The study was based on the Chinese Longitudinal Healthy Longevity Survey (CLHLS), which evaluated the health status and QoL of the older adults aged 65 and older via face-to-face home-based interviews, in randomly selected 23 out of the 31 Chinese provinces from 1998 to 2018 [17, 18]. The study mainly focused on the determinants of healthy aging and mortality in the oldest-old, by analysing aspects such as physical and mental health, socioeconomic characteristics, lifestyle, family dynamics and demographic details of older adults [17, 18]. The data for this cross-sectional analysis were derived from the 2018 wave of the CLHLS, which was released in 2020. All the 15,874 participants aged 65 and older from the CLHLS 2018 wave were included in the study. Following previous research [19, 20], having cataracts was determined using the following standardized question: “Are you suffering from cataracts?” To be eligible, participants without cataract records or complete records of demographic characteristics were excluded, leaving 1683 participants for the present study (see Fig. 1).

Fig. 1.

Flowchart for the selection of the analysed study sample from the Chinese Longitudinal Healthy Longevity Survey (CLHLS).

Socio-demographic information was collected to examine the risk factors for depression and QoL among older adults with cataracts. The data captured included age, gender, education level, marital status, living status, economic status and current smoking and drinking behaviors.

Severity of depression was evaluated using the validated Chinese version of the 10-item Center for Epidemiologic Studies Short Depression Scale (CESD-10), which has been validated in terms of its reliability and consistency across different age groups (Cronbach $\alpha$ = 0.78) [21]. The CESD-10 encompasses the following affective states: ‘Feeling Bothered’, ‘Feeling blue’, ‘Hopelessness’, ‘Feeling fearful’, ‘Unhappiness’, and ‘Loneliness’. Additionally, it comprises three cognitive states (‘Concentration difficulties’, ‘Everything was an effort’, and ‘Inability to get going’) and one sleep state (‘Sleep disturbances’) [13]. Participants rated the frequency of these symptoms over the preceding week on a four-point scale ranging from 0 (‘rarely or none of the time’) to 3 (‘most or all of the time’). The total score of CESD-10 ranges from 0 to 30, and the cut-off value of 10 has been validated in older adult populations to reliably identify partcipants as having depression.

The first two components of the World Health Organization Quality of Life scale – Brief version (WHOQOL-BREF) were extracted to provide a measurement of global QoL [11]. The first two components of the WHOQOL-BREF consisted of overall perception of QoL (item1), and satisfaction with general health facet (item2). Psychometric evaluation in Chinese older adults showed satisfactory properties (Cronbach’s $\alpha$ = 0.86) [22]. Both items were scored on a 5-point scale ranging from 1 to 5, where higher scores indicated better QoL.

Baseline characteristics between participants with depression and those without depression were compared using independent sample Mann-Whitney U tests or Pearson Chi-square tests, as appropriate. Analysis of covariance (ANCOVA) was applied to examine the independent relationship between QoL and depression, after adjusting for significant variables identified in the initial analyses. A two tailed p-value of p 0.05 was considered as statistically significant.

The package qgraph version 1.9.8 was utilized to visualize the network [23]. The network structure was computed using Extended Bayesian Information Criterion (EBIC) combined with the least absolute shrinkage and selection operator (LASSO) [15]. Nodes in the network represent various depressive symptoms or QoL, while each edge between two nodes indicates their association after accounting for the other nodes in the model. Stronger interactions are represented by thicker, more saturated edges. Edges in green denote positive relationships, whereas edges in red indicate negative ones. Mixed graphical models via nodewise regression was used to calculate the prediction ratio of a node based on all its neighboring nodes, which serves as an essential factor in assessing the practical significance of specific edges [24].

Node centrality is crucial for understanding individual node importance within a network model [15]. Centrality index of Expected Influence (EI), which quantifies the influence of a node with both positive and negative edges within a network, can predict how changes in one node relate to changes in others [25]. In the model, high EI nodes are more significant than those with low EI in terms of understanding mental disorder development, persistence, and remission within network theory contexts [25]. The packages bootnet version 1.5.6 was used to compute the above indices [15].

In addition, the “flow” function in the package qgraph version 1.9.8 was employed to designate QoL as a source node and identify direct and indirect connections to other nodes, thus maximizing predictive pathways while accounting for all variables in the model. Node-specific predictive betweenness, was calculated to determine the shortest predictive pathways between QoL and other nodes. Given that betweenness is typically an unstable centrality metric, the variability extent was calculated by both nonparametric and case-drop bootstraps [15].

The packages bootnet version 1.5.6 was employed to assess robustness and replicability of network [14, 15].

The correlation stability-coefficients (CS-coefficient) for EI and strength were calculated to investigate network stability by ensuring a minimum correlation of 0.7 with 95% probability between original and subset samples after the maximum drop proportions of cases were removed from the original sample. The correlation coefficent should not fall below 0.25 and preferably be above 0.5 [15]. Edge weights along with 95% confidence interval (CI) were computed using the non-parametric bootstrapping method, where narrower CIs suggest a more reliable network structure [15]. Additionally, the stability of EI and edge weights was further examined using bootstrapped difference tests [15]. Moreover, the stability analysis of average node-specific predictive betweenness was performed, which provided additional insights under case-dropping conditions.

All the statistical analyses were conducted using R program version 4.3.1 (Foundation for Statistical Computing, Vienna, Austria) [26]. The multivariate imputation by chained equations were carried out using package mice version 3.16.0 [27].

In total, 1683 older adults with cataracts were included, after excluding 14,191 individuals due to incomplete data on cataracts or essential demographic characteristics. Of the included participants, 642 (38.1%) were male, and the average age was 87.4 (standard deviation (SD) = 10.9) years. The prevalence of depression (CESD-10 total score $\ge$10) in older adults with cataracts was 16.5% (95% CI: 14.5–18.3%) in this study sample.

Table 1 shows the differences in the baseline demographic characteristics between participants with and without depression. Older adults with cataracts and concurrent depression were more likely to be female (p = 0.020), and have a lower education level (p = 0.022) and a lower economic level (p 0.001).

Older adults with cataracts and concurrent depression exhibited lower QoL scores (F = 130.1, p 0.001), compared to those without depression, after adjusting for covariates. The logistic regression analysis showed that a higher economic level (odds ratio (OR) = 0.309, p 0.001; OR = 0.13, p 0.001) was significantly linked to a reduced risk of depression (Table 2).

The network structure of depressive symptoms is shown in Fig. 2. The most influential node was CESD3 “Feeling blue” (EI = 1.9), followed by CESD4 “Everything was an effort” (EI = 0.9), and CESD9 “Inability to get going” (EI = 0.4) (Supplementary Table 1). There was significant difference in network structure between the poor economic level and the fair economic level (p of network invariance test = 0.917, p of global strength invariance test = 0.003) (Supplementary Fig. 1).

Fig. 2.

Network structure of depressive symptoms among older adults with cataract.

The flow network model of QoL and depressive symptoms is shown in Fig. 3. Only CESD3 “Feeling blue”, CESD4 “Everything was an effort”, CESD5 “Hopelessness”, CESD7 “Unhappiness” and CESD10 “Sleep disturbances” were directly connected to QoL. Among those, node-specific predictive betweenness from QoL showed that the strongest mediating node between QoL and other depressive nodes was CESD3 “Feeling blue” and CESD7 “Unhappiness” (Fig. 3). Moreover, the strongest linkages were observed in edge QoL - CESD7 “Unhappiness” (edge weight = –0.185), followed by QoL - CESD10 “Sleep disturbances” (edge weight = –0.127), and QoL – CESD3 “Feeling blue” (edge weight = –0.126).

Fig. 3.

Flow network for quality of life and depressive symptoms.

The network’s stability was assessed using the CS-coefficient for strength and EI, both of which were above 0.75, indicating high reliability (Supplementary Fig. 2). The accuracy of the network model was demonstrated by the narrow 95% CIs for estimated edge weights from non-parametric bootstrapping (Supplementary Fig. 3). Moreover, the model’s reliability was confirmed by the significant differences identified in bootstrapped edge weight tests (Supplementary Fig. 4). Average case-drop bootstraps of node-specific betweenness from QoL showed that overall the stability was not highly reliable, but CESD3 “Feeling blue” retained relatively high node-specific betweenness across case-drops (Supplementary Fig. 5).

To the best of our knowledge, this was the first study to explore the network structure of depressive symptoms among older adults with cataracts. We found that depression was common in this population, especially in those who had a poor economic status. In addition, we found that “Feeling blue” was the most central symptom within the network model and “Unhappiness” was the most significant direct association to QoL.

Among 1683 older adults with cataracts included in this study, the prevalence of depression (CESD-10 total score $\ge$10) was 16.5% (95% CI: 14.5–18.3%). In comparison to a recent meta-analysis of 27 studies, the prevalence of depression was 30% among individuals with visual impairment [28]. Our finding is slightly lower than a previous study in China which found that the prevalence of depressive symptoms was 23.9% (95% CI: 19.4–28.4%) among patients with cataracts [29]. The high prevalence of depression might be associated with poor visual acuity in patients with cataracts who did not have cataract surgery [30]. Having severe cataracts was also associated with higher levels of visual disability, poorer quality of life and more severe comorbidities [31], all of which might increase the risk of depression. Similar to previous findings that poorer economic status was associated with higher risk of depression in Chinese older adults [32], we found that older adults with cataracts who had a lower economic status also had an increased risk of depression. This might be attributed to having limited access to medical services due to lack of funds [33], which could lead to worsening cataracts and increasing risk of depression. Moreover, studies have reported that financial support for medical expenses among older adults with chronic diseases, such as cataracts, could reduce the risk of depression caused by economic pressure [34]. Our study further found that economic status had no impact on the network structure of depression. However, since the sample sizes for those with poor and fair economic status were different, this result should be interpreted with caution. The biological mechanism underlying the high risk of depression associated with cataracts remains unclear. It might be mediated by D-amino acids [35], particularly D-asp, which could alter the higher-order structure of lens crystallin proteins, potentially contributing to the development of cataracts [35, 36]. Meanwhile, previous research suggested that D-amino acids and their metabolites are involved in the pathophysiology of depression through the brain–gut–microbiota axis [37]. Further basic research is needed to elucidate the possible mechanisms involved.

In the network model “Feeling blue” was identified as the most influential (central) symptom, which is consistent with previous network analyses [38]. “Feeling blue” is defined as a pervasive feeling of sadness or emotional distress associated with depression [39]. Cataracts in older adults is characterized by clouding of the lens and visual impairment, which is likely to restrict physical mobility, daily activities, independence, and social engagement in older adult [9]. Such restrictions could worsen the feelings of sadness and “feeling blue”, which in turn might lead to the development of depression [9]. On the other hand, previous research revealed that “feeling blue” might constitute a reaction to life changes such as decreased mobility caused by physical illness [40]. In addition, “Everything was an effort” and “Inability to get going” were also the other key central symptoms, both of which might relate closely to cognition, underscoring the relevance of cognitive symptoms in activating and maintaining the network model of depression among older adults with cataracts. These symptoms could also reflect having fatigue or experiencing increased burden as part of depression, which could arise from decreased mobility as a result of impaired vision due to cataracts [9, 40].

In the flow network model assessing the interplay between QoL and depressive symptoms, the most negative correlation identified with QoL was characterized as “Unhappiness”. Prior research found that a lack of happiness could be a pivotal risk factor for having suicidal ideation, poor clinical outcomes, and functional impairment [41], all of which might lower QoL. Furthermore, “Sleep disturbances” was identified as another symptom negatively associated with QoL, which is consistent with prior findings [42] that highlighted the impact of disrupted sleep patterns on QoL in those with major depressive disorder. We also found that “Feeling blue” was another symptom negatively associated with QoL. The activation of depressive symptoms characterized as “Feeling blue” might precipitate a cascade of symptoms in the depression network model, which could further lower QoL. Notably, node-specific predictive betweenness analyses revealed that “Feeling blue” served as a mediator between QoL and other depressive symptoms. This implied a potential sequential activation that decreased QoL by initially triggering “Feeling blue”, and subsequently catalyzing the onset of other depressive symptoms. Given the important role of “Feeling blue” within the network of depression, our findings suggest that interventions targeting this symptom could potentially decrease the progression of other manifestations of depression.

This study had multiple strengths, chief among them being the considerable sample size and the representativeness of the participant group, which enhanced the generalizability of the findings within the defined demographic population. Another strength was the use of novel and sophisticated statistical analyses to examine the inter-relationships between depressive symptoms. However, there were several limitations. First, the reliance on self-reported data for both depressive symptoms and QoL could introduce recall bias, thus affecting the accuracy of the data collected. Secondly, the cross-sectional nature of the study restricted the ability to infer causality between depressive symptoms and QoL. Third, the findings might not apply to populations other than older adults over 65 years with cataracts.