1. Data and study sample

Data for this study was obtained from KLoSA, an ongoing longitudinal panel survey of community-dwelling elderly persons in Korea. The initial KLoSA survey was launched in 2006, in which information was collected on 10,254 participants from 6,171 households aged ≥45 years at baseline. It was implemented with biennial follow-ups, and the seventh wave had been completed by 2018. By the 7th follow-up survey, the existing panel showed a stable trend of 77.6%, with a total of 6,136 valid samples available for analysis.

This survey includes questionnaires that consist of demographics, family history, health, depression, deaths, etc. Survey of the deceased was conducted by interviewing their family members about pre-death situations. The participants of the KLoSA were selected through multistage stratified probability sampling based on geographical areas. Face-to-face interviews were conducted by trained interviewers using the Computer-Assisted Personal Interviewing methods for all interviewees. All processes were informed and signed agreements from each participant were collected accordingly, while the study was approved by the Institutional Review Board of Statistics, Korea (No. WMCSB 202006-43).

Among the initial 10,254 participants of the KLoSA in 2006, 6,090 participants were excluded due to age (<65 years), along with those who had been diagnosed with stroke (n=34). Participants who measured the 10-item Center for Epidemiological Studies Depression Scale (CES-D 10) only once during follow-up (n=1,293) were also excluded. To investigate the association between depressive symptoms and all-cause mortality among these individuals, we finally included 3,561 (1,483 men and 2,078 women) study subjects at baseline.

2. Measurements

To evaluate depressive symptoms of the subject, CES-D 10 was used in the present study. The CES-D 10, a commonly used assessment tool for evaluating depressive conditions in the elderly population, was proved to be highly correlated with the original 20-item version [13]. The subjects responded to each item of questions in regards to the presence of symptoms during the past week. A one-point scale was adopted for each question. The possible range of the 10-item was 0 to 10. The score of CES-D 10 was divided into four trajectory subgroups of both men and women.

Sociodemographic factors included age, educational level, marital status, and family income. The level of education was classified into four categories: elementary school or less, middle school, high school, and university or above. Marital status was grouped into living with and not living with a spouse. Not living with a spouse was comprised of divorce, bereavement, and separation for a variety of reasons such as long-term accommodation and long stays abroad. Family income was classified based on the annual household income as less than 2,000 (thousand Korean Won), 2,000 to 10,000 (thousand Korean Won), 10,000 to 25,000 (thousand Korean Won), and more than 25,000 (thousand Korean Won).

Health behavioral confounders included smoking status, alcohol consumption, and whether exercise was performed regularly at least once a week (yes or no). Smoking status was divided into groups of non-smokers, former smokers, and current smokers. Former smokers were defined as those who were not smoking at present but had consumed more than five packs (100 cigarettes) in total. Drinkers were classified according to alcohol consumption as non-drinkers and current drinkers. Subjects who drink sometimes or frequently or drank recently were classified as “current drinkers”, with others classified as “non-drinkers”.

Health status was assessed by chronic disease status, self-rated health, and functional status. Subjects with chronic diseases were evaluated for the presence of hypertension, diabetes mellitus (DM), cancer, chronic lung disease, chronic liver disease, and heart disease. Self-rated health was classified into five categories (very good, good, normal, poor, and very bad) as described in the Survey of Health, Aging and Retirement in Europe (SHARE) study, which was subsequently reclassified into three categories (good, moderate, and bad) in the present study.

The physical functional status of the subjects was measured by the seven-item Korean Activities of Daily Living Scale (K-ADL), which assesses a person’s ability to perform activities of daily living without assistance [14]. ADL functioning was assessed through the use of seven questions asking the subject whether a person requires assistance to perform activities including dressing, washing/brushing, bathing/showering, eating, leaving the room, using the toilet, and controlling stool. In the ADL index, the questionnaire assigned a value of “1” for subjects requiring partial or full assistance with the activity or “0” if they did not require any assistance, and all values were summed.

Handgrip strength was also measured as part of the physical function assessment. The grip strength of both hands was measured twice using a handgrip dynamometer (TANITA

Hand Grip Meter Blue 6103; Tanita Co., Tokyo, Japan), and the average was collected for the study.

Body mass index (BMI) was also calculated for the physical function measurement. It was computed as weight in kilograms divided by height in squared meters (kg/m²).

Cognitive function was assessed using the Korean version of the Mini-Mental State Examination, which is frequently implemented to detect the cognitive impairment of older adults in Korea [15].

3. Statistical analysis

Since depression and mortality are known to differ between men and women, we analyzed the baseline characteristics of the participants separately by gender. The baseline characteristics of the study participants were analyzed by chi-square test for categorical variables and by one-way analysis of variance for continuous variables. Values are presented in percentages or means±standard deviation (SD).

The Trajectories were defined with group-based modeling that had been designed to classify a population into subgroups based on the CES-D score. Group-based trajectory modeling was selected using the Traj command of the STATA procedure. With this method, we could identify homogeneous clusters of developmental trajectories within a sample population over time. The two-stage model selection process as suggested by Nagin [16] was conducted in stages. We determined the number of trajectory groups and the best polynomial trajectory function in the first stage, and the Bayesian Information Criteria (BIC) value was calculated in the second stage [17] To evaluate trajectory model fit, we used the average group posterior probability.

The Cox proportional hazard model was utilized to investigate the association between patterns for CES-D 10 score and all-cause mortality. We constructed two Cox proportional hazard regression models to adjust for various confounding factors. Model 1 was adjusted for age, BMI, chronic diseases (hypertension, DM, cancer, chronic lung disease, chronic liver disease, heart disease), and self-rated health. Model 2 was adjusted from model 1 covariates plus the level of education, marital state, smoking, alcohol, family income, regular exercise, ADL, MMSE, and hand grip strength (HGS). The results were expressed as hazard ratio (HR) with 95% confidence interval (CI). All data were analyzed with STATA (Stata/MP 13.1 for Windows; StataCorp LP, Lakeway, TX, USA). A P-value of <0.05 was considered statistically significant.

1. Trajectory model development

This study analyzed four trajectory groups for men and women according to the score of CES-D 10. The number of trajectories was determined from the lowest absolute BIC value as the best-fitting one and the presence of a minimum of 5% participants per trajectory [18]. The 5-group model had the lowest BIC value, but had a small sample size for each group (less than 5%). The BIC value for this model was -22590.95 in men and the best fit among all the other 4-group models (Supplementary Table 1). In women, the lowest absolute BIC value was -33900.39 (Supplementary Table 1). The model fit was evaluated through multiple diagnostics as guided by Nagin [16] and along with the other criteria such as reasonable sample size for each group membership (e.g., group size is reasonably large, >5%), four trajectory groups were identified as a result: the low trajectory (LT), the moderate high trajectory (MHT), the high, but decreasing trajectory (HDT), and high, but increasing trajectory (HIT) group.

The lowest average posterior probability was 0.75 in men and 0.71 in women, which is greater than the recommended value of 0.7. This indicates that the model assigned subjects to different trajectory groups with little ambiguity.

2. Demographic and comorbidity characteristics of trajectory groups

We identified four trajectories of depressive symptoms in both men and women characterized by maintaining a low CES-D 10 score throughout the study (LT: 138 [11.8%] in men; 866 [39.5%] in women), moderately high scores that steadily increased (MHT: 985 [60.1%] in men; 745 [34.8%] in women), high starting scores that steadily increased (HIT: 227 [17.4%] in men; 221 [12.3%] in women) and high starting scores but then decreasing throughout (HDT: 133 [10.8%] in men; 246 [13.4%] in women).

Characteristics of men and women by trajectories of depressive symptoms are presented in Tables 1, 2. Participants in the less depressed group (LT), tended to exercise more regularly, and their socioeconomic state was higher (P<0.001). Physical function was better than the other groups, the percentage of good self-rated health was higher (P<0.001), ADL score was lower (P<0.001), HGS was higher (P<0.001), and baseline cognitive function was more preserved (P<0.001).

Table 1.

Characteristics of male participants according to depressive symptoms trajectory

Table 2.

Characteristics of female participants according to depressive symptoms trajectory

LT in men was more likely to live with a spouse (P<0.001), but this tendency was not observed in women. The percentage of chronic diseases was lower in the LT in women (DM, heart disease, caner), but only DM was lower in the LT in men. LT in women was more educated (P<0.001), and the smoking rate was lower (P<0.001).

Figure 1 shows the change in CES-D 10 of participants during the study period. The HDT in both men and women(Figure 1) had the highest starting score but then decreased over time, whereas the other groups showed gradual increase (MHT, HIT, LT in men) or maintained (LT in women) scores throughout follow up.

Figure 1.

Depressive symptoms trajectories of the participants during study period (A) for men and (B) for women. CES-D 10 score decreased with age in the HDT group in both men (A) and women (B), while it increased in the other groups. Markers represent mean age at study visit. CES-D 10, the 10-item Center for Epidemiological Studies Depression Scale; LT, low trajectory; MHT, moderate high trajectory; HDT, high, but decreasing trajectory; HIT, high, but increasing trajectory.

3. Cox proportional HR of all-cause mortality according to CES-D 10 trajectories in participants

Tables 3, 4 show the association between depressive symptom trajectories and the Cox proportional HR of all-cause mortality in men and women, and the participants were adjusted for confounding variables (age, comorbidities, self-rated health, level of education, marital status, family income, smoking, alcohol, regular exercise, BMI, ADL, HGS, and MMSE). Compared with the LT, the MHT and HIT were associated with an increased mortality risk in men (Table 3), and HIT were associated in women (Table 4). The crude HRs of the MHT and HIT in men were 1.74 (95% CI, 1.22-2.48) and 2.83 (95% CI, 1.93-4.16) respectively, while adjusted HRs were 1.58 (95% CI, 1.10-2.26) and 2.12 (95% CI, 1.43-3.16) compared with the LT. In women, the crude HRs of the HIT compared with those of the LT was 2.02 (95% CI, 1.59-2.55) and the adjusted ratio was 1.62 (95% CI, 1.25-2.10). Among the confounding variables analyzed in model 2, the variables associated with mortality were age, DM, and smoking in both men and women. Figure 2 presents the Kaplan-Meier survival curves for these four groupings of depressive symptoms.

Table 3.

Cox proportional hazard ratio according to depressive symptoms trajectories in men

Table 4.

Cox proportional hazard ratio according to depressive symptoms trajectories in women

Figure 2.

Kaplan-Meier survival curves of cumulative incidence of mortality in men (A) and women (B) by depressive symptoms trajectories. Y-axis represents probability of survival (A, B). LT, low trajectory; MHT, moderate high trajectory; HDT, high, but decreasing trajectory; HIT, high, but increasing trajectory