Network analysis of anxiety and depressive symptoms during the COVID-19 pandemic in older adults in the United Kingdom

The health crisis caused by COVID-19 in the United Kingdom and the confinement measures that were subsequently implemented had unprecedented effects on the mental health of older adults, leading to the emergence and exacerbation of different comorbid symptoms including depression and anxiety. This study examined and compared depression and anxiety symptom networks in two specific quarantine periods (June–July and November–December) in the older adult population in the United Kingdom. We used the database of the English Longitudinal Study of Aging COVID-19 Substudy, consisting of 5797 participants in the first stage (54% women) and 6512 participants in the second stage (56% women), all over 50 years of age. The symptoms with the highest centrality in both times were: “Nervousness (A1)” and “Inability to relax (A4)” in expected influence and predictability, and “depressed mood (D1”; bridging expected influence). The latter measure along with "Irritability (A6)" overlapped in both depression and anxiety clusters in both networks. In addition, a the cross-lagged panel network model was examined in which a more significant influence on the direction of the symptom "Nervousness (A1)" by the depressive symptoms of "Anhedonia (D6)", "Hopelessness (D7)", and "Sleep problems (D3)" was observed; the latter measure has the highest predictive capability of the network. The results report which symptoms had a higher degree of centrality and transdiagnostic overlap in the cross-sectional networks (invariants) and the cross-lagged panel network model of anxious and depressive symptomatology.


Ethical considerations
Ethical clearances for ELSA were obtained through the London Multicenter Research Ethics Committee (MREC/01/2/91) 32 .More details on the ethical approval of each phase of ELSA can be found at https:// www.elsa-proje ct.ac.uk/ ethic al-appro val.The study, which involved only existing and anonymized data, was confirmed by the Research Ethics Committee of the Universidad Peruana Unión and did not require additional ethical clearance (Registration Number: 2023-CEUPeU-0014).The study was conducted in accordance with the Declaration of Helsinki.All ELSA participants gave informed consent before being included in the research.

Instruments
Center for Epidemiological Studies-Depression Scale (CES-D) Depression was measured with the CES-D developed by Radloff 33 in its 7-item version (e.g., "Much of the time during the past week: felt sad") under a unidimensional model.The scale is based on responses to seven questions in a dichotomous format, where in this study the values "no" (1) and "yes" (2) were assigned.This allows the generation of a continuous measure that spans a range from 7 to 14, where higher scores indicate higher levels of depressive symptoms experienced during the last week.
Generalized Anxiety Disorder scale (GAD-7) To measure anxiety, the GAD-7 created by Spitzer et al. 34 was used to detect the severity of generalized anxiety symptomatology according to DSM-V criteria over the last 2 weeks.It consists of 7 items (e.g., "Over the last 2 weeks: Feeling nervous, anxious, or on edge") within a unidimensional model, with a Likert-type response mode ranging from 0 to 3 points (never, several days, half of the days and almost every day).The total scale has a minimum score of 0 and a maximum of 21.
The network structure of partial correlations for both models was calculated with the bootnet package 39 , through the huge estimator 44,45 , a non-paranormal conversion of the data 46 and the Rotation of Information Criterion (RIC), was used for model selection 47 .
The stability and accuracy of each network model for each time was calculated through 5000 nonparametric Bootstrap and person-dropping samples, where a coefficient of stability (CS) greater than > 0.5 indicates strong stability and interpretability 48 .
Communities were explored with the spinglass clustering algorithm [49][50][51] through 500 spins at the two times.Centrality was explored through the expected influence step 1 (EI1; which is the sum of the weights of the axes at directly related nodes) and step 2 (EI2; which is the sum of the weights of the axes at indirectly related nodes) 52 .Also included was the bridge expected influence step 1 (BEI1; sum of the weights of the axes connecting each node with the nodes of other directly related communities) and step 2 (BEI2; sum of the weights of the axes connecting each node with the nodes of other indirectly related communities) 41 .
Subsequently, an analysis of the overlapping communities was performed with the cpAlgorithm function in the CliquePercolation package 40 , percolated items were found through the 'weighted CFinder' method 53 ; with a configuration of k = 4 cliques and an intensity of I = 0.08, for the cross-lagged panel network model.
Then, the networks of both times were compared with the NCT function of the NetworkComparisonTest package 42 , with the bonferroni-holm correction technique 54,55 and 1000 permutations.The similarity of the networks is explored through the correlations of the adjacency matrices and their centrality indices, if the result is 1, the networks have a perfect linear relationship, which means that the networks have the same structure; if the correlation is 0, the networks have no detectable linear relationship, so they have no correspondence and if the correlation coefficient is − 1, the networks are exact opposites 56 .
Finally, a two-step cross-lagged panel network (CLPN) analysis was performed for both estimation and summary results.Measures of centrality in the form of in-prediction and out-prediction were also calculated 57 .

Results
Table 1 shows the descriptive results at the item level, in addition to the expected influence and the final bridging expected influence scores.The item "Nervousness (A1)" obtained the highest scores at both times (T1; x = 1.49;SD = 0.77; T2; x = 1.57SD = 0.83).The skewness and kurtosis values exhibit some outliers.Although, for the analyses performed, the confirmation or not of multivariate normality is not essential, the non-normal data transformation method was used to mitigate this particularity.
Figure 2 shows the graph resulting from the CLPN analysis, in which the autoregressions can be observed, as well as the standardized regressions between the nodes.For the first case, the highest scoring autoregressions were "Anhedonia" (D6; β = 0.445) and "Happiness" (D4; β = 0.414), the regression matrix can be found in Appendix 4. On the other hand, the highest scoring inter-node regressions were from "Restless sleep" to "Nervousness (D3 → A1; β = 0.421)" and from "Fatigue" to "Fear (D2 → A7; β = 0.154)".As for the centralities (Appendix 3), the node with the highest in-prediction was "Nervousness" (A1; inPred = 0.38), meaning that this node was the most affected by the others in the model analyzed; while the node with the highest out-prediction was "Restless sleep" (D3; outPred = 0.21), i.e., this symptom was the one that most affected the others in the model.
The network comparison test between times T1 and T2 (Table 2) showed statistical significance in the differences only between network centrality indices (p < 0.05).Likewise, the correlation between adjacency matrices showed linearity (rho = 0.89; p < 0.001), as well as in the centrality indices.Additionally, at the item level, both axis strengths and centralities reported no significant differences.

Discussion
The current study's main objective was to examine and compare depression and anxiety symptom networks during the first (June-July) and second (November-December) period of COVID-19 confinement in a population of older adults in the UK.This observation is of great relevance because the emotional disorders examined were identified as the most common conditions in response to the emergence of COVID-19 in that context 5 .In addition, this finding provided a deeper understanding of how items related to depression and anxiety were interconnected and how they were influenced by stressful events such as social confinement.These results were supported by the confirmation of invariance in the overall structures of both networks.
Network clusters indicate that during the first period of confinement by COVID-19, there was a greater number of symptoms within the anxiety community, and this because the "restless sleep (D3)" item of depression clustered with anxiety symptoms.Based on the above, two premises can be asserted, the first is that this symptom  is associated with both depression and anxiety, as found in previous research 58 , and the second is that older adults in the UK who were evaluated reported restless sleep as an anxious symptomatology due to the stressful context of the COVID-19 pandemic.In addition, the clique-percolation method detected items from the anxiety group that were intertwined with the depression domain, which were: "Lack of control of worry (A2)", "Inability to relax (A4)", and "Irritability (A6)", as in the first state of social isolation by COVID-19, elderly individuals were exposed to greater anxious symptomatology due excessive worry and disorientation about the spread of viruses, the duration of social restrictions, or medical care 3,59 .In relation to the second network, only the item "Irritability (A6)" overlapped with the depression cluster, suggesting a reduced manifestation of anxious symptomatology in older adults and a possible more effective adaptation to social isolation due to COVID-19.This may be related to the use of technological tools to maintain interaction with family and friends, which may have contributed to greater emotional well-being in this group 60 .In addition, it was observed that this same item ("Irritability A6") was also present in the cross-lagged panel network model, which allowed us to infer that irritability was the only anxious symptom that remained stable during the two confinement periods.
In addition, items from the depression cluster were identified that overlapped with anxious symptomatology.It was observed that the symptoms of "Depressed mood (D1)" and "Unhappiness (D4)" were clustered within the anxiety community during the first confinement by COVID-19.These findings are consistent with the research of Kaiser et al. 61 , who also found that the depressive state item showed overlap with the anxiety domain in a group of psychiatric patients.Additionally, it is important to keep in mind that unhappiness is not an exclusive criterion of depression.For older adults with anxious symptomatology, they are likely to experience a sense of unhappiness due to constant worry about the possible consequences of the COVID-19 pandemic.These concerns may include health complications, lack of social interaction, and lifestyle changes, which contribute to their perception of unhappiness 62,63 .While in the second time of confinement by COVID-19, there was a greater number of overlapping items between both clusters, such as "Depressive mood (D1)", "Restless sleep (D3)", and "Anhedonia (D6)".Therefore, the presence of more pronounced comorbid symptoms with anxiety during this period may be attributed to the fact that older adults experienced greater uncertainty regarding the prolongation of social isolation during the second COVID-19 confinement, a situation like that experienced during the first quarantine 59,64 .It is important to recognize that the "Depressive mood (D1)" remains persistent in both evaluation periods and in the cross-lagged panel network model, which implies that it is a constant transdiagnostic symptom in both stages.In addition, this symptom has an overlapping presence in psychological distress (anxiety and depression) in the form of bridges, indicating an increased risk of connection with other symptoms in the network.This pattern is maintained in response to the effects of both COVID-19 quarantine periods.Additionally, in the crosslagged panel network model having "Sleep problems (D3)" had the largest network effect toward nervousness.
After a review of the literature, only two studies have been found that use the Clique-Percolation methodology to learn about overlapping symptoms within the anxiety and depression communities 30,61 .These studies differ from our findings because they report a greater number of groupings, without any diagnostic justification, and refer to symptoms that do not belong to any symptomatological domain.While in the current study, the two symptomatological clusters of anxiety and depression are maintained, which are necessary to identify because, from a transdiagnostic perspective, there are symptoms and causes that are shared equally for both disorders 65,66 .This is important since excessive clustering of symptoms in different domains can lead to a lack of clarity in the representation of the two mental health diagnoses considered by the DSM-V, which are represented as the interaction of all their signs and symptoms together within a given period 67 .The results refer to greater precision and interpretation based on transdiagnostic models between two emotional disorders such as anxiety and depression 28,68 , given that the methodology used allows studying these mental health problems as systems 17,69,70 .
Network findings point out that common components (items) underlying both disorders of "depressed mood" and "irritability" may arise in the face of emotional processing deficits 71,72 .Specifically, these symptoms may be related to inadequate regulation of these emotions and other negative feelings such as fear and unhappiness 73,74 .Deficits in this regulatory process may be the main cause of transdiagnostic (overlapping) symptoms reinforcing each other 89,92, which may interact jointly with other symptoms in the network with greater intensity and duration, e.g., they may reinforce insomnia and worry that are linked to memory problems in older adults 75 .On the other hand, the cross-lagged panel network shows that the symptom of depressed mood is related to a greater degree of lack of control of worry, which is reinforced by a greater feeling of loneliness.In addition, it was observed that the item "Inability to relax (A4)" within the anxious symptomatology had a higher centrality (expected influence) in the networks during both periods of confinement by COVID-19.This symptom showed a strong connection with "Lack of control of worry (A2)", "Nervousness (A1)", and "Excessive worry (A3)".www.nature.com/scientificreports/These findings are in accordance with the predictability values, given that the first four anxiety items presented the highest degree of explained variance (R2); therefore, both centrality and predictability allowed us to understand which items are of greater relevance and relative importance in the interaction of symptoms within the depression and anxiety network.Such results may be explained by the fact that the pandemic and social isolation generated a distressing and uncertain environment in elderly individuals 76,77 , who were considered a population at risk for the increase in the number of COVID-19 infections and deaths 9 .Studies conducted in China during the first quarantine were reported, with similar findings highlighting the centrality of such anxiety symptoms in distress symptomatology network with young 46 , and adult participants 78 .This is also consistent with other research conducted in participants from Norway, the United States, the UK, and Australia 79,80 .Previous work only considered one period, while the current study evaluated two networks in two periods of COVID-19, where there was no significant difference between the item centrality values of both networks using the NCT 55 .Therefore, the inability to relax can be considered an essential symptom in the manifestation of anxiety and depression in older adults in the UK because of the impact of COVID-19.
Regarding depressive symptoms, it was identified that, in the first network, the item "Depressed mood (D1)" was the most central and influential, given that it maintained high intensity relationships with "Fatigue symptoms (D2)", "Unhappiness (D4)", and "Loneliness (D5)".During this period, COVID-19 mandatory confinement status had been initiated in multiple countries, including the UK, which facilitated isolation and social distancing in various households and families 1 .This loss of social contact came to increase feelings of sadness, loneliness, and depression in older adults due to limited visits and gatherings by friends and family 81 .This measure also presented a higher index of bridge expected influence, i.e., a high level of depressed mood may act as one of the nodes of greater connection between the symptomatological domains (anxiety and depression) of the network.This is in line with other research in patients with epilepsy who were attending a hospital outpatient clinic in China, where they found that the depressed mood item was the most central and bridging symptom within a network of anxiety and depression 82 .
In the second network the results were similar with respect to depressive symptoms, where "Depressed mood (D1)" had higher EI and BEI centrality, these two measures are used to understand the importance and impact of nodes in a network globally and on a cluster basis 48 .Therefore, it is likely that older adults who experience high levels of depressive feelings are more vulnerable to exhibit other symptoms of depression to a greater degree.One example is that depressed mood and fatigue may elicit more pronounced anxiety responses in this age group.This could be because, due to their age and state of health, it is common for them to experience both physical and emotional exhaustion.These factors could contribute to increased concern for the future and for their own health, intensifying anxiety responses 83,84 .
After outlining the overlap and centrality of the items, it is important to consider the strongest relationship between symptoms of both emotional disorders.Based on the above, the findings showed that the symptom of "Restless sleep (D3)" and "Inability to relax (A4)" were more related in the two confinement times by COVID-19.In studies published during the early period of the pandemic, positive relationships were found to exist with such symptoms in the anxiety and depression network in Chinese university students 85 , as well as in Norwegian university students 80 .In the second COVID-19 confinement, there were also studies showing that restless sleep was directly related to difficulty relaxing in health care professionals during the last stage of the pandemic (October 2020) in China 86,87 .This may be since the inability to relax alters the reconciliation of sleep, since it increases or reduces the hours of deep sleep, in addition, it produces a poor night's rest that generates greater feelings of tiredness during the day and a greater nervousness 88,89 .
In the cross-lagged panel network spanning the two pandemic waves, a clear pattern of influence directed toward the symptom "Nervousness (A1)" is observed.This influence comes mainly from the depressive symptom related to "Sleep problems (D3)", as well as other symptoms such as "Anhedonia (D6)" and "Hopelessness (D7)".These symptoms also exerted a direct effect on the symptom of nervousness in the network.This dynamic may be explained by the impact of the COVID-19 pandemic, which caused more pronounced difficulties in falling and staying asleep in older adults.This situation generated fatigue and exhaustion, which in turn increased the propensity to nervousness, due to the difficulties to effectively regulate emotions in this age group 90 .In addition, sleep problems are intensified when the individual experiences a loss of pleasure from performing previously enjoyable activities, as well as hopelessness about life 8 .

Limitations
A limitation of this study lies in the fact that it addressed symptoms of depression and anxiety in older adults in the UK exclusively through surveys.Consequently, it would be important for future research to incorporate diagnostic evaluations to include clinical participants who present with specific mental problems, thereby broadening the understanding of these conditions in this population.Another limitation of this study lies in the analysis limited to the periods of confinement due to COVID-19 at the first and second moment.Therefore, it is essential to continue to investigate and examine these variables in subsequent periods, with the aim of strengthening the understanding of the stability of core and comorbid symptoms over time.Finally, it is important to note that few studies employ the Clique-Percolation approach in psychometric networks to represent clinical variables through clusters that share common underlying components.These models capture fundamental elements in transdiagnostic models of mental health, and their application could provide valuable insight into future research.

Conclusion
In conclusion, this study evidences the presence of items that share manifestations between the depression and anxiety communities in older adults during the first and second confinement in the UK.In particular, "Irritability (A6)" and "Depressed mood (D1)" were identified as symptoms with prominent comorbidity.In addition, the depressive symptom exhibited a higher centrality of interconnectedness between clusters in both evaluation periods.In the cross-lagged panel network, it was observed that the "Nervousness symptom (A2)" was the most predictable measure in this structure.In addition, the item related to "Sleep problems (D3)" was identified as having a significant predictive impact, according to centrality indicators over time.These results also reaffirmed the presence of a transdiagnostic overlap between irritability and depressed mood.