INTRODUCTION
Interpersonal sensitivity is defined as having a heightened awareness and response to the feelings, actions, or criticism of others, often leading to individual distress. This can exacerbate social difficulties and contribute to the development of various mental health conditions [1-3]. Moreover, it is associated with a heightened risk of mood disorders [4]. The Interpersonal Sensitivity Measure (IPSM), a validated tool, has been employed to measure interpersonal sensitivity. A correlation was observed between higher IPSM scores and poorer outcomes of depressive episodes [5], and a study indicated a relationship between higher IPSM scores and treatment-resistant depression in individuals compared to the remission group [6]. Additionally, individuals diagnosed with major depressive disorder were more likely to return to work after a period of sick leave when they had lower interpersonal sensitivity [7]. While prior research indicates mood states can influence interpersonal sensitivity, the reverse can also be true, with interpersonal sensitivity impacting the course of mood disorders.
However, our understanding of interpersonal sensitivity remains incomplete, particularly regarding how its different symptoms interact in individuals with mood disorders. Traditional methodologies examine variables in isolation, thereby neglecting the intricate web of connections among symptoms. The network theory of psychiatric disorders, which views psychiatric disorders as interconnected networks of causally interacting symptoms, has gained attention [8]. The use of network analysis as an analytical tool in psychopathology research has rapidly expanded [9]. Through network analysis, it is possible to identify the core symptoms that strongly affect other symptoms [10,11]. This approach holds promise for guiding and assessing therapeutic interventions, opening up possibilities for more targeted and effective treatments [10].
We conducted a network analysis the IPSM to comprehend interpersonal sensitivity in patients with mood disorders. The objectives were to 1) explore the network structure of interpersonal sensitivity in patients with mood disorders (major depressive disorder [MDD], bipolar I disorder [BD I], and bipolar II disorder [BD II]) and 2) identify the central symptoms of interpersonal sensitivity.
METHODS
Participants
A total of 837 participants (MDD, n=265; BD I, n=126; BD II, n=446) were included in this cross-sectional study conducted between July 2013 and February 2021. The patient group was recruited from the mood disorder clinic of Seoul National University Bundang Hospital.
Data including age, sex, and diagnosis of psychiatric disorders were collected, and all participants in the patient group were diagnosed using the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition [12] criteria. Diagnoses were confirmed by board-certified psychiatrists (THH and WM) based on the Mini-International Neuropsychiatric Interview [13] or review of case records.
The study protocol was approved by the Institutional Review Board of Seoul National University Bundang Hospital (B-2205-756-111). Data were acquired by reviewing patients’ medical records; therefore, patient consent was not required.
Clinical instrument: IPSM
The IPSM [2], a validated 36-item self-report checklist, evaluates interpersonal sensitivity, a psychosocial factor that can contribute to depression [14,15]. The questionnaire is completed on a 4-point Likert scale, with 1=very unlike me, 2=moderately unlike me, 3=moderately like me, and 4=very like me. The total score ranges from 36 to 144, with higher scores indicating greater interpersonal sensitivity. In our study, we used the Korean version of the IPSM, validated in 2013, with robust internal reliability (Cronbach’s alpha: 0.73-0.83, 0.94) and test-retest reliability (coefficients: 0.80-0.94) [14]. The internal consistency of the questionnaire was excellent in this study (McDonald’s ω=0.93). However, because the patient sample’s nature limited repeated measures, a test-retest analysis was not conducted.
Statistical analysis
Network estimation and community clustering
We constructed a network of patients with mood disorders consisting of 36 nodes representing each of the 36 IPSM items. The edges in the networks are undirected and reflect the Spearman’s correlation between the two nodes. The networks were visualized using a Gaussian graphical model, which can accommodate ordinal as well as continuous data [16]. The Gaussian graphical model has advantages in visualizing and identifying the centrality of a certain node in the network. To avoid network complexity, which induces spurious relationships and causes problems in network interpretation and replication, the graphical least absolute shrinkage and selection operator (glasso) [17] with the extended Bayesian information criterion model [18] is utilized when building networks. The introduction of this model elicits a shrinkage of weak associations and pseudo-edges between nodes, thereby constructing a parsimonious network [18]. The package glasso (version 1.11) was used to implement such regularization [19,20]. Additionally, the qgraph package (version 1.92) [21] was selected for visualization of a network and its centrality, and the bootnet package (version 4.1) was utilized to test network stability [22]. Moreover, to estimate the number of dimensions and identify communities, we used the relatively recent exploratory graph analysis (EGA) technique which implements the Walktrap algorithm [23]. This method is based on graphical lasso with the regularization parameter specified via the extended Bayesian information criterion, and has been found to outperform typical factor-analytic methods [24]. We conducted an additional bootstrap EGA (bootEGA) involving 1,000 bootstrapped samples to assess the stability of the items and the dimensional structure of the network. These analyses were performed using the EGA and bootEGA functions in the EGAnet package [25].
Node centrality
Several centrality indices are used to measure a node’s importance in networks: 1) node strength, which assesses the sum of weighted connections with other nodes, indicating strong and direct connections [26], 2) betweenness, reflecting the node’s ability to connect to other nodes through the shortest path [26], and 3) closeness, quantifying the node’s relationship considering indirect connections and shorter average distance to all other nodes for higher values [26]. All measures are reported as standardized z-scores [26].
Network accuracy and stability
To evaluate the precision of the network inferences, we performed a nonparametric bootstrap test with 2,000 samples to examine the edge-weight accuracy with 95% confidence intervals (CIs). Additionally, the stability of the centrality indices was investigated using a case-dropping subset bootstrap test to estimate the correlation stability coefficient. This coefficient indicates the maximum proportion of cases that can be excluded while ensuring a 0.7 or higher correlation between the original centrality indices and the centrality of networks based on subsets with a 95% probability [22]. If removing a significant number of samples from the original data does not result in a significant decrease in the centrality index of the node, the network is considered stable, preferably with a value greater than 0.50 [22]. Additionally, edge weight and node centrality index differences were examined to test whether they differed significantly from each other.
Network comparison test
The network comparison test (NCT), a two-tailed permutation test, was conducted to investigate the differences in network structure between the sub-diagnosis groups (i.e., MDD and bipolar disorder) [26,27]. We evaluated differences between the two networks using several invariance measures, including network structure invariance and global strength invariance [27,28]. We conducted the NCT by permutating 1,000 times at a significance level of 0.05. All statistical analyses were performed using the R software version 4.1 (R Foundation for Statistical Computing, Vienna, Austria).
RESULTS
Clinical and demographic characteristics
Table 1 presents the participants’ demographic characteristics. The mean age for the patient group was 36.23 years (standard deviation=11.56). The patient group consisted of 232 men (27.7%) and 605 women (72.3%). The scores and mean differences of the IPSM items among the different mood disorder subgroups are shown in Table 2.
Network estimation and community clustering
Figure 1 presents the estimated networks, including the color-coded clusters of the patient group. Using the EGA method, the patient group network was composed of five communities (1: Separation Anxiety, 2: Apprehension about Others’ Reactions, 3: Low Self-Esteem, 4: Timidity, and 5: Interpersonal Worry/Dependency). Supplementary Table 1 shows the grouping of items, indicating which items are part of each community. The bootstrapping procedure using bootEGA function reveals that the five-community solution was most frequently replicated, occurring in 51.8% of cases (518 out of 1,000 samples). Additionally, about 25% of the IPSM items exhibited low replication level (Supplementary Figure 1). This result is consistent with previous studies on the IPSM using factor analysis, which also showed complex factor loading patterns (Supplementary Figure 2). The mean node predictability was 0.458, whereas the mean edge weight for the patient group was 0.022.
Centrality analysis and edge weights
Figure 2 presents the centrality indices of the IPSM network. Item 10 (“I worry about being criticized for things that I have said or done”) showed the highest value in strength. Regarding strength centrality, no significant difference was found with item 10 and items 5 (“If others knew the real me, they would not like me”), 25 (“I always expect criticism”), 26 (“I can never be really sure if someone is pleased with me”), 30 (“I worry about what others think of me”), 36 (“I care about what people feel about me”), 31 (“I do not feel happy unless people I know admire me”), and 33 (“I worry about hurting others’ feelings”) through the bootstrapped difference test (Supplementary Figure 3). Item 19 (“I fear that my feelings will overwhelm people”) had the highest betweenness score, and item 30 (“I worry about what others think of me”) was ranked first for closeness centrality. Supplementary Table 2 provides the IPSM items along with their corresponding centrality parameters (i.e., strength, betweenness, and closeness).
Furthermore, when examining the differences in strength centrality within the diagnosis groups (MDD and BD), we found that item 25 (“I always expect criticism”) had the highest value in the MDD group, but it did not significantly differ from 14 other items (i.e., 3, 5, 9, 10, 17, 19, 21, 24, 28, 30, 31, 33, 34, 36), including item 10, which showed the highest network centrality of all subjects. In contrast, item 10 (“I worry about being criticized for things that I have said or done”) exhibited the highest strength centrality value in the BD group; however, it did not exhibit a significant difference with other six items (i.e., 26, 5, 31, 30, 25, 36) (Supplementary Figures 4 and 5).
In terms of edge weight, the strongest edge weight was observed between items 1 and 17 (“I feel insecure when I say goodbye to people” and “I feel anxious when I say goodbye to people”), which exhibited a significant difference when compared to other edges (Supplementary Figure 6).
Network accuracy and stability test
The network stability indices for network strength, betweenness, and closeness are shown in Supplementary Figure 7. The results revealed that strength centrality displayed a satisfactory level of stability, allowing for the removal of up to 75% of the sample while maintaining a correlation of 0.7 with an unchanging network structure compared to the original configuration. However, the betweenness and closeness centrality indices exhibited robust correlation stability coefficients only up to the elimination of 28.3% and 51.6% of the sample, respectively, suggesting that they may be less dependable measures for ensuring stability in cases of substantially smaller sample sizes.
Supplementary Figure 8 displays the results of visualizing the 95% CIs using bootstrapping samples for all edges, arranged in descending order based on their magnitudes within the IPSM networks. The comparatively slender bootstrapped CIs indicated the accuracy of the estimates.
NCT according to mood disorder diagnosis
There were no statistically significant differences in the overall strength of the networks between patients with MDD and those with BD (16.197 vs. 16.770; S=0.151, p=0.857). Additionally, the network invariance test revealed that the networks between the two groups displayed similar structures (M=0.164, p=0.660). No significant differences in strength centrality were found between the MDD and BD groups across the 36 items, apart from item 26 (“I can never be really sure if someone is pleased with me”). Within mood disorder subgroups, network invariance and global strength tests showed no significant differences. Analyses comparing MDD vs. BD I, MDD vs. BD II, and BD I vs. BD II are provided in Supplementary Material
DISCUSSION
We conducted a network analysis of the IPSM in patients with mood disorders. Our findings revealed a network structure comprising five communities or clusters (Figure 1). Notably, item 10 (“I worry about being criticized for things that I have said or done”) emerged as the node with the highest strength centrality. Several other items also displayed high centrality and demonstrated no significant difference from item 10; these were associated with “Interpersonal Worry/Dependency” and “Low Self-Esteem.” Furthermore, NCTs between patients with MDD and BD revealed no major differences, underscoring the consistency of centrality patterns across these mood disorders. Our study contributes to the understanding of the use of the IPSM for people with mood disorders and offers valuable insights into therapeutic strategies targeting these specific symptoms.
In network analysis, central symptoms act as focal points that trigger the activation of adjacent symptoms [10]. Our network analysis revealed that item 10 was the node with the highest strength centrality, consistent across the diagnostic categories of MDD and BD. Additionally, several other items related to “Interpersonal Worry/Dependency” and “Low Self-Esteem” displayed high centrality in the network. Our findings suggest that focused interventions for these core symptoms may be essential to reduce interpersonal sensitivity in mood disorders [10]. Mental health professionals may consider implementing cognitive-behavioral approaches to help patients manage sensitivity to criticism and negative self-evaluation [29]. This supports the inclusion of specific approaches, such as positive self-affirmation and cognitive restructuring, within treatment protocols to improve patient outcomes [30]. Intervention programs such as dialectical behavioral therapy and interpersonal psychotherapy have already integrated elements focused on addressing the interpersonal challenges experienced by individuals with mood disorders, and numerous studies have validated their effectiveness [31-33]. In addition to these programs, further research is necessary to investigate whether interventions that incorporate the aspects highlighted in this study, specifically the core symptoms identified within the categories of “Interpersonal Worry/Dependency” and “Low Self-Esteem”, could lead to more effective approaches. Furthermore, given that interpersonal sensitivity is associated with treatment outcomes in mood disorders [4], our findings suggest that targeted interventions for this core symptom could be pivotal not only in reducing interpersonal sensitivity but also in enhancing the overall treatment outcomes of mood disorders.
We identified five distinct communities within the IPSM network. Previous factor analytic methods to classify interpersonal sensitivity questionnaires have yielded varying results, grouping items into either five or three factors (Supplementary Figure 2). Moreover, bootEGA results (Supplementary Figure 1) indicated the classifications of items lacked robustness, underscoring the IPSM items’ close interrelations and the challenge of categorizing them into specific symptom clusters. Despite these challenges, EGA uncovered a more streamlined and consistent pattern compared to methods such as Walktrap and Spinglass (Supplementary Figure 2). The process of naming the communities grouped by EGA involved assessing the congruence between IPSM items within identified clusters and those associated with factors from earlier research. Where there was a significant overlap, we adopted the names of established factors. In cases of unique clusters, names were assigned based on shared item traits. This deliberative process, conducted in consultation with two psychiatrists (YK and WM) and two psychologists (HSK and JP), led to the current cluster names as detailed in Figure 1 and Supplementary Table 1.
Compared to prior research, in the EGA, items 1 and 17 assigned to the “Separation Anxiety” cluster were consistently categorized under the same factor in Boyce and Parker [2] and You et al.’s [34] studies. Moreover, the “Timidity” cluster included four items that partially corresponded with the “Timidity” factor in these studies, sharing three items [2,34]. The “Interpersonal Worry/Dependency” cluster, with 11 items, aligned with eight items from the “Interpersonal Worry/Dependency” factor in Harb et al.’s [35] and Doğan and Sapmaz’s [36] three-factor models, supporting the validity of the five-factor model and incorporating aspects of the three-factor model. However, the “Apprehension about Others’ Reaction” cluster did not show as consistent a correspondence with previous research, indicating a composition from a wider set of characteristics that merited its own classification. This variability in item grouping could result from sociocultural and demographic differences among study populations or methodological variances. Overall, the EGA findings closely paralleled those in Lee et al.’s [14] study. Notably, seven of the nine items in our EGA’s “Low Self-Esteem” cluster coincided with Lee et al.’s [14] “Low self confidence” factor, and most items in the “Interpersonal Worry/Dependency” cluster were also present in Lee et al.’s [14] “Interpersonal awareness” and “Interpersonal vulnerability” factors. Except for item 32, all items in our EGA’s “Separation Anxiety” and “Timidity” clusters were included in Lee et al.’s [14] “Lack of assertiveness/separation anxiety” factor. These similarities likely reflect shared cultural and societal contexts, while the observed discrepancies may indicate differences in demographic characteristics, such as gender distribution and clinical profiles—Lee et al.’s study [14] predominantly involved healthy male controls, in contrast to this study’s focus on mostly female mood disorder patients. Despite clustering diversity, EGA extends beyond traditional factor analysis by pinpointing key clusters with central nodes through network analysis, offering valuable clinical insights for targeting “Interpersonal Worry/Dependency” and “Low Self-Esteem.”
In the NCT analysis, no significant differences were observed in terms of the overall network strength or underlying structure between the MDD and BD groups. Furthermore, items with high network strength appeared similarly across both mood disorder categories. Additional subgroup analyses, specifically MDD vs. BD II, MDD vs. BD I, and BD I vs. BD II, reinforced these findings, revealing no significant differences in these key network parameters. These consistent patterns in the IPSM network across diagnoses suggest that the fundamental mechanisms affecting interpersonal sensitivity are similar within patient mood disorder groups. This consistency implies that interventions aimed at addressing interpersonal sensitivity may possess cross-diagnostic utility and potential efficacy in treating both MDD and BD.
Previous studies have mainly compared IPSM scores between those with depressive episodes and healthy individuals [2], focused on social anxiety disorder patients [35], or non-clinical groups [14]. These studies have not provided evidence for differing factor characteristics across mood disorder diagnoses. While mood lability and irritability are key symptoms linking unipolar to bipolar transitions among adolescents via network analysis [37], network studies on interpersonal traits in mood disorders remain unreported to date. The lack of statistically significant differences in network structures between MDD and BD in our study suggest that IPSM item correlations and core symptom variability might be minimal between mood disorders. The differences between mood disorders may not lie as much in the structure of the network but potentially in the severity of interpersonal sensitivity. Referring to Table 2, certain items showed statistically higher scores in BD II over MDD, and in BD II over BD I. These similar structures across mood disorders suggest that clinically recognizing potential differences in severity of interpersonal sensitivity might be more crucial than distinctions in therapeutic approaches between MDD and BD.
Our study has several limitations. First, the cross-sectional design precluded us from establishing direct causality between interventions targeting core symptoms and the comprehensive alleviation of interpersonal sensitivity symptoms. Future studies employing longitudinal or randomized controlled trial designs will be essential to address this issue. Second, the results may vary depending on the state of a patient’s mood disorder. Although interpersonal sensitivity is generally considered a stable personality trait, it may also act as a state-dependent measure subject to fluctuations based on the course of the illness and the patient’s current condition. Third, our study was conducted in tertiary healthcare settings in East Asia, which could limit the generalizability of the findings to other populations [9,38,39]. Fourth, although our study had an adequate sample size, a larger sample could yield more statistically significant NCT results. Despite these limitations, this study is valuable in elucidating the structure and core symptoms of interpersonal sensitivity in mood disorders through network analysis.
In conclusion, our findings indicated that “Interpersonal Worry/Dependency and “Low Self-Esteem” emerged as prominent features in the IPSM network for patients with mood disorders. Further research is needed to investigate whether such interventions have a positive impact on the course or outcomes of mood disorders.
PDF Links
PubReader
ePub Link
XML Download
Full text via DOI
Download Citation
Supplement1
Print
