The sample consisted of 483 adolescents aged 12 to 17 years (M = 15.52, SD = 1.305; 58% female) from a psychiatric hospital and two secondary schools in China. The hospital sample included 216 outpatients who had one-on-one mental health assessments conducted by psychiatrists. The school sample consisted of 267 randomly selected middle school students who completed self-rating assessments while being monitored by their teachers in their respective classrooms (i.e., group testing).

Inclusion criteria were adolescents aged 12 to 17 years, capable of understanding and completing the assessments, and who consented (from their guardians) to participate. Exclusion criteria included severe cognitive impairments that might interfere with understanding or completing the assessments, and failure to pass the manipulation checks, designed to assess sincerity and attention to the questionnaire’s instructions.

For the hospital sample, patients were briefly introduced to the purpose of our study and asked to complete the relevant scales based on the previous week, after entering the psychological assessment room. They could ask their psychiatrist for help if they had any questions. Prior to the mental health assessment, the patient and parent/guardian were informed about the assessment process and the scales to be used, followed by the signing of an informed consent form by the parent/guardian and the patient, as required by local hospital ethics regulations. When the task was completed, the psychiatrist checked to make sure that all portions were completed, and the subject was asked to leave the assessment room.

The school sample conducted their self-rating assessments while being monitored by their teachers in their respective classrooms (i.e., group testing). All participants signed an informed consent form and were explained the rules regarding anonymity, confidentiality, and their right to quit the experiment.

The respondents from the hospital were all included because the psychiatrists ensured carefulness in the one-on-one tests. However, considering that some students from schools may not have taken the group test seriously, we used a manipulation check in their questionnaires [26]. Students were asked to select D for “This question has other uses, please select D.” If students did not select D, they were excluded from the data analysis. We included two similar questions serving as checks; those failing one of the items were excluded. Of the total 267 respondents, 115 were excluded because they failed the manipulation check. Therefore, a total of 368 participants (M = 15.43, SD = 1.22; 60% female) were included in the data analysis.

The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008. All procedures involving human subjects/patients were approved by IRB in Wenzhou Seventh People’s Hospital (EC-KY-2,022,048).

We used the Self-Injurious Thoughts and Behaviors Interview (SITBI) to assess adolescents in the clinical setting. The SITBI is a structured interview that assesses the presence, frequency, and characteristics of a wide range of self-injurious thoughts and behaviors, including SI, suicide plans, gestures of suicide, suicide attempts, and NSSI [27]. A previous study [28] reported a Cronbach’s α of 0.77 for the Chinese version of the SITBI in a sample of adolescents, demonstrating good internal consistency reliability. To merge the data for analysis in the present study, both groups of the participants answered yes or no to questions “Have you ever had thoughts of killing yourself?” regarding SI and “Have you ever had thoughts of purposely hurting yourself without wanting to die? (for example, cutting or burning)” regarding NSSI. Note that these are about their lifetime prevalence of NSSI and SI.

The Hospital Anxiety and Depression Scale (HADS) assesses both anxiety and depression, which commonly coexist [29]. The measure contains a total of 14 items, including seven for depressive symptoms (HADS-D) and seven for anxiety symptoms (HADS-A), with a focus on non-physical symptoms. An example item from the HADS-D is “I still enjoy the things I used to enjoy,” rated on a 4-point Likert scale (0 = definitely as much, 3 = hardly at all). Higher scores indicate more severe symptoms. One previous study [30] reported Cronbach’s α of 0.76 and 0.79 for the anxiety and depression subscales in Chinese adolescents, and confirmed the two-factor structure of the HADS through confirmatory factor analysis, supporting its construct validity.

The Difficulties in Emotion Regulation Scale (DERS) is a 36-item self-report measure of six facets of ER. Items are rated on a scale of 1 (“almost never” [0–10%]) to 5 (“almost always” [91–100%]) [31]. Higher scores indicate more difficulty with ER. The psychometric properties of the DERS and its subscales are described herein. The Cronbach’s alpha values of the scale and six subscales range from 0.88 to 0.96 and test-retest reliability from 0.52 to 0.77. The results of the measurement conform to the necessary psychometric properties. It is suitable for measuring the degree of difficulty with emotion regulation in Chinese adolescents [32].

There have been several revised versions of the SPSI-R for use in the Chinese language [33]. The present study used the Chinese version, which shows both good reliability and validity. The scale consists of 52 items rated on a 5-point Likert scale (0 = not at all true of me, 4 = extremely true of me). An example item is “When my first efforts to solve a problem fail, I become uneasy about my ability to handle the situation.” The overall Cronbach’s alpha is 0.85, and the rational problem-solving (RPS), avoidance style (AS), negative problem orientation (NPO), positive problem orientation (PPO), and impulsivity/carelessness style (ICS) subscale values are 0.85, 0.82, 0.70, 0.66, and 0.69, respectively. The SPSI-R uses a five-point Likert-type scale ranging from 0 to 4.

Network analyses with frequency variables of NSSI and SI were conducted in R using the mgm package, which allows for the explicit estimation of graphical models using mixed variable types [34]. The parameters were specified as lambdaGam = 0.25 and alphaGam = 0.25 in the mgm function. The glmnet package was used to provide regression analysis with L1 and/or L2 regularization. With this algorithm, one variable was taken as Y and the others as predictors, resulting in a coefficient matrix illustrating the relationships among these variables. Based on this coefficient matrix, the network could be plotted (see Fig. 1).

We calculated several indices of node centrality to identify the symptoms or components most central to the network. For each node in the male and female networks, we calculated the strength (i.e., the absolute sum of edge weights connected to a node), closeness (i.e., the average distance from the node to all other nodes in the network), betweenness (i.e., the number of times a node lies on the shortest path between two other nodes), and expected influence (i.e., the sum of edge weights connected to a node) (see Fig. 2).

Additionally, we conducted logistic regression analyses to examine the relationships between NSSI, SI, and their related factors (i.e., depression, anxiety, emotion regulation difficulties, and problem-solving abilities) separately for males and females. The dependent variables were the presence of SI (0 = no, 1 = yes), while the independent variables were the scores on the NSSI, HADS-D, HADS-A, DERS, and SPSI-R subscales.