A cohort comprising 823 adolescent patients, aged between 10 to 18 diagnosed with MDD was recruited from both inpatient and outpatient departments of the Psychiatric Department at the First Affiliated Hospital of Kunming Medical University over the period from June 2021 to December 2023. Diagnosis of MDD was conducted in accordance with DSM-5 criteria by attending psychiatrists or higher-ranking physicians (47). Each participant completed a self-assessment questionnaire, and informed consent was obtained from both participants and their guardians prior to survey administration. Participation was voluntary, and all participants and guardians were informed about the anonymity and confidentiality of the questionnaire. Exclusion criteria encompassed (1): current diagnosis of organic diseases, schizophrenia spectrum disorders, other psychiatric disorders (e.g., neurodevelopmental disorders), bipolar disorder, or psychoactive substance use disorder; (2) incomplete or invalid questionnaire data; (3) inability to understand or cooperate with the completion of questionnaires; (4) not within the specified age range of 10 to 18 years; (5) patients taking multiple psychotropic drugs for treatment.

For this study, the Suicidal Behavior Questionnaire-Revised (SBQ-R) (48), assessed past-year suicidal ideation frequency, specifically querying participants on the frequency of suicidal thoughts in the past year. The response options for this item were: “Never,” “Rarely (1 time),” “Sometimes (2 times),” “Often (3–4 times),” and “Very often (5 or more times).” Participants reporting “never” were excluded. Suicidal ideation occurring 1–2 times was classified as lower frequency, while 3 or more times was categorized as higher frequency. Previous research has demonstrated that this item from the SBQ-R serves as a robust clinical screening tool, aiding in the identification of individuals at elevated risk for suicide and informing subsequent evaluation and intervention strategies (49, 50). The network analysis included 733 MDD-diagnosed children/adolescents with suicidal ideation. Ethical approval was obtained from the Ethics Committee of the First Affiliated Hospital of Kunming Medical University (2021 Ethical Approval L No. 25).

Descriptive statistics were computed using SPSS, and the network structure estimation was conducted in R (version 4.4.1) employing several dedicated packages, namely bruceR, qgraph, networktools, bootnet, and mgm.

The study enrolled a total of 733 subjects (N = 733), with a mean age of 14.81 years ± standard deviation of 1.63 years. Among the 733 individuals in this cohort, 548 (74.8%) were of Han ethnicity. Among the 548 people in this cohort, 74.8% are of Han ethnicity. In addition, 536 participants (73.1%) were urban residents, and 341 participants (46.5%) were only children. Detailed demographic characteristics are presented in Table 1 .

Figure 1 illustrates the network structure of depression, anxiety, and childhood abuse in Depressed Adolescents with suicidal ideation. Out of 210 possible edges, 109 (51.90%) non-zero edges were observed, with an average weight of 0.04. The network demonstrated overall positive correlations among symptoms. Predictability of the symptoms is represented in Figure 1 as a ring pie chart, with an average predictability score of 0.46. This indicates that adjacent nodes in the model can explain 46% of the variance of each node on average.

In the network model, as depicted in Figure 1 , the nodes representing “emotional neglect” (EN) and “physical neglect” (PN) exhibited the strongest direct relationship within the childhood abuse symptom group (r = 0.48). This was followed by the association between “emotional abuse” (EA) and “physical abuse” (PA) (r = 0.32), and between EA and EN (r = 0.31). Meanwhile, within the depressive symptom group, the node representing “anhedonia” (PHQ1) and the node “sad mood” (PHQ2) were most directly connected (r = 0.31). This was followed by connections between the PHQ1 and the “fatigue” node (PHQ4) (r = 0.28), the PHQ4 and the “appetite” node (PHQ5) (r = 0.25), and the “worthless” node (PHQ6) and the “Death” node (PHQ9) (r = 0.23). In the anxiety symptom group, the connection between node “nervous” (GAD1) and node “Control worry” (GAD2) (r = 0.36) is the most direct. This is followed by the connection between node GAD2 and node “worry too much” (GAD3) (r = 0.31), and the connection between GAD3 and node “relax” (GAD4) (r = 0.19). Furthermore, there are numerous interconnections between symptoms across the three communities. For instance, the node “restless” (GAD5) is most closely associated with the node “motor” (PHQ8) (r = 0.23). Additionally, there is a notable connection between PHQ9 and EA (r = 0.13). Supplementary Table 1 details all edge weights within the network.

The intensity of nodes in children and adolescents with MDD and suicidal ideation is depicted in Figure 2 . Among these nodes, GAD2 emerges as the most robust, with nodes PHQ4 and EA also demonstrating statistically significant strength compared to most other nodes in the network. This indicates that these nodes are more central to the overall network structure. Additional centrality indicators can be found in Supplementary Figure 1 .

To gain a deeper understanding of the relationships between different symptom communities, the bridge strength of each node was calculated, resulting in the bridge network illustrated in Figure 1 . According to the bridge strength analysis shown in Figure 3 , the nodes PHQ8, PHQ9, GAD5, and EA exhibit higher bridge strength compared to most other nodes. This signifies that these four nodes serve as critical bridging symptoms, linking various symptom communities within the network. Table 2 presents the raw values for the strength of all symptoms and their respective bridge strengths. We estimated networks for the Han and urban subgroups, and their network structurekey, central symptoms and bridge symptoms were largely consistent with the overall sample ( Supplementary Figures 5 , 6 ).

As illustrated in Supplementary Figure 2 , the bootstrap 95% CI is narrow, indicating that the estimates of edge weights are both accurate and stable. Results from the bootstrap difference test for edge weights ( Supplementary Figure 3 ) reveal that the strongest edges in the network are found between items from the childhood abuse scale. Supplementary Figure 4 shows the difference test of node strength. The stability of the CS for node strength and bridge strength is reflected in a CS coefficient of 0.75 and 0.67, indicating that these estimates are sufficiently stable (see Figure 4 ). Additionally, the CS coefficients for the remaining centrality indicators are all greater than 0.25 (see Supplementary Figure 7 ). Additionally, the subgroup networks showed good stability ( Supplementary Figures 8 , 9 ). Given that the estimates of node strength are more reliable, the interpretation of the results in this study primarily focuses on node strength.

We have found that depression and anxiety symptoms in adolescents with suicidal ideation are highly interconnected and form a cluster. Within the depression and anxiety network, we observed that “control worry” and “fatigue” show high strength, making them core symptoms of the network, which is consistent with previous research conducted in child and adolescent populations (63, 64). Fatigue not only serves as one of the important symptoms for the diagnostic criteria of depression (47), but has also been mentioned in numerous studies of depressive symptoms. For instance, a depression symptom network study involving 3,463 outpatients with depression in the United States revealed that lack of energy has the strongest node centrality (65). Another survey conducted among college students demonstrated that fatigue is associated with more severe depressive symptoms and a higher risk of suicidal and anxiety tendencies (66).

Empirical evidence indicates that adolescents with depression exhibit a distinct depressive symptom profile characterized by pronounced fatigue, in contrast to adult populations (67). From a developmental psychopathology perspective, this phenomenon may be attributed to the interplay of multiple biological and psychosocial factors, including pubertal hormonal fluctuations and circadian rhythm dysregulation, which collectively predispose adolescents to increased vulnerability to persistent fatigue and energy depletion (68, 69). This finding prompts us to investigate the sensation of fatigue in depressive symptoms, which may uncover the complex relationships between depression, anxiety, and suicide. Interestingly, related neuroimaging studies have also shown that reduced neuronal activity in relevant brain regions of individuals with depression may explain the symptom of fatigue (70).

Furthermore, we found that “control worry” consistently exhibits high node centrality across almost all network models studying depressive and anxiety symptoms. Despite using a sample of patients with depression in this study, we obtained similar results. Childhood and adolescence are critical periods for brain structural changes and development (71), potentially related to cognitive formation. According to the cognitive theory of suicide (72), the occurrence of suicidal behavior is associated with negative cognitions among individuals with depression. Children and adolescents with depression often have negative self-cognitions, manifested as pessimistic expectations about themselves, their environment, and the future, which may lead to persistent tension and worry, and subsequently, suicidal ideation.

Within the overarching framework of the depression-anxiety-childhood abuse network, EA emerges as a node with greater strength, alongside “control worry” and “fatigue”, echoing analogous findings in prior research on childhood trauma networks (42). A deeper exploration reveals that the most pronounced edges are predominantly observed among the various forms of childhood maltreatment, with a notable emphasis on EA. Our findings well with those reported in earlier studies (73). Indeed, previous scholars have documented the frequent coexistence of psychological abuse with PA (74). This concurrence supports our research conclusions and further substantiates the intricate interplay between diverse manifestations of childhood maltreatment. We postulate that, within the familial sphere, the occurrence of childhood maltreatment, including sexual abuse, may exacerbate familial conflicts. Both the familial milieu and maltreatment experiences have the potential to impact children’s depressive and anxiety symptoms (75). A comprehensive meta-analysis indicates that individuals exposed to childhood maltreatment exhibit a twofold increased risk of recurrent depressive episodes compared to those without such histories (76). Hence, irrespective of the maltreatment type, any such experience poses a significant threat to the mental health of children and adolescents.

Conversely, in the realm of psychiatric symptoms, the strongest edges are predominantly found within the respective communities of depressive and anxiety symptoms, rather than between them, aligning with extensive prior research. Across these investigations, we consistently identify the most robust edges within depressive or anxiety symptom clusters, using PHQ-9 and GAD-7 scales to identify connections such as GAD1-GAD2, GAD2-GAD3, and PHQ1-PHQ2 (77, 78). However, differences in the pattern of strongest edges become evident when our findings are compared with certain studies (79), potentially attributed to our focus on healthcare workers. In the post-pandemic era, the frequent revisions of infection control protocols and guidelines have introduced an element of uncertainty. Compared with the general populace, healthcare workers confront unique challenges, including the implementation of infectious disease prevention measures and the management of critically ill patients (80), which may manifest in distinct depressive and anxiety profiles.

Nonetheless, among the top ten strongest edges identified, only one bridges the depressive and anxiety symptom communities, specifically PHQ8 and GAD5, resonating with prior research (77). This cross-community edge underscores the potential of these symptoms as candidate bridge symptoms within the entire network. From a clinical standpoint, the strong association between the “Motor” and “Restless” nodes suggests that interventions targeting the somatic manifestations of internal distress, particularly psychomotor agitation, may simultaneously alleviate symptoms of both depression and anxiety.

Regarding the connection between adverse childhood experiences and depressive and anxiety symptoms in adolescents, we found nodes within the communities of childhood trauma, depression, and anxiety that collectively serve as crucial bridges. Specifically, PHQ8, PHQ9, GAD5, and EA exhibit the highest bridge strength. As illustrated in Figure 1 , PHQ9 and EA link the two communities of depression and childhood maltreatment (81). Previous researchers posited that emotional abuse constitutes a persistent, repetitive, and inappropriate emotional response to children’s emotional expressions and their accompanying behaviors.

According to the cumulative risk hypothesis, negative events in early childhood, such as maltreatment, may have a cumulative effect, leading to increased internalizing and externalizing behavior problems in the future (82), which may ultimately result in suicidal ideation. The despair theory is often utilized by researchers to explain the linkage between negative events, emotions, and suicide (83). On one hand, individuals with depressive disorders may develop suicidal ideation due to cognitive changes that lead to feelings of hopelessness and diminished Reasons for Living (84). On the other hand, individual negative life events, such as emotional abuse, can also induce feelings of despair. Specifically, despair mediates the relationship between emotional abuse and depressive symptoms, and negative life events combined with negative cognition can predict despair and depressive symptoms (85), potentially leading to the emergence of suicidal ideation. Our network analysis identified cognitive-affective symptoms as central mediators that transmit the impact of emotional abuse toward suicidal ideation, highlighting a potential psychopathological pathway underlying this association.

In China, parents often see their children as “private property” and believe they have the right to beat their children when they try to educate and “regulate” them (86). Therefore, it is necessary to fundamentally change this concept and phenomenon. In addition, the incidence of child abuse can be reduced to some extent through training (87). Target adolescents exposed to childhood abuse to develop positive personality traits and improve emotional regulation skills (88). This measure is essential for children who have already been diagnosed with mdd. It is important to note that when children experience more abuse, even more positive experiences (such as social support, peer care, etc.) do not significantly improve the exposure of adverse experiences to mental health problems (89), so it is very possible to prevent childhood abuse. However, there are documented reasons why child abuse has not yet been recognized as a social problem worthy of public attention in China (90), and this phenomenon needs to be greatly changed. On the one hand, correct guidance of public opinion can be established through media and Internet to reduce the occurrence of childhood abuse. On the other hand, strengthen the formulation of relevant laws and regulations, restrict the parenting behavior, and maximize the protection of children and adolescents from domestic abuse.

This study identified key bridge nodes linking depression, anxiety, and childhood abuse through bridge network analysis, offering a new perspective for clinical identification and intervention. PHQ8, GAD5, and EA emerged as important hubs connecting different psychological problems. In outpatient screening, elevated scores on these bridge nodes may indicate the presence of comorbid depression and anxiety as well as a history of childhood abuse, warranting further assessment of suicide risk (91). In terms of treatment, for individuals with high scores on bridge nodes such as EA and GAD5, different strategies may be applied. For example, targeting restlessness (GAD-5) directly may be effective, as studies have shown that early improvement in restlessness is associated with remission in major depression (92). For patients with high EA scores, it is essential to assess for a potential history of childhood trauma. Interventions focusing on trauma and family functioning may help reduce the mutual reinforcement between depression and anxiety, potentially lowering overall suicide risk (93).

The strength of this study lies in its focus on clinically diagnosed depressive populations, where it identifies children with high-risk suicidal ideation and incorporates childhood maltreatment, along with depressive and anxiety symptoms, into the model analysis.

However, the study also has certain limitations. Firstly, it only includes childhood maltreatment as an environmental factor in the network model, neglecting other risk factors. Confounding variables (e.g., socioeconomic status, pharmacologic treatment, or trauma recency) were not controlled, which may subtly influence certain symptom-to-symptom associations. Nonetheless, we believe the core findings—such as the strong bridge connections between abuse and symptoms—are robust. We recommend future studies include more risk factors and control for confounders to further validate these associations. Secondly, as a cross-sectional study, we cannot establish the temporal sequence and causal relationship between childhood maltreatment experiences and depressive/anxiety symptoms. Results from a single time point are inadequate for understanding the entire process. Future studies should conduct dynamic observations of variables, with longitudinal data and cross-lagged network models potentially providing better explanations. Thirdly, The SBQ-R was used for measurement. However, the information obtained from individual item scores is relatively limited, as the screening was based solely on questions related to suicidal ideation. Future research could further investigate the causal inference of abuse and depressive symptoms on suicidal ideation within network structures (94). We also considered combining the SBQ-R with the Columbia Suicide Severity Rating Scale to assess different dimensions of suicidality. Lastly, as our sample was drawn from a single hospital, the generalizability of our findings is limited to similar clinical populations. Future studies will aim to recruit adolescents with depression from multiple hospitals and diverse regional backgrounds.