DOI: https://doi.org/10.21203/rs.3.rs-2008037/v1
Given the acceleration and deepening of China's aging process and the relatively high prevalence of depressive symptoms in Chinese elderly population, this study aimed to identify the trajectories of depressive symptoms and factors associated with trajectory class to gain a better understanding of the long-term course of depressive symptoms in Chinese elderly population.
Data were obtained from four waves’ survey of China Health and Retirement Longitudinal Study (CHARLS). A total of 3646 participants who aged 60 or older during baseline survey and completed all follow-ups were retained in this study. Depressive symptoms were measured using the 10-item version of the Centre for Epidemiologic Studies Depression Scale (CES-D-10). Growth mixture modelling (GMM) was adopted to identify the trajectory classes of depressive symptoms, and both linear function and quadratic function were considered. Multivariate logistic regression model was performed to calculate adjusted odds ratios (ORs) of associated factors to predict trajectory class of the participants.
The four-class quadratic function model was the best fitting model of the trajectories of depressive symptoms in Chinese elderly population. The four trajectories were labelled increasing (16.70%), decreasing (12.31%), high and stable (7.30%) and low and stable (63.69%) according to their trends. Except low and stable trajectory, other trajectories were almost above the critical line of depressive symptoms. Multivariate logistic regression model suggested that trajectories of chronic depressive symptoms could be predicted by being female, living in village, having lower education level and suffering from chronic diseases.
This study identified four depressive symptoms trajectories in Chinese elderly population and analysed associated factors of trajectory class. These findings can provide references for the prevention and intervention work to reduce chronic course of depressive symptoms in Chinese elderly population.
Depression (Major depressive disorders, MDD) among the elderly population is a worldwide public health issue. According to the World Health Organization (WHO) (WHO, 2021), elderly population have relatively higher prevalence of depression than adults and the number still has a growing trend (1). Late life depression can cause the elderly to suffer greatly and function poorly in the family, resulting in significantly lower quality of life and higher mortality (2, 3). In addition, in contrast to depression in younger adults, late life depression is closely related to a variety of chronic diseases, combined with which may lead to a more serious health condition (4).
Depressive symptoms are major manifestations of depressive disorders. The occurrence of depressive symptoms in the elderly population are associated with various demographic factors, social factors and health condition (5, 6). The significant level of depressive symptoms among the elderly is also highly prevalent and associated with high co-morbidity and increased mortality risk (7). As depressive symptoms are easier and more convenient to diagnose than clinical depression, their measures are used more frequently in mental health primary care and other epidemiological investigations to identify people who are at high risk of depression or more likely to have clinical depression.
However, the level of depressive symptoms is not always stable in a long term, the result of a single measurement may be affected by recent physical conditions or life events. Considering this limitation, longitudinal studies can be adopted to describe the changes of depressive symptoms over time for each individual. This person-centred approach enables to shift the focus from diagnosis of depressive symptoms to their changes over time. Summarizing the changes into several patterns that show relatively similar developmental trajectories can make an addition to the understanding of development and of premorbid course depressive symptoms.
Several studies have reported the trajectories of depressive symptoms over time in elderly population, e.g. Rote et al. (8) identified three patterns of depressive symptoms trajectories in elderly Mexican Americans aged 75 and older, including low throughout, increasing, and high but decreasing. Through separately modelling the trajectories of depressive symptoms in elderly women and men aged 65 and over, Carrière et al. (7) observed two different patterns of trajectories in both genders, one was high with an increasing trend, the other was relatively lower with a decreasing trend. Three trajectory classes of normative, subclinical and chronic symptom were identified in the English older adults aged 65 and over (9). Nevertheless, there still remains relatively less work examining the trajectories for depressive symptoms among the elderly population, and the existing studies of different populations have also yielded inconsistent results (10).
To our knowledge, few study has investigated the changes of depressive symptoms over time in the elderly Chinese population except a study of elderly people living in rural China (11), and a one-year follow-up study among the elderly in Hong Kong (12). To address this gap in knowledge, this study aimed to analyse the latent growth trajectory and its heterogeneity of depressive symptoms among the Chinese elderly population, and identify the factors associated with trajectory belongings.
Data was drawn from four waves’ survey of the China Health and Retirement Longitudinal Study (CHARLS) conducting in 2011, 2013, 2015 and 2018, respectively. CHARLS is an ongoing longitudinal study aims to collect the data of demographic, social, economic, as well as health status of a nationally representative sample of middle-aged and elderly Chinese residents. It adopted a multi-stage stratified probability-proportional-to-size (PPS) sampling method, and the data was collected through face-to-face interviews by trained investigators. The survey data of all waves are publicly available online. At the present study, participants aged 60 or over during the baseline survey and completed all subsequent follow-ups were selected first. Then those with missing values on any of the main demographic variables, or more than two questions in depressive symptoms measurement were unanswered or with the responses of "I don't know" or "I refuse to answer" were removed. After that, a total of 3646 participants were retained as the final sample for analysis.
Depressive symptoms were quantified using the 10-item version of the Centre for Epidemiologic Studies Depression Scale (CES-D-10) (13). The 10 items include three on depressive affect, five on somatic symptoms, and two on positive affect. Each item has four options of “Rarely or none of the time (< 1 day)”, “Some or a little of the time (1 − 2 days)”, “Occasionally or a moderate amount of the time (3 − 4 days)”, and “Most or all of the time (5 − 7 days)”. Participants were asked to recall and choose the most appropriate frequency of specific feelings and behaviours during the past week. Each item was rated on a 3-point Likert scale. The total score of the CES-D-10 ranged from 0 to 30, and participants who scored ≥ 10 points were considered to have significant depressive symptoms (14). The Cronbach’s alpha of the Chinese version of this scale among Chinese elderly population was 0.813 (15).
Growth mixture modelling (GMM)(16) was adopted to identify the trajectory class of depressive symptoms. GMM is a combination of latent growth curve model (LGCM) and latent class model (LCM), which can simultaneously describe the individual trajectory and distinguish its heterogeneity. It is a kind of group-based trajectory modelling that allows individual variation within each trajectory class. At present study, four-time CES-D-10 scores were used for the GMM analysis to identify the growth trajectories and classes of depressive symptoms.
Former studies have demonstrated that basic demographic characteristics including gender, age, marital status, education level, and residence are associated with depressive symptoms among Chinese elderly population (5, 17). Moreover, physical health status, especially the conditions about chronic diseases, is also closely related to depressive symptoms in the elderly (5, 18, 19). Accordingly, we collected demographic information and the number of chronic diseases suffered of the participants to investigate how these variables influence the latent growth trajectories of depressive symptoms over time among the Chinese elderly population.
First, we described the distribution of the participants and calculated four-wave CES-D-10 scores of each subgroups. Then, GMM analysis was performed to select the best fitted model, which determined the number of distinct trajectory classes. It contained two steps, the first was to conduct an unconditional model with one class and calculate the model fit indices, including Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), adjusted Bayesian Information Criterion (aBIC), Entropy and adjusted Lo-Mendell-Rubin likelihood ratio test (aLMR)(20). Among these indices, lower values of AIC, BIC and aBIC indicate a better fitted model, higher Entropy statistic represents higher classification accuracy, and significant result on aLMR suggests that the model with one more trajectory class fits the data better. The second step was to perform successive analyses by adding a trajectory class each time, and compare the model fit indices across models. We considered trajectories fitted with both linear function and quadratic function. All GMMs were specified 200 random sets of starting values to be generated with 10 final stage optimizations. Next, the parameters of the best fitted model, and the distribution of latent growth trajectories of the participants were calculated. Finally, a multivariate logistic regression model was specified to calculate the adjusted odds ratios (ORs) of associated factors to predict the trajectory belongings of the participants. The descriptive analysis and multivariate logistic regression model were performed using IBM SPSS Statistics version 24 and GMM was established using Mplus 8. All missing values were assumed to be missing at random and accounted by using regression interpolation. Statistical significance was set to P < 0.05, and all P values were two-sided.
Table 1 presents the demographic characteristics and four-wave CES-D-10 scores of the participants. Among the 3646 participants, males and females accounted for about 50% each, the average age was 66 years old, and nearly 80% were aged 60–69 years old during the baseline survey. Over 80% of the participants were married and about three quarters lived in village. Uneducated participants accounted for about 30% and nearly half of the participants completed the education no further than elementary level. About three quarters reported that they were diagnosed with at least one type of chronic disease. The average CES-D-10 scores of the four waves were statistically different (P < 0.001), among which the fourth wave’s score was the highest.
Variables | N (%) | CES-D-10 scores (Mean ± SD) | |||
---|---|---|---|---|---|
Wave 1 | Wave 2 | Wave 3 | Wave 4 | ||
Gender | |||||
Male | 1834 (50.30) | 7.74 ± 5.87 | 7.15 ± 5.24 | 7.33 ± 5.97 | 8.13 ± 6.19 |
Female | 1812 (49.70) | 10.07 ± 6.77 | 9.24 ± 6.20 | 9.87 ± 6.98 | 10.61 ± 7.21 |
Age at baseline (years old) | |||||
60–69 | 2788 (76.47) | 8.98 ± 6.45 | 8.24 ± 5.89 | 8.68 ± 6.63 | 9.33 ± 6.83 |
70–79 | 791 (21.69) | 8.51 ± 6.29 | 8.01 ± 5.56 | 8.35 ± 6.54 | 9.42 ± 6.75 |
≥ 80 | 67 (1.84) | 9.84 ± 7.61 | 8.09 ± 6.37 | 7.87 ± 6.80 | 9.87 ± 7.62 |
Marriage status | |||||
Married | 3038 (83.33) | 8.62 ± 6.31 | 7.95 ± 5.66 | 8.42 ± 6.53 | 9.15 ± 6.70 |
Divorced / Widowed / Never married | 608 (16.67) | 10.29 ± 6.92 | 9.39 ± 6.48 | 9.46 ± 6.95 | 10.42 ± 7.33 |
Birthplace | |||||
Rural | 3374 (92.54) | 9.05 ± 6.45 | 8.31 ± 5.85 | 8.75 ± 6.64 | 9.54 ± 6.68 |
Urban | 272 (7.46) | 6.95 ± 6.02 | 6.63 ± 5.38 | 6.65 ± 6.06 | 7.12 ± 5.94 |
Residence | |||||
Main city zone | 440 (12.07) | 6.28 ± 5.24 | 6.27 ± 4.82 | 6.44 ± 5.92 | 6.98 ± 5.94 |
Town | 418 (11.46) | 7.96 ± 6.04 | 7.39 ± 5.71 | 7.62 ± 6.17 | 8.48 ± 6.48 |
Village | 2788 (76.47) | 9.45 ± 6.56 | 8.61 ± 5.92 | 9.08 ± 6.70 | 9.87 ± 6.92 |
Education level | |||||
Illiterate | 1114 (30.55) | 10.25 ± 6.89 | 9.21 ± 6.25 | 10.04 ± 6.89 | 10.79 ± 7.10 |
Elementary school or lower | 1797 (49.29) | 9.02 ± 6.26 | 8.17 ± 5.69 | 8.62 ± 6.53 | 9.45 ± 6.80 |
Middle school | 487 (13.36) | 6.93 ± 5.62 | 6.87 ± 5.14 | 6.43 ± 5.83 | 7.18 ± 5.78 |
High school / Vocational school or greater | 248 (6.80) | 5.78 ± 5.06 | 6.27 ± 5.12 | 6.14 ± 5.51 | 6.55 ± 5.67 |
Number of chronic diseases suffered† | |||||
None | 929 (25.48) | 6.88 ± 5.82 | 6.68 ± 5.21 | 7.02 ± 5.89 | 7.96 ± 6.40 |
One | 1086 (29.79) | 8.23 ± 6.13 | 7.93 ± 5.66 | 8.27 ± 6.46 | 8.84 ± 6.64 |
Two | 796 (21.83) | 9.64 ± 6.48 | 8.67 ± 5.94 | 8.90 ± 6.59 | 9.75 ± 6.70 |
Three or more | 835 (22.90) | 11.30 ± 6.58 | 9.73 ± 6.15 | 10.46 ± 7.12 | 11.22 ± 7.19 |
Total | 3646 (100.00) | 8.90 ± 6.44 | 8.19 ± 5.83 | 8.59 ± 6.62 | 9.36 ± 6.83 |
†A total of 14 types of chronic diseases were asked: 1. Hypertension; 2. Dyslipidemia (elevation of low density lipoprotein, triglycerides (TGs), and total cholesterol, or a low high density lipoprotein level); 3. Diabetes or high blood sugar; 4. Cancer or malignant tumor (excluding minor skin cancers); 5. Chronic lung diseases, such as chronic bronchitis, emphysema (excluding tumors, or cancer); 6. Liver disease (except fatty liver, tumors, and cancer); 7. Heart attack, coronary heart disease, angina, congestive heart failure, or other heart problems; 8. Stroke; 9. Kidney disease (except for tumor or cancer); 10. Stomach or other digestive disease (except for tumor or cancer); 11. Emotional, nervous, or psychiatric problems; 12. Memory-related disease; 13. Arthritis or rheumatism; 14. Asthma. |
Table 2 reports the fit indices of twelve growth mixture models of trajectories of depressive symptoms, including seven linear models (Models 1–7) and five quadratic function models (Models 8–12). First, the linear models were fitted. Three indicators of AIC, BIC and aBIC prompted to choose the model with a larger number of classes, while the indicator of aLMR suggested that Model 7 was not significantly better than Model 6, and the entropy statistic of Model 6 was also better than Model 7. Therefore, Model 6 was considered to be the best fitting linear model. Then the quadratic function models were fitted. Also, a larger number of model classes improved the model fit on indicators of AIC, BIC and aBIC. The aLMR statistics indicated that each model was significantly better than the previous one until the number of model classes increasing to four. Besides, there was little difference in entropy statistics between Model 11 and Model 12. Consequently, Model 11 was selected to be the best fitting quadratic function models. Finally, the model fit indices of Model 6 and Model 11 were compared. Since the entropy statistics of the two models were equal, the final determination was dependent on the comparison results of AIC, BIC and aBIC. Thus, the four-class quadratic function model was the best fitting model of the trajectories of depressive symptoms.
Model | LL | AIC | BIC | aBIC | Entropy | aLMR |
---|---|---|---|---|---|---|
Linear | ||||||
Model 1 | -45583.34 | 91184.68 | 91240.49 | 91211.89 | - | - |
Model 2 | -45341.49 | 90706.98 | 90781.40 | 90743.26 | 0.778 | 464.808*** |
Model 3 | -45263.56 | 90557.18 | 90650.14 | 90602.48 | 0.772 | 149.768 |
Model 4 | -45130.41 | 90296.81 | 90408.43 | 90351.24 | 0.756 | 255.912*** |
Model 5 | -45102.99 | 90247.97 | 90378.20 | 90311.47 | 0.758 | 52.696** |
Model 6 | -45062.56 | 90173.12 | 90321.95 | 90245.69 | 0.762 | 81.000** |
Model 7 | -45034.64 | 90123.29 | 90290.71 | 90204.92 | 0.752 | 63.262 |
Quadratic | ||||||
Model 8 | -45516.99 | 91059.98 | 91140.60 | 91099.29 | - | - |
Model 9 | -45263.95 | 90560.95 | 90666.37 | 90612.35 | 0.755 | 492.032*** |
Model 10 | -45126.22 | 90294.45 | 90424.67 | 90357.94 | 0.777 | 266.383* |
Model 11 | -45008.01 | 90066.01 | 90221.04 | 90141.60 | 0.762 | 229.443* |
Model 12 | -44949.54 | 89957.08 | 90136.92 | 90044.77 | 0.767 | 113.470 |
*P < 0.05; **P < 0.01; ***P < 0.001 | ||||||
LL = Log likelihood, AIC = Akaike Information Criteria, aBIC = Sample-Size Adjusted Bayesian Information Criteria, aLMR = Lo–Mendell–Rubin Adjusted LRT Test, Bold = Selected model. |
Table 3 shows the parameters of quadratic function, slope and intercept for each trajectory of the four-class quadratic function model. Figure 1 displays the estimated means and individual values of the four trajectories. The trends of the trajectories suggest that the four trajectories could be labelled “increasing”, “decreasing”, “high and stable” and “low and stable”. Among the four trajectories, “low and stable” trajectory comprised the majority of the participants (63.69%). The estimated mean CES-D-10 score of elderly in “increasing” trajectory (16.70%) began below the cut-off point of depressive symptoms and increased to above the point at wave 3, by which time the increase of CES-D-10 score accelerated. The “decreasing” trajectory (12.31%) had an initially high CES-D-10 score that above the cut-off point and decreased to near the judgment level at wave 4. The “high and stable” trajectory had the highest initial CES-D-10 score and stabilized at the high level throughout four waves.
Classes | N (%) | Quadratic Function | Slope | Intercept |
---|---|---|---|---|
Class 1: Increasing | 609 (16.70) | 1.660*** | -2.233** | 9.196*** |
Class 2: Decreasing | 449 (12.31) | 0.757* | -4.684*** | 16.628*** |
Class 3: High & stable | 266 (7.30) | 0.532 | -1.068 | 19.164*** |
Class 4: Low & stable | 2322 (63.69) | -0.082 | 0.231 | 5.771*** |
*P < 0.05; **P < 0.01; ***P < 0.001 |
Table 4 presents the results of a multivariate model for the prediction of trajectory belonging. Each of the three trajectories of “increasing”; “decreasing”, “high and stable” was compared to the “low and stable” trajectory. The adjusted ORs showed that females were more likely to experience the other three trajectories of depressive symptoms compare with males. The marital status of divorced, widowed or never married significantly increased the odds of belonging to “decreasing” trajectory (OR = 1.580, 95%CI: 1.211–2.062) and “high and stable” trajectory (OR = 1.626, 95%CI: 1.172–2.256). Participants who lived in town were more likely to be in “decreasing” or “high and stable” trajectories, and living in village resulted in a higher likelihood of belonging to the other three trajectories. Compared with illiterate participants, those with the education level of elementary school or lower were less likely to be in the “high and stable” trajectory (OR = 0.734, 95%CI: 0.543–0.992). Moreover, middle school and higher education level decreased the likelihood of going through the other three trajectories. The participants who suffered from one kind of chronic disease were more likely to be in “high and stable” trajectory, and the more types of chronic diseases suffered, the greater likelihood to be on other trajectories than “low and stable”.
Variables | Odds Ratio (95% CI) (“low and stable” as reference) | ||
---|---|---|---|
increasing | decreasing | high and stable | |
Gender (Male as reference) | |||
Female | 1.731 (1.416–2.117) | 1.704 (1.354–2.146) | 2.798 (2.051–3.817) |
Age (60–69 years old as reference) | |||
70–79 years old | 1.143 (0.566–2.310) | 0.867 (0.665–1.128) | 0.729 (0.514–1.036) |
≥ 80 years old | 1.036 (0.828–1.297) | 1.349 (0.643–2.832) | 1.788 (0.777–4.115) |
Marital status (Married as reference) | |||
Divorced / Widowed / Never married | 1.120 (0.872–1.438) | 1.580 (1.211–2.062) | 1.626 (1.172–2.256) |
Birthplace (Rural as reference) | |||
Urban | 0.872 (0.554–1.371) | 1.073 (0.639–1.804) | 1.331 (0.666–2.660) |
Residence (Main city zone as reference) | |||
Town | 1.351 (0.879–2.077) | 1.850 (1.089–3.143) | 3.091 (1.360–7.030) |
Village | 1.754 (1.211–2.543) | 2.639 (1.648–4.226) | 5.906 (2.769–12.596) |
Education level (Illiterate as reference) | |||
Elementary school or lower | 0.889 (0.717–1.102) | 0.906 (0.710–1.157) | 0.734 (0.543–0.992) |
Middle school | 0.420 (0.287–0.614) | 0.497 (0.326–0.757) | 0.576 (0.342–0.969) |
High school / Vocational school or greater | 0.560 (0.345–0.908) | 0.413 (0.219–0.780) | 0.413 (0.168–1.013) |
Number of chronic diseases suffered (None as reference) | |||
One | 1.253 (0.974–1.612) | 1.279 (0.935–1.749) | 1.826 (1.177–2.835) |
Two | 1.606 (1.230–2.097) | 2.003 (1.458–2.752) | 2.538 (1.616–3.986) |
Three or more | 1.964 (1.500–2.570) | 3.498 (2.573–4.732) | 6.164 (4.056–9.368) |
† The ORs were adjusted for gender, marriage status, residence, education level, number of chronic diseases suffered | |||
OR: odds ratio and CI: confidence intervals |
China has the largest elderly population in the world, and its aging process is still intensifying. Moreover, the prevalence of depressive symptoms among the Chinese elderly population was high according to nation survey reports (21). This study aimed to explore how depressive symptoms evolve in Chinese elderly population and identify the predictors of the trajectory belonging. The findings were expected to provide references for the prevention and early intervention of the depressive symptoms in the elderly.
Based on the longitudinal data of the seven-year follow-up research, we revealed four trajectories of depressive symptoms in Chinese elderly population, including low and stable, increasing, decreasing, and high and stable, accounting for 63.69%, 16.70%, 12.31% and 7.30%, respectively. These heterogeneous patterns were consistent with the findings of Kuchibhatla et al.(22) and Liang(11), in which four similar trajectories were identified in the elderly, with a majority of participants in the low and stable group and less than 10% in the high and stable group. It is worth noting that in the study of Kuchibhatla et al., the trajectories of both increasing and decreasing groups were below the threshold of positive depressive symptoms, while in the present study, participants of the increasing group had relatively higher initial CES-D-10 scores, which were close to the cut-off point of depressive symptoms, and the estimated scores of decreasing group was also constantly above critical line of depressive symptoms, which indicated that although the degree of depressive symptoms showed a decreasing trend, it failed to relieve to negative. Thus, the four trajectories can represent four classes: stable non-onset, onset after gradual deterioration, improved but not recovered, and severe chronic condition.
There were also other studies that reported four trajectory class, but the trajectory patterns were not exactly the same as our findings, e.g., in the study of Hsu et al.(23), instead of a high and stable group, there was a medium group, and in the study of Kuo et al.(24), there was a persistent mild group that replaced the decreasing group. Besides, some studies have revealed six (e.g. (25, 26)) or three (e.g. (27, 28)) trajectory classes. The variation among these research results may be mainly caused by different demography of the population, and the differences in measurement tools of depressive symptoms or follow-up interval may also be the source of variation.
Given that depressive symptoms progress differently among individuals, the exploration of predictors for the trajectory class can provide information on how to stratify the elderly population. Our findings suggest that all the increasing, decreasing, high and stable depressive trajectories could be predicted by being female, living in village, lower education level and comorbidity of chronic diseases. Besides, the decreasing, high and stable trajectories could also be predicted by marital status of divorced, widowed or never married. These results were consistent with the general conclusion of previous studies that explored factors associated with depressive symptoms in the elderly (e.g. (29–31)). In addition, it is noteworthy that the differences in CES-D-10 scores of the four trajectories had already been appeared at baseline. Therefore, it can be inferred that those who got higher CES-D-10 score may be experiencing an accumulating exposure, or demonstrating an increased sensitivity, to the risk factors of depressive symptoms, which prevented them from recovering to non-onset level.
Several limitations of the present study should be considered. First, although depressive symptoms of the same individual were repeatedly measured, the follow-up interval was long, so the between-wave transitory changes of depressive symptoms couldn’t be monitored. Second, although the sample size was sufficient to obtain reliable model estimates, there were still a considerable number of elderly were excluded from the study because their failure in completion of all follow-ups. And the loss of follow-up may be associated with the deterioration of depressive symptoms, which could affect the model estimation. Third, only limited factors were included in the analysis of predictors, and the associations between the factors and depressive trajectory classes might be partly due to their associations with baseline depressive levels.
This was the first study to explore seven-year trajectories of depressive symptoms in Chinese elderly population. Four depressive symptoms trajectories including low and stable, increasing, decreasing, high and stable, were identified by using growth mixture modelling. Except low and stable trajectory that showed a persistent low level of depressive symptoms, other trajectories were almost above the critical line of depressive symptoms, including the decreasing trajectory. There were significant demographic predictors of these trajectories such as gender, marital status, residence and educational level. In addition, suffering from chronic diseases was associated with a worse course of depressive symptoms. Our results demonstrate that depressive symptoms were hard to relieve to negative naturally once they occurred in Chinese elderly population, early intervention efforts can be undertaken in specific populations according to the trajectory predictors to avoid long-term exposure to depressive symptoms, which can lead to an increased risk of clinical depression.
Akaike information criterion
adjusted Bayesian information criterion
adjusted Lo-Mendell-Rubin likelihood ratio test
Bayesian information criterion
China Health and Retirement Longitudinal Study
the 10-item version of the Centre for Epidemiologic Studies Depression Scale
growth mixture model
latent class model
latent growth curve model
odds ratio
probability-proportional-to-size
the 95% confidence intervals
World Health Organization
The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.
Not applicable.
The datasets analyzed during the current study are available in the CHARLS repository: http://charls.pku.edu.cn/index/en.html.
The authors declare that they have no competing interests.
None.
Y.X. conducted data analysis and is a major contributor in writing the manuscript. M.M. contributed in raw data processing and helped to complete the software application. W.W. revised the manuscript. The authors have read and approved the final manuscript.
The research team of CHARLS are acknowledged for their work.