Article Highlight | 20-Nov-2026

Shared symptom architecture in adolescent self-injury and suicide attempts: A 2,496-participant network analysis

Health Data Science

Background

Non-suicidal self-injury (NSSI) refers to deliberate, repetitive bodily harm without lethal intent, often serving emotion regulation; suicide attempts (SA) involve an intent to end one's life. The two are conceptually distinct yet frequently co-occur in adolescents. Global incidence of NSSI in children and adolescents is approximately 19.5%, rising to 27.4% among Chinese adolescents aged 13 to 18.

Prior work has largely treated the two as separate diagnostic entities. DSM-5 added NSSI as a distinct behavioral disorder in 2017, but such classifications face challenges in reliability, validity and clinical utility, and a "common cause" model explains comorbidity and dynamic interaction poorly. Network theory offers an alternative: psychopathology as a network of central nodes and edges, where high-centrality nodes may serve as intervention targets.

Methods

An online survey was conducted between September and December 2022 across three high schools in a city in Jiangxi Province, China, reaching 9,140 students. After removing responses with completion times beyond three standard deviations from the mean, 7,030 remained (participation rate 76.91%); after excluding incomplete responses, 2,496 adolescents reporting a history of NSSI entered the network analysis, of whom 23.03% had a history of SA.

Instruments included the Adolescent NonSuicidal Self-Injury Assessment Questionnaire (ANSAQ, 12 items), the Patient Health Questionnaire-9 (PHQ-9), the Chinese version of the Difficulties in Emotion Regulation Scale (DERS, 36 items, six dimensions), the Chinese Mobile Phone Addiction Tendency Scale (MPATS, 16 items, four dimensions), and the Chinese Interpersonal Reactivity Index (C-IRI, 28 items, four dimensions).

Regularized partial correlation networks were fitted separately for the two groups (with and without SA) using EBICglasso in R 4.3.0. Expected influence (EI) and bridge expected influence (BEI) were computed. Stability was assessed with bootstrapping (1,000 iterations, 95% CIs) and networks were compared using the Network Comparison Test (1,000 permutations), with Holm–Bonferroni correction for multiple comparisons.

Results

Network density was comparable. Non-zero edges accounted for 43.36% (258 of 595) and 38.82% (231 of 595) respectively, mean edge weight was 0.024 in both, and global strength did not differ significantly (S = 16.498 and 16.080; P = 0.136). The network invariance test indicated a significant difference in average degree (M = 0.173, P = 0.002), suggesting at least one edge differs.

Centrality indices were stable: the centrality stability coefficient was 0.75 (NSSI-NoSA) and 0.67 (NSSI-SA), above both the recommended minimum of 0.25 and the preferable threshold of 0.5.

The strongest EI node in both networks was "limited access to emotion regulation strategies," the DERS dimension reflecting a restricted repertoire of adaptive coping (E = 1.389 and 1.393). The strongest BEI node in both was C-IRI "personal distress," the emotional reaction to others' misfortune (BEI = 0.497 and 0.552).

After Holm–Bonferroni correction, only "perspective taking" remained significant among centrality differences (E(Δ) = −0.215, P < 0.001), higher in the group without SA. For edge weights, 44 edges differed significantly before correction; none remained significant afterwards.

The EI of NSSI 4 ("intentionally hitting walls, tables, and other hard objects") differed significantly between groups, higher in the NSSI-SA group. The edge weight between NSSI 4 and NSSI 5 ("intentionally hurting oneself by hitting with a fist, palm, or hard object") also differed significantly, roughly twice as high in the NSSI-SA group.

Implications

The authors note that the high centrality of limited strategies may primarily reflect strong within-construct correlations among DERS dimensions rather than cross-domain influence. Still, adolescence is a sensitive period for emotion regulation development, and improving adaptive strategy use may indirectly affect comorbid symptoms. The paper cites evidence that cognitive behavioral therapy and dialectical behavioral therapy can modify emotion regulation strategies in adolescents, and that school-based suicide intervention measures can effectively reduce attempts.

On bridge nodes, the authors likewise counsel caution: the centrality of "personal distress" may mainly derive from connections within the DERS community rather than cross-community transmission.

Limitations

The authors list several limitations: the study is exploratory and clinical interpretation should be cautious; the design is cross-sectional with a nonclinical sample and limited sociodemographic detail; data were collected in the late stages of the COVID-19 pandemic and rely on self-report, which may overestimate the number with a history of SA; scale total scores were not included in the network. Future work will employ a longitudinal design and cross-cultural, hierarchical network comparisons.

Disclaimer: AAAS and EurekAlert! are not responsible for the accuracy of news releases posted to EurekAlert! by contributing institutions or for the use of any information through the EurekAlert system.