This cross-sectional study, conducted from January 13 to March 18, 2022, was supported by the Psychological Health Education Center at Zhengzhou University. In China, the surveyed location was a Ministry of Education-designated world-class university and a first-class discipline construction university, consistently ranking among the top five nationwide in student enrollment each year, thereby ensuring the representativeness of the survey sample. We employed a multistage sampling method, using seven academic categories (e.g., Medicine, Science, Engineering) as primary units, and degree types (undergraduate and graduate) as secondary units, with stratified sampling throughout. Subsequently, classes served as third-stage units, with cluster sampling randomly selecting students from several classes within each combination of discipline and degree type. Similar to other studies (29, 30), the “Questionnaire Star” program integrated into WeChat was adopted to collect data. An electronic questionnaire was disseminated to 4,952 currently enrolled university students, resulting in the collection and inclusion of 4,698 responses, yielding a response rate of 94.87%. The questionnaire was distributed via the WeChat and Tencent QQ platforms. To ensure data quality, measures implemented: restricting each IP address to a single submission, excluding responses with completion times shorter than 240 seconds or longer than 10 minutes, eliminating incomplete questionnaires, and discarding those with contradictory answers. As a result, a total of 3,896 eligible participants were included. Prior to participation, all subjects provided electronic informed consent. This study received approval from the Ethics Committee of Zhengzhou University (ZZUIRB2022–06).

Power analysis was commonly employed to ascertain the requisite sample size for a study. According to Epskamp and Fried’s research (31), we utilized the netSimulator function to simulate data within a given network model and anticipated network structure, thereby exploring the appropriate sample size necessary for detecting genuine effect sizes. As the network encompasses a greater number of nodes, a larger sample size is necessitated to achieve an equivalent level of reproducibility. Consequently, we conducted data simulations for a network model encompassing comorbid symptoms and health-promoting lifestyles, with the result shown in Supplementary Figure S1 . And when N = 1600, the network’s correlation and strength were above 0.9, and the effects on other metrics were also acceptable (above 0.7).

All statistical analyses were performed using the R-Studio program (version 4.3.2). First, descriptive statistical methods were used to summarize demographic information and item scores from each scale. Then, we constructed a comorbid network of anxiety, depression, and sleep problems, in addition to a comprehensive network of comorbid symptoms and their association with health-promoting lifestyles. Network analysis can establish bivariate relationship among multiple variables and reveal the structure and nature of entire complex system of mental disorders caused by the causal interactions among variables (including core components, co-occurrence relationships, and critical nodes), which cannot be explained by regression analysis and latent variable analysis (42).

Before constructing the networks, potential item redundancy was checked using the goldbricker function from the R package networktools (version 1.5.1). Following Jones’ manual (43), if the proportion of significantly different central correlations between two variables and other items is less than 25%, then it can be confirmed that these two items measure the same trait or symptom (i.e., redundancy).

The demographic characteristics of the overall sample are presented in Table 1 . Among the 3,896 participants included in this study, the mean age was 21.92 years (SD = 2.56), with 1,587 (40.7%) being male and 2,309 (59.3%) female. Of these, 1,824 (46.8%) were undergraduates, and 2,072 (53.2%) were graduate students. Altogether, 1,485 (38.1%) reported experiencing anxiety (GAD-7 total score ≥ 5), 1,745 (44.8%) reported experiencing depression (PHQ-9 score ≥ 5), and 629 (16.1%) reported experiencing sleep problems (PSQI total score ≥ 8). Supplementary Table S1 displays the mean, standard deviation (SD), skewness, kurtosis, strength, bridge strength, and predictability of all items across the scales. In both the comorbid network of anxiety, depression, and sleep problems, and the comprehensive network of comorbid symptoms and health-promoting lifestyles, no item was found to be redundant with any other item (i.e., showing less than 25% significantly different correlations).

As illustrated in Figure 1 , within the network structure encompassing anxiety, depression, and sleep problems, there are a total of 131 non-zero edges out of 231 possible edges (network density = 0.57), with an average weight of 0.043. The five most prominent edges are all within specific communities: Sleep duration - Sleep efficiency (PSQI3 - PSQI4), Subjective sleep quality - Daytime dysfunction (PSQI1 - PSQI7), Nervousness - Uncontrollable worry (GAD1 - GAD2), Anhedonia - Low energy (PHQ1 - PHQ4), and Abnormal behavior & speech - Self-harm even suicide (PHQ8 - PHQ9) ( Supplementary Table S2 ). Significant edges connecting different communities include Self-harm even suicide - Use of sleep medication (PHQ9 - PSQI6), Low energy - Daytime dysfunction (PHQ4 - PSQI7), and Feeling afraid - Self-harm even suicide (GAD7 - PHQ9) ( Supplementary Table S2 ). Additionally, Excessive worry (GAD3) and Trouble relaxing (GAD4) exhibit the highest predictability (0.747, 0.742), whereas Use of sleep medication (PSQI6) has the lowest predictability (0.173). The average predictability is 0.56, indicating that, on average, more than half of the variance in the nodes can be explained by their adjacent nodes ( Supplementary Table S1 ). Network stability and accuracy tests are shown in Supplementary Figure S4 , with both strength and bridge strength centrality being 0.75, exceeding the recommended threshold of 0.5.

The centrality plot ( Figure 2A ) indicated that Low energy (PHQ4) exhibits the highest strength centrality, followed by Daytime dysfunction (PSQI7) and Trouble relaxing (GAD4), suggesting that these nodes occupy the most significant and influential positions within the comorbidity symptom network. According to Jones’ perspective (50), bridge symptoms are selected using the 80th percentile threshold for bridge centrality. As shown in Figure 2B , Daytime dysfunction (PSQI7), Self-harm even suicide (PHQ9), Sad mood (PHQ2), Low energy (PHQ4), and Feeling afraid (GAD7) exhibit the highest bridge strength, identifying them as the pivotal bridge symptoms linking anxiety, depression, and sleep problems comorbidity. The bootstrapped difference test ( Supplementary Figure S4 ) for node strength and bridge strength further corroborate that these nodes are statistically significantly stronger than other nodes within the network.

The network of comorbid symptoms and health-promoting lifestyles is illustrated in Figure 3 , with stability and accuracy tests presented in Supplementary Figure S5 . Regarding strength centrality, Low energy (PHQ4), Daytime dysfunction (PSQI7), and Trouble relaxing (GAD4) remain the most significant and influential comorbid symptoms, while Physical activity, Health responsibility, and Stress management are identified as the top three central HPLs ( Supplementary Figure S2 ). In terms of correlations, Spiritual growth and Physical activity exhibit the most negative edges with comorbid symptoms. Moreover, by evaluating the individual connection between each specific protective HPL node and the wider community of comorbid symptoms, Physical activity demonstrated the highest bridge strength. Specifically, Spiritual growth showed the strongest bridge strength with the depression community, Physical activity with the sleep problems community, and both Health responsibility and Stress management with the anxiety community ( Supplementary Table S3 ; Supplementary Figure S3 ). However, as Health responsibility showed mostly positive correlations with anxiety symptoms. Thereby, Physical activity, Spiritual growth, and Stress management were identified as the central mental health-promoting behaviors.

NCT revealed the invariances of global strength and network structure among the four groups ( Supplementary Figures S6–S9 ). In terms of global strength comparisons across the four subsamples, significant difference was only found in mental health status (non-mentally disordered: 8.72 vs at least one mental disorder: 13.21, p = 0.01). Other comparisons showed no significant differences (male: 13.19 vs female: 12.99, p = 0.634; undergraduate: 12.98 vs graduate: 13.43, p = 0.139; only-child: 13.23 vs non-only-child: 13.52, p = 0.733). Regarding network structure comparisons, significant differences were observed in familial sibling status (M = 0.16, p = 0.030) and mental health status (M = 0.28, p = 0.01); The other two subsamples showed no significant differences (gender: M = 0.13, p = 0.129; educational level: M = 0.12, p = 0.158). Additionally, after applying the Bonferroni-Holm correction, all edge weights in the three subsamples remained non-significant (p > 0.05).

Low energy (PHQ4) exhibited the highest strength centrality, indicating its crucial role in activating and sustaining the comorbid network of depression, anxiety, and sleep problems. This phenomenon was also widely observed among Chinese nurses and nursing students and closely associated with Anhedonia (PHQ1) (55, 56). Low energy, manifesting as fatigue or exhaustion, is a core feature of depression (57). During the late stages of the COVID-19 pandemic, the prevalence of fatigue among nursing students reached 67.3% (58). Evidence suggested that this may be associated with the somatization of psychological distress, particularly prevalent among Asian adolescents. This cultural tendency favors physical expression over verbal expression as a coping strategy (55). The cumulative impact of intense academic pressures, limited recreational and social activities, and inadequate sleep on their physical and mental health often manifested as a state of low energy. Ariel et al. found that low energy as a crucial predictor across multiple domains of health and functional impairment, corroborating our findings (59). It may also serve as an early indicator of anhedonia. Neurological evidence indicated that energy depletion may inhibit the neural systems responsible for enhancing the reception of rewarding stimuli (60). Furthermore, as the second most central symptom, Daytime dysfunction (PSQI7) showed strong edge connectivity with Low energy (PHQ4), serving as bridge symptoms. This supported findings from studies on Macau residents, migrant Filipino domestic workers, and clinicians with depression symptoms (61–63). It has been identified as a prominent bridge symptom linking childhood trauma, sleep disorders, and depression in adolescents (64). Exploratory research into sleep structure revealed that sleep problems could be categorized into nocturnal symptoms and daytime consequences, both contributing to dissatisfaction with sleep quantity or quality. However, interference with daytime functioning was centrally positioned and may be a critical outcome measure in insomnia treatment (65). Related studies indicated that students with low energy are more sensitive to stressors, amplifying the negative consequences of insomnia rather than the experience of insomnia itself (66, 67). The clinical significance of Trouble relaxing (GAD4) has been well-documented in insomnia and psychopathology (59). Notably, trouble relaxing was highly correlated with restlessness and clusters within the psychomotor community. Our results also showed that these symptoms have the highest predictability within the network, corroborating Bai et al.’s finding that difficulty relaxing ranks second in EI, with restlessness showing the highest BEI after sleep problems (63). The inability to relax as a core nexus symptom in comorbid PTSD, depression, and anxiety among trauma-exposed samples (68). Outside the classroom and internship settings, interventions such as mindfulness-based approaches and relaxation exercises can alleviate students’ physical and mental stress (69).

Node bridge strength centrality may provide insights for identifying critical bridge symptoms in the connection and progression of mental disorders. Beyond low energy and daytime dysfunction, Self-harm even suicide (PHQ9), Sad mood (PHQ2), and Feeling afraid (GAD7) also emerged as bridge symptoms within the current network. As a global health concern, Self-harm even suicide (PHQ9) has consistently been recognized as a key intervention target within adolescent mental disorders (70). Numerous scholars have probed into the nexus between adolescent mental health and suicidal ideation, revealing anxiety as a prospective predictor of suicidal ideation; over 50% of completed suicides and 20%-48% of suicide attempts are attributed to depression (71). We also identified that within the depression community, suicidal ideation is most strongly associated with abnormal behavior & speech (i.e., psychomotor agitation and impulsivity). Among adults experiencing major depression episodes, risk-taking behavior and psychomotor impulsivity were believed to facilitate the transition from suicidal ideation to suicide attempts (72). Our study also identified Feeling afraid (GAD7) as another significant bridge symptom, closely related to suicidal ideation. Epidemiological investigations signify that 75% of lifetime mental disorders manifest before the age of 24, with anxiety disorders typically emerging from early adolescence to young adulthood (73). The factors unique to college students, such as student debt, employment uncertainty, academic pressure, and separation from family, showed that feeling afraid doubles the likelihood of suicidal ideation among those experiencing anxiety symptoms (74). Additionally, Sad mood (PHQ2) played a distinct bridge role in linking anxiety disorders, as corroborated by a bayesian network of anxiety-depression bridges (75). Stigma toward mental illnesses could also lead to the emergence of adolescent emotional problems (76).

The “Healthy China Action (2019–2030)” plan, formulated by the Healthy China Initiative Committee emphasizes the “big health” concept, integrating prevention and treatment through 15 major health promotion actions, shifting the focus from disease-centric to health-centric approaches. This offered valuable insights for our research. We found that physical activity occupied a central position in the network, potentially at the intersection of mental disorders and health. It remained the most prominent health node when quantifying the degree to which specific health-promoting behaviors correlate with changes in comorbid symptoms. Specifically, physical activity showed a significant negative correlation with sleep problems, potentially reducing the use of hypnotics, alleviating daytime dysfunction, and improving subjective sleep quality. Numerous scholars have corroborated our findings. It enhanced dopamine levels, altering responses to emotional events, and serving as a stress relief mechanism to help individuals return to pre-stress levels more quickly (77, 78). Moreover, adhering to scientific and effective physical exercise can regulate serum cortisol to optimal levels, boost immunity, and serum cortisol is closely linked to the body’s circadian rhythm (79). Studies also indicated that while body temperature decreases when falling asleep, evening exercise initially raises deep body temperature and accelerates the rate of temperature decline, thereby improving sleep (80). It’s noteworthy that whether a U-shaped relationship exists between physical activity and mental health (81), or the optimal frequency and intensity, requires further research to ascertain.

Spiritual growth can significantly reduce levels of depression. It’s strongly negatively correlated with self-harm even suicide, and worthlessness, marking these as the strongest negative edges within the network. Spiritual growth encompasses pursuing long-term life goals, experiencing oneself as positively developing and changing, gaining a sense of belonging, and maximizing the potential for a healthy life through goal-directed efforts (41, 82). Importantly, signs of depression are also characteristic of a lack of spirituality, Leung found that spiritual development significantly reduced depression, anxiety, and stress in adolescents while enhancing their self-confidence (83). It has been observed that young individuals experiencing depression often exhibit intrusive rumination, characterized by involuntary and repetitive contemplation of negative thoughts, which exacerbates their suffering (84). In contrast, spiritual growth can facilitate deliberate rumination, where individuals purposefully reflect on and analyze traumatic events to understand their meaning and impact, thereby promoting psychological growth (85). Additionally, self-efficacy is closely related to spiritual growth and can be viewed as an outward manifestation of it. It also serves as a mediator through which spiritual growth positively influences mental health, significantly reducing symptoms of depression, anxiety, and externalizing behaviors (86). As the third major health-promoting node, stress management showed a significant negative correlation with anxiety symptoms, alleviating trouble relaxing and restlessness. In this study, trouble relaxing was identified as a core symptom of anxiety, and restlessness was closely linked to it. This illustrates how stress management, by targeting core symptoms, can effectively and directly mitigate anxiety. Adolescents who frequently employ stress management strategies and adopt more positive coping and defense mechanisms exhibit fewer anxiety symptoms over time, underscoring the importance of incorporating stress management techniques into mental health programs (87).

The NCT analysis and quantification of centrality indices have confirmed the universal applicability of stress management, physical activity, and spiritual growth as key health-promoting factors for students. Nonetheless, some distinctions emerged when comparing the intensity levels of health-promoting behaviors’ effects on mental health symptoms across subgroups. Regarding gender, we found no association between stress management and anxiety in males, while physical activity significantly reduced anxiety and improved sleep quality. This implies that men often employ ineffective strategies for managing stress. Traditional masculine traits, which emphasize strength and control, condition men to suppress vulnerable emotions like fear and sadness, a habit persisting into adulthood (88). Additionally, a report found that men more frequently engage in regular exercise compared to women (89). For graduate students, stress management was the most central health-promoting behavior, surpassing physical activity. Among only-children, stress management strategies were more effective in alleviating anxiety compared to those with siblings. We hypothesize that for individuals experiencing greater and more complex stress, targeted stress management strategies are more direct and effective in reducing psychological issues in the short term. Our findings align with epidemiological surveys on students’ mental health (90–92). A higher proportion of women (41.14%) than men (33.71%) experienced anxiety. Anxiety severity was higher among graduate students (8.32) compared to undergraduates (8.09), and higher among only-children (8.75) compared to those with siblings (8.03). Notably, maintaining health responsibility (i.e., continuously attending to physical and mental health and remaining vigilant for potential illnesses) can counterintuitively exacerbate anxiety among women and graduate students. While this represents a form of health responsibility, over attention to health issues in high-stress populations can amplify anxiety, a phenomenon known as “health anxiety”. When classified by clinical scores, university students with at least one mental disorder exhibited significantly higher network density (0.44) compared to those without mental disorders (0.33). This elevated connectivity reflects the complexity and interconnection of mental disorders. Moreover, within the network of students experiencing mental disorders, stronger associations were observed among Abnormal behavior & speech - Self-harm even suicide, Spiritual growth - Worthlessness, and Stress management - Restlessness. Self-harm even suicide emerged as the second most central symptom node, following Low energy. Consistent with study on female nursing students (55), this finding suggested that psychomotor symptoms (such as psychomotor agitation/impulsivity and restlessness) and worthlessness may trigger connections within the network and significantly predict suicidal ideation. This also indirectly supports the notion that targeted interventions in stress management and spiritual growth are more directly effective under severe mental disorders.