Content uploaded by Maarten C Eisma
Author content
All content in this area was uploaded by Maarten C Eisma on Feb 04, 2020
Content may be subject to copyright.
Journal Pre-proof
Bereavement or Breakup: Differences in Networks of Depression
Julian Burger , Margaret S Stroebe , Pasqualina Perrig-Chiello ,
Henk A.W Schut , Stefanie Spahni , Maarten C Eisma ,
Eiko I Fried
PII: S0165-0327(19)33045-9
DOI: https://doi.org/10.1016/j.jad.2020.01.157
Reference: JAD 11606
To appear in: Journal of Affective Disorders
Received date: 1 November 2019
Revised date: 10 January 2020
Accepted date: 26 January 2020
Please cite this article as: Julian Burger , Margaret S Stroebe , Pasqualina Perrig-Chiello ,
Henk A.W Schut , Stefanie Spahni , Maarten C Eisma , Eiko I Fried , Bereavement or
Breakup: Differences in Networks of Depression, Journal of Affective Disorders (2020), doi:
https://doi.org/10.1016/j.jad.2020.01.157
This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition
of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of
record. This version will undergo additional copyediting, typesetting and review before it is published
in its final form, but we are providing this version to give early visibility of the article. Please note that,
during the production process, errors may be discovered which could affect the content, and all legal
disclaimers that apply to the journal pertain.
©2020 The Authors. Published by Elsevier B.V.
This is an open access article under the CC BY-NC-ND license.
(http://creativecommons.org/licenses/by-nc-nd/4.0/)
Running head: DIFFERENCES IN NETWORKS OF DEPRESSION
Highlights
Patterns of depression can be linked to specific adverse life events
Network analyses allow exploring differences in symptom relations after life events
Differences in symptom patterns give rise to tailored treatment hypotheses
Widowhood/separation are primarily linked to loneliness, followed by other
symptoms
DIFFERENCES IN NETWORKS OF DEPRESSION
2
Bereavement or Breakup: Differences in Networks of Depression
Julian Burger, Margaret S. Stroebe, Pasqualina Perrig-Chiello, Henk A.W. Schut, Stefanie
Spahni, Maarten C. Eisma and Eiko I. Fried
Author Note
Julian Burger, Department of Psychiatry, University Medical Center Groningen,
Groningen, the Netherlands and Department of Psychological Methods, University of
Amsterdam, Amsterdam, the Netherlands; Margaret S. Stroebe, Department of Clinical
Psychology, Utrecht University, Utrecht, the Netherlands and Department of Clinical
Psychology and Experimental Psychopathology, University of Groningen, Groningen, the
Netherlands; Pasqualina Perrig-Chiello, Department of Developmental Psychology,
University of Bern, Bern, Switzerland; Henk A.W. Schut, Department of Clinical
Psychology, Utrecht University, Utrecht, the Netherlands; Stefanie Spahni, Department of
Health Psychology and Behavioral Medicine, University of Bern, Bern, Switzerland; Maarten
C. Eisma, Department of Clinical Psychology and Experimental Psychopathology, University
of Groningen, Groningen, the Netherlands; Eiko I. Fried, Department of Clinical Psychology,
Leiden University, Leiden, the Netherlands.
Correspondence concerning this article should be addressed to Julian Burger,
Department of Psychiatry, University Medical Center Groningen, Hanzeplein 1, 9713
GZ Groningen, The Netherlands. Contact: j.burger@uva.nl
DIFFERENCES IN NETWORKS OF DEPRESSION
3
Abstract
Background
Prior network analyses demonstrated that the death of a loved one potentially precedes
specific depression symptoms, primarily loneliness, which in turn links to other depressive
symptoms. In this study, we extend prior research by comparing depression symptom
network structures following two types of marital disruption: bereavement versus separation.
Methods
We fitted two Gaussian Graphical Models to cross-sectional data from a Swiss survey of
older persons (145 bereaved, 217 separated, and 362 married controls), and compared
symptom levels across bereaved and separated individuals.
Results
Separated compared to widowed individuals were more likely to perceive an unfriendly
environment and oneself as a failure. Both types of marital disruption were strongly linked to
loneliness, from where different relations emerged to other depressive symptoms. Amongst
others, loneliness had a stronger connection to perceiving oneself as a failure in separated
compared to widowed individuals. Conversely, loneliness had a stronger connection to
getting going in widowed individuals.
Limitations
Analyses are based on cross-sectional between-subjects data, and conclusions regarding
dynamic processes on the within-subjects level remain putative. Further, some of the
estimated parameters in the network exhibited overlapping confidence intervals and their
order needs to be interpreted with care. Replications should thus aim for studies with multiple
time points and larger samples.
Conclusions
DIFFERENCES IN NETWORKS OF DEPRESSION
4
The findings of this study add to a growing body of literature indicating that depressive
symptom patterns depend on contextual factors. If replicated on the within-subjects level,
such findings have implications for setting up patient-tailored treatment approaches in
dependence of contextual factors.
Keywords: Depression, Divorce, Network Analysis, Bereavement, Marital Disruption
DIFFERENCES IN NETWORKS OF DEPRESSION
5
Bereavement or Breakup: Differences in Networks of Depression
1. Introduction
1.1 Marital transition and mental health
One of the most well-known wedding vows suggests a long-term perspective on a
relationship, with death being the only cause for its termination: ―Till death do us part.‖
Demographic data, however, suggest that the end of a marriage is not always marked by the
death of a partner. Marital disruption, the termination of a marriage due to separation or
divorce, has been well-established as a frequent life event. In the USA, the probability that a
first marriage is still intact after 20 years has been calculated at approximately 52% for
women and 56% for men aged 15–44 (Copen et al., 2012).
Both spousal loss and separation are associated with major psychological distress,
increasing the risk of severe long-term detriments to well-being and health. One of the most
frequent consequences of spousal loss and separation are mood-related disorders, and more
specifically, depression (Sbarra, 2015; Wójcik et al., 2019). The Diagnostic and Statistical
Manual of Mental Disorders 5 (DSM-5; American Psychiatric Association, 2014)
characterizes depression through nine criteria, namely, depressed mood, diminished
interest/pleasure, weight/appetite increase/decrease, insomnia/hypersomnia, psychomotor
agitation/retardation, fatigue, feelings of worthlessness or inappropriate guilt, lack of
concentration or indecisiveness, and suicidal ideation. The presence of at least five of the
symptoms (at least one of which have to be either sad mood or anhedonia) qualifies for the
diagnosis Major Depressive Disorder (MDD). Taking into account all possible combinations
of sub-symptoms, this results in over 10,000 hypothetical symptom combinations for the
same diagnosis, and empirical studies have observed that many of these are realized in
patients with a diagnosis of MDD (Fried & Nesse, 2015; Zimmerman, Ellison, Young,
Chelminski, & Dalrymple, 2015). Crucially, different life events have been associated with
DIFFERENCES IN NETWORKS OF DEPRESSION
6
differences in depressive symptomatology (Cramer, Borsboom, Aggen, & Kendler, 2012;
Fried, Nesse, Guille, & Sen, 2015). Based on this finding, the present study uses a network
approach to investigate whether the two types of loss introduced above are differentially
related to depression symptoms.
1.2 The network perspective to depression following bereavement
The network approach to psychopathology conceptualizes symptoms and other factors
of mental health as causally interacting entities (Borsboom and Cramer, 2013). Network
analyses have been applied to the field of bereavement, through the study of depression and
complicated grief symptoms (Robinaugh et al., 2016, 2014) and their interrelations (Djelantik
et al., 2019; Malgaroli et al., 2018). Specifically, as discussed above, Fried et al. (2015) fitted
several models to a dataset to compare elderly bereaved versus still-married participants.
Loneliness was much more strongly related to spousal loss than other depression symptoms,
and in turn was associated with a host of other symptoms. We aim to extend this finding to
compare the effects of spousal loss to marital breakup.
1.3 Bereavement versus breakup
There are reasons to assume differences in the symptom dynamics of depression
following spousal bereavement versus marital breakup. Wrzus, Hänel, Wagner, and Neyer
(2013) classify widowhood as an expected life event, usually accompanied by a supportive
social environment, especially after an initial phase of social withdrawal. Bereavement is
predominantly associated with feelings of grief over the loss of the loved person, alongside a
variety of related manifestations (Stroebe et al., 2017). While stigmatizing responses towards
bereaved individuals with a diagnosis of prolonged grief disorder have been experimentally
demonstrated (Eisma, 2018), conclusive evidence regarding the prevalence of stigmatization
in spousal loss is scarce; a systematic review of social support in bereaved individuals found
that most studies conducted on this issue face several methodological and sampling
DIFFERENCES IN NETWORKS OF DEPRESSION
7
limitations (Logan et al., 2018). In a previous network study, Fried and colleagues (2015)
found that people who had lost a loved one primarily developed loneliness over other
depressive symptoms; loneliness, in turn, was related to a host of other depressive symptoms.
The authors speculated that loneliness might thus be a gateway symptom which prevention
strategies for depression could focus on to disrupt relations with other symptoms following
spousal loss.
While one can make similar predictions about loneliness following marital breakup
(especially perhaps for those who did not initiate the separation, cf. Hewitt and Turrell,
2011), other symptoms of depression would seem likely to be important as well.
Wrzus et al. (2013) noted that separation (specifically: divorce) can be especially stressful
due to the reduction in a person‘s social network, through the partial loss of in-laws and
spouse‘s friends. Given that breakup is associated with adverse interpersonal relationship
experience (Sbarra, 2015), items representing the perceived negative opinions and social
responses of others might thus be as or even more apparent, compared to loneliness.
Measures of depression include relevant items; the CES-D (Radloff, 1977) items ―I thought
my life had been a failure‖ and ―People were unfriendly‖ (in the following referred to
as failure and unfriendly, respectively) thus arguably capture the experience of breakup better
than bereavement.
Following these contrasts in marital transition, crucial differences in the nature of
mental health-related difficulties could be expected: For bereaved individuals, one could
argue that loneliness as a consequence of spousal loss (Fried et al., 2015) is accompanied
with symptoms related to grief work. Separated individuals on the other hand are more liable
to evaluate their life plan as a failure, with their social environment often compounding this
due to lack of support and/or understanding (Wrzus et al., 2013).
1.4 The current study
DIFFERENCES IN NETWORKS OF DEPRESSION
8
We estimated network models and compared symptom levels following widowhood
and separation, compared to a still-married sample and tested three hypotheses:
H1. CES-D sum-scores are higher among both bereaved and separated individuals
compared to married individuals.
H2. Separated individuals show higher levels of failure and unfriendly compared to
widowed individuals.
H3. Both loss types are primarily linked to loneliness, which in turn is associated with
other CES-D symptoms.
A note on exploratory analyses. Network analysis at present is largely used to gain
exploratory insight into multivariate dependencies. These structures can generate hypotheses
about putative causal relations. To this end, we extend our investigation to interesting
relations that have not been hypothesized. These exploratory analyses are distinguished from
our confirmatory findings (the latter include the respective hypothesis in brackets). Most
importantly, we are interested in how loneliness is differentially related to other CES-D
symptoms, comparing bereaved with separated individuals.
2. Methods
2.1 Participants
We analyzed data from the Swiss project ―Relationships in later life‖
(http://www.kpp.psy.unibe.ch/forschung/projekte/nccrlives/index_ger.html). In this project,
information on marital transitions and related mental health components were collected over
three waves (2012, 2014, and 2016). The Swiss Federal Statistical Office identified a random
sample (stratified by gender, age, and marital status) of 6889 married, widowed, divorced and
separated individuals aged 40 – 90. These individuals subsequently received letter mail with
an invitation to the study and the paper-and-pencil questionnaire. Additionally,
advertisements were placed on various platforms (radio, newspaper, and online). Participants
DIFFERENCES IN NETWORKS OF DEPRESSION
9
were informed regarding the purpose of the prospective longitudinal data-collection (changes
and stability of relationships in later life). In total, data on 1276 married, 566 widowed, 721
divorced, and 250 separated individuals were collected, from which we derived two marital
status sub-samples. A schematic overview of the sampling procedure in this study can be
seen in Figure 1.
[Figure 1]
2.1.1 Widowed and separated individuals. We sampled widowed and separated individuals
from all three waves, if they met two inclusion criteria: First, the loss/breakup occurred
within two years prior to assessment, and second, the widowed/separated person did not have
a new partner at the time of assessment.
The former criterion was chosen on the basis of two considerations: On the one hand,
due to the way data was collected (time distance of two years in between waves), extending
the time criterion to more than two years would mean that participants who experienced
loss/breakup more than two years prior to wave 2 and 3 would be sampled multiple times
(from several waves). On the other hand, decreasing time-intervals to less than two years
would have led to rather low sample sizes in the present dataset. We therefore faced a trade-
off between statistical power and capturing experiences in close approximation to the life
event, and opted for a compromise of two years. We hope that future research will investigate
effects of different time distances to the life event to capture both, adaptation over longer
periods including more complex processes of loss and depression, as well as experiences in
close approximation to the life event).
The second criterion was chosen to account for protective influences that a new
partnership might have on an individual‘s grief (de Jong Gierveld, 2004). This resulted in 145
widowed and 217 separated individuals.
[Table 1]
DIFFERENCES IN NETWORKS OF DEPRESSION
10
2.1.2 Samples for network analysis. We see two main possibilities for constructing
networks to tackle our research questions: a) adding married participants as controls/contrast
to both the widowed and the separated sample, and estimating two networks for the
respective samples (using a similar logic to Fried et al. 2015), or b) estimating three separate
networks for the three groups widowed, separated and married. The main difference between
these approaches is that the networks estimated in method a) allow us to include the life event
as a node in the network, which is not possible for networks estimated in method b). This is
because in method b), the samples are set up in a way that each participant experienced the
same life event within one sample. The variable ‗life event‘ thus has no variance,
consequently making it impossible to estimate (partial-)correlations between the life event
and other variables.
Since the focus of our analysis is to examine differences in how widowhood and
separation are (differentially) related to depressive symptoms, we estimated two networks
according to option a), while providing the networks resulting from the estimation method b)
in the supplemental material (Figure S1). The networks estimated according to method b) can
be relevant in focusing on structural differences of depressive symptoms within each sample,
if relations to the life event are not of interest. Accordingly, we randomly sampled 362
married controls who did not previously experience spousal loss or separation/divorce, and
constructed two samples that were then used to estimate the networks. The first sample
consisted of the 145 widowed individuals introduced above combined with 145 married
controls, the second sample of 217 separated individuals combined with the remaining 217
married controls. Table 1 compares demographic characteristics across the widowed,
separated and married sample.
We decided to sample married controls randomly as opposed to making use of
matching procedures, since several demographic variables of interest had many missing
DIFFERENCES IN NETWORKS OF DEPRESSION
11
observations. To ensure that estimated network structures were not dependent on the seed
chosen to sample married controls, we repeated the sampling procedure four times with other
random seeds, and correlated the adjacency matrices of the resulting network with the one
discussed below. Correlations ranged from .89 to .92 for the widowed, and from .92 to .94 for
the separated network, indicating that the network structures had high consistency for
different compositions of the married sample.
2.2 Outcome measures
Depressive symptoms were assessed using the German short version of the Center for
Epidemiologic Studies Depression scale (CES-D; Radloff, 1977; German: Allgemeine
Depressions-Skala, ADS-K; Meyer and Hautzinger, 2001). Participants rated 15 items with
respect to the frequency with which they occurred in the last week, with the four response
categories ―rarely or none of the time (less than 1 day)‖, ―some or a little of the time (1-2
days)‖, ―occasionally or a moderate amount of time (3-4 days)‖ and ―most or all of the time
(5-7 days)‖. The German version of the CES-D has been found to be reliable with
Cronbach‘s Alpha between .89 and .92 (Hautzinger and Geue, 2016). In line with these
findings, we obtained a Cronbach‘s alpha of .90 for our study sample. While the CES-D is
used as a screening-tool and does not allow to determine diagnostic status, it provides useful
information regarding our proposed differences in comparison to other scales. Specifically,
the CES-D items ―I thought my life had been a failure‖ and ―People were unfriendly‖ are
relevant to investigate the above discussed differences in social support and evaluation of
one‘s life.
One major challenge in the extant network literature in psychopathology is that some
items modeled in networks might measure the same construct (Fried & Cramer, 2017). This
poses a problem for inferences because edges in network models should only be interpreted
as putative causal relations if the nodes are indeed distinct entities. At present, there are no
DIFFERENCES IN NETWORKS OF DEPRESSION
12
clear guidelines to differentiate between a correlation that arises from items measuring the
same construct and a correlation due to two items being related, but originating from distinct
constructs. Since purely data-driven approaches cannot account for theoretical
considerations, we combined items if they met two criteria. Items were combined if the items
showed correlations of r ≥ .50, and if the items could be understood to measure the same
construct. Accordingly, we combined the items mood, upset and depressed into the new item
mood, and happy and enjoy into the new item happy, resulting in 12 instead of 15 items. The
final list of items is presented in the supplemental materials, Table S1. The item-pairs
depressed – concentration, concentration – exhausted, lonely – mood, lonely – depressed,
sad – depressed, getgo – depressed, getgo – exhausted and lonely – sad all exhibited
correlations of r ≥ .50, however, for the purpose of this paper, we understand them as
theoretically separate constructs.
2.3 Statistical analyses
2.3.1 Symptom level comparison. Prior to the network analyses, widowed, separated
and married individuals were compared with respect to differences in the item sum-score
using a one-way ANOVA and post-hoc tests. Furthermore, overall differences with respect to
specific symptoms were analyzed in a MANOVA and symptoms were examined individually
with respect to group differences.
2.3.2 Network analysis. Following the group comparisons, we estimated two
separate networks. Both networks consisted of the combined set of 12 CES-D items and one
node to the life event (network 1: spousal loss versus marriage, network 2: marital breakup
versus marriage). We estimated regularized partial correlation networks (Epskamp and Fried,
2018) based on Spearman‘s rank correlation, due to the ordinal nature of items. We chose
Spearman correlations over polychoric correlations, since polychoric correlations led to
highly unreliable parameter estimates; as explained elsewhere (Epskamp et al., 2018), this
DIFFERENCES IN NETWORKS OF DEPRESSION
13
can happen when the sample size is small, items have few response options, and are
considerably skewed. To account for potential spurious relations, we used a regularization
approach with the tuning parameter γ (specifying the level of sparsity) set to 0.5 (Foygel and
Drton, 2010). Recent literature suggests that non-regularized networks might be preferable in
some cases, especially for very large sample sizes (Williams et al., 2019). Since this is not the
case for our sample, we present the non-regularized partial correlation networks in the
supplemental material (Figure S2).
It is good practice to determine the accuracy and stability of estimates and inferences
in the networks. To this end, we conducted the stability/accuracy routine using the bootnet
package in R described elsewhere (Epskamp et al., 2018). The networks were estimated using
the bootnet and the qgraph package (Epskamp et al., 2012). Additionally, we compared the
two networks using the NetworkComparisonTest (van Borkulo et al., 2015). Since this
procedure might yield biased results if the network samples are unequal in size (van Borkulo
et al., 2017), we additionally correlated the weight matrices to obtain a measure of similarity,
and subtracted the weight matrices to examine the largest absolute differences between edge
weights.
Contrary to many network analyses conducted in the field of psychopathology, we did
not calculate centrality measures for our networks. Most centrality measures are metrics
based on summarizing edge weight information in respect to a given node, degree centrality
for instance is calculated by summing all absolute edge weights going into a node. Our
networks are composed of both, CES-D items and a node coding a life event, consequently
making the interpretation of centrality measures as indicative of central to the network of
symptoms problematic. This is because centrality metrics in our case would favor items that
exhibit large relations to the life event over items that are unrelated to the life event. For that
reason, we focused on comparing specific edges rather than centrality measures.
DIFFERENCES IN NETWORKS OF DEPRESSION
14
3. Results
3.1 Symptom level comparison
3.1.1 Sum-Score and diagnosis of depression. Widowed (n = 145), separated (n =
217) and married (n = 362) individuals differed in their overall CES-D sum-score, F(2, 609)
= 52.93, p < .001, Cohen‘s f = 0.34. More specifically, sum-scores of married individuals
(Mmar = 6.67, SDmar = 6.07) were lower than those of widowed individuals (Mwid = 11.65,
SDwid = 6.72; t(194.50) = 6.98, p < .001, Cohen‘s d = 0.78, CI [3.58, 6.39]) and separated
individuals (Msep = 13.47, SDsep = 9.91; t(293.48) = 8.62, p < .001, Cohen‘s d = 0.83, CI
[5.24, 8.35]), but the widowed and separated groups did not differ from each other (t(306.52)
= 1.93, p = .055, Cohen‘s d = 0.21, CI [-3.66, 0.04]), supporting our first hypothesis (H1).
While the CES-D does not allow for determining diagnostic status, prior psychometric
analyses (Lehr et al., 2008) suggested a score of 18 for a putative diagnosis. Following this
cutoff, 6.04% of the married, 17.95% of the widowed and 29.95% of the separated
individuals met the screening criterion of the scale.
3.1.2 Differences in specific symptoms. A MANOVA revealed overall differences
between widowed (n = 145) and separated (n = 217) individuals with respect to specific CES-
D items, T2(12, 301) = 4.91, p < .001. In particular, as can be seen in Figure 2, differences
emerged only for specific symptoms.
As hypothesized (H2), and after accounting for multiple-testing using Bonferroni-
correction, separated individuals showed higher levels of failure (t(343) = 5.56, p < .001,
Cohen‘s d = 0.58 , CI [.27, .57]) and unfriendly (t(343) = 3.59, p < .001, Cohen‘s d = 0.36, CI
[.09, .30]) compared to widowed individuals. Furthermore, there were differences for the
symptoms afraid (t(345.98) = 3.17, p = .002, Cohen‘s d = 0.33, CI [.10, .41]; separated >
widowed) and mood (t(319.35) = 3.03, p = .003, Cohen‘s d = 0.33 , CI [.09, .43]; separated >
widowed).
DIFFERENCES IN NETWORKS OF DEPRESSION
15
Some other symptoms indicated significant differences between separated/widowed
individuals (exhaust, t(318.96) = 2.78, p = .006, Cohen‘s d = 0.30 , CI [.08, .45], separated >
widowed; sleep, t(321.96) = 2.04, p = .043, Cohen‘s d = 0.22, CI [.01, .38], separated >
widowed; happy, t(281.39) = 2.60, p = .010, Cohen‘s d = 0.28, CI [.07, .47], separated >
widowed), however these did not remain significant after controlling for multiple testing.
Given that some of these p-values were close to the traditional significance threshold of 5%,
we want to call for caution in interpreting these effects as either clear positive or negative
effects (Amrhein et al., 2019); more conclusive evidence will require replicating our study.
[Figure 2]
3.2 Network analysis
3.2.1 Network accuracy and stability. Graphical results of the stability and accuracy
analysis can be found in the supplemental materials (Figure S3-S5). In general, the edge
weights exhibit rather large confidence intervals, and some of the lower absolute edge
weights do not differ significantly from other edges, indicating that the order of edges should
be interpreted with some caution.
3.2.2 Network inferences. Figure 3 shows the estimated networks for the
widowed/married (a, left) and the separated/married (b, right) sample.
Widowhood. As hypothesized (H3), and in line with prior findings of Fried et al.
(2015), experiencing spousal loss was primarily associated with loneliness (partial correlation
of r = .30), and additionally with sadness (r = .26). In turn, loneliness was linked to several
CES-D symptoms (sorted by decreasing partial-correlation): talk (r = .17), getgo (r = .16),
mood (r = .11), afraid (r = .09), happy (r = – .06), and failure (r = .06). In contrast to Fried et
al. (2015), this analysis additionally revealed a strong direct relation between spousal loss and
sad (r = .22) and weaker associations with unfriendly (r = – .01) and happy (r = – .01).
DIFFERENCES IN NETWORKS OF DEPRESSION
16
Separation. As hypothesized (H3), and similar to the widowed network, separation
was also strongly linked to loneliness (r = .33). Loneliness was in turn associated with other
CES-D symptoms (sorted by decreasing partial-correlation): sad (r = .29), failure (r = .16),
mood (r = .14), talk (r = .10), happy (r = –.07), getgo (r = .04), unfriendly (r = .04), and
exhausted (r = .01). Next to loneliness, this network also exhibited somewhat weaker direct
relations to the life event: sad (r = .10), getgo (r = –.08), unfriendly (r = .04), and happy (r =
.02).
[Figure 3]
3.2.3 Network Comparison. To compare the networks globally, we first calculated
the correlation of the adjacency matrices to obtain a measure of similarity, and second
conducted the NetworkComparisonTest. The correlation between the adjacency matrices was
r = .75, indicating that overall, the two network structures were largely similar. The
NetworkComparisonTest revealed a significant result for the global invariance test (p = .005),
indicating that there were some differences in the overall structure between the networks.
Of specific interest for our hypotheses (H3) was the extent to which loneliness
following the two life events was differentially related to other CES-D symptoms. In an
exploratory analysis, we investigated for which edges the two network structures showed the
maximum difference, through subtracting their weight matrices. We visualized the largest
absolute differences between edges in a network (Figure 4). The largest absolute differences
between estimates were obtained for the edges happy – mood (Δr = .15), exhaust –
concentration (Δr = .15), afraid – sad (Δr = .15), getgo – concentration (Δr = .13),
separation/widowhood – sad (Δr = .12), afraid – unfriendly (Δr = .12), lonely – getgo (Δr =
.12), lonely – failure (Δr = .11), sad – failure (Δr = .11), and getgo – failure (Δr = .11). With
respect to our hypotheses (H3), differential associations with loneliness could be found to
failure and getgo.
DIFFERENCES IN NETWORKS OF DEPRESSION
17
[Figure 4]
4. Discussion
Different life events may lead to different depressive symptoms, not only in overall
quantity — some life events have more severe consequences than others — but also in
quality. Since episodes of major depressive disorder are often preceded by severe stress or
adverse life events (Hammen, 2005), the idea that different life events lead to different
symptom profiles could explain a large part of the dramatic heterogeneity of depression
symptoms (Fried et al., 2015; Zimmerman et al., 2015).
To our knowledge, this is the first study to investigate potential differences in
depressive symptomatology between spousal loss and marital breakup by comparing
symptom profiles and modeling the relationship between life events and symptoms via
network models. We showed that one of the main differences between the two life events is a
stronger feeling of experiencing an unfriendly environment and oneself as a failure within
separated compared to widowed individuals. This finding is consistent with literature
regarding consequences of the reduction in social network following separation and its effect
on the individual‘s psychosocial well-being (Wrzus et al., 2013).
The network of bereaved individuals is largely consistent with previous findings of
Fried et al. (2015), indicating that spousal loss is primarily connected to loneliness, in turn
connecting to other depressive symptoms. Additionally, we found a strong link between
spousal loss and sadness. The present study extends this finding to a different type of marital
disruption; similar to spousal loss, marital breakup was also primarily linked to loneliness.
Overall, the two networks showed largely similar structures, as indicated by a large
correlation between their weight matrices.
In an exploratory analysis, we investigated the largest differences in edges between
the two networks. Experiencing oneself as a failure revealed a stronger connection to
DIFFERENCES IN NETWORKS OF DEPRESSION
18
loneliness in separated compared to widowed individuals. For widowed individuals, we
obtained stronger links for lonely – getgo, getgo – exhaust, and getgo – concentration.
Keeping in mind the exploratory nature of this analysis, these findings give rise to two
hypotheses: 1) Loneliness in separated compared to widowed individuals is more strongly
associated with symptoms related to the normative evaluation of the life event (stronger
relation of loneliness with experiencing oneself as a failure), and 2) loneliness in widowed
compared to separated individuals is more strongly associated with symptoms related to the
person‘s level of activity and cognitive capacities (stronger relations of loneliness with
getting going, and getting going with exhaustion and concentration).
4.1 Implications for future research and clinical practice
In line with previous research (Cramer et al., 2012; Fried et al., 2015), our study
provides further evidence of the importance of contextual information in explaining
depressive symptom patterns. In clinical practice, this could provide important information in
conceptualizing a patient‘s case, in understanding the etiology of depression, and in
identifying potential treatment targets. This study indicates that the main difference in
widowed compared to separated individuals might be characterized through a) differences in
the intensity of specific symptoms (i.e., experiencing oneself as a failure and an unfriendly
environment), and b) differences in specific relations to for example loneliness (e.g., failure
and get going). These findings can help tailoring treatment approaches to characteristics of a
given life event.
For both groups, prevention strategies targeting loneliness might be promising. For
widowed and separated individuals specifically, one could try to disrupt relations between
loneliness and other symptoms, if these can be replicated in other work. For instance, this
study suggests that separated individuals would additionally benefit from learning that
experiencing loneliness does not mean that their life plan is a failure (i.e., disrupting the
DIFFERENCES IN NETWORKS OF DEPRESSION
19
association between loneliness and failure), and widowed individuals could benefit from a
stronger focus on helping them ―getting going‖, for instance through behavioral activation
(Papa et al., 2013).
4.2 Limitations
The results of this study must be interpreted in the light of some limitations. First, we
analyzed cross-sectional data, any conclusions regarding dynamics remain thus putative.
Further, the time-scale on which depressive episodes unfold may differ between participants,
depending on the complexity of their depressive patterns. In a follow-up study, it would be
important to include several time points to aim to estimate Granger-causal relations between
life events and symptoms, and test effects of varying time-distances to the life events of
interest.
Second, as became evident in the accuracy and stability analysis, many parameters are
estimated with at best moderate precision. Our study faced a trade-off between sample size
and the time passed since the critical life event, and we opted for a compromise of less than
two years. We hope to replicate our finding in larger datasets of bereaved and separated
individuals—once these become available—which will allow for stricter screening. This
would also allow us to differentiate between potentially meaningful subgroups, such as
initiators and non-initiators of separation (Hewitt and Turrell, 2011).
Third, separated individuals were significantly younger widowed individuals in this
study. This might be considered a potential confound and limit the extent to which results can
be generalized to other age groups. Demographic data (Copen et al., 2012) suggest that
separation is indeed more prevalent among younger individuals, whereas elderly individuals
are more likely to experience spousal loss compared to separation. The precise role of age in
expressing specific symptoms thus remains a topic for future research.
DIFFERENCES IN NETWORKS OF DEPRESSION
20
Fourth, when applying network analyses to psychological scales, the choice of the
scale and the topological overlap of its items might drastically influence the structure of the
resulting network (Fried & Cramer, 2017). In the present dataset, we identified variables that
could have been potentially relevant to add to our network investigation, more specifically
contextual information regarding the cause of death in widowed participants, reasons for
separation, and the Prolonged Grief Disorder-13 (PG-13; Prigerson et al., 2009) tool,
however, these variables have unfortunately not been assessed at all three waves, and
therefore were not suitable to be included in our analyses. Since reactions to loss experience
have been linked to these specific symptoms of Prolonged Grief Disorder (PGD; Prigerson et
al., 2009), we encourage to include such variables in future studies. Furthermore, since the
network structure is based on partial correlations, excluding or combining items will lead to
different network structures. This is why we, unlike most prior studies in the field, decided to
thoroughly study item content, and modified the constructs under investigation based on a
thresholding rule. However, this issue needs more attention from both clinical theories and
empirical research, and decisions should in the best case be guided by both statistical tests
and theoretical considerations.
Lastly, we used the CES-D for this analysis. The CES-D contains the items loneliness
and experiencing oneself as a failure, which were important for our research questions. On
the other hand, it is a screening tool for depression but is not used for the actual diagnosis of
depression according to the DSM-5 (American Psychiatric Association, 2014), and differs
considerably from other depression scales in terms of content (Fried, 2017). It would thus be
interesting to model a broader range of depressive symptoms in future studies.
5. Conclusions
This study provides further evidence for the relation between specific adverse life
events and different symptom patterns of depression. Network models are a promising tool in
DIFFERENCES IN NETWORKS OF DEPRESSION
21
understanding these differential relations, and can be used to compare spousal loss with
marital disruption in this regard. A better understanding of these differences can in turn help
in tailoring interventions to specific contextual factors.
Approval of authors
All authors have seen and approved the final version of the manuscript being
submitted. The article is the authors' original work, hasn't received prior publication and is
not under consideration for publication elsewhere.
Funding
This research did not receive any specific grant from funding agencies in the public,
commercial, or not-for-profit sectors.
Author’s contributions
MS and PP developed the study concept; with them, EF and HS derived the
hypotheses and empirical testing procedure. PP and SS had collected the data as part of a
larger project awarded to PP. JB and EF conducted the statistical analyses and drafted the
manuscript, with substantive expertise from PP, SS, HS, ME, and MS. All authors
contributed to revisions of the manuscript and approved the final version for submission.
Declarations of interest: none.
The authors declare that there is no conflict of interest regarding the publication of
this article.
DIFFERENCES IN NETWORKS OF DEPRESSION
22
References
American Psychiatric Association, 2014. Diagnostic and statistical manual of mental
disorders : DSM-5. American Psychiatric Association, DSM.
https://doi.org/10.1176/appi.books.9780890425596.744053
Amrhein, V., Greenland, S., McShane, B., 2019. Scientists rise up against statistical
significance. Nature. https://doi.org/10.1038/d41586-019-00857-9
Borsboom, D., Cramer, A.O.J., 2013. Network Analysis: An Integrative Approach to the
Structure of Psychopathology. Annu. Rev. Clin. Psychol.
https://doi.org/10.1146/annurev-clinpsy-050212-185608
Copen, C.E., Daniels, K., Vespa, J., Mosher, W.D., 2012. First marriages in the united states:
Data from the 2006-2010 national survey of family growth. Natl. Health Stat. Report.
Cramer, A.O.J., Borsboom, D., Aggen, S.H., Kendler, K.S., 2012. The pathoplasticity of
dysphoric episodes: Differential impact of stressful life events on the pattern of
depressive symptom inter-correlations. Psychol. Med.
https://doi.org/10.1017/S003329171100211X
de Jong Gierveld, J., 2004. Remarriage, Unmarried Cohabitation, Living Apart Together:
Partner Relationships Following Bereavement or Divorce. J. Marriage Fam.
https://doi.org/10.1111/j.0022-2445.2004.00017.x
Djelantik, A.A.A.M.J., Robinaugh, D.J., Kleber, R.J., Smid, G.E., Boelen, P.A., 2019.
Symptomatology following loss and trauma: Latent class and network analyses of
prolonged grief disorder, posttraumatic stress disorder, and depression in a treatment-
seeking trauma-exposed sample. Depress. Anxiety. https://doi.org/10.1002/da.22880
Eisma, M.C., 2018. Public stigma of prolonged grief disorder: An experimental study.
Psychiatry Res. https://doi.org/10.1016/j.psychres.2017.12.064
Epskamp, S., Borsboom, D., Fried, E.I., 2018. Estimating psychological networks and their
DIFFERENCES IN NETWORKS OF DEPRESSION
23
accuracy: A tutorial paper. Behav. Res. Methods. https://doi.org/10.3758/s13428-017-
0862-1
Epskamp, S., Cramer, A.O.J., Waldorp, L.J., Schmittmann, V.D., Borsboom, D., 2012.
qgraph : Network Visualizations of Relationships in Psychometric Data. J. Stat. Softw.
https://doi.org/10.18637/jss.v048.i04
Epskamp, S., Fried, E.I., 2018. A tutorial on regularized partial correlation networks.
Psychol. Methods. https://doi.org/10.1037/met0000167
Foygel, R., Drton, M., 2010. Extended Bayesian information criteria for Gaussian graphical
models, in: Advances in Neural Information Processing Systems 23: 24th Annual
Conference on Neural Information Processing Systems 2010, NIPS 2010.
Fried, E.I., 2017. The 52 symptoms of major depression: Lack of content overlap among
seven common depression scales. J. Affect. Disord.
https://doi.org/10.1016/j.jad.2016.10.019
Fried, Eiko I., Bockting, C., Arjadi, R., Borsboom, D., Amshoff, M., Cramer, A.O.J.,
Epskamp, S., Tuerlinckx, F., Carr, D., Stroebe, M., 2015. From loss to loneliness: The
relationship between bereavement and depressive symptoms. J. Abnorm. Psychol.
https://doi.org/10.1037/abn0000028
Fried, E.I., Cramer, A.O.J., 2017. Moving Forward: Challenges and Directions for
Psychopathological Network Theory and Methodology. Perspect. Psychol. Sci.
https://doi.org/10.1177/1745691617705892
Fried, E.I., Nesse, R.M., 2015. Depression is not a consistent syndrome: An investigation of
unique symptom patterns in the STAR∗D study. J. Affect. Disord.
https://doi.org/10.1016/j.jad.2014.10.010
Fried, E. I., Nesse, R.M., Guille, C., Sen, S., 2015. The differential influence of life stress on
individual symptoms of depression. Acta Psychiatr. Scand.
DIFFERENCES IN NETWORKS OF DEPRESSION
24
https://doi.org/10.1111/acps.12395
Hammen, C., 2005. Stress and Depression. Annu. Rev. Clin. Psychol. 1, 293–319.
https://doi.org/10.1146/annurev.clinpsy.1.102803.143938
Hautzinger, M., Geue, K., 2016. Allgemeine Depressionsskala. Diagnostische Verfahren der
Psychother.
Hewitt, B., Turrell, G., 2011. Short-term functional health and well-being after marital
separation: Does initiator status make a difference? Am. J. Epidemiol.
https://doi.org/10.1093/aje/kwr007
Lehr, D., Hillert, A., Schmitz, E., Sosnowsky, N., 2008. Screening depressiver störungen
mittels allgemeiner depressions-skala (ADS-K) und state-trait depressions scales
(STDS-T). Diagnostica. https://doi.org/10.1026/0012-1924.54.1.
Logan, E.L., Thornton, J.A., Breen, L.J., 2018. What determines supportive behaviors
following bereavement? A systematic review and call to action. Death Stud.
https://doi.org/10.1080/07481187.2017.1329760
Malgaroli, M., Maccallum, F., Bonanno, G.A., 2018. Symptoms of persistent complex
bereavement disorder, depression, and PTSD in a conjugally bereaved sample: A
network analysis. Psychol. Med. https://doi.org/10.1017/S0033291718001769
Meyer, T.D., Hautzinger, M., 2001. Allgemeine Depressions-Skala (ADS): Normierung an
Minderjährigen und Erweiterung zur Erfassung manischer Symptome (ADMS).
Diagnostica. https://doi.org/10.1026//0012-1924.47.4.208
Papa, A., Rummel, C., Garrison-Diehn, C., Sewell, M.T., 2013. Behavioral Activation for
Pathological Grief. Death Stud. https://doi.org/10.1080/07481187.2012.692459
Prigerson, H.G., Horowitz, M.J., Jacobs, S.C., Parkes, C.M., Aslan, M., Goodkin, K.,
Raphael, B., Marwit, S.J., Wortman, C., Neimeyer, R.A., Bonanno, G., Block, S.D.,
Kissane, D., Boelen, P., Maercker, A., Litz, B.T., Johnson, J.G., First, M.B.,
DIFFERENCES IN NETWORKS OF DEPRESSION
25
Maciejewski, P.K., 2009. Prolonged grief disorder: Psychometric validation of criteria
proposed for DSM-V and ICD-11. PLoS Med.
https://doi.org/10.1371/journal.pmed.1000121
Radloff, L.S., 1977. The CES-D Scale. Appl. Psychol. Meas.
https://doi.org/10.1177/014662167700100306
Robinaugh, D.J., LeBlanc, N.J., Vuletich, H.A., McNally, R.J., 2014. Network analysis of
persistent complex bereavement disorder in conjugally bereaved adults. J. Abnorm.
Psychol. https://doi.org/10.1037/abn0000002
Robinaugh, D.J., Millner, A.J., McNally, R.J., 2016. Identifying highly influential nodes in
the complicated grief network. J. Abnorm. Psychol. https://doi.org/10.1037/abn0000181
Sbarra, D.A., 2015. Divorce and health: Current trends and future directions. Psychosom.
Med. https://doi.org/10.1097/PSY.0000000000000168
Stroebe, M., Stroebe, W., Schut, H., Boerner, K., 2017. Grief is not a disease but
bereavement merits medical awareness. Lancet. https://doi.org/10.1016/S0140-
6736(17)30189-7
van Borkulo, C., Boschloo, L., Borsboom, D., Penninx, B.W.J.H., Waldorp, L.J., Schoevers,
R.A., 2015. Package ‗NetworkComparisonTest.‘ JAMA Psychiatry.
https://doi.org/10.1001/jamapsychiatry.2015.2079
van Borkulo, C.D., Boschloo, L., Kossakowski, J.J., Tio, P., Schoevers, R.A., Borsboom, D.,
Waldorp, L.J., 2017. Comparing network structures on three aspects: A permutation test.
Manuscr. Submitt. https://doi.org/10.13140/RG.2.2.29455.38569
Williams, D.R., Rhemtulla, M., Wysocki, A.C., Rast, P., 2019. On Nonregularized
Estimation of Psychological Networks. Multivariate Behav. Res.
https://doi.org/10.1080/00273171.2019.1575716
Wójcik, G., Zawisza, K., Jabłońska, K., Grodzicki, T., Tobiasz-Adamczyk, B., 2019.
DIFFERENCES IN NETWORKS OF DEPRESSION
26
Transition out of Marriage and its Effects on Health and Health--Related Quality of Life
among Females and Males. COURAGE and COURAGE-POLFUS--Population Based
Follow-Up Study in Poland. Appl. Res. Qual. Life. https://doi.org/10.1007/s11482-019-
09742-z
Wrzus, C., Hänel, M., Wagner, J., Neyer, F.J., 2013. Social network changes and life events
across the life span: A meta-analysis. Psychol. Bull. https://doi.org/10.1037/a0028601
Zimmerman, M., Ellison, W., Young, D., Chelminski, I., Dalrymple, K., 2015. How many
different ways do patients meet the diagnostic criteria for major depressive disorder?
Compr. Psychiatry. https://doi.org/10.1016/j.comppsych.2014.09.007
DIFFERENCES IN NETWORKS OF DEPRESSION
27
Figure 1. Schematic set-up of the samples and analyses used in this study. Inclusion criteria
for separated/widowed individuals were a) a maximum time-distance to the respective life
event of two years, and b) that the participant was not living in a new partnership. Married
controls were randomly sampled from the pool of married participants. In order to be able to
model the loss-type in the networks, an equal amount of married controls was added to both
samples.
separated < 2 years,
n= 217
all separated
participants
(3 waves),
n= 250
all divorced
participants,
(3 waves),
n= 721
all widowed
participants
(3 waves),
n= 566
widowed
< 2 years,
n= 145
network sample 1, n= 290
widowed and married controls,
nwidowed = 145 , nmarried =145
network sample 2, n= 334
separated and married controls,
nseparated = 217 , nmarried = 217
all married
participants
(3 waves),
n= 1276
MANOVA,
ANOVA,
post-hoc tests
Network
Analysis
married controls
(randomly sampled)
n= 362
Running head: DIFFERENCES IN NETWORKS OF DEPRESSION
Table 1
Demographics of the widowed, separated and married sample.
Widowed
< 2 years,
n = 145
Separated
< 2 years,
n = 217
Married
controls,
n = 362
Comparing widowed against separated sample
M
SD
M
SD
M
SD
Difference tests
Significance
Effect size, confidence interval
1. Gender,
(% female)
79.31
-
76.04
-
52.76
-
2(1) = 0.53
p = .466
w = 0.001
2. Age
71.80
11.90
51.88
8.43
64.69
13.64
t(238.72) = 17.44
p < .001***
d = 1.93, CI [17.67, 22.17]
3. Duration of
marriage (years)
16.58
9.97
21.86
11.03
11.52
6.72
t(12.54) = 1.78
p = .100
d = 0.50, CI [-1.17, 11.73]
4. Time since
separation
(months)
11.95
7.29
11.23
7.20
-
-
t(306.15) = 0.93
p = .352
d = 0.10, CI [-2.26, 0.81]
5. CES-D sum
score
11.65
6.72
13.47
9.91
6.67
6.07
t(306.52) = 1.93
p = .055
d = 0.21, CI [-3.66, 0.04]
Running head: DIFFERENCES IN NETWORKS OF DEPRESSION
Figure 2. Post-hoc comparisons for all CES-D symptoms between separated and widowed
individuals, sorted by decreasing mean differences. 95% confidence intervals are indicated.
Note that we only indicated significance levels for items that were significant after correcting
for multiple testing using the Bonferroni method.
*** significant at .001; ** significant at .01; * significant at .05.
DIFFERENCES IN NETWORKS OF DEPRESSION
2
Figure 3. Regularized partial correlation network of the combined set of CES-D symptoms
and spousal loss (a, 145 widowed individuals and 145 married controls) and marital breakup
(b, 217 separated individuals and 217 married controls). Solid blue lines represent positive
edges, dashed red lines represent negative edges.
DIFFERENCES IN NETWORKS OF DEPRESSION
3
Figure 4. Network indicating the ten largest absolute differences in edge weights for the
widowed network compared to the separated network, based on the difference scores of the
respective weight matrices.
concentr
exhaust
failure
afraid
sleep
talk
lonely
sad
unfriend
getgo
life event
mood
happy
