Introduction
Depression is a common mental health issue and is often associated with reduced life satisfaction, higher healthcare service use and poor functional status (Beekman et al., Reference Beekman, Penninx, Deeg, de Beurs, Geerling and van Tilburg2002). The prevalence of depression has dramatically increased globally since the outbreak of the COVID-19 pandemic (Galea et al., Reference Galea, Merchant and Lurie2020; Xiang et al., Reference Xiang, Zhao, Liu, Li, Zhao, Cheung and Ng2020), particularly among older adults (Luppa et al., Reference Luppa, Sikorski, Luck, Ehreke, Konnopka, Wiese, Weyerer, Konig and Riedel-Heller2012; World Health Organization, 2020). It is estimated that nearly 14% of people older than 55 years have depression, including 2% with major depression (Beekman et al., Reference Beekman, Copeland and Prince1999), with the highest prevalence observed between ages 45 and 64 (Romanoski et al., Reference Romanoski, Folstein, Nestadt, Chahal, Merchant, Brown, Gruenberg and McHugh1992). Late-life depression carries unique burdens, including poorer health outcomes, higher disability and greater societal costs compared with younger populations (Beekman et al., Reference Beekman, Penninx, Deeg, de Beurs, Geerling and van Tilburg2002; Koivumaa-Honkanen et al., Reference Koivumaa-Honkanen, Kaprio, Honkanen, Viinamaki and Koskenvuo2004; Kok and Reynolds, Reference Kok and Reynolds2017; Taylor, Reference Taylor2014; Veltman et al., Reference Veltman, Kok, Lamers, Stek, van der Mast and Rhebergen2020). Sociodemographic factors such as female sex, increasing age, being single or divorced, lower educational attainment, unemployment and low income are often associated with late-life depression (Luppa et al., Reference Luppa, Sikorski, Luck, Ehreke, Konnopka, Wiese, Weyerer, Konig and Riedel-Heller2012; Schaakxs et al., Reference Schaakxs, Comijs, Lamers, Kok, Beekman and Penninx2018).
Exposures to stressors and their stress response are known to be one of the major causes of depression onset (Su et al., Reference Su, D’Arcy, Li, O’Donnell, Caron, Meaney and Meng2022). The COVID-19 pandemic, as well as its control measures (such as lockdowns, social distancing, wearing masks, monitoring pandemic information and others), have produced both acute and chronic health consequences (Brooks et al., Reference Brooks, Webster, Smith, Woodland, Wessely, Greenberg and Rubin2020; The Lancet Public, 2020). In Canada, the pandemic control measures included nationwide lockdowns, travel restrictions and phased reopening strategies introduced from March 2020 onwards. Like many other countries and regions, these control measures have effectively controlled the transmission of SARS-CoV-2; some of these measures have also introduced additional stressors by disrupting daily routines, economic activities and access to healthcare services (Ahmed et al., Reference Ahmed, Shafer, Malla, Hopkins, Moreland, Zviedrite and Uzicanin2024). These COVID-19-related stressors, such as financial strain, domestic conflicts and family rearrangements, have made people more prone to depression and other mental disorders (Foa et al., Reference Foa, Gilbert and Fabian2020; Schmitz et al., Reference Schmitz, Holley, Meng, Fish and Jedwab2020; Siddaway, Reference Siddaway2020; The Lancet Psychiatry, 2021; Zacher and Rudolph, Reference Zacher and Rudolph2021).
Even though COVID-19 disproportionately affects older people, compared to their younger counterparts, adults aged 65 or older were less affected by the psychological impact of the COVID-19-related stressors (Czeisler, Reference Czeisler2020; De Rubeis et al., Reference De Rubeis, Anderson, Khattar, de Groh, Jiang, Oz, Basta, Kirkland, Wolfson and Griffith2022; Heid et al., Reference Heid, Cartwright, Wilson-Genderson and Pruchno2021). Furthermore, females reported higher prevalence rates of depression and other internalising disorders compared to males (Bangasser and Valentino, Reference Bangasser and Valentino2014; Grant et al., Reference Grant, Dawson, Stinson, Chou, Dufour and Pickering2004; Kendler et al., Reference Kendler, Kessler, Walters, MacLean, Neale, Heath and Eaves1995; Tozzi et al., Reference Tozzi, Garczarek, Janowitz, Stein, Wittfeld, Dobrowolny, Lagopoulos, Hatton, Hickie, Carballedo, Brooks, Vuletic, Uhlmann, Veer, Walter, Bulow, Volzke, Klinger-Konig, Schnell, Schoepf, Grotegerd, Opel, Dannlowski, Kugel, Schramm, Konrad, Kircher, Juksel, Nenadic, Krug, Hahn, Steinstrater, Redlich, Zaremba, Zurowski, Chy, Dima, Cole, Grabe, Connolly, Yang, Ho, LeWinn, Li, Groenewold, Salminen, Walter, Simmons, van Erp, Jahanshad, Baune, van der Wee, van Tol, Penninx, Hibar, Thompson, Veltman, Schmaal and Frodl2020). These disparities are partially explained by the fact of the nature and magnitude of stressors faced by each sex. During the pandemic, females disproportionately shouldered caregiving responsibilities, experienced greater employment instability and experienced increased domestic stressors. All of these increased their risk of mental health problems (Harman, Reference Harman2016; Wenham et al., Reference Wenham, Smith, Davies, Feng, Grepin, Harman, Herten-Crabb and Morgan2020a). Therefore, age and sex should be considered when preparing and responding to the pandemic, as well as in the evaluation of public health control measures (Wenham et al., Reference Wenham, Smith, Morgan and Group2020b).
Recently, a novel disease model – “common systems model” has been proposed to understand that mental disorders are influenced by dynamic interactions between symptoms and environmental or social influences, rather than explained by latent variables (Borsboom, Reference Borsboom2017). This notion is in contrast with the traditional “common cause model”, which argues that symptoms are merely manifestations of an underlying factor and tends to overlook the interdependence among symptoms (Fried and Nesse, Reference Fried and Nesse2015). The “common systems model” recognises a system-based framework that highlights the potential for feedback loops and mutual reinforcement between symptoms, offering a more ecologically valid understanding of psychopathology development (Borsboom, Reference Borsboom2017). To operationalise this model, the network analysis was proposed to examine the relationships between individual symptoms in a visual network and quantify their relative importance or centrality, indicating which symptoms are most influential within the system, also known as core symptoms (McNally et al., Reference McNally, Robinaugh, Gwy, Wang, Deserno and Borsboom2015). These core symptoms could serve as targets for clinical management by reducing their interconnections. Age and sex differences are also presented in the structure and centrality of symptom networks (O’Shields et al., Reference O’Shields, Graves and Mowbray2023). Older adults may have different central symptoms and stressors compared with younger individuals. Among older adults, stressors related to physical health might be more central and strongly connected to depressive symptoms (Solmi et al., Reference Solmi, Koyanagi, Thompson, Fornaro, Correll and Veronese2020). However, most studies are cross-sectional, only interrelationships are learned, not temporal or predictive relationships (Zavlis et al., Reference Zavlis, Butter, Bennett, Hartman, Hyland, Mason, McBride, Murphy, Gibson-Miller and Levita2022).
Despite longitudinal studies guided by the common cause model suggesting significant relationships between COVID-19 stressors and depression (Raina et al., Reference Raina, Wolfson, Griffith, Kirkland, McMillan, Basta, Joshi, Oz, Sohel and Maimon2021), there remains a paucity of research examining how these stressors are connecting with depression at the symptom level using the novel disease model – “common symptoms model”. Cross-lagged panel network analysis (CLPN), integrating cross-lagged panel modelling and network analysis, was recently developed (Rhemtulla et al., Reference Rhemtulla and Cramer2022). This analytic approach enables researchers to simultaneously examine both the temporal dynamics and directional influences among multiple variables across different time points (Funkhouser et al., Reference Funkhouser, Chacko, Correa, Kaiser and Shankman2021).
A long-standing debate between social causation and health selection hypotheses in both health sciences and social sciences has been made to argue whether social factors such as financial struggles lead to the development of depressive symptoms or whether the occurrence of depressive symptoms increases financial difficulties (Johnson et al., Reference Johnson, Cohen, Dohrenwend, Link and Brook1999). The verification of the directionality between the social causations and health selection hypotheses is tied to clarifying the root cause of depression, which ensures targeted and specific interventions are designed and implemented effectively and impactfully. Given limited longitudinal research on the directionality between COVID-19-related stressors and depressive symptoms, the present study aimed to: (1) examine whether the pattern and strength of intercorrelations between COVID-19-related stressors and depressive symptoms differ by sex and age groups; and (2) clarify whether specific stressors predict subsequent depressive symptoms over time, and identify the most central stressors and symptoms that shape these dynamic interrelations.
Methods
Study cohort
The data analysed are from multiple waves of the Canadian Longitudinal Study on Aging (CLSA), which is a national longitudinal study that tracks over 50,000 Canadians aged 45 and over for a minimum of 20 years. The CLSA comprises two cohorts: the comprehensive cohort, randomly selected from individuals residing within 25 km (or 50 km in lower-density areas) of 11 data collection sites across Canada and the Tracking Cohort, randomly recruited from all 10 Canadian provinces via telephone interviews. The CLSA baseline data were collected from 51,338 participants between 2011 and 2015, with follow-ups every 3 years (Raina et al., Reference Raina, Wolfson, Kirkland, Griffith, Balion, Cossette, Dionne, Hofer, Hogan and E2019). At Follow-up 1 (FUP1; 2015-2018), 48,893 participants (95% retention) remained enrolled, with 44,817 providing complete follow-up data. A total of 42,511 eligible participants from the cohort were invited to participate in the CLSA COVID-19 sub-study during Canada’s initial nationwide lockdown. It was administered between April and May 2020 and the follow-up survey (COVID-exit survey) took place between September and December 2020, a period characterized by partial reopening and the anticipation of vaccine approval. This study specifically analyzed participants from the comprehensive cohort who completed both CLSA COVID-19 baseline and exit surveys, yielding a final sample of 12,957 individuals. Figure S1 presents the flowchart of the selection process of the study cohort. Ethical approval for the CLSA was granted by the Hamilton Integrated Research Ethics Board, and the secondary analysis conducted for this study received approval from the Institutional Research Ethics Board at the Douglas Research Centre.
Measures
Depressive symptoms were measured with the Centre for Epidemiologic Studies Short Depression Scale (CES-D10), which includes three items on depressed affect, five items on somatic symptoms and two items on positive affect (Andresen et al., Reference Andresen, Malmgren, Carter and Patrick1994). Specifically, it consists of the symptoms of feeling depressed, everything is an effort, restless sleep, feeling happy (reverse-scored), feeling lonely, enjoying life (reverse-scored), feeling sad, could not get going, trouble concentrating and feeling fearful. Each symptom was rated on a four-point Likert scale, with higher total scores indicating greater depressive symptoms. It is reliable and valid in assessing depressive symptoms in adults, with an internal consistency of 0.86, a test–retest reliability of 0.85, convergent validity of 0.91 and divergent validity of 0.89 (Bjorgvinsson et al., Reference Bjorgvinsson, Kertz, Bigda-Peyton, McCoy and Aderka2013; Miller et al., Reference Miller, Anton and Townson2008).
Five COVID-19-related stressors were measured using a self-reported questionnaire at the COVID-19 baseline survey. They were selected based on their availability in existing validated questionnaires and had been previously used in research involving adolescents, which supports their empirical relevance and face validity (Su et al., Reference Su, Li, Schmitz and Meng2024): (1) Health-related stressors were identified by asking participants to indicate whether they were ill or if someone close to them was ill or had died due to COVID-19 or non-COVID-19-related reasons; (2) Difficulties with accessing resources was identified by asking participants to indicate whether they had experienced loss of income, and difficulties in accessing necessary supplies, food and usual healthcare including prescription medications and treatments; (3) Conflict was identified by asking participants to report whether they had experienced increased verbal or physical conflict; (4) Separation from family was identified by asking participants to report whether they were separated from family during the pandemic; and (5) Caregiving experience assessed whether participants had spent increased time in caregiving or whether they were unable to care for people who required assistance due to a health condition or limitation.
Statistical analysis
Descriptive analyses were used to demonstrate the characteristics of the study cohort. We first used cross-sectional network analysis to examine relationships between stressors and symptoms at a single time point. The results of cross-sectional network analysis generated contemporaneous networks, provided a snapshot of interconnectedness between the studied variables, and identified central symptoms. CLPN was then used to explore these relationships across time. They captured temporal dynamics and identified core variables that influence future changes within the network. Specifically, Gaussian Graphical Models (GGMs) were employed to visualize the network structures, in which each node represents an observed variable and each edge represents a statistical association between two nodes, typically a partial correlation coefficient (Costantini et al., Reference Costantini, Epskamp, Borsboom, Perugini, Mõttus, Waldorp and Cramer2015). The edge weights reflect the strength of these associations, and the global strength refers to the overall connectivity in the network (i.e., the sum of absolute edge weights). To ensure a parsimonious and interpretable network structure, the graphical least absolute shrinkage and selection operator (LASSO) was used to estimate a sparse partial correlation network by shrinking small coefficients towards zero and setting many to exactly zero, thereby enhancing model interpretability (Friedman et al., Reference Friedman, Hastie and Tibshirani2008). This regularisation approach helps address potential multicollinearity and overfitting. Model convergence was carefully monitored to ensure estimation stability and robustness. The determination of the shrinkage and selection operator degree was based on the extended Bayesian Information Criterion (EBIC) (Epskamp et al., Reference Epskamp, Borsboom and Fried2018). In line with conventional standards in psychological research (Gignac and Szodorai, Reference Gignac and Szodorai2016), we consider coefficients of about 0.10 as small, 0.20 as moderate and 0.30 or higher as large effects. To compare the overall structures of networks and the cumulative strength of connections within networks across groups, the network comparison test (NCT) was conducted using the R package NCT (Van Borkulo et al., Reference Van Borkulo, van Bork, Boschloo, Kossakowski, Tio, Schoevers, Borsboom and Waldorp2023).
CLPN was also conducted to assess how individual items influence each other over time by examining cross-lagged and autoregressive effects. Likewise, to assess the cross-lagged effects of different symptoms, we set the autoregressive paths to zero to emphasize the directional associations between distinct variables across time. This approach aligns with the core aim of CLPN in identifying predictive relationships between nodes while simplifying model complexity (Rhemtulla et al., Reference Rhemtulla and Cramer2022). In addition, we examined two measures of node centrality to determine the relative importance of each symptom within the cross-lagged network: in-prediction, which indicates how much each symptom is predicted by others in the network; and out-prediction, which shows how much each symptom predicts others in the network (Team, Reference Team2013). In addition, to assess the accuracy and stability of the contemporaneous and temporal networks, nonparametric bootstrapping was employed to calculate the 95% confidence interval (CI) around each edge. Additionally, case-drop procedures were performed to compute the correlation stability coefficient (CS-coefficient) of the centrality, using the R package bootnet. CS-coefficients above 0.50 for centrality indices indicate good stability, consistent with recommended thresholds (Epskamp et al., Reference Epskamp, Borsboom and Fried2018). In addition, to examine the robustness of our findings, we conducted a sensitivity analysis in which the autoregressive paths of each node (e.g., depressive symptoms and stressor domains) were freely estimated across time points, instead of being fixed to zero as in the primary model. This allowed us to assess whether the dynamic associations observed in the cross-lagged panel network were influenced by the exclusion of temporal stability components. The network was estimated using the psychonetrics package with the same regularisation and model selection criteria.
The percentage of missing data in the study cohort varied between 0.1% and 12.1%. The missing data pattern was determined to be missing completely at random (MCAR). This means that the probability of data being missing is unrelated to any observed or unobserved data within the study, indicating no discernible pattern in the missingness. Therefore, missing data were imputed using multiple imputations with the mic package (Van Buuren and Groothuis-Oudshoorn, Reference Van Buuren and Groothuis-Oudshoorn2011), and analyses were conducted on the imputed datasets according to best practices. The results reported in the present study were obtained by pooling estimates across multiply imputed datasets, ensuring robust inference and reducing bias associated with missing data. All analyses were conducted using R version 4.1.3.
Results
Characteristics of the study cohort
The study cohort at baseline included 6,136 males and 6,657 females with diverse characteristics. Most participants were between 55 and 64 years (37.0% in males vs. 38.2% in females). In addition, 71.3% of males and 63.3% of females were married or in common-law relationships. Regarding education, a majority had post-secondary degrees or diplomas (83.2% in males vs. 79.7% in females). Income levels also varied, with 5.5% of males and 17.4% of females earning less than $20,000. In terms of retirement status, 40.7% of males and 44.1% of females were completely retired.
Cross-sectional/contemporaneous network
Figures S2 and S3 present the estimated networks of stressors and depressive symptoms for males and females, respectively. Contemporaneous networks for males and females were densely connected with 160 and 162 edges. The stability of the centrality indices indicates that node strength in these two networks was accurately estimated. Stability analyses observed sufficient correlation-stability (CS) coefficient for node strength both networks (males = 0.75; females = 0.75) and for expected influence (males = 0.75; females= 0.75). In the network of males, the nodes with the highest strength centrality were feeling depressed, easily bothered and feeling unhappy. Similarly, feeling depressed, feeling everything is an effort and feeling unhappy were the nodes with the highest strength centrality for the network of females (Figure S4-S5). The NCT analysis found statistical significance in global strength between these two networks (S = 0.48, P = 0.01), with females having a stronger network connection (global strength = 5.29 for females vs. global strength = 4.82 for males). Additionally, significant differences in the overall structures of the networks for males and females were observed (M = 0.10, P = 0.01), suggesting distinct network structures between males and females.
Similarly, cross-sectional networks of stressors and depression symptoms for participants aged less than 65 years old and more than 65 years old were estimated, respectively (Figures S6 and 7). The contemporaneous networks among participants aged less than 65 years exhibited a stronger connection with 160 edges, compared to 150 edges observed among those aged more than 65 years. The stability of the centrality indices suggested that node strength in these two networks was accurately estimated with sufficient strength CS coefficients (CS ≥ 65 years = 0.75; CS < 65 years = 0.75) and expected influence CS coefficients (CS ≥ 65 years = 0.75; CS <65 years = 0.75). The central nodes with the highest strength centrality for those aged less than 65 years were feeling depressed, feeling that everything is an effort and feeling unhappy. For people aged more than 65 years, the nodes exhibiting the highest strength centrality were feeling depressed, easily bothered and feeling that everything is an effort. Details can be found in Figures S8-S9. Additionally, the NCT was performed to compare the differences in the network structures and strength between these two groups. There were statistical differences in the global strength of the connections within these two networks, with participants aged less than 65 years exhibiting more dense connections compared to their counterparts (global strength < 65 years = 5.11, global strength ≥ 65 years = 4.68, S = 0.43, P = 0.01). In addition, we also found statistical differences in the overall structures in the network estimations of different age groups (M = 0.11, P = 0.01), which suggests that the overall structures of these two networks were different.
Temporal/longitudinal network
Overall sample estimation
Longitudinal associations between depressive symptoms and COVID-19-related stressors were tested using the cross-lagged panel network (see Fig. 1). For the total study cohort, the strongest cross-lagged edges were feeling unhappy → hopeless (β = 0.20), feeling everything is an effort → could not get going (β = 0.14) and could not get going→ feeling everything is an effort (β = 0.13). These findings may reflect an emotional decline, where persistent low mood diminishes hope and deepens depressive states. The mutual reinforcement between “feeling everything is an effort” and “could not get going” reflects a vicious cycle of exhaustion and inactivity, where lack of energy leads to withdrawal, which in turn worsens fatigue. Figure S10 displays the centrality estimations from the cross-lagged panel network model. The most impactful nodes with high out-prediction and low in-prediction values were feeling unhappy and could not get going. In contrast, the least influential nodes with low out-prediction and high in-prediction values were easily bothered and separated from family, suggesting they minimally predicted other symptoms in the network but were strongly prospectively predicted by other symptoms. Figure S11 shows that the strongest bridge symptoms between stressors and depressive symptoms for the overall sample are feeling unhappy and feeling fearful or tearful.
Figure 1. Estimated network of COVID-19-related stressors and depressive symptoms in the overall sample.
Sex differences in the temporal/longitudinal networks of COVID-19-related stressors and depressive symptoms
The temporal networks on cross-time effects among males and females are presented in Figs. 2 and 3. For males, feeling unhappy → hopeless (β = 0.18), feeling everything is an effort → could not get going (β = 0.15) and could not get going→ feeling everything is an effort (β = 0.14) had the strongest cross-lagged edges. Similarly, the most impactful nodes with high out-prediction and low in-prediction values were feeling unhappy and feeling depressed. In contrast, the least influential nodes with low out-prediction and high in-prediction values were easily difficult to concentrate and separate from family (Figure S12). The strongest bridge symptoms between stressors and depressive symptoms for males are health-related stressors and feeling unhappy (Figure S13).
Figure 2. Estimated network of COVID-19-related stressors and depressive symptoms among males.
Figure 3. Estimated network of COVID-19-related stressors and depressive symptoms among females.
For females, feeling unhappy → hopeless (β = 0.21), feeling everything is an effort → could not get going (β = 0.15) and could not get going→ feeling everything is an effort (β = 0.12) had the strongest cross-lagged edges (Fig. 3). In addition, the most influential nodes with high out-prediction and low in-prediction values were feeling unhappy and increased verbal or physical conflict. In contrast, the least influential nodes with low out-prediction and high in-prediction values were easily bothered and feeling fearful. Details can be found in Figure S14. Figure S15 depicts that the strongest bridge symptoms between stressors and depressive symptoms for females are feeling unhappy and feeling fearful or tearful. The cross-lagged panel network was re-estimated, allowing autoregressive paths to be freely estimated in males and females. The results indicated that key cross-lagged associations and centrality indices remained largely stable across these analyses, demonstrating that the main findings are robust and consistent across different sexes (Figures S16-S21).
Age differences in the temporal/longitudinal networks of COVID-19-related stressors and depressive symptoms
For individuals less than 65 years old, feeling unhappy → hopeless (β = 0.20), could not get going→ feeling everything is an effort (β = 0.13) and feeling everything is an effort → could not get going (β = 0.12) had the strongest cross-lagged edges (Fig. 4). Figure S22 displays the centrality estimations from the cross-lagged panel network model. The most impactful nodes with high out-prediction and low in-prediction values were feeling unhappy and feeling that everything is an effort. In contrast, the least influential nodes with low out-prediction and high in-prediction values were easily bothered and could not get going. In addition, the strongest bridge symptoms between stressors and depressive symptoms across the network are separation from family and feeling unhappy (Figure S23).
Figure 4. Estimated network of COVID-19-related stressors and depressive symptoms among people aged under 65 years.
For those more than 65 years old, feeling unhappy → hopeless (β = 0.19), feeling everything is an effort → could not get going (β = 0.17) and could not get going→ feeling everything is an effort (β = 0.12) had the strongest cross-lagged edges (Fig. 5). Figure S24 presents the centrality estimations from the cross-lagged panel network model for people aged over 65 years old. Likewise, the most impactful nodes with high out-prediction and low in-prediction values werefeeling unhappy and could not get going. In contrast, the least influential nodes with low out-prediction and high in-prediction values were easily bothered and restless sleep. Additionally, Figure S25 presents that the strongest bridge symptoms between stressors and depressive symptoms are feeling unhappy and feeling fearful or tearful. Similarly, we re-estimated the cross-lagged panel network with autoregressive paths freely estimated for the two age groups (<65 and ≥ 65 years). The results showed that the main cross-lagged relationships and key centrality measures were similar across both groups (Figures S26-S31).
Figure 5. Estimated network of COVID-19-related stressors and depressive symptoms among people aged 65 years and older.
Discussion
This study provides one of the first pieces of evidence on the unique temporality and directionality between COVID-19-related stressors and depressive symptoms in a community-based Canadian longitudinal cohort of middle-aged and older adults. Our results unveil differences in the network structures and strengths of connections between stressors and depressive symptoms across different age and sex groups. Males and females exhibited significant statistical differences both cross-sectionally and longitudinally in global strength, overall structure, the most and least influential nodes and bridge symptoms in the networks of COVID-19 stressors and depressive symptoms. As expected, age differences were also observed in their cross-sectional and longitudinal networks. These findings highlight the importance of targeted interventions addressing age/sex specific- core symptoms and stressors to effectively mitigate the negative impact of stressors during the pandemic.
This study found that individuals less than 65 years old and females exhibited more dense connections compared to their counterparts. In line with previous research, those less than 60 were more susceptible to depression and anxiety symptoms, especially in response to stressful events in the context of the COVID-19 pandemic (Whitehorne-Smith et al., Reference Whitehorne-Smith, Mitchell, Bailey, Agu, Williams, Oshi, Harrison and Abel2021). Especially, young females showed higher levels of depressive symptoms when facing significant stressors such as financial strain and social isolation during the pandemic. This may partly be attributed to their disproportionate caregiving responsibilities and economic pressures during the pandemic (Penna et al., Reference Penna, de Aquino, Pinheiro, Do Nascimento, Farias-Antúnez, Dabs, Mita, Machado and Castro2023).
Notably, feeling unhappy was the only central symptom consistently observed in the networks between sex and age groups. The rest of the central symptoms varied in terms of manifestation and intensity in the networks across groups. For example, feeling depressed was a central symptom for males, whereas increased verbal or physical conflict was a key symptom for females. Previous research has shown that females exposed to physical or verbal abuse were more likely to report higher rates of depressive symptoms (Piccinelli and Wilkinson, Reference Piccinelli and Wilkinson2000). During the pandemic, females reported more stress and anxiety symptoms compared to males, highlighting sex differences in psychological responses to stressors (Arcand et al., Reference Arcand, Bilodeau-Houle, Juster and Marin2023). Moreover, sex-specific coping mechanisms may also play a role in the observed difference. Females are more likely to use less effective coping strategies, such as rumination, which can exacerbate depressive symptoms during the pandemic (Piccinelli and Wilkinson, Reference Piccinelli and Wilkinson2000). Furthermore, feeling that everything is an effort and could not get going were central symptoms for people aged less than 65 years and those aged more than 65 years, respectively. Studies highlight that such symptoms are significant indicators of depressive states and often relate to broader issues such as physical frailty and sarcopenia, especially for older adults (Zhu et al., Reference Zhu, Ding, Zhang, Wang and Chen2024). These two symptoms suggest a deeper sense of perceived fatigue, which is a critical component of depressive states. The exhaustion manifests as a persistent feeling that even the simplest tasks require significant effort and a profound difficulty in initiating activities. In line with our findings, feeling depressed and everything was an effort were identified as the most central symptoms among those around 60 years old in the English Longitudinal Study of Ageing (ELSA) study (Sun et al., Reference Sun, Zhang, Si, Bai, Chen, Lam, Lok, Su, Cheung, Ungvari, Jackson, Sha and Xiang2024). During the pandemic, older adults were required or encouraged to stay home all day; thus, the pandemic-induced lack of motivation has been observed in various aspects of older adults’ lives, including reduced engagement in exercises and social activities (Donovan and Blazer, Reference Donovan and Blazer2020). Their diminished motivation likely hindered their ability to seek and maintain social connections, further exacerbating depressive symptoms while facing additional stressors during the pandemic. Among individuals aged over 65 years, could not get going was identified as a significant symptom. This is consistent with findings observed among Chinese older adults (Zhang et al., Reference Zhang, Wang, Zhou, Dong, Guo, Wang, He, Wang, Wu, Yao, Hu, Wang, Zhang and Sun2023). Such symptoms not only illustrate the physical toll of depression but also underscore the pervasive impact of mental fatigue on the daily functioning and overall well-being of older adults during the pandemic.
Concerning the bridge symptoms between stressors and depressive symptoms, females and individuals aged more than 65 both had feelings of unhappiness and feelings of fear or tearful as the strongest bridge symptoms. In contrast, health-related stressors and separation from family act as key bridge symptoms for males and those under 65, respectively. Consistent with our results, Tsai and Chang also found that perceived health-related stressor was associated with depressive symptoms among older adults. They also observed that long-term high levels of health-related stressors had a more significant impact on depressive symptoms in males compared to females (Tsai and Chang, Reference Tsai and Chang2016). Studies have shown that stressful events impact depressive symptoms differently in older males and older females (Chou and Chi, Reference Chou and Chi2000). Health-related stressors had a greater impact on males aged 60 and older (Kim and Park, Reference Kim and Park2012). This may be partly explained by the traditional roles of males and females. Older males were less likely to seek emotional support or access to healthcare services proactively due to sexed health beliefs and norms around stoicism and self-reliance (Courtenay, Reference Courtenay2000). Additionally, declining physical health in later life may pose a greater threat to identity and autonomy in males, especially if their self-worth is closely tied to functional independence or productivity (Tannenbaum and Frank, Reference Tannenbaum and Frank2011). The fact that males have a shorter life expectancy may also contribute to health-related stressors having greater impacts on them. Moreover, family separation plays a significant role in one’s mental health, leading to increased levels of depression (Tang et al., Reference Tang, Liang, Zhang, Kelifa, He and Wang2021). These findings indicate that health-related stressors and family separation can exacerbate depressive symptoms, particularly among males and individuals aged between 45 and 65, thus acting as bridge symptoms that link depression symptoms.
Strengths and limitations
The findings of the study address the knowledge gap on the dynamic relationships and directionality between COVID-19-related stressors and depressive symptoms in a population-based cohort of middle-aged and older adults. Using a longitudinal cohort design and the advanced CLPN technique, this study illustrates both autoregressive and cross-lagged effects within these relationships , providing insights into the temporal dynamics of symptom networks across sex and age groups.
Additionally, this study elucidates distinct contemporaneous and temporal network relationships for different sexes and age groups. The findings shed light on directing the development and implementation of targeted and specific intervention programmes aimed at mitigating the impact of these studied stressors. These insights are pivotal for informing targeted strategies that address the unique needs of different age and sex groups.
There are limitations to be noted. First, this study focuses primarily on a national cohort of middle-aged and older adults in Canada, which may limit the generalizability of the findings to populations with distinct demographic distributions, such as low- or middle-income countries. Second, depressive symptoms were assessed using the self-reported CES-D questionnaire. Such measures may also induce biases, with participants potentially underreporting or overreporting their symptoms. Third, our analysis was limited by the small sample size in certain groups, such as individuals aged 80 and above, which restricted comparisons due to insufficient statistical power for extreme groups. Future research is still warranted to explore whether those aged 80 or older would have different symptom networks compared to the rest. Finally, our CLPN analysis did not control for potential confounders (such as pre-COVID depressive symptoms, chronic health conditions or socioeconomic status). The CLPN models focus on the dynamics of interconnections between symptoms; they cannot take multiple covariates at a time. Therefore, the findings of CLPN models may be subject to residual confounding. Future research is encouraged to develop a relevant methodology that can incorporate multiple covariates at a time for more precise modelling.
Conclusions
Overall, the study unfolds both cross-sectional and longitudinal relationships and directionality between COVID-19-related stressors and depressive symptoms across sex and age groups. The identified differences in network structures and strengths, as well as the bridge symptoms across groups, could serve as targets when developing and implementing mental health intervention programmes for specific populations. Strategies are warranted to manage health-related stressors among males. Coping strategies and social support for dealing with conflicts should be provided to females. Additionally, issues of family separation should be screened and dealt with among those aged under 65 years. Beyond individual-level interventions, these findings may have direct implications for policy and service delivery. Sex-specific psychosocial screening should be integrated into primary care and community health settings, with resource allocation tailored to the needs of males experiencing health-related stressors, females facing interpersonal conflict and adults under 65 affected by family separation. Preventive programmes that focus on stress management and family support for these vulnerable groups could help reduce the burden of depressive symptoms at the population level. Such measures are able to inform evidence-based mental health policy in the post-pandemic era.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S2045796025100279.
Availability of data and materials
Data are available from the Canadian Longitudinal Study on Aging (www.clsa-elcv.ca) for researchers who meet the criteria for access to de-identified CLSA data.
Acknowledgments
This research was made possible using the data/biospecimens collected by the Canadian Longitudinal Study on Aging (CLSA). Funding for the Canadian Longitudinal Study on Aging (CLSA) is provided by the Government of Canada through the Canadian Institutes of Health Research (CIHR) under grant reference: LSA 94473 and the Canada Foundation for Innovation, as well as the following provinces: Newfoundland, Nova Scotia, Quebec, Ontario, Manitoba, Alberta and British Columbia. This research has been conducted using the CLSA dataset Baseline Comprehensive Dataset (Version 7.0), Follow-up 1 Comprehensive Dataset version 5.0, Follow-up 2 Comprehensive Dataset version 1.0 and the COVID-19 questionnaire study dataset version 1.0, under Application Number 23CA014. The CLSA is led by Drs. Parminder Raina, Christina Wolfson and Susan Kirkland. Funding for support of the CLSA COVID-19 questionnaire-based study is provided by the Juravinski Research Institute, Faculty of Health Sciences, McMaster University, the Provost Fund from McMaster University, the McMaster Institute for Research on Aging, the Public Health Agency of Canada/CIHR grant reference CMO 174125 and the government of Nova Scotia. The opinions expressed in this manuscript are the author’s own and do not reflect the views of the Canadian Longitudinal Study on Aging.
Financial support
This work was supported by a Catalyst Grant: Analysis of CLSA Data, from the Canadian Institutes of Health Research, grant number 23CA014.
Competing interests
The authors declare no competing interests.
Ethical standards
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 2000.
Open access
