Introduction
Addressing the psychological challenges of older adults is a critical public health concern, especially in China with its rapidly aging population. There were around 209.78 million Chinese older adults by 2022, which accounts for 14.9% of the entire population and may reach 450 million in 2050 (National Bureau of Statistics of China, 2022; United Nations, 2019). This demographic transition, alongside growing financial and health-related pressures, could place older adults at greater risk of poorer mental health. Depression is a particular concern, as its prevalence has significantly risen from 3.9% to 17% in the past two decades (Li et al., Reference Li, Zhang, Shao, Qi and Tian2014; Zhang et al., Reference Zhang, Xu, Nie, Zhang and Wu2012). Therefore, understanding the underlying mechanism is crucial for mitigating depression in Chinese adults in mid- and later life for fostering successful aging.
Although past studies have identified some risk factors of mid- and later life depression, there remains a paucity of sound evidence to establish an optimal threshold for intervention. This limitation may stem from the traditional approach in psychological research that views symptoms as mere manifestations of diseases (Borsboom, Reference Borsboom2008). Besides, the prevailing idea that symptom amelioration relies primarily on treatment lacks comprehensive consideration; for instance, treating depression as an overall condition but not on its constituent symptoms such as depressed mood (Borsboom, Reference Borsboom2008). This constrains the potential to improve existing interventions and therapeutic strategies, and to explore the unidentified psychopathological and pathogenetic mechanisms underlying mental illnesses.
The network approach is a paradigm shift in psychopathology research as it provides a novel perspective to understand the dynamics of depressive symptoms by conceptualizing mental disorders as emerging from dynamic interactions among symptoms, which moves beyond the traditional view assuming these could be adequately represented by summed symptom scores (Borsboom, Reference Borsboom2008, Reference Borsboom2017; Cramer & Borsboom, Reference Cramer, Borsboom, Scott and Kosslyn2015; Cramer, Waldorp, van der Maas, & Borsboom, Reference Cramer, Waldorp, van der Maas and Borsboom2010; Fried & Cramer, Reference Fried and Cramer2017). It posits that the onset of one symptom can catalyze the activation of others, create feedback loops, and lead to a self-sustained activation of the whole network – the pathogenesis of mood disorders (Robinaugh, Hoekstra, Toner, & Borsboom, Reference Robinaugh, Hoekstra, Toner and Borsboom2020). Recent methodological advancements in network analysis allow capturing and visualizing the putative cascade of depressive symptoms as networks (Jones, Mair, & McNally, Reference Jones, Mair and McNally2018), where the strength and nature of connections between symptoms (or nodes) are depicted through the thickness and color of lines (or edges) (Opsahl, Agneessens, & Skvoretz, Reference Opsahl, Agneessens and Skvoretz2010) – thicker and red edges indicate stronger connections and negative relationships (Epskamp, Borsboom, & Fried, Reference Epskamp, Borsboom and Fried2018; Epskamp & Fried, Reference Epskamp and Fried2018; Jones et al., Reference Jones, Mair and McNally2018). Analyzing measures of the centrality of each node allows for the identification of central symptoms (most influential) in the network (Levinson et al., Reference Levinson, Zerwas, Calebs, Forbush, Kordy, Watson, Hofmeier, Levine, Crosby, Peat, Runfola, Zimmer, Moesner, Marcus and Bulik2017; Smith et al., Reference Smith, Mason, Crosby, Cao, Leonard, Wetterneck, Smith, Farrell, Riemann and Wonderlich2019; Snippe et al., Reference Snippe, Viechtbauer, Geschwind, Klippel, de Jonge and Wichers2017).
Most studies have used cross-sectional network modeling to analyze depressive symptoms in the older population across different countries (Belvederi Murri, Amore, Respino, & Alexopoulos, Reference Belvederi Murri, Amore, Respino and Alexopoulos2020; Castellanos et al., Reference Castellanos, Ausín, Bestea, González-Sanguino and Muñoz2020; Malgaroli, Calderon, & Bonanno, Reference Malgaroli, Calderon and Bonanno2021; McWilliams, Sarty, Kowal, & Wilson, Reference McWilliams, Sarty, Kowal and Wilson2017; Park, Reference Park2021; van Borkulo et al., Reference van Borkulo, Boschloo, Borsboom, Penninx, Waldorp and Schoevers2015), which has failed to thoroughly elucidate the temporal and causal dynamics between symptoms. For example, a data analysis from the Survey of Health, Ageing, and Retirement in Europe (SHARE) identified that death wishes, loss of interest, depressed mood, and pessimism are the central symptoms in the depressive symptoms network among older adults (Belvederi Murri et al., Reference Belvederi Murri, Amore, Respino and Alexopoulos2020), while a study on the Korean older population found anxiety, insomnia, and depressive mood instead (An et al., Reference An, Park, You, Park, Park, Woo and Son2019). These undirected cross-sectional networks provide only a static image that offers limited insights into the development and temporal predictive patterns of symptom interrelations (Borsboom & Cramer, Reference Borsboom and Cramer2013; Contreras et al., Reference Contreras, Nieto, Valiente, Espinosa and Vazquez2019; Fried et al., Reference Fried, van Borkulo, Cramer, Boschloo, Schoevers and Borsboom2017; Malgaroli et al., Reference Malgaroli, Calderon and Bonanno2021; Robinaugh et al., Reference Robinaugh, Hoekstra, Toner and Borsboom2020). Despite the potential of longitudinal network analysis to reveal dynamics between depressive symptoms, few studies have used it to test the sequence of consequences among depressive symptoms. Building on this perspective, existing research collectively substantiated the conceptualization of depression as a complex network of interrelated symptoms, where each symptom not only contributes to the overall disorder but also interacts with and influences each other, helping identify central symptoms that may serve as potential targets for intervention while highlighting the need to consider how symptoms are embedded within broader contextual systems (Cramer et al., Reference Cramer, Van Borkulo, Giltay, Van Der Maas, Kendler, Scheffer and Borsboom2016; De Boer, de Bruin, Geurts, & Glas, Reference De Boer, de Bruin, Geurts and Glas2021). Therefore, this study aims to employ longitudinal data to investigate temporal predictive patterns of depression symptoms among Chinese middle-aged and older adults.
According to De Boer et al. (Reference De Boer, de Bruin, Geurts and Glas2021), individuals are embedded within a multilayered network integrating environmental factors, for which the full explanatory capacity of network theory is met only with additional assumptions and prerequisites, such as clarifications of the associations between symptoms and environmental factors, and how network structure elucidates the emergence and development of symptoms, thus highlighting the necessity to integrate both symptoms and their environmental contexts.
Furthermore, the impact of social relationship networks, especially familial ties, on individual mental health should not be overlooked (Bronfenbrenner, Reference Bronfenbrenner1979; Ungar, Reference Ungar and Ungar2021; Weiss-Gal, Reference Weiss-Gal2008). The family systems theory posits that interdependent subsystems formed by family members can significantly and mutually affect each other’s thoughts, feelings, and moods (Cox & Paley, Reference Cox and Paley1997, Reference Cox and Paley2003; Minuchin, Reference Minuchin1985). For example, research on emotional contagion has shown that couples who share emotions would unconsciously mimic their partners’ facial expressions, vocalizations, and bodily postures during dyadic interactions (Hatfield, Cacioppo, & Rapson, Reference Hatfield, Cacioppo and Rapson1993; Herrando & Constantinides, Reference Herrando and Constantinides2021; Mazzuca, Kafetsios, Livi, & Presaghi, Reference Mazzuca, Kafetsios, Livi and Presaghi2019), highlighting the potential of cross-spousal associations between couples’ moods.
With the gradual contraction of middle-aged and older adults’ social networks (Zheng & Chen, Reference Zheng and Chen2020), the spousal relationship begins to become central. However, existing research tends to focus on individual-level symptom networks and aggregated depression scores (An et al., Reference An, Park, You, Park, Park, Woo and Son2019; Malgaroli et al., Reference Malgaroli, Calderon and Bonanno2021), which overlook symptom-level associations between couples. Given their shared daily environments and repeated interactions, depressive symptoms may be linked across spouses, suggesting wives’ and husbands’ symptom networks should be examined together. A study using dyadic network analyses to study Chinese postpartum couples, findings revealed significant depressive symptom interactions within and between partners, emphasizing the importance of considering both partners when assessing and intervening for postpartum depression (Huang, Zheng, & Guo, Reference Huang, Zheng and Guo2023).
However, the co-occurrence and interaction of depressive symptoms between middle-aged and older spouses is critical but still remains under-researched. The network theory suggests that symptoms connecting two symptom communities (known as bridge symptoms) may help explain how psychopathological states are linked from one community to another (Castro et al., Reference Castro, Ferreira, de Castro, Rodrigues, Correia, Ribeiro and Ferreira2019). Identifying such bridge symptoms in couples’ networks could provide clinical implications for preventing the contagion of depression and tailoring cost-effective interventions to deactivate specific symptoms (Borsboom & Cramer, Reference Borsboom and Cramer2013; Fried et al., Reference Fried, van Borkulo, Cramer, Boschloo, Schoevers and Borsboom2017; Montazeri, de Bildt, Dekker, & Anderson, Reference Montazeri, de Bildt, Dekker and Anderson2020), and reduce the development of comorbidity (Castro et al., Reference Castro, Ferreira, de Castro, Rodrigues, Correia, Ribeiro and Ferreira2019; Garabiles, Lao, Xiong, & Hall, Reference Garabiles, Lao, Xiong and Hall2019; Jones, Ma, & McNally, Reference Jones, Ma and McNally2021). Hence, the present study aims to better understand the intercorrelations between depressive symptoms of this dynamic psychopathology system by examining the directional predictive patterns of the depression symptom network in Chinese middle-aged and older couples from a dyadic perspective.
Moreover, gender differences in depressive symptom networks are also important to consider, as symptoms may express differently within close relationships. Although the family systems framework emphasizes mutual dependency within couples, it does not suggest identical symptom networks for both. Biological, psychological, and social factors (e.g. genetic predispositions, hormonal influences, gender roles, and societal attitudes) may all contribute to gender-specific patterns in depressive symptoms (Salk, Hyde, & Abramson, Reference Salk, Hyde and Abramson2017). Prior research has indicated that women are more likely to exhibit internalizing symptoms, while men show more externalizing symptoms (Price, Gregg, Smith, & Fiske, Reference Price, Gregg, Smith and Fiske2018; Sigmon et al., Reference Sigmon, Pells, Boulard, Whitcomb-Smith, Edenfield, Hermann, LaMattina, Schartel and Kubik2005). In line with this, other findings revealed divergent symptom network patterns, with women’s networks centered around depressed mood and men’s around fatigue and sleep difficulties (Castellanos et al., Reference Castellanos, Ausín, Bestea, González-Sanguino and Muñoz2020). However, whether similar gender-specific patterns of symptom networks exist within Chinese middle-aged and older couples, and whether cross-spousal predictive patterns differ by gender, still remains unclear.
The present study
Although some scholars have begun applying the network approach to intra-individual depressive symptoms (Belvederi Murri et al., Reference Belvederi Murri, Amore, Respino and Alexopoulos2020; Malgaroli et al., Reference Malgaroli, Calderon and Bonanno2021; McWilliams et al., Reference McWilliams, Sarty, Kowal and Wilson2017), few studies have been conducted on Chinese middle-aged and older adults. Despite the limitations of previous cross-sectional studies in determining the directionality of depressive symptom influences, they provide a solid foundation for understanding the complex interplays. Notably, until the recent work by Huang et al. (Reference Huang, Zheng and Guo2023), prior research had not yet extensively investigated dyadic symptom networks in depression within couples. This oversight represents potential underestimation of the complexity and emergence of comorbidity structures, especially among the Chinese population. Inspired by the network approach in psychological science, this exploratory study aims to investigate the intra-individual correlation pattern of depressive symptoms, the inter-individual interaction pattern, and gender differences in Chinese middle-aged and older couples over time using Cross-Lagged Panel Network Modeling (CLPN).
Method
Participants
The data for this study were from the China Health and Retirement Longitudinal Study (CHARLS), which is a nationally representative survey of the middle-aged and elderly, recruiting households with members aged 45 years or above. CHARLS included one age-eligible member of each eligible household and the spouse of the member (if available). The survey data were collected via face-to-face, computer-assisted personal interviews. This study included the baseline cohort recruited in 2011 and the second cohort recruited in 2013, both of which were followed 2 years later. We excluded respondents who had missing values on any depressive symptoms and whose responses were provided by proxy reporters. The final analytic sample consisted of 2,578 couple dyads, whose demographic information is reported in Table 1.
Demographic information and descriptive statistics of depressive symptoms
Table 1. Long description
The table is divided into two main sections.
Section 1: Demographic Information.
- Age (years): Husband M = 56.65 (SD = 8.22); Wife M = 54.45 (SD = 7.85).
- Education: The majority of husbands completed elementary (27.04%) or middle school (30.33%), while the largest group of wives were illiterate (32.00%).
- Duration of marriage: M = 32.23 years (SD = 9.07).
- Residence: 64.04% Urban, 35.96% Rural.
Section 2: Depressive Symptoms (N and % reporting symptoms).
This section tracks 10 items at T 1 and T 2 for both spouses.
1. Bothered by things: Husband T 1 (45.85%), T 2 (34.33%); Wife T 1 (55.51%), T 2 (45.81%).
2. Trouble keeping mind on tasks: Husband T 1 (38.83%), T 2 (38.83%); Wife T 1 (32.89%), T 2 (48.95%).
3. Felt depressed: Husband T 1 (42.75%), T 2 (35.26%); Wife T 1 (55.59%), T 2 (43.83%).
4. Everything was an effort: Husband T 1 (40.96%), T 2 (29.95%); Wife T 1 (46.70%), T 2 (41.04%).
5. Felt hopeful (reverse coded): Husband T 1 (47.71%), T 2 (41.78%); Wife T 1 (47.60%), T 2 (34.17%).
6. Felt fearful: Husband T 1 (10.94%), T 2 (13.7%); Wife T 1 (28.86%), T 2 (18.27%).
7. Restless sleep: Husband T 1 (39.53%), T 2 (35.14%); Wife T 1 (48.88%), T 2 (53.80%).
8. Was happy (reverse coded): Husband T 1 (48.41%), T 2 (46.35%); Wife T 1 (50.43%), T 2 (36.73%).
9. Felt lonely: Husband T 1 (16.95%), T 2 (18.93%); Wife T 1 (26.76%), T 2 (19.71%).
10. Could not get going: Husband T 1 (11.33%), T 2 (13.73%); Wife T 1 (21.68%), T 2 (14.31%).
Note: Depressive symptoms were coded as 0 = no symptoms, 1 = with symptoms.
Measure
Depressive symptoms were assessed with the 10-item Centre for Epidemiological Studies-Depression Scale (CES-D-10). Participants rated the items based on their feelings and behaviors during the last week on a 4-point Likert scale: 1 = rarely or none of the time (<1 day), 2 = some or a little of the time (1–2 days), 3 = occasionally or a moderate amount of the time (3–4 days), and 4 = most or all of the time (5–7 days). In the sample of this study, more than half of the respondents for most items indicated minimal levels of depressive symptoms, leading to deviations from the normal distribution (see Supplementary Table S1). Therefore, we converted the responses into binary (i.e. recoding 1 as 0 and recoding 2, 3, 4 as 1) to indicate the absence and presence of the rated symptom. The Cronbach’s α of the scale was ɑ = 0.75 in this study.
Data analysis
There are several methods of estimating networks from longitudinal data. This study adopted cross-lagged panel network (CLPN) models because of their flexibility in handling panel data with non-normal distributions, including binary data. In CLPN with binary data, edges represent nodewise cross-lagged effects (i.e. logistic regression coefficients after considering autoregressive effects and controlling for all other network nodes and covariates at T1). To remove spurious edges from the network, the regression models were regularized with LASSO with 10-fold cross-validation, which shrunk weak regression coefficients to zero. To enhance interpretability of the network, we converted log odds to odds ratios (ORs) and further omitted weaker edges (ORs within 0.65–1.35) and covariates in network visualization following the practices of Funkhouser, Chacko, Correa, Kaiser, and Shankman (Reference Funkhouser, Chacko, Correa, Kaiser and Shankman2021). As CLPN does not currently accept missing data, all observations with missing reports were excluded from the analyses. The network analyses were conducted with the glmnet package (Friedman, Hastie, & Tibshirani, Reference Friedman, Hastie and Tibshirani2010), and network visualizations were conducted with the qgraph package in R (Epskamp et al., Reference Epskamp, Cramer, Waldorp, Schmittmann and Borsboom2012). We computed two individual networks (i.e. consisting of 10 depressive symptoms of the individual) separately for husbands and wives and a dyadic network (i.e. consisting of 20 depressive symptoms, with 10 of the husband and 10 of the wife) in the couple dyad. Covariates included participants’ age, education level, and residence (i.e. rural or urban China) as they may confound associations of depressive symptoms across developmental, socioeconomic, and living contexts (Liu et al., Reference Liu, Fan, Luo, Zhou, Shen, Hu and Zhai2021; Tang, Jiang, & Tang, Reference Tang, Jiang and Tang2021). Duration of marriage was also controlled to account for shared length of marriage exposure for couples, given the dyadic nature of the study.
To assess the relative influences of the nodes within the network, we computed centrality indices of out-expected-influence (out-EI) and in-expected-influence (in-EI) of each symptom node in each network. Out-EI takes the sum of values of outgoing edges connected to the node, while in-EI takes the sum of values of incoming edges connected to the node (Robinaugh, Millner, & McNally, Reference Robinaugh, Millner and McNally2016). A positive out-EI indicates that the focal node exerts positive influences on the network (or receives positive influences from the network, in the case of in-EI), while a negative out-EI indicates negative influences. In the dyadic network, we defined symptoms of the husband as one community and those of the wife as another community and computed bridge strength and bridge EI of each symptom. Bridge strength takes the sum of absolute values of outgoing and incoming edges between a specific node and nodes from another community, indicating the focal node’s total connectivity with other communities. Bridge EI only takes the sum of values (not absolute values) of outgoing edges from a specific node to other communities, which measures the resulting changes in other communities of the focal node’s increase (Jones, Heeren, & McNally, Reference Jones, Heeren and McNally2017). The estimation of centrality indices did not involve covariates and was conducted with the qgraph package (Epskamp et al., Reference Epskamp, Cramer, Waldorp, Schmittmann and Borsboom2012) and the networktools package in R (Jones et al., Reference Jones, Heeren and McNally2017).
To evaluate the stability of centrality indices, we computed the correlation stability coefficient (CS-coefficient) using a case-dropping subset bootstrap method. The CS-coefficient represents the maximum proportion of cases that can be dropped such that the correlation between the original strength centrality and the ones of networks based on subsets reaches 0.70 or higher with 95% probability. A CS-coefficient above 0.25 was considered to be acceptable, and a CS-coefficient above 0.50 was considered to be preferable (Epskamp et al., Reference Epskamp, Borsboom and Fried2018). To evaluate network accuracy and test pairwise differences in node centralities, we conducted nonparametric bootstrap 1,000 times to compute the confidence intervals of edge values and centrality indices (Supplementary Figures S1–S3). These bootstraps were conducted with the bootnet package in R (Epskamp & Fried, Reference Epskamp and Fried2018).
Results
Figure 1a displays the estimated individual network of depressive symptoms in husbands. Results yielded a relatively dense network, with 81 (out of 100 possible edges) edges being estimated as significant after regularization. For the full adjacency matrix of the network, please see Supplementary Table S2 (see edge accuracy in Supplementary Figure S1). Among these significant edges, nine were relatively strong (OR ≥ 1.35 or ≤ 0.65) and are presented in Figure 1a. As shown, Happiness, as a reverse symptom, was related to lower odds of Loneliness (OR = 0.65), Could not get going (OR = 0.55), Fear (OR = 0.59), and Depressed feelings (OR = 0.65) in the next year. The symptom of Depressed feelings predicted higher odds of Bothered (OR = 1.39), Concentration troubles (OR = 1.37), and Strenuous feelings (OR = 1.59). Moreover, Could not get going predicted higher odds of Strenuous feelings (OR = 1.36), and Loneliness predicted higher odds of Depressed feelings (OR = 1.41).
Dynamic network.
Figure 1. Long description
A four-panel network diagram labeled a through d. A legend at the bottom identifies blue circles as Husband and red circles as Wife. Ten symptoms are numbered 1 through 10.
Panel a shows blue nodes H 1 through H 10 in a circle. Arrows indicate relationships: H 2 to H 3, H 3 to H 4, H 4 to H 10, H 8 to H 3, H 8 to H 6, H 8 to H 9, and H 9 to H 10.
Panel b shows red nodes W 1 through W 10 in a circle. Arrows originate mostly from W 6, pointing to W 1, W 3, W 4, W 9, and W 10. Other arrows connect W 1 to W 3 and W 10 to W 9.
Panel c and d show integrated networks. In panel c, thick dark blue arrows show strong cross-partner influences: W 6 points to H 6, H 8, and H 10. W 1 points to H 10. H 8 points to W 6. Light blue and red arrows show secondary connections between various H and W nodes.
Panel d shows a similar integrated network but with fewer connections, focusing on the strongest paths from W 6 to H 6, H 8, and H 10, and from H 7 to H 10.
Symptom key: 1: bothered by things, 2: trouble concentrating, 3: felt depressed, 4: everything was an effort, 5: felt hopeful, 6: felt fearful, 7: restless sleep, 8: was happy, 9: felt lonely, 10: could not get going.
The z-scored EIs of symptom nodes in the individual network of husbands are presented in Figure 2, and the original EI indices are presented in Supplementary Table S5. The two reverse symptoms (i.e. Hope and Happiness) had negative out-EIs and in-EIs, indicating not only their inhibitory impacts but also their vulnerabilities to adverse influences within the depression network. The other symptoms had positive out- and in-EIs, except that Could not get going had a small but negative in-EI. Bootstrap node-wise difference tests on EIs (Supplementary Figure S5) showed that Happiness had a distinctive out-EI. Meanwhile, the in-EIs of Happiness and Hope significantly differed from those of other nodes. Worth noting, concentration troubles and strenuous feelings showed leading in-EIs in husbands’ individual network. These suggest that they were most susceptible to the change of other depressive symptoms. Both out-EI (CS [cor = 0.70] = 0.59) and in-EI (CS [cor = 0.70] = 0.75) in the individual network of husbands showed good stability (Supplementary Figure S4).
EI and bridge centralities for network.
Figure 2. Long description
A multi-panel figure with three line graphs and a legend. All graphs share a vertical Y-axis numbered 1 to 10, corresponding to a list of symptoms. The X-axis represents Z-scores ranging from -2.5 to 3.5.
Top-Left Graph: Z-scored E I in the individual network. It shows fluctuating lines for Husband and Wife In-E I and Out-E I. A notable peak occurs for Wife Out-E I at symptom 6 with a Z-score above 2.0.
Top-Right Graph: Z-scored E I in the dyadic network. Similar to the first graph, it tracks In-E I and Out-E I. The Wife Out-E I line shows a significant spike at symptom 6, reaching a Z-score of approximately 3.5.
Bottom-Left Graph: Z-scored bridge centralities in the dyadic network. This graph tracks Bridge E I and Bridge strength. Most data points cluster between -1.0 and 1.0, except for a sharp peak in Wife Bridge E I and Bridge strength at symptom 6, reaching Z-scores near 3.5.
Line Definitions Legend:
Blue lines represent Husband. Red lines represent Wife.
Solid lines with squares: Out-E I.
Dashed lines with squares: In-E I.
Solid lines with triangles: Out-E I (Wife).
Dashed lines with triangles: In-E I (Wife).
Dash-dot lines: Bridge E I.
Dotted lines: Bridge strength.
Symptoms List:
1: I was bothered by things that don’t usually bother me.
2: I had trouble keeping my mind on what I was doing.
3: I felt depressed.
4: I felt everything I did was an effort.
5: I felt hopeful about the future.
6: I felt fearful.
7: My sleep was restless.
8: I was happy.
9: I felt lonely.
10: I could not get going.
Network analysis also revealed a dense individual network in wives, with 94 out of 100 possible edges estimated as significant after regularization. Supplementary Table S3 presents the full adjacency matrix (see edge accuracy in Supplementary Figure S2) and Figure 1b displays the relatively strong edges (OR ≥ 1.35 or ≤ 0.65). The individual network of wives demonstrated substantive differences from that of husbands. In wives, Fear predicted higher odds of Bothered (OR = 1.35), Depressed feelings (OR = 1.58), Strenuous feelings (OR = 1.63), and Loneliness (OR = 1.58). Could not get going and Fear showed a mutually predictive effect (OR = 1.73 and 1.37). Bothered also predicted higher odds of Depressed feelings (OR = 1.46). The z-scored EIs are displayed in Figure 2 and the original EI indices are presented in Supplementary Table S5. Similar to the individual network in husbands, Hope and Happiness had negative out-EIs and in-EIs, while other symptoms had positive out- and in-EIs. Bootstrap node-wise difference tests on EIs (Supplementary Figure S6) showed that the out-EI of Fear was significantly higher than those of most nodes (except for Bothered) within the network. The two reverse symptoms, Hope and Happiness, had distinctively lower negative out- and in-EIs compared with other nodes, which suggested that the reverse symptoms exerted strong negative influence on, as well as received strong negative influences from, the depression network. Out-EI (CS [cor = 0.70] = 0.67) and in-EI (CS [cor = 0.70] = 0.75) in the individual network of wives showed good stability (Supplementary Figure S4).
The dyadic network consisting of the husband’s and the wife’s symptoms is presented in Figure 1c (see Supplementary Table S4 for the full adjacency matrix and Supplementary Figure S3 for accuracy). Most strong edges in the individual networks remained substantive in the dyadic network. Network analysis identified several strong cross-community edges bridging the husband and the wife, which are highlighted in Figure 1d by omitting within-individual edges. All bridges, except for one (i.e. husbands’ Loneliness predicting higher odds of wives’ Loneliness; OR = 1.36), started from the wife and pointed to the husband. Wives’ Loneliness (OR = 1.60), Could not get going (OR = 1.49), and Strenuous feelings (OR = 0.57) significantly predicted husbands’ Fear. Wives’ Happiness, as a reverse depressive symptom, predicted higher odds of husbands’ Happiness (OR = 1.77), Loneliness (OR = 1.42), and Could not get going (OR = 1.79). Wives’ Fear demonstrated relatively strong connections with husbands’ symptoms, including Loneliness (OR = 2.51), Could not get going (OR = 2.62), and Fear (OR = 1.98). Interestingly, it also predicted higher odds of husbands’ Happiness (OR = 1.98). The z-scored bridge centralities are displayed in Figure 2, and the original bridge centralities are displayed in Supplementary Figure S6 and Table S6. For husbands, Happiness showed relatively strong negative bridge EI. For wives, Fear demonstrated distinctively high positive bridge EIs and bridge strength, followed by Happiness. Strenuous feelings and Sleep difficulties had relatively strong negative bridge EIs. Bootstrap node-wise difference tests on node strength (Supplementary Figure S7) and bridge centralities (Supplementary Figure S8) showed that wives’ Fear had distinctively higher bridge EI and bridge strength in the dyadic network. The bridge EI of wives’ Happiness was significantly higher than most nodes in the network, while that of wives’ Strenuous feelings was negative and lower than most nodes. Bridge EIs (CS [cor = 0.70] = 0.75) and bridge strength (CS [cor = 0.70] = 0.75) in the dyadic network demonstrated good stability (Supplementary Figure S4).
Discussion
In network analysis, the interrelations and dynamic changes among individual symptoms provide a more dynamic perspective on symptom progression than traditional cross-sectional methods. This approach highlights symptom interactions over time and thus discovers complex patterns of symptom development and change. However, these relationships should be understood with caution. Although using panel data provides significant insights, their capacity to directly establish causation is limited (Borsboom, Reference Borsboom2017; Cramer & Borsboom, Reference Cramer, Borsboom, Scott and Kosslyn2015; Fried & Cramer, Reference Fried and Cramer2017; Fried et al., Reference Fried, van Borkulo, Cramer, Boschloo, Schoevers and Borsboom2017). As most previous research has predominantly examined depressive symptoms using cross-sectional network analysis in older adults (Belvederi Murri et al., Reference Belvederi Murri, Amore, Respino and Alexopoulos2020), to address this gap, the current study leverages longitudinal data to parse the temporal precedence and directionality of depressive symptoms both at individual and dyadic levels in a nationally representative sample of Chinese middle-aged and older couples, which provides a more sophisticated understanding of the complex dynamics of depression and highlights the importance of both individual and relational factors in its etiology.
Modeling these temporal relationships offers an opportunity to elucidate the gender differences in the etiological underpinnings of depression within couples. It is observed that the predictive relationship between depressive symptoms varied between husbands and wives at the individual level. At the dyadic level, the current study identified fearfulness in wives as a key bridge symptom, potentially linking to husbands’ depressive symptomatology. This insight is pivotal in explaining the high rates of co-occurrence of depressive symptoms in couples and underscores the need for targeted interventions that consider the relational dynamics in addressing mental health challenges in this demographically significant group (Forsman, Schierenbeck, & Wahlbeck, Reference Forsman, Schierenbeck and Wahlbeck2011).
Furthermore, by elucidating these gender differences and dyadic interactions, this study extends symptom-level research on couples. Our findings inform more focused approaches to understanding depressive symptoms, particularly relevant for geriatric populations and diverse cultural settings.
Individual depressive symptom network in husbands and wives
Happiness had the highest out-EI estimates in the husbands’ network, indicating it as a strong protective factor against other depressive symptoms over time (i.e. loneliness, could not get going, fearfulness, and depressed feelings). This supports contemporary therapeutic approaches that emphasize incorporating positive affect as a key strategy for mitigating depressive symptoms (Tolentino & Schmidt, Reference Tolentino and Schmidt2018). Notably, this protective effect appeared to be non-significant in the wives’ network, implying the role of gender in the dynamics of emotional networks. This may reflect gender differences in the expression and processing of emotions, or different needs and experiences of social support and emotional communication between men and women (Brody, Reference Brody1985; Kret & de Gelder, Reference Kret and de Gelder2012; Proverbio, Reference Proverbio2023). For example, women may rely more on extensive connections in their social networks for emotional support, whereas men may rely more on interaction with their partners for emotional stability. The relatively higher but non-significant predictive role of depressed feelings (e.g. I felt depressed) on other symptoms indicates a partial alignment with prior studies for European older adults, highlighting possible cultural variations (Savelieva, Komulainen, Elovainio, & Jokela, Reference Savelieva, Komulainen, Elovainio and Jokela2021). The inconsistency in the predictive role of depressed feelings across different studies may be attributed to sample variability, methodological approaches, or cultural factors, necessitating further exploration (Buil, de Chernatony, & Martínez, Reference Buil, de Chernatony and Martínez2012). Another potential reason for these findings could be the different focuses of studies. For example, some research primarily concentrates on psychopathological aspects and may not sufficiently consider how positive emotions affect the outcomes (Blodgett et al., Reference Blodgett, Lachance, Stubbs, Co, Wu, Prina and Cosco2021). Aligning with the recent shifts toward integrating the disciplines of positive and clinical psychology (Duckworth, Steen, & Seligman, Reference Duckworth, Steen and Seligman2005; Seligman & Peterson, Reference Seligman, Peterson, Aspinwall and Staudinger2003; Wood & Tarrier, Reference Wood and Tarrier2010), the current study further provides new evidence on how ‘positive’ symptoms protect against the worsening of ‘negative’ symptoms in husbands’ networks (Wasil et al., Reference Wasil, Gillespie, Park, Venturo-Conerly, Osborn, DeRubeis and Jones2021; Wood & Tarrier, Reference Wood and Tarrier2010).
Moreover, results revealed that symptoms of concentration troubles (e.g. I had trouble keeping my mind on what I was doing) and strenuous feelings (I felt everything I did was an effort) had the highest in-EI estimates in husbands’ networks but not in wives’, highlighting intriguing gender differences in the symptomatology and dynamics of couple relationships, in which these symptoms were particularly susceptible to deterioration influenced by other symptoms (Bringmann et al., Reference Bringmann, Vissers, Wichers, Geschwind, Kuppens, Peeters and Tuerlinckx2013; van Borkulo et al., Reference van Borkulo, Boschloo, Borsboom, Penninx, Waldorp and Schoevers2015). This suggests that negative affect is significantly associated with cognitive abilities, consistent with literature on affective states and cognitive impairments (Schweizer et al., Reference Schweizer, Kievit, Emery, Cam and Henson2018). The strong association from depressed feelings to Strenuous feelings supports the ABC model, suggesting that interventions for reducing one’s sad moods may be effective in increasing their motivation and behaviors (Mezulis, Hyde, Simonson, & Charbonneau, Reference Mezulis, Hyde, Simonson, Charbonneau, Strauman, Costanzo and Garber2011; Millgram et al., Reference Millgram, Joormann, Huppert, Lampert and Tamir2019).
Interestingly, fearfulness (e.g. I felt fearful) exhibited the highest out-EI estimate in wives’ networks, suggesting it as a precursor to other depressive symptoms. This might reflect the societal and life challenges specific to middle-aged and older women, such as lower social status, less pension, and a heightened sense of fear and anxiety about death and future uncertainties (Fortner & Neimeyer, Reference Fortner and Neimeyer1999; Kiely, Brady, & Byles, Reference Kiely, Brady and Byles2019; Lewis, Reference Lewis1997). Observed significant associations between fearfulness and could not get going (e.g. I could not get going), loneliness (e.g. I felt lonely), strenuous feelings, and depressed feelings suggest strong correlations between these emotions and depressive symptoms, which could be attributed to their concerns about loss of mobility, reduced self-control, and independence (Bai, Lai, & Liu, Reference Bai, Lai and Liu2020; Chen, Giles, Wang, & Zhao, Reference Chen, Giles, Wang and Zhao2018; Fiske, Wetherell, & Gatz, Reference Fiske, Wetherell and Gatz2009; Li et al., Reference Li, Zhang, Shao, Qi and Tian2014; Pei et al., Reference Pei, Cong, Silverstein, Li and Wu2022; Schafer & Shippee, Reference Schafer and Shippee2010; Walsh, Scharf, & Keating, Reference Walsh, Scharf and Keating2017). Thus, further research needs to unravel the underlying mechanisms behind these interesting directional associations.
Besides, happiness and hope had distinctive in-EI estimates in both husbands’ and wives’ individual networks, indicating their susceptibilities to adverse influences within the depression network, which extended previous findings on the strong association between depressive symptoms and happiness (Sin & Lyubomirsky, Reference Sin and Lyubomirsky2009; Wasil et al., Reference Wasil, Gillespie, Park, Venturo-Conerly, Osborn, DeRubeis and Jones2021). For Chinese middle-aged and older wives, addressing one negative symptom could potentially improve their overall positive mental health, thus creating substantial changes in the network of symptoms (Bos et al., Reference Bos, Snippe, De Vos, Hartmann, Simons, Krieke and Wichers2017).
The present study reveals gender differences in the network of depressive symptoms among middle-aged and older Chinese couples. Specifically, happiness, concentration troubles, and strenuous feelings were all crucial in husbands’ networks but non-significant in wives’. This absence indicates different coping mechanisms or support structures and emphasizes the necessity of gender-specific approaches in understanding and addressing the interplay between emotional and cognitive symptoms (Brody, Reference Brody1985). However, fearfulness was a key predictor of other depressive symptoms in wives’ networks, which aligns with recent research indicating gender-specific manifestations in depressive symptom networks and highlights the need for gender-sensitive approaches in both research and treatment of depression (Castellanos et al., Reference Castellanos, Ausín, Bestea, González-Sanguino and Muñoz2020). Additionally, a Korean study found that women are more vulnerable to atypical depressive symptoms, whereas men are more vulnerable to depressive feelings (Kim et al., Reference Kim, Cho, Hong, Bae, Cho, Hahm, Lee, Park, Lee, Jeon and Chang2015). This could be attributed to aging as a gendered experience, with older women facing unique challenges over time (Cicirelli, Reference Cicirelli2002). Previous studies have identified fear as a present-oriented mood state associated with a threat, one of the core symptoms of anxiety (McLean, Asnaani, Litz, & Hofmann, Reference McLean, Asnaani, Litz and Hofmann2011; Ströhle, Gensichen, & Domschke, Reference Ströhle, Gensichen and Domschke2018). Given the high prevalence of comorbidity of anxiety and depression in Chinese middle-aged and older women (Guo et al., Reference Guo, Meng, Huang, Fan, Zhou, Ling and Su2016; Li et al., Reference Li, Shi, Yang, Gao, Li, Tao, Wang, Zhang, Gao, Liu, Li, Li, Liu, Wang, Zhang, Lv, Wang, Chen, Hu and Zhang2012; Zhao et al., Reference Zhao, Zhang, Liu, Yue, Hou, Xia and Ge2020), anxiety symptoms such as Fear may be more likely to be linked to the occurrence of other depressive symptoms in wives’ networks than in their husbands’. Such cross-spousal associations align with previous findings in older adults in Poland during COVID-19 (Gambin et al., Reference Gambin, Sękowski, Woźniak-Prus, Wnuk, Oleksy, Cudo and Maison2021). Future studies should examine gender differences in the comorbidity of depression and anxiety at symptom levels among middle-aged and older Chinese couples using directional network analysis (Borsboom, Reference Borsboom2017; Bringmann et al., Reference Bringmann, Vissers, Wichers, Geschwind, Kuppens, Peeters and Tuerlinckx2013; Cramer et al., Reference Cramer, Van Borkulo, Giltay, Van Der Maas, Kendler, Scheffer and Borsboom2016).
Furthermore, the study underscores the significance of examining cultural aspects of the depressive symptom network and its gender variations. For middle-aged and older Chinese couples, culturally specific role expectations and societal structures may be pivotal. Future investigations could probe more deeply into how these cultural factors sculpt the manifestation of depressive symptoms and the emotional contagion between spouses, and how to effectively address these challenges across different cultural settings. In addition, the study highlights the importance of developing gender-sensitive interventions to promote successful aging in middle-aged and older married couples. For instance, considering the crucial role of happiness in husbands’ network, a happiness-enhancing practice may predict the remission of other depressive symptoms, activating positive feedback loops within the network and subsequently reducing the persistence of their later depression (Greenawalt et al., Reference Greenawalt, Orsega-Smith, Turner, Goodwin and Rathie2019; Ho, Yeung, & Kwok, Reference Ho, Yeung and Kwok2014; Sin & Lyubomirsky, Reference Sin and Lyubomirsky2009; Wasil et al., Reference Wasil, Gillespie, Park, Venturo-Conerly, Osborn, DeRubeis and Jones2021). Together, these novel findings contribute to the existing knowledge by identifying the most critical individual depressive symptoms that drive the progression of the mental health challenges (Dablander & Hinne, Reference Dablander and Hinne2019; Robinaugh et al., Reference Robinaugh, Hoekstra, Toner and Borsboom2020).
Dyadic depressive symptom network within couples
Regarding the inter-individual predictive pattern of depressive symptoms for middle-aged and older couples, our research found that wives’ fearfulness had the highest bridge EI and bridge strength, indicating that it may serve as a key bridge symptom associated with their husbands’ subsequent depressive symptoms (Jones, Ma, & McNally, Reference Jones, Ma and McNally2021), including loneliness, could not get going, and fearfulness. These results demonstrate the cross-spousal predictive associations of depressive symptoms and of one partner’s emotional state on the other (Baek, Youm, & Kim, Reference Baek, Youm and Kim2021; Borsboom & Cramer, Reference Borsboom and Cramer2013; Pradeep & Sutin, Reference Pradeep and Sutin2015). Our study contributes preliminary evidence to the understanding of cross-spousal predictions of emotional contagion in middle-aged and older couples, particularly highlighting the role of wives’ depressive symptoms in their husbands’ depressive symptoms. This may be attributed to gender differences in emotional expression and processing, in that men are more likely to exhibit lower emotional literacy and are less likely to seek support from intimate relationships (Oliver, Pearson, Coe, & Gunnell, Reference Oliver, Pearson, Coe and Gunnell2005; Wilhelm, Reference Wilhelm2009). One possible explanation is that the tendency of wives to seek reassurance due to feelings of fear may place emotional burdens on their partners, potentially creating a deterioration in other symptoms over time; however, this remains exploratory, and further research is needed to clarify this dyadic process (Horenstein & Heimberg, Reference Horenstein and Heimberg2020).
Surprisingly, several cross-spousal findings presented a more complex pattern compared to the generally expected outcome explained by a conventional emotional contagion framework: wives’ fearfulness was associated with higher odds of husbands’ subsequent happiness, whereas wives’ happiness was associated with higher odds of husbands’ subsequent negative symptoms (e.g. loneliness and could not get going), and positive symptoms (e.g. happiness). These mixed results suggest that cross-spousal associations in symptoms may involve more nuanced dyadic processes in terms of the interplay of emotions in marital relationships (Butler, Reference Butler2011). The potential positive impact of wives’ reassurance-seeking on their partners’ satisfaction (Abe & Nakashima, Reference Abe and Nakashima2021) and the tendency of middle-aged and older individuals to experience mixed emotions (e.g. greater co-occurrence of positive and negative emotions) may provide additional explanations and add further complexity to these findings compared to the traditional framework that conceptualizes emotions dichotomously (An et al., Reference An, Park, You, Park, Park, Woo and Son2019). Previous research has indicated that middle-aged and older men often experience a growing sense of disconnection from their community post-retirement, consequently placing greater emotional reliance on their marital relationships (Noh et al., Reference Noh, Kwon, Lee, Oh and Kim2019; Takashima, Onishi, Saeki, & Hirano, Reference Takashima, Onishi, Saeki and Hirano2020). This shift in emotional focus may contribute to more complex dynamics in the marriage. For example, wives’ happiness may be linked to husbands’ happiness, but it may not necessarily buffer husbands’ loneliness or motivational difficulties. Such dynamics could inadvertently heighten feelings of loneliness for men, particularly when emotional support, perceived understanding, or satisfaction of marriage is limited (Hsieh & Hawkley, Reference Hsieh and Hawkley2018). Meanwhile, alternative interpretations may also be considered. For example, single-item symptoms may be interpreted differently across gender and relationship contexts. In addition, contextual variables (e.g. marital satisfaction, physical health, caregiving burden, stressful life events, etc.) not included in measurement may confound the dyadic process. Besides, as the symptom-focused models remain exploratory, these unexpected patterns should be replicated across different contexts, samples, and methodological approaches. Future research could further identify potential factors, such as the buffering role of marital satisfaction in cross-spousal associations between wives’ and husbands’ depressive symptom contagion, to explain these incongruent findings.
Limitations
Although the present study provides an innovative methodological approach and insights into understanding Chinese middle-aged and older couples’ dyadic interactions of depressive symptoms, there were still several limitations that should be addressed: (1) men are more likely to exhibit externalizing symptoms of depression such as aggression, substance abuse, and risk-taking; these were not captured by standard screening tools (Smith, Mouzon, & Elliott, Reference Smith, Mouzon and Elliott2018), potentially underestimating their mental health challenges (Borsboom, Reference Borsboom2008; McNally, Reference McNally2021). Future research should systematically assess both internalizing and externalizing symptoms for both partners to improve validation. (2) The traditional reliance on self-report data may limit robustness; thus, using a multimodal approach such as clinician-rated and physiological measures including electroencephalography (EEG) would enhance validity. (3) Findings may not generalize to other racial or ethnic populations, as the sample consisted solely of middle-aged and older Chinese couples (Jones, Williams, & McNally, Reference Jones, Williams and McNally2021). Further research should generalize studies to broader samples in diverse cultural backgrounds for improving robustness and applicability. (4) Although this study adjusted for several demographic and relational covariates, unmeasured individual, dyadic and contextual factors, physical health, caregiving burden, marital satisfaction, socioeconomic conditions, and stressful life events, etc., may also contribute to the variations in the observed cross-spousal associations (Maier, Riedel-Heller, Pabst, & Luppa, Reference Maier, Riedel-Heller, Pabst and Luppa2021). Overall, these findings highlight the need for more comprehensive, culturally sensitive, and gender-specific research on depressive symptoms among middle-aged and older couples, advocating more tailored and holistic interventions including targeted psychoeducation, training in emotional availability and expression, and community resources.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/S0033291726106382.
Data availability statement
The dataset that supports the findings of this study is available on the CHARLS website (https://charls.charlsdata.com). The code supporting the findings of this study may be made available from the corresponding author upon reasonable request.
Author contributions
Diyang Qu and Anni Zhu contributed equally as co-first authors. Diyang Qu and Anni Zhu conceived and designed the study, conducted statistical analyses, interpreted the data, and drafted the manuscript. Junkang Lin contributed to data interpretation, manuscript drafting, and revisions. Jiaxun Yu contributed to manuscript review and editing. Jinfeng Zhang provided methodological guidance, contributed to data interpretation, and critically revised the manuscript for important intellectual content. Jiasheng Huang supervised the entire project, contributed substantially to the study design, data interpretation, critical revision of the manuscript, and provided final approval for publication. All authors have read and approved the final manuscript and agree to be accountable for all aspects of the work.
Funding statement
This work was supported by the National Natural Science Foundation of China (Grant No. 82103956).
Competing interests
Diyang Qu, Anni Zhu, Junkang Lin, Jiaxun Yu, Jinfeng Zhang, and Jiasheng Huang declare no conflicts of interest related to the manuscript titled ‘Mapping Emotional Contagion: Directional Predictive Patterns of Depressive Symptoms Among Middle-Aged and Elderly Chinese Couples’.
Ethical standards
Ethical approval for all waves of the CHARLS study was obtained from the Institutional Review Board at Peking University (IRB00001052-11015). All participants provided written informed consent as part of the CHARLS study procedures. The current study used anonymized CHARLS data for secondary analysis and did not involve any direct interaction with participants.
Open access
