Page 36 - JCBP-2-4

Journal of Clinical and
Basic Psychosomatics Core depressive symptoms of peripartum women

(Zhongda Hospital, affiliated to the Southeast University, metrics. These metrics included strength, betweenness,
30
and Nanjing Hospital, affiliated to the Nanjing University of closeness, and expected influence (EI), which are integral
Chinese Medicine) between May 2022 and March 2023. The to characterizing the network’s architecture. Strength was
18
participants were categorized into three groups based on their calculated as the aggregate weight of edges connected to each
pregnancy phase: the second-trimester group (161 women), node. Betweenness quantified the frequency with which
third-trimester group (248 women), and postpartum a node lay on the shortest path between pairs of nodes.
group (110 women). This study was approved by the Ethics Closeness was determined as the inverse of the mean distance
Committee of Zhongda Hospital (No. 2020ZDSYLL230-P01; from a particular node to all other nodes. EI represented the
October 27, 2020). All participants had to provide and sign a cumulative weight of the edges emanating from a given node.
written informed consent available online.
2.3.3. Network stability
2.2. Instrument The resilience of the network solution, including the
The Edinburgh Postnatal Depression Scale (EPDS) is a well- precision of edge weights and the reliability of centrality
24
established instrument for detecting depressive symptoms measures, was assessed using the “bootnet” R package.
26
in both pregnant and postpartum women. This 10-item Network stability was gauged through a bootstrapping
questionnaire utilizes a self-assessment scale wherein each procedure that involved 2500 resampling iterations, with
item is rated from 0 (most of the time) to 3 (not at all). The 95% confidence intervals (CIs). A wider CI indicated a
16
total score can vary from 0 to 30. A total score of ≥10 suggests lower precision in edge weights, whereas a narrower CI
the presence of depression, and an elevated total score is indicated a more dependable network structure. The
29
indicative of more pronounced depressive symptoms. 25 inclusion of “0” within the range of the constructed CIs
signified the absence of statistically significant differences in
2.3. Statistical analyses edge weights (or node strength) among distinct symptoms.
Statistical analyses were conducted using SPSS software To assess the stability of the centrality indices using a
(version 21.0; 1IBM Corporation, Armonk, NY, USA). case-dropping subset bootstrap method, the correlation
16
One-way analysis of variance (ANOVA) was used stability (CS) coefficient was determined. The network’s
to compare the age and EPDS scores among groups. structure was deemed stable if the CS coefficient
Bonferroni correction was used to perform a post-hoc remained high even after eliminating a subset of cases. An
analysis of the groups. The Chi-square test was employed exemplary CS value should approach 0.7, indicating a 95%
to assess differences in depression prevalence. A p-value of likelihood that the correlation would remain 0.7 even after
<0.05 was indicative of statistical significance. All network removing the maximum allowable proportion of cases.
29
analyses were performed using R (version 4.2.1). Furthermore, the CS value should not drop below 0.25,
2.3.1. Network estimation and it should ideally be >0.5.
The symptom networks were depicted through network 3. Results
diagrams, including the overall EPDS network for all
participants as well as the EPDS networks specific to each 3.1. Demographic characteristics and depressive
time period. Within these diagrams, “nodes” corresponded symptoms
to the variables, whereas “edges” signified the correlations Figure 1 and Table S1 depict the age and EPDS scores
existing between them. A least absolute shrinkage and during the three time periods. No significant difference in
26
selection operator Gaussian graphical model was applied age was found between the three groups (P > 0.05). The
to infer the structure of the symptom networks, and prevalence of depression in all participants was 32.18%,
pairwise correlations delineated the associations among with the highest prevalence in the postpartum group
the variables. The selection of the optimal network (45.45%), followed by the second trimester (27.33%) and
27
configuration was guided by the extended bayesian third trimester (29.44%) groups. Similarly, the postpartum
information criterion. For executing this analysis, the group exhibited the highest total scores, both depression
28
“bootnet” R package’s “estimateNetwork” function was and anxiety scores, and the highest scores for each item of
used, employing “EBICglasso” as the standard approach the EPDS (all P < 0.05, Bonferroni correction), except for
for network estimation. 29 items 5, 7, and 10 (P > 0.05).

2.3.2. Network centrality 3.2. Network estimation and centrality
The “qgraph” package in R, specifically its “centrality plot” Figure 2 depicts the depression network and the EI of
function, was used to ascertain the network’s centrality all participant data. Except for EPDS10 (“self-harm”)

Volume 2 Issue 4 (2024) 3 doi: 10.36922/jcbp.4089