From January to February 2023, 7,503 college students (freshmen to seniors) from 10 regions in China were invited to participate via the Questionnaire Star platform. After excluding questionnaires with abnormal response times (253 samples, <5 min or >30 min) and invalid responses (20 samples, including inconsistent answers to validation questions or logical contradictions in key items), 7,230 valid questionnaires were retained (effective recovery rate: 96.3%) (Figure 1).

The study was approved by the Ethics Committee of the first author’s university [Southern Medical University Audit (2023) No. 46]. A convenience sampling method was used to distribute an anonymous electronic questionnaire through the Questionnaire Star platform¹, with the following inclusion criteria: (1) currently enrolled college students; (2) consistent answers to 2 validation questions. Exclusion criteria: (1) self-reported cognitive impairment or inability to complete the questionnaire independently; (2) unwillingness to provide informed consent. Participants completed demographic, sociological, and lifestyle information, as well as the GAD-7, PHQ-9, and e-Health Literacy Scale, which took approximately 15 min.

In this study, data analysis was conducted using Mplus version 8.3 and SPSS 23.0². To determine the heterogeneity of negative emotional symptom presentation, we assessed the respondents’ anxiety and depression symptoms using the GAD-7 and PHQ-9, and performed Latent Profile Analysis (LPA). Models containing 1 to 5 latent classes were ordered to identify the most meaningful and parsimonious model. Six fit indices were used to evaluate the models: Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), sample size-adjusted BIC (aBIC), Lo–Mendell–Rubin likelihood ratio test (LMRT), bootstrap likelihood ratio test (BLRT), and entropy. Generally, the best-fitting model should exhibit lower AIC, BIC, and aBIC values; higher entropy; significant LMRT and BLRT values; and each subgroup should account for at least 5% of the total population. In addition to fit statistics, model selection should also consider previous research and the clinical significance of the latent classes.

Secondly, some variables were used as predictors for the optimal latent profile analysis model. We performed binary logistic regression using the three-step command in Mplus 8.3 to evaluate whether the selected variables as predictors would lead to a higher probability of a college student belonging to one latent class over another. The LPA results served as the dependent variable, while gender, grade ranking, family type, per capita monthly family income, adherence to physical activity, smoking status and alcohol consumption were used as independent variables. Binary logistic regression analysis was conducted, and odds ratios (OR) and their 95% confidence intervals (95% CI) were calculated based on the β values and standard errors, and reported in the results section. The OR values reflect the predictive effects of gender, grade ranking, family type, per capita monthly family income, adherence to physical activity, smoking status and alcohol consumption on the negative emotions of different profile classes.

Finally, the BCH (Bias-Corrected Bootstrap with a Huber-White estimator) method was applied to analyze e-HL as an outcome measure of negative emotional states. All statistical analyses were performed using Mplus version 8.3 and SPSS 23.0 software, with a two-sided test p < 0.05 considered statistically significant. The design and analysis of this study were not pre-registered.

The average GAD-7 score was 3.72 ± 4.00, and the average PHQ-9 score was 4.90 ± 5.05, as detailed in Table 1. Correlation analysis showed that negative emotional symptoms (anxiety and depression) were significantly positively correlated with sex (anxiety: r = 0.126, p < 0.001; depression: r = 0.115, p < 0.001) and alcohol consumption (anxiety: r = 0.058, p < 0.001; depression: r = 0.069, p < 0.001). Negative emotional symptoms were significantly negatively correlated with family residential background (anxiety: r = −0.078, p = 0.002; depression: r = −0.024, p = 0.040), per capita monthly family income (anxiety: r = −0.092, p < 0.001; depression: r = −0.086, p < 0.001), adherence to physical activity (anxiety: r = −0.119, p < 0.001; depression: r = −0.146, p < 0.001), and e-HL (anxiety: r = −0.205, p < 0.001; depression: r = −0.214, p < 0.001). Collinearity analysis showed variance inflation factors (VIF) < 10 for all independent variables, indicating no multicollinearity (Table 2).

This study began with an initial model of one class and progressively increased the number of model classes to five. We evaluated and tested the fit and differences of the latent profile models using various indices and integrated theoretical rationale, clinical interpretability, and statistical fit to determine the optimal model. The indices for different LPA models are presented in Table 3.

Considering the BLRT, the p-values for each class model were significant at the α = 0.05 level, and the entropy values for all 2–5 class models were acceptable (all above 0.90). As the number of latent classes increased, the AIC, BIC, and aBIC values continued to decrease. Therefore, these fit indices might not be very helpful in selecting the appropriate model.

However, the p-values of the LMR were significant (p < 0.01) in the two-class and three-class models, but not significant in the four-class (p = 0.125) and five-class models (p = 0.104). This indicates that the two-class model is better than the one-class model, and the three-class model is better than the two-class model, but the four-class model is not better than the three-class model, and the five-class model is not better than the four-class model.

The three-class solution included low negative emotional symptoms (56.61%), moderate negative emotional symptoms (39.27%), and high negative emotional symptoms (4.12%). Although the LMRT was significant, the smallest class (high negative emotional symptoms) accounted for only 4.12% of the sample. Clinically, this intermediate group lacked distinct characteristics to guide targeted interventions, as their mental health needs could not be differentiated from the low or high groups. Additionally, the 4.12% small class, while statistically detectable in a large sample (n = 7,230), had limited generalizability and practical value for population-based mental health screening.

The two-class solution was supported by strong theoretical and clinical rationale. Theoretically, it aligns with the “subclinical-clinical” dichotomy of mental health in general populations, which is widely used in public health screening. Clinically, this classification clearly distinguishes high-risk groups that require targeted interventions from low-risk groups that need preventive education—an essential distinction for college mental health management. Based on the integrated evaluation of statistical fit, theoretical consistency, and clinical applicability, the two-class model was identified as the optimal solution.”

The two-class characteristics were low/no negative emotional symptoms (61.49%) and high negative emotional symptoms (38.51%), as shown in Table 3. The average latent class probabilities for each group in the two-class model were all greater than 0.90, as shown in Table 4. The scores for each item of the GAD-7 and PHQ-9 in the two-class model are shown in Figure 2. Analysis of variance showed that there were significant differences in each item of the GAD-7 and PHQ-9 (p < 0.001), as shown in Table 5.

Binary logistic regression showed that sex, per capita monthly family income, adherence to physical activity, and alcohol consumption were significant predictors of profile membership (Table 6).

Females were more likely to be in the high negative emotional symptoms group than males (OR = 1.448, p < 0.001); Lower per capita monthly family income was associated with higher odds of being in the high negative emotional symptoms group (OR = 0.850, p < 0.001); Regular physical activity was a protective factor (OR = 0.544, p < 0.001); Alcohol consumption was associated with higher odds of being in the high negative emotional symptoms group (OR = 1.558, p < 0.001). Grade ranking and family residential background were not significant predictors (p > 0.05).

In this study, e-HL was analyzed as a variable associated with negative emotional symptom characteristics using the BCH approach. As shown in Table 7, there were significant differences in e-HL between the two-class models. The average score of e-HL was higher in the low/no negative emotional symptom group, indicating that the higher the severity of negative emotional symptom, the lower the level of e-HL.

Firstly, we adopted a person-centered approach to assess the heterogeneity of negative emotional symptom presentation (Hu et al., 2022). On the one hand, we found that the differences between the classification models were only in the severity of negative emotion symptoms, rather than the type of symptoms. In short, the classification model can only identify based on the severity of symptoms rather than the absence or presence of specific symptoms. This reflects a high similarity in the pattern of symptom severity response for these distress disorders, which can provide references for clinical assessment and/or treatment. Our study also found that within each latent profile category, the severity of individual negative emotions exhibited flat differences, meaning that the severity reflected by depression and anxiety in negative emotions is consistent with each other, and there are no subgroups dominated by depressive symptoms or anxiety symptoms. This result is consistent with previous studies (Dai et al., 2024), suggesting that in practice, for patients with negative emotions, the possibility of comorbidity of anxiety and depression should be considered (Wang et al., 2020).

Additionally, item 1 of the GAD-7 (“Feeling nervous, anxious, or on edge”) scored significantly higher than other anxiety-related items. This may be because academic pressures such as performance requirements, exams, study materials, and time management make students feel tense and uneasy. Academics are an indispensable part of every college student’s life, and without a healthy attitude towards academic goals, students may experience severe stress. While academics can be seen as a positive challenge with the potential to enhance learning abilities and skills, if viewed negatively, this pressure may adversely affect students’ mental health (Beiter et al., 2015); For the PHQ-9, item 1 (“Little interest or pleasure in doing things”) and item 4 (“Feeling tired or having little energy”) scored significantly higher than other depression-related items. This may be because sleep disorders (SD) are prevalent among college students (Nyer et al., 2013). Poor sleep quality and insufficient rest lead to a lack of interest in activities and feelings of tiredness or lack of energy. Sleep disorders (SD) are common somatic symptoms in patients with major depressive disorder (MDD), including excessive or insufficient sleep and restless sleep, and are also one of the DSM-IV diagnostic criteria for depression (Vanwoerden and Stepp, 2022); It is worth noting that item 9 of the PHQ-9 (“Thoughts that you would be better off dead, or of hurting yourself in some way”) scored significantly lower than other items, indicating that suicidal ideation among college students in this study was relatively low, consistent with previous research on suicidal ideation among Chinese college students (Kang et al., 2023; Cong et al., 2021). Suicidal ideation is the most important predictor of suicidal behavior (Liao et al., 2022), and represents the individual’s thoughts and intentions of ending their life (Grasdalsmoen et al., 2020). Notably, the average GAD-7 (3.72 ± 4.00) and PHQ-9 (4.90 ± 5.05) scores in this sample were relatively low (below the clinical cutoff of ≥10), which may reflect the overall mental health status of Chinese college students in non-clinical settings. This finding suggests that our two-profile model is particularly suitable for population-based early screening, as it identifies high-risk individuals within a generally low-severity population, this is more clinically meaningful than a median split, which would arbitrarily divide the population into two groups regardless of clinical relevance. Compared to median splits, LPA identifies latent subgroups based on empirical response patterns, avoiding arbitrary cutoff points and better capturing the true heterogeneity of symptom presentation.

Our study identified predictive factors associated with various latent profile categories, including sex, per capita monthly family income, adherence to physical activity, and alcohol consumption. First, our study revealed that girls are more likely to experience severe negative emotional states compared to boys, consistent with previous research findings (Tao et al., 2024; Liu et al., 2019; Zhang X. et al., 2024). Secondly, college students with lower per capita monthly family income are likely to experience more severe negative emotional states. With the accelerating pace of life, college students from economically disadvantaged families face not only academic pressures but also concerns about future career development and family financial responsibilities, making them more susceptible to anxiety or depression (Markoulakis et al., 2020). Thirdly, the results indicate that college students who have not exercised regularly in the past month may exhibit more severe negative emotional states. This may be because regular physical activity enhances emotional stability and overall emotional resilience, positively influencing college students’ emotional states (Mu et al., 2024). Fourthly, college students who have consumed alcohol in the past month are more likely to experience severe negative emotional states. Previous studies have shown a significant association between alcohol consumption and negative emotions among college students, with those who engage in heavy drinking not only harming their health but also experiencing anxiety and depression (Nourse et al., 2017). These findings suggest that female students, those with lower per capita monthly family income, those who lack regular exercise, and those with drinking alcohol habits should receive focused attention. Early identification and intervention for students likely to experience severe negative emotional states are crucial, and symptomatic treatment should be provided when necessary.

This study chose e-HL as an associated variable of negative emotion characteristics because previous research has shown that college students’ e-HL is closely related to anxiety and depression symptoms (Zhang S. et al., 2024; Lin et al., 2024). The more severe the anxiety and depression symptoms, the lower the e-HL level among college students. When college students experience anxiety or depression, their self-management behaviors become less regulated, affecting their psychological help-seeking behaviors and health behaviors, which in turn impacts their e-HL level. Notably, this relationship may be confounded or moderated by digital addiction—a prevalent issue among college students that frequently co-occurs with negative emotional symptoms (Zhang and Wang, 2024; Zhang J. et al., 2024). Recent evidence indicates that digital addiction is strongly associated with both depression/anxiety and impaired e-HL: individuals with digital addiction may exhibit compulsive online behaviors that reduce the quality of health information processing while exacerbating negative emotions through social comparison or information overload (Zhang and Wang, 2024). Conversely, negative emotional symptoms may increase the risk of digital addiction as a coping mechanism, creating a bidirectional cycle that indirectly impairs e-HL (Zhang J. et al., 2024). Our study revealed differences in e-HL across different latent profile categories, with individuals in the high negative emotions group having lower e-HL levels compared to those in the low/no negative emotions group. Research indicates that good e-HL skills are conducive to individuals purposefully acquiring health knowledge and improving health behaviors. The application competency dimension represents an individual’s ability to obtain health information online, while the judgment and decision-making ability dimension reflects an individual’s capacity to discern the authenticity of information based on sufficient health knowledge and skills. In the era of rapid information technology and internet development, the threshold for obtaining health knowledge from mobile phones and computers has become more accessible with the widespread use of the internet. Individuals with high e-HL are more likely to adopt healthy behaviors and make greater use of online health resources. When individuals experience negative emotions, they are more likely to seek scientifically correct methods to alleviate their negative emotions, thereby preventing or mitigating depressive symptoms to some extent (Shudayfat et al., 2023). However, the failure to account for digital addiction in the current study means we cannot rule out the possibility that the observed negative correlation between negative emotional symptoms and e-HL is partially driven by this confounding factor. For example, the high negative emotional symptom group may have higher rates of digital addiction, which independently reduces their e-HL, leading to an overestimation of the direct effect of negative emotions. Therefore, in the process of conducting mental health interventions for college students, it is essential to further strengthen students’ personal self-management, enhance their self-efficacy, encourage them to change unhealthy behaviors, and consolidate personal health-promoting behaviors to better improve their e-HL level. From a theoretical perspective, exploring the significant differences in e-HL among different subgroups helps deepen our understanding of the influencing factors related to negative emotions. Clinically, direct treatment methods focusing solely on negative emotions are not very effective. Our study suggests that improving college students’ e-HL levels may enhance the effectiveness of negative emotion treatments, providing a reliable direction for the clinical treatment of negative emotions.

Based on the study findings, we put forward the following evidence-based and targeted intervention strategies for college mental health management to address the identified high-risk groups and key influencing factors:

First and foremost, implement targeted stratified mental health screening: Prioritize high-risk subgroups (females, low-income students, inactive individuals, and drinkers) for bi-annual screening using GAD-7 and PHQ-9. For students categorized into the high negative emotional symptoms group, conduct quarterly re-screening and initiate a 72-h response mechanism—matching positive cases with a dedicated mental health counselor for a preliminary assessment and personalized support plan within 3 working days. For the low/no negative emotional symptoms group, deliver annual universal mental health checks combined with preventive education to maintain psychological resilience.

Secondly, carry out lifestyle-oriented intervention programs: For the high negative emotional symptoms group: Design a 12-week structured physical activity program (3 sessions/week, 40 min/session) combining aerobic exercise and mindfulness training, led by certified fitness instructors and psychological counselors. Simultaneously, implement a peer support program for alcohol reduction—pairing students with drinking habits with non-drinking peers for weekly check-ins, and offering a “low-alcohol social skills” workshop to address social drinking triggers. For the low/no negative emotional symptoms group: Promote campus-wide fitness challenges with incentives and launch “healthy socialization” campaigns to reinforce positive lifestyle norms.

Thirdly, strengthen e-HL enhancement initiatives: For the high negative emotional symptoms group: Deliver an 8-week integrated workshop (1 session/week, 90 min/session) combining e-HL and digital addiction prevention. Core modules include: identifying credible mental health resources (e.g., national psychological service platforms, academic databases), critical evaluation of online health misinformation, and practical skills for controlled online use (e.g., app-based screen time management, setting information-seeking time limits). Provide 1-on-1 coaching (3 sessions total) for students with low e-HL scores (<25 points) to address individual gaps. For the low/no negative emotional symptoms group: Develop a self-paced online course (6 modules, 20 min/module) on e-HL basics (e.g., efficient health information search strategies, utilizing online counseling resources) and issue a “digital health passport” for course completion, encouraging proactive use of online mental health tools.

Fourthly, provide targeted support for students from low-income families: Establish a “financial-psychological integration support package” for low-income students in the high negative emotional symptoms group: Combine need-based scholarships (with simplified application procedures and 7-day feedback) with monthly one-on-one psychological counseling (focused on economic stress coping and future planning) and peer mentorship programs (pairing with senior students who have overcome similar challenges). For low-income students in the low/no negative emotional symptoms group: Offer part-time work-study opportunities related to mental health promotion (e.g., campus mental health ambassadors) to enhance self-efficacy, and provide free access to e-HL workshops and fitness program resources.

Fifthly, integrate digital addiction intervention into profile-specific mental health care: For the high negative emotional symptoms group: Embed digital addiction screening into routine mental health evaluations. For students with moderate-to-high digital addiction scores, add 4 specialized sessions on digital behavior regulation to their personalized support plan. For the low/no negative emotional symptoms group: Launch campus-wide “digital well-being” campaigns to prevent the onset of problematic digital use and maintain healthy online behaviors.”

This study has several limitations: (1) Self-report measures may introduce bias, and future studies should combine clinical interviews for validation; (2) The cross-sectional design cannot establish causality, and longitudinal studies are needed to explore bidirectional relationships between negative emotions and e-HL; (3) Convenience sampling with uneven gender and grade distribution limits generalizability, so future studies should use stratified sampling to ensure representativeness; (4) Key variables such as social support and academic stress were not included in the study, and future research should incorporate these to explore more comprehensive mechanisms. (5) Digital addiction was not measured, which may confound the relationship between negative emotional symptoms and e-HL. Future studies should include digital addiction scales to disentangle these relationships.