1970s [18], is renowned for its reliability and validity in the context of the general population and primary care patients [19].
The 10-item version of the CES-D (CESD-10), based on the work of Andresen et al., was extrapolated from the original 20-item version by applying item–total correlations and eliminating redundant items [20]. CESD-10 score, the 10-item screening tool we used for depressive symptoms consist of 10 points for the 10-item version. The CES-D has proven to be a useful indicator of depression in older adults.
The CESD-10, which has demonstrated good predictive accuracy in comparison to the full-length 20-item version, assesses three factors: depressive effects (“the blues”, “depressed mood”, “fear”, and “loneliness”), somatic retardation (“bothered”, “sleepy”, “go-getting”, and “attentive”) and positive effects (“happy” and “hopeful”). Depressive symptoms were assessed for 7 days prior to the interview. In this study, we treated depressive symptoms as a continuous measure.
Analytical approach and statistics
Analysis of variance (ANOVA) and a generalized linear mixed model were used to investigate the impact of age differences between husbands and wives on depressive symptoms. For all analyses, p ≤ 0.05 was taken to indicate statistical significance for two-tailed tests. All analyses were conducted using the SAS statistical software package (ver 9.2; SAS Institute Inc., Cary, NC, USA).
Mixed effects model (SAS® Proc Mixed)
Mixed model was required in order to handle the unbalanced data with correlated outcomes and missing data. In all mixed models presented, only the intercept was allowed to vary between subjects, and the regression slopes were assumed to be fixed effects; random intercept models were applied to our data.
A repeated-measurement model using Proc mixed procedure was performed for this analysis. Depressive symptoms as a continuous variable was the outcome in all mixed models. Covariates of interest from all subjects were added to the model to determine their effects on the increased depressive symptoms. To determine whether the increased depressive symptoms changed over time, we included time (year) in the model as a categorical covariate; the regression coefficient was used to estimate both the change in increased depressive symptoms and independent variables, annually.