Network Analysis of Borderline Personality Features in Adolescence Using a Screening Tool in a Chinese Community Sample

Abstract

Introduction

This study investigated the structure of the borderline personality features (BPFs) network and the most central BPF in adolescence.

Methods

Cross-sectional self-report data from 4,866 Chinese adolescents (M = 13.96, SD = 1.64; 61.3% girls) were included in the network analysis models. BPFs were assessed with the McLean Screening Instrument for Borderline Personality Disorder.

Results

Identity disturbance and affective instability emerged as the most central BPF in the current adolescent sample. In addition, chronic emptiness was also found with high centrality. The general networks of BPF were very similar between adolescent boys and girls, although some differences were detected.

Discussion

This study further supports the necessity of BPD assessment and diagnosis in adolescence and identifies the distinctive importance of identity and affective dysregulation in the early development of BPD. The findings provide empirical insights into the interconnections of BPF, which resonate with therapeutic mechanisms of evidence-based treatments for BPD. However, the research was limited in its use of a screening measurement rather than a diagnostic tool. Future studies can further explore BPD psychopathology in adolescence with longitudinal data and clinical interviews.

Introduction

Borderline personality disorder (BPD) is a severe mental disorder composed of emotional dysregulation, interpersonal instability, identity disturbance, and behavioral impulsivity [1]. As adolescence is a critical period for developing BPD symptoms [2], early diagnosis and preventive interventions are crucial for BPD treatment. Yet, the diagnosis of BPD in adolescence has been long debated [3]. Clinicians are concerned about the negative effect of labeling and stigma and confounding the normal developmental trajectory of emotional swaying with BPD symptoms [4]. Meanwhile, longitudinal studies supported the continuity and stability of BPD from adolescence to adulthood, and thus a growing number of researchers support BPD as a reliable and valid diagnosis in adolescents [5–7].

Several models of the core deficits of BPD have been proposed. Linehan’s [8] biosocial model outlines that the transaction of biological vulnerability and invalidating parenting results in difficulty in recognizing and regulating emotions, which then leads to unintegrated self and disturbed relationships. In the psychodynamic theory, individuals with BPD cannot integrate the representations of self and others and, thus, experience intense effects and unrealistic expectations of others [9]. In addition, results from a longitudinal study [10] indicated that behavioral impulsivity might be a crucial feature in the maintenance and prediction of BPD development. Finally, symptoms relevant to problematic relationships have been suggested as the best predictor of severe BPD [11, 12], which can refer to Bowlby’s [13] attachment theory. This theory suggests that individuals with BPD may exhibit attachment styles such as being insecure, anxious/ambivalent, and disorganized/disoriented [14].

However, the above models were mostly validated in adult patients, whereas lifespan studies found that the underlying psychopathology of BPD varies for adolescents. Videler and colleagues [15] reviewed 33 articles and reported that while BPD in adolescents is characterized by affective instability, behavioral impulsivity, and suicidality, functional impairments in relationships are more prevalent in later life stages. Thereby, in-depth investigations on the psychopathology of BPD in adolescents are needed to construct the psychopathological model of youth BPD to inform early assessment, prevention, and intervention.

Moreover, most existing studies follow the diagnostic criteria in DSM-5 [1], which is subjected to several limitations. For example, the traditional categorical model requires the presence of five out of nine symptoms, allowing numerous possibilities of combinations and, thus, substantial heterogeneity among patients [16]. Discriminant validity and comorbidity among personality disorders are also questioned [17].

Network Analysis of BPD Psychopathology

Current diagnosis of mental disorders tends to follow the traditional medical model of psychopathology, conceptualizing mental disorders in a homogenous way. In this model, clinical symptoms are mere reflections and outcomes of the latent disease, which are similarly important indicators for the diagnosis. However, the traditional model ignores the heterogeneity among symptoms and, thus, hampers the effectiveness of diagnosis and interventions [18].

Alternative to the traditional medical model’s view of mental disorders as an underlying pathological entity, the network approach deems mental disorders as complex, dynamic networks that consist of symptoms with solid connectedness. The development and maintenance of mental disorders may attribute to the interactions and feedback cycles among symptoms [19, 20]. The network structure can be visualized with nodes and edges. Nodes represent observed variables such as symptoms, traits, attitudes, feelings, and emotions, while edges between nodes delineate their pairwise relationships, and the edge’s thickness demonstrates the strength of the association. The activation or attenuation of a symptom can predict that of another symptom connected with a thick edge.

In contrast to the latent model focusing on the total score of inventories and symptoms, network analysis highlights identifying symptoms with high centrality. A symptom with high centrality can activate other connected symptoms, resulting in the arousal and sustainment of psychological illness, or be mostly influenced by other symptoms [21–23]. As a result, identifying central symptoms provides vital implications for designing prevention and intervention programs, which may manifest higher efficacy if targeting core symptoms or connected edges rather than the overall disorder [24]. Strength is the most replicable and generalizable index for centrality, which refers to the extent of how much a node is directly connected to other nodes in the network and is calculated by summing the absolute correlations of all edges [25]. Expected influence (EI) [26] is an additional measure for centrality, representing the sum of all edge weights connected to a particular node.

The composition of a cross-sectional network model estimates the partial correlations between each pair of symptoms in the condition of controlling the effects of other nodes. To further simplify the interpretation of the mode, we applied the graphical least absolute shrinkage and selection operator [27], in which positive and negative correlations are treated equally, and the edges with weak connections are deleted.

The network approach has demonstrated promising practicability in the research on BPD psychopathology. For instance, a network analysis conducted on college students and clinical patients found that for both samples, emotional instability, self-diffusion, and avoidance of abandonment were the most central symptoms, deserving more attention in the diagnosis and intervention of BPD [28]. However, the similarity between the samples should be interpreted with caution due to the small sample size of the clinical sample (n = 96). To expand on this discovery, Southward and Cheavens [29] conducted a network analysis of the core deficits of BPD from a dimensional perspective. They reported that participants who scored high on BPD had central symptoms of loneliness, impulsivity, and intense emotions, which come from the subscale of negative relationships, self-harm, and affectivity instability, respectively. In contrast, among participants scoring low on borderline features, symptoms with the highest centrality were chronic emptiness from the identity disturbance subscale, and intense moods and mood shifts from the affectivity instability subscale. This result suggested the potential difference between networks for the clinical and community populations, which was on contrary to the findings of Richetin et al. [28]. Moreover, the authors did not find a significant difference in the networks of male and female participants. Thus, further analysis of the role of self-image in BPD psychopathology or gender difference in BPD networks is required. In addition to central symptoms, Peckham et al. [30] administered a screening measure to a large cross-sectional sample (N = 5,212) to investigate the age difference of BPD psychopathology in adults. Researchers used the networktree algorithm to determine the optimal split at the age of 46, and results found that for the older half, the association between self-harm and emptiness is weaker, while the association between interpersonal difficulty and anger becomes stronger.

While these studies examined adult participants with self-report questionnaires only, recently, Peters et al. [31] recruited two adult patient samples and a mixed adolescent sample (85% inpatients, 15% healthy controls) and assessed BPD symptoms with structural interviews, which allowed for a more comprehensive and robust network analysis of BPD psychopathology. As in previous studies, affective stability and identity disturbance yielded to be central symptoms across the three samples. Despite network models being highly correlated and similar, unstable relationship was more central only in adult models, while stress-related dissociation/paranoia was more central only in the adolescent model, requiring further investigation on age and severity specificity of BPD pathology. This finding is also consistent with Videler et al.’s [15] summary that the course of BPD evolves from affect dysregulation, behavioral impulsivity, and self-harm in youth to interpersonal and other functional impairments throughout age. However, the authors [31] noted the limited generalizability due to the majorly white (62%) and inpatient sample (85%), while the racial difference in the phenomenology of BPD is evident [32, 33]. Thereby, further replications recruiting community samples of adolescents from other racial backgrounds are necessary to explore the psychopathology of BPD in adolescents and potential gender differences. Overall, existing network analyses indicated that rather than a single core domain, the development and maintenance of BPD might attribute to the interplay of multiple deficits, especially emotional dysregulation and identity disturbance.

The Current Study

Network analysis has introduced a systematic, holistic, and interactive perspective for the research on psychopathology and is increasingly applied in this field. Previous studies focusing on the psychopathology of BPD in adulthood have suggested emotional instability, identity diffusion, and effort to avoid abandonment as central symptoms [28]. However, little research investigated adolescent BPD with samples of other races. In this case, the present study aimed to estimate the psychopathological network of BPD in a large sample of Chinese adolescents, identify the high centrality features, and investigate the potential structural differences between BPD networks of boys and girls. Since our study recruited a nonclinical sample, nodes are referred as borderline personality features (BPFs) rather than symptoms. As an instrument that measures BPD features in adolescence with satisfactory psychometric properties is rare in China, we decided to administer McLean Screening Instrument for Borderline Personality Disorder (MSI-BPD) [34] to our participants, as it is a short and effective inventory and the most commonly studied self-report questionnaire for BPD [35], thus meeting our research purpose for large-scale analysis. Based on existing theories and research findings [8, 31], it is hypothesized that our network model would support emotional dysregulation and identity disturbance as core deficits.

Materials and Methods

Participants

Altogether 4,866 teenage students recruited from 23 middle schools in southwest China completed the survey. Ages ranged from 10 to 18 (M = 13.96, SD = 1.64), and the majority of participants were identified as female adolescents (61.3%). All students had sufficient language and cognitive abilities to provide informed consent and inventories; their parents also provided informed consent.

Procedure

We first trained local teachers to instruct the participants on the project information and the use of the questionnaires. Then after advising parents and students about the confidentiality of the data and their freedom to withdraw, students completed the paper-form surveys, which local teachers then recorded electronically. Feedback on the surveys was sent to participants through email. Ethical approval was granted by Peking University (project number: #2022-02-06). This study was not preregistered.

Materials

The MSI-BPD [34] is a self-report questionnaire developed based on the Diagnostic Interview for DSM-Ⅳ Personality Disorder [36]. There are 10 items requiring yes-or-no answers. The first eight diagnostic criteria of BPD in DSM-5 are each represented by one item. The last criterion, stress-related paranoia and dissociation, is assessed with two items [1]. A score higher than 7 suggests a diagnostic consideration of BPD, as the logistic regression of a 7-point cutoff yielded the best sensitivity and specificity. The Chinese version of MSI-BPD demonstrated good psychometrics (Cronbach’s α > 0.80, RMSEA <0.80) for adolescents [37, 38]. Cronbach’s α for the current sample was 0.85.

Analysis

Descriptive statistical analysis was performed with SPSS 26.0. Subjects containing missing values were excluded, and the data were tested for normality. Network analysis was performed using R 4.1.2 [39]. The goldbricker function of the R package networktools was used for estimation with the threshold limit as 25%, assessing whether 25% or fewer correlations with other symptoms significantly differed across node pairs to avoid the effect of topological overlap on the network.

We then used the bootnet package for the main analysis and the qgraph package for visualization [25]. A node represented each symptom assessed by one item of MSI-BPD. The Ising model [40] was applied because the data in this study were dichotomous and estimated connectedness between each pair of nodes through logistic regression while controlling for the effect of all other nodes. Likewise, to achieve the simplicity and clarity of the model, the extended Bayesian information criterion was applied by shrinking weak edges to zero.

Node strength and EI were calculated using the R package bootnet [25]. To improve the reproducibility of the network and obtain more precise results, we also conducted accuracy and stability tests for the network structure with the bootnet R package [25, 41]. The stability of centrality was quantified with the correlation stability (CS) coefficient, which was calculated by drawing bootstraps and comparing the centrality indices between the original sample and the subsets. The CS coefficient would be acceptable if above 0.25, preferred if above 0.5, and good if above 0.7. The accuracy of the edge weights was estimated by drawing bootstraps and calculating the 95% CI; a higher CI indicated weaker accuracy.

Furthermore, network models were constructed for male and female participants, respectively. Then, the R package NetworkComparisonTest [42] was used to compare the structure, edge weights, and centrality indices between these networks with 1,000 permutations. Finally, the Bonferroni-Holm correction was applied to avoid type-I error inflation.

Results

Descriptive Statistics

Table 1 shows the proportions for all modeled BPD features. Proportions were highest for dissociation (BPD7; 40.4%), affective instability (BPD4; 36.8%), and inappropriate intense anger (BPD5; 33.8%). BPD symptoms with the lowest prevalence were self-harm/suicide (BPD2; 17.9%), chronic emptiness (BPD8; 23.7%), and abandonment avoidance (BPD10; 23.6%). Of the adolescent sample, 679 (14.0%) participants scored above the cutoff threshold on the MSI-BPD (i.e., 7 points), indicating the likely presence of BPD.

Table 1.

Descriptive statistics for BPD features in adolescents

Symptom	Code	Mean	SD	Prevalence, %
Unstable relationships	BPD1	0.27	0.44	26.6
Self-harm/suicide	BPD2	0.18	0.38	17.9
Impulsivity	BPD3	0.26	0.44	26.2
Affective instability	BPD4	0.37	0.48	36.8
Inappropriate intense anger	BPD5	0.34	0.47	33.8
Distrust of others	BPD6	0.32	0.47	32.5
Dissociation	BPD7	0.40	0.49	40.4
Chronic emptiness	BPD8	0.24	0.43	23.7
Identity disturbance	BPD9	0.32	0.47	32.1
Abandonment avoidance	BPD10	0.24	0.42	23.6
Total score	BPDSum	2.93	2.92	–

Network Structure Analysis

All MSI-BPD items passed the normality test, and the goldbricker function test showed no redundant nodes. In this case, all items were included in the network analysis. The Ising model was applied as data in the current study were dichotomous. Figure 1 presents the network construction of nonclinical BPD symptoms among adolescents. The nodes represent the 10 items of MSI-BPD, the edges linking nodes indicate the relationships between the BPD symptoms, and the thickness of the edges reflects the strength of the pairwise connections (i.e., the size of edge weights). Forty-five possible edges between the 10 nodes, of which 43 edges were not equal to zero, constitute this adolescent BPD network. The average edge weight in the network is 0.52. Results of the partial correlation matrix estimated by the Fruchterman-Reingold algorithm [43] demonstrate strong associations between the following symptom pairs: identity disturbance (BPD9) and dissociation (BPD7); identity disturbance (BPD9) and chronic emptiness (BPD8); unstable relationships (BPD1) and abandonment avoidance (BPD10); and self-harm/suicide (BPD2) and impulsivity (BPD3).

Fig. 1.

Network model of BPD features in adolescents. BPD1 – unstable relationships, BPD2 – self-harm/suicide, BPD3 – impulsivity, BPD4 – affective instability, BPD5 – inappropriate intense anger, BPD6 – distrust of others, BPD7 – dissociation, BPD8 – chronic emptiness, BPD9 – identity disturbance, BPD10 – abandonment avoidance.

The strengths of the nodes in the adolescent BPD network are shown in Table 1, which was highest for identity disturbance (BPD9), affective instability (BPD4), and chronic emptiness (BPD8). In addition, to replicate and examine the outcomes of strength centrality, we also measured the EI of the nodes. As shown in Table 1, EI was highest for identity disturbance (BPD9), affective instability (BPD4), and chronic emptiness (BPD8), which were similar to the results of node strength.

Predictability estimates for the nodes in the adolescent BPD network revealed that the symptoms with high predictability were identity disturbance (BPD9, R² = 0.33), chronic emptiness (BPD8, R² = 0.31), inappropriate intense anger (BPD5, R² = 0.30), and dissociation (BPD7, R² = 0.30). The average predictability of all nodes was 0.28. Node predictability was significantly positively correlated with the centrality indices (r = 0.93, p < 0.001, 95% CI = [0.72–0.98]).

Finally, non-parametric bootstrapped difference tests were conducted to examine whether there is a significant difference between the centrality of any two nodes or the edge weights. These evaluations targeted the node strength, demonstrating the highest stability in the current study, as suggested by previous research [25, 44]. As shown in online supplementary Figure S1 (for all online suppl. material, see https://doi.org/10.1159/000535586), the node strength of identity disturbance (BPD9) was significantly higher than 50% of the other symptoms; the node strengths of affective instability (BPD4), dissociation (BPD7), and chronic emptiness (BPD8) were significantly higher than 30% of the other symptoms. Online supplementary Figure S2 displays results for tests on edge weights. The following edge weights were significantly different from most other edge weights: identity disturbance (BPD9) – dissociation (BPD7); identity disturbance (BPD9) – chronic emptiness (BPD8); unstable relationships (BPD1) – abandonment avoidance (BPD10); and affective instability (BPD4) – inappropriate intense anger (BPD5).

Stability and Accuracy of the Network Model

The bootnet R package was used to conduct the network structure’s accuracy and stability tests to examine the reproducibility of the results [25, 41]. First, the accuracy of the edge weights was evaluated by estimating the confidence intervals through non-parametric bootstrapping; the results are shown in online supplementary Figure S3. The 95% confidence interval of the bootstrapped edge weights was wide, indicating that caution should be taken when interpreting the node rankings as most differences were not significant.

Next, the stability of the centrality indices was assessed through the case-dropping subset bootstrap method. It was set so that the correlation between centrality indices in the subset network and the original network has a 95% probability of reaching the default value of 0.7. Online supplementary Figure S4 presents the results of randomly selecting different proportions of subsets of the observed data. Since the CS coefficient would be acceptable if above 0.25, preferred if above 0.5, and good if above 0.7 [25], the results suggested that stability was good for node strength (CS_(cor=0.7) = 0.75 > 0.5).

Network Comparison between Male and Female Adolescents

Table 2 displays proportions and results for χ² tests for item scores and an independent t test of the total score of MSI-BPD between male and female adolescents. The most frequently found feature was dissociation for both boy (40.9%) and girl (40.1%) participants. Compared to female adolescents, male adolescents reported significantly more borderline features, with a significantly higher proportion of reporting unstable relationships, inappropriate intense anger, chronic emptiness, and abandonment avoidance.

Table 2.

Descriptive and independent comparison results for BPD features in adolescent boys and girls

	Boys (n = 1,882)			Girls (n = 2,984)		t
	mean	SD		mean	SD
Unstable relationships	0.31	0.46		0.24	0.43	5.20***
Self-harm/suicide	0.19	0.39		0.17	0.38	1.39**
Impulsivity	0.25	0.44		0.27	0.44	−0.99*
Affective instability	0.37	0.48		0.36	0.48	0.78
Inappropriate intense anger	0.37	0.48		0.32	0.47	3.84***
Distrust of others	0.33	0.47		0.32	0.47	1.13*
Dissociation	0.41	0.49		0.40	0.49	0.55
Chronic emptiness	0.29	0.45		0.20	0.40	6.95***
Identity disturbance	0.33	0.47		0.31	0.46	1.56**
Abandonment avoidance	0.28	0.45		0.21	0.41	6.01***
Total score	3.14	3.01		2.80	2.85	3.98***

*p < 0.05,

**p < 0.01,

***p < 0.001, two-sided.

To illustrate the network structures of BPD features in adolescent boys and girls, we first plotted the networks with the R package Bootnet (see Fig. 2a, b). Features with the highest strength were chronic emptiness (BPD8, 1.46) and affective instability (BPD4, 1.23) for boys and girls, respectively. Next, the R package NetworkComparisonTest was used to compare the network structures for BPD features between male and female adolescents. Results indicated a non-significant difference in the global strengths (24.07 for males, 24.00 for females; S = 0.067, p = 0.91) and the maximum difference in edge weights (M = 0.49, p = 0.38) between the two networks, suggesting similar patterns of connectivity and structure between male and female adolescents.

Fig. 2.

Network model of BPD features in adolescent boys (a) and girls (b). BPD1 – unstable relationships, BPD2 – self-harm/suicide, BPD3 – impulsivity, BPD4 – affective instability, BPD5 – inappropriate intense anger, BPD6 – distrust of others, BPD7 – dissociation, BPD8 – chronic emptiness, BPD9 – identity disturbance, BPD10 – abandonment avoidance.

In addition, the results indicated significant differences between the edge weights of the two networks: unstable relationships (BPD1) – impulsivity (BPD3), affective instability (BPD4) – inappropriate intense anger (BPD5), and self-harm/suicide (BPD2) – distrust of others (BPD6). The EI values were also compared: the EI of affective instability (BPD4) was significantly higher for female adolescents (1.23) than male adolescents (−0.0099), while the EI of distrust of others (BPD6) was significantly lower for the females (−1.24) than the males (−0.97).

Discussion

In a large sample of Chinese adolescents, the current study employed network analysis to investigate the psychopathology of BPF in youth. The 10-node BPD network suggested the most central adolescent BPD features were identity disturbance, affective instability, and chronic emptiness. This result supported the psychopathological model of emotional dysregulation and identity disturbance as the core impairments in BPD, which have long been proposed to be central difficulties of BPD in different theoretical orientations. For instance, from the perspective of cognitive-behavioral therapy, childhood abuse, neglect, and invalidation cause the impairment of emotional recognition and regulation, representing the core deficit of BPD [8]. On the other hand, the object relation model proposes that the chaos of early interactions between the individual and the caregivers leads to unintegrated representations of self and other, ultimately resulting in borderline pathology [45].

This finding largely replicated previous results with discrepancies worthy of attention. In addition to the current adolescent community sample, both emotional dysregulation and identity disturbance have been previously considered central symptoms in adolescent psychiatric patients [31], adult score low on BPD features [29], and an adult sample that comprised mostly college students [28]. Nevertheless, identity disturbance was not the central symptom for the high BPD group in Southward and Cheavens’ [29] study. A critical goal of network analysis is to identify core nodes and connected edges that can be targets for preventions and interventions [24]. Consistently, of the two evidence-based treatments for BPD, dialectical-behavioral therapy [8] prioritizes reducing self-destructive behaviors and enhancing emotional regulation as a key mechanism, while transference-focused psychotherapy [46] operates through the integration of self and others. Both treatment programs have been validated for adolescent patients [47, 48]. In this case, a thorough clinical assessment of patients’ pathology may match them with more suitable psychotherapy and achieve higher efficacy.

Another critical difference is the role of chronic emptiness, which was reported as the central symptom in the low BPD group only in Southward and Cheavens’ [29] study. These researchers also found recklessness a core feature in the high BPD group. This transition is consistent with the developmental course of BPD. Moreover, according to a systematic review [49], chronic emptiness is an antecedent risk factor for behavioral impulsivity and self-harm, as patients try to fill their internal void with external stimuli from sensation-seeking actions. As adult studies have proposed impulsivity and aggression as an endophenotype of BPD [50, 51], future research may investigate chronic emptiness as a potential early indicator for preventing the development of behavioral symptoms specific in adolescents.

The present study did not find symptoms relevant to interpersonal relationships to be central, while an effort to avoid abandonment [28], loneliness [29], and relationship difficulties [31] were reported as central symptoms in adults. Consistently, Peters et al. [31] did not find unstable interpersonal relationships a central symptom in adolescent psychiatric inpatients. As suggested in prior literature [15, 31], a potential reason may be that the current sample was relatively too young to develop romantic relationships, and their social circles were firmly restricted in school and family settings.

The network structures of BPD features between adolescent males and females were compared in the present study to examine the gender differences in BPD among adolescents. While gender differences exhibited in the prevalence of BPD features, no significant differences were found in terms of the global strengths and the maximum difference of edge weights. These findings suggest that despite the greater level of BPD symptomatology in male adolescents, the overall network structures of BPD features did not differ significantly between genders. Adolescent males and females may exhibit similar BPD psychopathology as a whole. Consistently, epidemiological studies reported that three-quarters of the individuals diagnosed with BPD are females [1], and network analysis of BPD symptoms conducted in a large mixed population found no significant gender differences between the male and female networks of BPD features [29].

Despite no differences in the overall network structure, the networks of BPD features showed disparities in certain parts of the network construction between male and female adolescents. In the two networks, several edge weights (e.g., unstable relationships [BPD1] – impulsivity [BPD3] and affective instability [BPD4] – inappropriate intense anger [BPD5]) were significantly different. Concerning the centrality estimates, affective instability (BPD4) exhibited higher importance among the BPD features in female adolescents than in males, whereas distrust of others (BPD6) played a more central role in males than females. These findings indicated that male and female adolescents were likely to show several different BPD core features, suggesting the necessity of gender-sensitive prevention and intervention with particular attention to certain BPD features [52].

Using the widely recognized and psychometrically satisfying MSI-BPD measuring 10 features of BPD [34], results suggested that 679 (14.0%) participants scored higher than the 7-point cutoff, which was much higher than the prevalence rate reported by Leung and Leung (7.7% for females, 5.0% for males) [37]. This finding further defended the early diagnosis of BPD in adolescents and the need for in-depth investigations into its psychopathology, to which our discoveries regarding the central borderline features can offer insights.

There are potential cultural influences on our findings that are worth noting. First, our sample reported the highest prevalence of dissociation among borderline features. Similarly, a Hong Kong college sample was found with a higher prevalence of pathological dissociation than other nonclinical samples [53]. It is possible that items measuring dissociation were understood differently in Chinese culture: “feeling unreal” may be interpreted as a mental resistance to reality rather than dreamlike or distorted experience of derealization. However, unlike in Peters et al. [31], dissociation was not found with high centrality in the network, which may be attributed that our study recruited a community sample, while Peters et al. [31] investigated adolescent inpatients with more severe pathology. That chronic emptiness was a highly central feature may be subjected to profound cultural impact. Compared to Westerners, individuals from the traditional Chinese culture gain less access to exciting and challenging activities and are thus prone to feelings of emptiness and boredom [54]. Moreover, as Chinese parents show a high tendency to control and overemphasize academic achievements [55], especially in an invalidating family (which is an etiological environment in Linehan’s [8] biosocial model), their children will have limited time and opportunities to explore personal interests and then develop a sense of meaning.

The results from the present study should be interpreted in the context of limitations. First and foremost, this study employed a self-report screening instrument (i.e., MSI-BPD based on the DSM-5 diagnostic taxonomy) to assess adolescent BPD features [1, 34]. Therefore, MSI-BPD demonstrates a high sensitivity to symptoms and is hard to accurately reflect the clinical significance of the symptoms. For example, the endorsement rate of BPD7 (dissociation) at 40.4% was an implausible prevalence from a general community sample. It was possible that the inflated prevalence would disrupt the network model. The questionnaire also lacks a specific time frame, while DSM-5 [1] requires that BPD features must last for at least 1 year to diagnose individuals under 18. The items of dissociation and paranoia do not clarify that these experiences need to be stress-related, further increasing the risk of false positives. While Peters et al. [31] have administered clinical interviews to assess BPD symptoms, to validate BPD networks in more culturally diverse samples, future studies should integrate various approaches (e.g., developmental, categorical, and dimensional) and measures (e.g., diagnostic questionnaires, clinical interviews) of BPD psychopathology to examine better the inner structure of BPD and the specific relationships between different features or aspects of BPD [52, 56]. Instruments that comprehensively examine BPF in adolescents with high reliability and validity are in high need.

Additionally, the undirected cross-sectional networks of adolescent BPD features offered limited insights into the possible causal and temporal relationships between different BPD features [57]. Furthermore, the 1-year duration of BPD features is a critical criterion for BPD diagnosis in adolescents [1, 52]. Networks based on longitudinal data are thus needed to investigate further the underlying mechanisms of adolescent BPD and the developmental course of BPD [29]. Second, the study was conducted with an adolescent sample in a large community but did not recruit adolescent patients with clinical diagnoses, thus restricting the generosity of the discoveries. Future researchers should include a separate clinical sample to better illustrate the etiology and pathogenesis of BPD in adolescents.

Finally, adolescents with BPD often have other comorbid mental disorders, such as eating disorders [3]. The current study solely focused on the network construction of BPD features, which may limit its theoretical and clinical values. Karatzias et al. [58] recently used network analysis to examine the distinctiveness and connectedness of four stress-related disorders and supported the diagnostic boundaries of disorders. Thereby, network analysis can be further conducted to explore the associations between adolescent BPD features and relevant comorbid disorders [59].

Conclusion

This study is one of the first attempts to examine the network structure of BPD features in a large Chinese adolescent sample. Through encouraging network analysis, we clarified the core BPD features, elaborated on the relationships between BPD features, and achieved results that supported the early identification of BPD features in adolescents. Moreover, it is suggested that identity disturbance and affective instability may play a key role in developing and maintaining BPD. This finding may inform prevention and intervention targeting these central features of adolescent BPD. In addition, significant gender differences were exhibited in a few edge weights and EIs of BPD features but not in the overall network structure. Finally, we provided recommendations for future studies to further elucidate BPD’s nature, structure, and development and improve psychological services and clinical treatment for adolescent BPD.

Statement of Ethics

This research was conducted ethically in accordance with the World Medical Association Declaration of Helsinki. Ethical approval was granted by the Ethical Committee of Peking University (project reference number: #2022-02-06). Participants and their parents provided written informed consent before participation.

Conflict of Interest Statement

The authors have no conflicts of interest to declare.

Funding Sources

This research was supported by the National Science Foundation of China (project number: 30900401). The funding was awarded to Dr. Jie Zhong. There is no restriction regarding the submission of the report for publication.

Author Contributions

Qian Wang: conceptualization, data curation, formal analysis, investigation, methodology, project administration, software, validation, visualization, and writing – original draft, review, and editing. Zirong Li: formal analysis, methodology, software, validation, visualization, and writing – original draft, review, and editing. Jie Zhong: conceptualization, funding acquisition, investigation, methodology, project administration, resources, supervision, validation, and writing – review and editing.

Funding Statement

This research was supported by the National Science Foundation of China (project number: 30900401). The funding was awarded to Dr. Jie Zhong. There is no restriction regarding the submission of the report for publication.

Data Availability Statement

The data and programming codes of this study are available on request from the corresponding author. The data were not publicly available as that was not stated in the informed consent.