This editorial refers to “Network perspective on rumination and non-suicidal self-injury among adolescents with depressive disorders” by Zhang et al, 2026; https://doi.org/10.5498/wjp.v16.i1.113130.
INTRODUCTION
Over the past half-century, psychiatric research and clinical practice have been predominantly guided by two overarching frameworks: The categorical approach, as epitomized by diagnostic systems such as the Diagnostic and Statistical Manual of Mental Disorders and International Classification of Diseases, which classify mental disorders into discrete entities based on symptom checklists; and the dimensional approach, which conceptualizes psychopathology along continuous spectra of severity. While these paradigms have provided a common language and advanced the field considerably, their limitations in fully capturing the complexity, heterogeneity, and dynamic nature of mental illness have become increasingly apparent. Critiques often center on issues related to diagnostic co-occurrence, within-diagnosis symptom heterogeneity, and the elusive nature of the latent disease entities presumed to underlie symptomatic presentations.
In response to these limitations, network theories relating to psychological constructs have undergone marked development over recent decades within the field of psychiatry[1]. These theories are predicated on the premise that psychological phenomena, such as emotions and psychopathology, can be conceptualized as complex systems of interacting components[2]. Established constructs are reconceptualized not as latent entities but as emergent properties arising from the interplay within these multivariate systems in network theories[3]. Within this conceptual framework, the central concept is that an episode of mental disorder arises from causal interactions between its symptomatic elements. This concept has especially resonated with those clinical scientists who are disenchanted with traditional categorical and dimensional approaches to mental illness[1].
A psychological network is composed of nodes, which represent variables that constitute the construct of interest, and edges that connect these nodes, thus reflecting the conditional associations between pairs of variables[1]. The dynamic nature of a psychological network is recognized as another promising path for the acquisition of causal knowledge relating to how different variables maintain mental disorders[4]. Building on these methodological foundations, Zhang et al[5] published a study in the World Journal of Psychiatry constructed an association network between depression-related cognitions (i.e., rumination subtypes) and behaviors [i.e., non-suicidal self-injury (NSSI)]. Their findings indicated that “symptom-focused rumination” functioned as a core bridge node connecting with various NSSI methods, while “scratching oneself” emerged as the most influential NSSI behavior. When considering the psychopathological network proposed by Zhang et al[5], it appears that negative cognitions associated with depression and maladaptive behaviors are intricately linked in a specific manner, thus implying that targeted interventions that address these pivotal nodes may yield promising therapeutic outcomes.
However, translating these statistical indicators into effective clinical interventions necessitates the navigation of three critical considerations. First, distinction between the statistical centrality of a specific symptom and its causal role in maintaining the network, which may lead to inefficient targeted interventions. Second, the directional ambiguity of bridge nodes, particularly in terms of key drivers between network groups, which render clinical decision-making uncertain. Third, the potential influence of unmeasured external variables as primary drivers of the entire system of symptoms, which could result in the observed network being a spurious representation, thereby misdirecting therapeutic focus. Collectively, these issues constrain the clinical utility of network-based findings and precision interventions for NSSI. In this editorial, we provide a detailed analysis of these gaps and key directions for addressing them, while offering practical clinical recommendations feasible within current methodological constraints.
CRITICAL ISSUES FOR PSYCHOLOGICAL NETWORK ANALYSIS
The statistical centrality of a symptom does not necessarily indicate its causal influence within a network. Zhang et al[5] identified “scratching oneself” as the most central node in terms of expected influence, thus highlighting its high degree of interconnectivity within the network. However, this statistical prominence does not inherently imply that scratching acts as a primary causal driver that sustains the system. An alternative hypothesis is that scratching may be a common and low-threshold expression of distress. Scratching is easily accessible and frequently co-occurs with other symptoms without necessarily activating them. Interventions targeted to scratching behavior may prove insufficient for effectively modulating the broader symptom network if the behavior merely represents a distal manifestation (or downstream consequence) of distress. A more robust method for inferring causal centrality within a network is to collect longitudinal data from study participants and perform a Cross-Lagged Panel Network (CLPN) analysis. The core principle of CLPN is to model the relationships between all variables in a network across at least two or more time points. It does this by simultaneously estimating two key types of effects. Cross-lagged paths estimate whether the level of one variable at time “t” prospectively predicts the level of another variable at time “t + 1”, after controlling for the auto-regressive stability of the latter variable. This temporal precedence is a fundamental criterion for establishing potential causality. The contemporaneous networks represent the relationships between variables within the same time point, after accounting for the temporal and auto-regressive effects from the previous measurement. This helps identify which symptoms cluster together at a given moment. Therefore, within the CLPN, if scratching possesses genuine causal influence, we would expect to see significant cross-lagged paths from scratching at one time point to multiple other symptoms at the next time point. Conversely, if scratching is merely a downstream consequence, the dominant cross-lagged paths would flow from other symptoms to scratching, with little to no predictive power from scratching to the rest of the network. However, we need to acknowledge that such methods still carry the inherent risk of conflating statistical centrality within probabilistic networks with causality in symptom dynamics[6].
The cross-sectional nature of data introduces directional ambiguity, particularly for bridge nodes. Zhang et al[5] emphasized “symptom-focused rumination” as the key bridge that links cognitive clusters of rumination with the behavioral clusters of NSSI. This finding aligns with the emotional cascade model[7], which posits that intense rumination on negative feelings drives the desire for emotional escape, a function that NSSI fulfills. Yet, the reverse pathway is equally plausible: Engaging in self-injurious behaviors, along with the resulting physical sensations and emotional aftermath, may provoke intense rumination relating to symptoms, pain, and existential feelings of brokenness. In this cross-sectional network, it is impossible to identify whether cognition drives behavior or vice versa. Such statistical inference is inadequate and troubling for a clinical psychologist, as this strategy provides limited guidance for intervention planning. If rumination precedes NSSI, therapeutic priorities might focus on cognitive strategies such as attentional deployment or interventions based on mindfulness[8]. In contrast, if NSSI triggers rumination, the emphasis should shift toward alternative learning and stabilizing behaviors[9]. Given the inherent difficulty in obtaining longitudinal data, some researchers have opted to utilize Bayesian directed acyclic graphs (DAGs) to characterize directionality. Unlike the undirected edges in a standard association network, which simply show that two variables are related, a DAG uses directed edges (arrows) to represent proposed causal pathways. An edge issuing from node X and incident on node Y means that the presence of node Y more strongly implies the presence of node X rather than vice versa. In this manner, cross-sectional data can be modeled as a causal system. However, the Bayesian DAGs appear overly idealized, as they rely on statistical assumptions about the absence of bidirectional relationships between nodes that contradict clinical observations in psychopathology[1]. Nevertheless, this remains a feasible approach when working with cross-sectional data, as cyclic relationships do not always exist. Exploring the directional relationships between variables to accumulate robust evidence is also a prudent choice. Addressing this challenge may ultimately require the development of methodologies that integrate dynamic network systems, constructed via ecological momentary assessment, with emerging computational intervention simulations to provide more robust evidence[10].
It should also be acknowledged that network analysis currently faces a notable crisis in the form of reproducibility, particularly with regards to the consistency of network edges across different samples[11]. The findings reported by Zhang et al[5] were derived from a specific population (adolescents diagnosed with depressive disorders); furthermore, the network structure may differ substantially between community and clinical samples[12]. The generalizability of these specific connections between symptoms for other populations remains a key empirical question. It is challenging for a single research team to conduct diverse and cross-cultural sampling. Instead, a more promising avenue may be to enhance open data access and meta-analytic techniques. Finally, similar to several studies, the network characterized by Zhang et al[5], inevitably faced the “third variable problem”[13]. Experiences such as childhood trauma, ongoing family conflict, peer victimization, or academic pressure may act as primary drivers of the entire symptom system[14]. In such cases, these coupling within-syndrome symptoms may actually be attributed to external factors in the observed network structure.
SENSIBLE IMPLEMENTATION OF THE NETWORK APPROACH IN THE CLINIC
Application of network approaches in depression as reference
Despite the methodological challenges described earlier, network perspective can still yield valuable contributions to clinical practice when employed in a judicious manner. In depression research, application of the network approaches can extend far beyond descriptive scenarios, such as identifying core symptoms and potential subtypes[15]. The network analysis is now integrated into interventional strategies, with preliminary evidence suggesting that targeting central depressive symptoms may induce cascading therapeutic effects across the network, thus leading to broader symptomatic relief[16]. Network analysis has also been utilized to determine which symptoms improve (or fail to improve) following specific interventions for depression[17]. Compared with the examination of symptom sum scores, investigating the treatment-related change of a specific symptom may facilitate a more nuanced understanding of how a particular treatment works. In terms of prevention, network analysis has previously identified difficulty falling asleep as a risk factor for first-onset major depressive disorder, thus providing an actionable and sensible target with which to combat the global burden of depression[18].
The potential of applying network approaches in NSSI
The potential application of network approaches in the clinic for NSSI has yet to be fully investigated. While there is a growing body of NSSI literature utilizing the network analysis, most of the existing literature focused predominantly on the interconnectivity between NSSI behaviors[19], or merely investigated widely correlated factors[20]. Following this novel conceptualization of NSSI psychopathology, a critical next step is to translate these insights into precision psychological medicine[21]. Conducting randomized controlled trials to test the causality of identified NSSI network hubs represents a key initial undertaking. Furthermore, establishing formal theories that are capable of elucidating these phenomena will significantly facilitate the development of specific psychotherapy programs for NSSI.
A network approach-based treatment model for NSSI
The primary clinical application of the network approach lies in identifying the optimal focus of intervention plans, a goal that most network analysis studies of symptom interrelations strive to achieve. However, it is important to combine network configurations with clinical judgment, a theoretical understanding of symptom mechanisms, and the consideration of intervention feasibility. A symptom may be statistically influential yet difficult to modify directly; instead, a less central but more malleable symptom might serve as a more practical intervention point. Clinicians must remain vigilant to network centrality as a target for intervention. Indeed, recent simulation studies have shown that removal of the central node does not always yield the greatest reduction in overall network connectivity[22]. Targeting a moderately central node that occupies a critical structural position may be more effective. Therefore, for future interventions for NSSI, clinically viable targets may include both central and moderately central nodes, such as scratching and cutting, as identified by robust existing evidence[5,19].
In addition, the network approach is an option for monitoring NSSI treatment efficacy and preventing relapse. By repeatedly assessing network structures throughout the duration of treatment, clinicians can track how interventions alter interconnections between symptoms[17]. Precedent exists for employing network intervention analysis to investigate the temporal sequence and symptom specificity of interventions (e.g., cognitive-behavioral therapy for insomnia) targeting depressive symptoms[23]. Tracking this process over time has elucidated the evolution of symptom changes and their underlying association structure, distinguishing between symptoms directly impacted by the intervention and those indirectly affected through the improvement of mediating symptoms. This approach can similarly be applied to the clinical treatment of NSSI. Successful treatment might manifest as reduced network density, along with fewer and weaker connections between symptoms[24]. For psychological issues prone to recurrence, such as NSSI, leveraging network intervention analysis to establish the temporal sequence of symptom change is highly beneficial for relapse prevention. Monitoring for the early re-emergence of previously central connections could serve as a key warning of relapse.
Finally, the integration of the network approach and evidence-based practice may be particularly advantageous for designing multi-component interventions for complex mental health problems (e.g., NSSI), which frequently involve multifaceted etiologies[25]. Recognizing that distinct optimal intervention strategies may exist for different symptom structures, network meta-analysis will be instrumental in providing precise and personalized treatment recommendations for individuals with specific NSSI subtypes and varying illness trajectories.
CONCLUSION
Zhang et al[5] represent the current use of the network approach to identify potential targeted symptoms for interventions. However, translating statistical correlations into clinical practice within psychological networks is limited by gaps in our empirical and methodological knowledge. This underscores the need for further accumulation of evidence on causality, interventions, and cross-cultural dimensions. Optimizing the application of network analysis for personalized mental healthcare also requires the integration of clinical expertise. In this way, network science can better serve the treatment of complex psychological disorders.
ACKNOWLEDGEMENTS
We would like to express our gratitude to Ran Zhang (South China Normal University) for his voluntary assistance in improving the language quality of this article.
Peer review: Externally peer reviewed.
Peer-review model: Single blind
Specialty type: Psychiatry
Country of origin: China
Peer-review report’s classification
Scientific quality: Grade B, Grade B, Grade D
Novelty: Grade A, Grade B, Grade B
Creativity or innovation: Grade A, Grade B, Grade B
Scientific significance: Grade A, Grade A, Grade B
P-Reviewer: Abbasi S, Lecturer, Researcher, Pakistan; Li Y, Doctorate Student, China S-Editor: Bai Y L-Editor: Webster J P-Editor: Lei YY
