INTRODUCTION
University represents a critical period during which students gradually mature and transition into independent adult roles—entering the workforce or postgraduate training, assuming financial responsibility, and navigating new social roles. At this stage, mental health exerts a substantial influence on life development. With changes in the social model and increasing pressures from academic work, finances, and employment, psychological problems among college students can become increasingly severe, with anxiety being a common mental health concern. Recent post-pandemic studies (2022-2024) in China and internationally report comparable or higher detection of anxiety symptoms among university students [1-4].
The detection rate of anxiety among college students in China is approximately 13.7%. Between 2010 and 2020, this figure showed a substantial upward trend, with detection rates among individuals with different characteristics ranging between 10% and 20% [5]. A global meta-analysis reported a detection rate of 39.0% for anxiety symptoms among college students, with variation across different regions and national income levels [6].
Relevant studies have found that academic pressure, campus bullying, and other adverse events can induce self-harming behaviour through anxiety. This is closely associated with both intentional and unintentional harm, smoking, drinking, substance abuse, lack of physical activity, skipping breakfast, irregular sleep, and other undesirable behaviours [7,8]. At the same time, anxiety may be influenced by physiological, psychological, and environmental factors.
Given these findings, identifying college students with high levels of anxiety and proposing targeted interventions may yield positive outcomes in anxiety management [9]. Therefore, this paper aims to systematically review previous studies on the influencing factors of anxiety among Chinese college students, summarize a wide range of findings through meta-analysis, and incorporate a broader set of potential influencing factors. The objective is to provide high-level evidence, identify key control points within the causal network or causal chain specific to Chinese college students, and offer a reference for the prevention and mitigation of anxiety in this population. China-specific evidence remains fragmented due to measurement non-equivalence and inconsistent cut-offs across commonly used instruments (e.g., the Self-rating Anxiety Scale [SAS], the 7-item Generalized Anxiety Disorder Scale [GAD-7], and the Depression Anxiety Stress Scale-21 [DASS-21]) [10,11], uneven provincial coverage, and campus-level contextual differences [12], as well as post-pandemic cohort shifts in risk structure [13]. Recent campus-based surveys (2023-2024) further document elevated anxiety and its links with sleep and healthbehavior patterns among Chinese university students [14-17]. To address these gaps, we harmonised measures, prespecified moderators pertinent to Chinese campuses, and reported heterogeneity (including I² and 95% prediction intervals) to improve the generalisability and policy relevance of pooled estimates, while noting the limited representation of post-pandemic evidence in eligible studies.
METHODS
This study followed the PRISMA guidelines.
Database and retrieval strategy
We searched Chinese and English databases, including PubMed, Embase, Cochrane, CNKI, Wanfang, and VIP, for studies on anxiety among Chinese college students published between January 1, 2010 and June 1, 2024. Searches covered both Chinese- and English-language sources within this window, using three concept blocks (Anxiety ∧ Students ∧ China). Database-specific subject headings (e.g., MeSH/Emtree) were combined with free-text synonyms. Records were deduplicated before screening. Two reviewers independently screened titles/abstracts and full texts, with discrepancies resolved by a third reviewer.
Search terms were as follows: Chinese—“Anxious/Anxiety/Anxiety disorders,” “College students/Colleges/Undergraduates,” “China/Chinese,” “Influencing factors/Risk factors/Correlative factor”; English—“Anxious/Anxiety/Anxiety disorders,” “College students/Colleges/Undergraduates,” “China/Chinese,” “Influencing factors/Risk factors/Correlative factor.”
The PubMed strategy was as follows: (“Influencing factors” OR “Risk factors” OR “Correlative factor”) AND (“College students” OR “Colleges” OR “Undergraduates”) AND (“China” OR “Chinese”) AND (“Anxious” OR “Anxiety” OR “Anxiety disorders”).
We also screened reference lists and emailed domain experts for grey literature but received no responses and note this as a minor study limitation.
Inclusion and exclusion criteria
The following served as inclusion criteria: 1) studies on anxiety among Chinese college students (from January 1, 2010 to June 1, 2024); 2) validated, reliable instruments; 3) clear case definitions enabling the identification of anxiety symptoms; 4) cross-sectional design; and 5) complete data allowing the extraction of detection rates and/or factors. To ensure construct comparability across studies, we prespecified a focus on generalized anxiety assessed by validated instruments (e.g., SAS, GAD-7, DASS-21 anxiety, Symptom Checklist [SCL-90] anxiety).
The following served as exclusion criteria: 1) irrelevant topic; 2) special/emergency settings (e.g., earthquakes, influenza); 3) conference abstracts, case reports, or systematic reviews; 4) insufficient outcomes for analysis; 5) duplicates; 6) full text unavailable; 7) domain-specific anxieties (e.g., English learning, test/exam, mathematics, other situation-bound types); and 8) postgraduate participants included >10% of the sample.
Literature screening and data extraction
Literature screening
Two investigators independently screened the records following the Cochrane Handbook (v5.0.2). Titles/abstracts were screened first, followed by full-text assessment against the inclusion criteria. Disagreements were resolved by a third reviewer.
Data extraction
The following information was extracted from the included articles: first author, publication date, province, region, sample size, detection rate, research factors, and diagnostic tools. Additionally, where reported, we recorded study-level characteristics (instrument and cut-off, sampling approach, proportion of female participants, response rate, survey period including pandemic-related semesters, and study quality) to help contextualise sources of heterogeneity.
Literature quality evaluation
Quality evaluation was conducted using the cross-sectional study quality assessment checklist developed by the Agency for Healthcare Research and Quality, which includes 11 items. Each item was scored as 0 for “no” or “unclear” and 1 for “yes.” Specifically, articles scoring 0-3 points were considered low quality, those scoring 4-7 points were considered medium quality, and those scoring 8-11 points were considered high quality.
Statistical methods
We used Review Manager (RevMan) version 5.4 (The Cochrane Collaboration) for all analyses. Effect sizes were odds ratios (ORs) with 95% confidence intervals (CIs). Heterogeneity was assessed using the χ² test and the I² statistic. If p≥0.1 and I²<50%, we applied a fixed-effect model; otherwise (p<0.1 and/or I²≥50%), we used a random-effects model. A fixed-effect model assumes a common true effect and attributes differences to sampling error, whereas a random-effects model allows true effects to vary and accounts for within-study error and between-study variance.
For factors with high heterogeneity (I²≥50%), we conducted subgroup analyses by survey year (≤2014 vs. ≥2015) and region (East vs. other). Subgroup differences were tested with the Q statistic. Publication bias was assessed with funnel plots. Because subgroups included few studies, and within-subgroup I² often remained high, we interpreted non-significant subgroup tests cautiously and emphasized narrative considerations concerning potential moderators.
To interpret residual heterogeneity, we narratively considered study-reported characteristics (e.g., instrument and cutoff, sampling approach, proportion of female participants, response rate, survey period including pandemic-related semesters, and study quality).
We have confirmed that all included studies have been ethically approved.
RESULTS
Characteristics of the literature
A total of 1,506 articles were identified, including 964 Chinese articles and 542 English articles. After two rounds of screening, 17 Chinese articles and 3 English articles [18-37] were included based on the inclusion and exclusion criteria, resulting in 20 articles for final analysis. The literature screening process is shown in Figure 1. The total sample size of the included articles was 34,672, of which 8,135 included anxiety symptoms, yielding a detection rate of 23.46%, as shown in Table 1. Although the search extended to June 1, 2024, relatively few post-2020 studies met our eligibility criteria; therefore, the final pool reflects the studies that provided comparable instruments and extractable data.
Meta-analysis results
Among the studies included in this analysis, 14 used the SAS, 2 used the GAD-7, 2 used the SL-90, 1 used the DASS-21, and 1 used the Beck Anxiety Inventory.
The results of the meta-analysis indicated that grade level (I²=84%, OR=1.52, p=0.020) suggested that the anxiety rate among graduating students was 1.52 times that of non-graduating students, with a statistically significant difference (Figure 2A). Due to the high heterogeneity observed, subgroup analyses were performed. When stratified by survey year (≤2014 vs. ≥2015), heterogeneity remained high within subgroups (≥2015: I²=90.6%, χ²=21.19, p<0.001; ≤2014: I²=79%, χ²=9.53, p=0.009), with no significant difference between subgroups (random-effects χ²=0.82, df=1, p=0.364) (Supplementary Figure 1). Given the small number of studies per subgroup and overlapping CIs, these subgroup-difference tests were likely underpowered; therefore, non-significance should not be overinterpreted. Similarly, stratification by region (East vs. other) showed persistent heterogeneity (East: I²=84.7%, χ²=19.61, p<0.001; other: I²=91.1%, χ²=11.2, p<0.001) without significant subgroup differences (random-effects χ²=0.46, df=1, p=0.497) (Supplementary Figure 2). Residual heterogeneity may reflect unmeasured moderators beyond region and year, such as instrument/cut-off non-equivalence, survey mode, academic-calendar stressors, local pandemic disruptions, and campuslevel counselling capacity. Dissatisfaction with major (I²=0%, OR=0.55, p<0.001) indicated that the anxiety rate among students satisfied with their major was 0.55 times that of those who were dissatisfied (Figure 2B).
Only-child status (I²=0%, OR=1.25, p=0.007) was associated with an anxiety rate 1.25 times that of non-only children (Figure 3A). A single-parent family structure (I²=63%, OR=2.41, p<0.001) indicated that the anxiety rate among students from single-parent families was 2.41 times that of those from non-single-parent families (Figure 3B). Subgroup analysis by survey year revealed that all of the included studies were conducted after 2015, precluding temporal comparison (Supplementary Figure 3). Regional stratification showed moderate heterogeneity in the East (I²=55.2%, χ²=4.47, p=0.107) and higher heterogeneity in other regions (I²=83.5%, χ²=6.04, p=0.014), with no significant difference between regions (randomeffects: χ²=0.55, df=1, p=0.459) (Supplementary Figure 4). Similarly, subgroup differences may be underpowered, and remaining heterogeneity could arise from the unmeasured moderators listed above. The results further showed that college students with left-behind experience (I²=80%, OR=2.05, p< 0.001) had an anxiety rate 2.05 times that of those without such experience (Figure 3C).
An introverted personality (I²=29%, OR=1.45, p<0.001) suggested that the anxiety rate of introverted students was 1.45 times that of extroverted students (Figure 4A). Drinking (I²=0%, OR=2.05, p=0.040) also indicated that the anxiety rate among students who drink was 2.05 times that of those who do not, with a statistically significant difference (Figure 4B).
Narrative exploration suggested that variability in measurement instruments/cut-offs and the timing of surveys during pandemic-related semesters likely contributed to the persistent heterogeneity observed for graduation status, whereas differences in sampling approach, female proportion, response rate, and quality scores appeared to have had a smaller impact. These observations are consistent with the subgroup patterns and are intended to contextualise residual heterogeneity without adding new analyses or figures.
To further validate these findings, we conducted additional pooled analyses using RevMan 5.4 software and used a layout consistent with standard meta-analysis presentation. These analyses confirmed the robustness of our results: personality characteristics maintained low heterogeneity (I²=28.6%, τ²=0.0132, χ²=5.60, p=0.231) with consistent effect size (OR=1.45, 95% CI: 1.28-1.63, random-effects: Z=3.61, p<0.001) (Supplementary Figure 5). Similarly, drinking habits showed no heterogeneity (I²=0%, τ²=0, χ²=0.01, p=0.938) with a significant association (OR=2.05, 95% CI: 1.02-4.12, random-effects: Z=2.02, p=0.043) (Supplementary Figure 6).
The results of the meta-analysis did not suggest statistically significant differences in anxiety levels among college students based on romantic relationship status, ethnicity, place of origin, academic stream (arts or sciences), smoking habits, gender, medical major status, or whether their current major was their first choice, and different family financial status. These findings are presented in Table 2. The forest plots for the various influencing factors are shown in Figure 5.
Publication bias
In this study, more than 10 articles were included for the two factors of gender and students’ hometowns, allowing for the assessment of publication bias. The funnel plots for both factors showed that the scatter was almost evenly distributed on either side of the axis, indicating a low likelihood of publication bias in the included articles and enhancing the reliability of the findings. Consistent with our prespecified plan, statistical tests for small-study effects (e.g., Egger’s test) were considered only when a factor had ≥10 contributing studies and heterogeneity was low-to-moderate. In this dataset, although gender and hometown met the ≥10 threshold, betweenstudy heterogeneity was substantial, for which Egger’s test can be unreliable; therefore, we relied on visual funnel-plot inspection and interpreted these findings cautiously, acknowledging that small-study effects cannot be ruled out.
DISCUSSION
To ensure the comparability of results and the applicability of the analysis, articles published between 2010 and 2024 were selected, with only those presenting qualitative research results included to maintain the consistency of outcomes and validity of influencing factors. The 20 articles included in this study showed notable variation in the detection rates of anxiety symptoms, with an overall rate of 23.46%, which is comparable to the 26.4% reported in the 2021 China National Mental Health Development Report [38]. However, detection rates varied considerably across studies, ranging from 3.98% to 33.41%, suggesting that anxiety is widespread among Chinese college students, with particularly high incidence in some regions. Although the search extended to June 1, 2024, relatively few post-2020 studies met our inclusion criteria (cross-sectional design, validated general-anxiety instruments, extractable data), which may partly explain the limited representation of post-pandemic evidence in the pooled dataset. Nonetheless, contemporary cross-sectional data from Chinese campuses in 2023-2024 report substantial anxiety burden, particularly when sleep disturbance and health-risk behaviors co-occur [14-16]. Consequently, the pooled estimates are likely to reflect pre-pandemic baselines more than post-pandemic risk structures.
The detection rate of anxiety symptoms among graduates was substantially higher than that of non-graduates. This may be because graduates are in a critical transition period from university to the labour market and independent adult life. They face new life choices, such as employment or further education, along with pressures from graduation theses and projects. Specifically, “employment anxiety” refers to a negative emotional state characterised by tension, irritability, and unease or panic, arising from misconceptions or uncertainty about employment prospects before graduation. The employment pressure currently faced by college students has led to widespread anxiety, posing substantial challenges for employment support services in higher education institutions [39].
Graduation status showed high heterogeneity in the study (I²=84%). To explore the potential sources of this heterogeneity, we conducted subgroup analyses by survey year and region. However, heterogeneity remained substantial within all subgroups (I²=79%-91%), and tests for subgroup differences were non-significant (p>0.36), suggesting that neither temporal nor regional factors adequately explain the observed heterogeneity. This persistent heterogeneity may reflect unmeasured factors such as differences in measurement tools, the timing of data collection within the academic year, and institution-specific characteristics. Plausible moderators include instrument/cut-off non-equivalence, survey mode (online vs. in-class), academic-calendar stressors (e.g., thesis, internships, job-search), local pandemic disruptions, and campus-level counselling capacity. When interpreting heterogeneity and external validity, we considered survey timing (pandemic-related vs. other semesters), recognising that risk structures among student cohorts may have shifted in the post-pandemic period. This interpretation aligns with recent campus surveys linking anxiety with pandemic-period stressors and lifestyle factors [14-17]. From an implementation perspective, universities and counselling centres may interpret screening rates in light of instrument choice and academic-calendar timing (e.g., thesis submission and job-seeking periods) to anticipate service demand and plan service capacity accordingly. Notably, sensitivity analysis suggested that the study conducted by Zheng et al. [19] may be an outlier contributing to the heterogeneity, particularly in the “Other” regional subgroup.
In terms of dissatisfaction with their major, students who were dissatisfied with their field of study showed a high detection rate of anxiety symptoms. This may be because, upon entering university, many students discover that their actual major differs substantially from what they had anticipated. This mismatch can lead to disappointment about their future and a sense of helplessness in changing their situation, gradually increasing their psychological burden and leading to anxiety or depression.
The family environment is also critical for physical and mental development, influencing personality, academic performance, and overall mental health. College students from intact families and with no left-behind experience tend to exhibit lower rates of anxiety symptoms. Empirical studies support the harmful impact of parental neglect in childhood on poor adjustment in adulthood. Longitudinal studies on college students with left-behind experience have found a positive correlation between parental neglect and poor adaptation, including depression, anxiety and stress, in which anxious attachment and perceived social support act as mediators. These students often develop a fragile psychological state and reduced selfesteem, making them more vulnerable to anxiety, depression and stress-related symptoms [40]. Accordingly, institutions may prioritise proactive outreach and mentorship, enhanced counselling access, and links to social support for students with leftbehind experience or from single-parent families.
Additionally, adolescents from incomplete family structures often face both systemic and personal disadvantages, increasing their susceptibility to poverty and discrimination. They are also more likely to experience chronic depression, anxiety, social isolation, and a lack of emotional support [41].
Our subgroup analyses for family structure revealed an interesting pattern. While the overall association remained significant, regional differences in heterogeneity were observed, with Eastern regions showing moderate heterogeneity (I²=55%) compared with other regions (I²=84%). The absence of studies conducted before 2015 for this factor limited our ability to assess temporal trends. These findings suggest that the impact of family structure on anxiety may be moderated by regional socioeconomic factors, though further research with more granular geographic and temporal data is needed. The analysis of personality characteristics demonstrated one of the most consistent findings in our study, with low heterogeneity (I²=29%) across the included studies. This robust association between introversion and anxiety aligns with established psychological theories and provides strong evidence for personality as a stable predictor of anxiety symptoms among Chinese college students.
Individuals who suppress emotional expression have been found to tend to express fewer positive emotions while experiencing more negative ones, which is negatively correlated with overall well-being [42]. This may result in fewer opportunities for emotional release among introverted students, causing them to internalise negative emotions and increasing the risk of anxiety. Although some individuals may turn to alcohol for emotional relief, alcohol abuse does not resolve underlying issues but instead exacerbates both physical and psychological burdens, creating a vicious cycle that intensifies anxiety [43].
Anxiety is common among college students in China and Western countries, with both sharing some commonalities. Consistent with this cross-national pattern, U.S. studies also report a high prevalence of anxiety; for example, one study found that anxiety was more common among senior college students [44]. In Spain, a survey of college students similarly found regular drinking to be substantially associated with anxiety symptoms [41]. However, there are also differences in anxiety-related factors; for example, in Western countries, college students who are not good at socialising, lack partners, are transfer students or live off campus tend to experience more pronounced anxiety symptoms [40,45,46]. For factors without statistically significant pooled effects in our study (e.g., smoking, romantic relationship status, academic stream, gender), non-significance likely reflects a mix of definitional heterogeneity (e.g., current vs. having ever smoked; relationship status vs. relationship quality), the limited number of studies contributing to each factor (rather than small within-study samples), unmeasured confounding factors (e.g., depression, sleep, physical activity), and potential dose/context variation, rather than any definitive absence of association.
College students’ anxiety can be alleviated by identifying the sources of their anxiety, adjusting cognitive models, establishing action strategies, and seeking social support. According to different sources of anxiety, school, family, and social collaborative interventions (e.g., psychological counselling, adjusting work and rest, increasing exercise time and cultivating interpersonal relationships) can all help to effectively reduce anxiety levels among college students. For institutional implementation, practical steps include: targeted screening and stepped counselling during graduation seasons; the integration of career-counselling with mental-health referral pathways for final-year students; proactive mentorship and support for only-child, left-behind, and single-parent-family students; brief alcohol-use interventions and health education; and routine monitoring that records instrument/cut-off and calendar timing to anticipate high-risk periods. In clinical services, brief low-intensity options (psychoeducation, brief cognitive-behavioural strategies, and digital self-help) can be offered as firstline with stepped escalation for persistent or severe symptoms; for students with alcohol use, embedding brief alcohol interventions and clear referral pathways is advisable. For high-risk groups identified here (final-year students, only-child, leftbehind or single-parent family backgrounds, and introverted traits), proactive outreach, mentorship, and facilitated access to counselling should be routine. To support institutional planning, simple dashboards that routinely record instrument/cut-off and survey timing (by academic calendar) can inform staffing and service capacity; standardized reporting templates will facilitate cross-campus benchmarking and quality improvement.
In this study, publication bias was assessed through two factors: students’ gender and hometown. The results were relatively low; however, this study could not include unpublished studies, which may have resulted in some publication bias. This could lead to overestimation of effects for frequently reported factors and underestimation—or non-detection—of effects for understudied factors. In addition, no restrictions were placed on the measurement tools used to assess anxiety levels. Although the SAS scale was primarily used, six included articles utilised other scales, which may have introduced measurement bias. Non-equivalence of instruments and cutoffs likely produced non-differential misclassification, biasing pooled ORs toward the null and inflating heterogeneity (I²). Furthermore, the survey did not cover all the provinces in China, potentially leading to selection bias. Over-representation of Eastern provinces may skew pooled estimates toward Eastern settings and limit generalizability to under-represented regions. Due to differences in measurement tools, it was difficult to combine quantitative values; therefore, quantitative analysis and bias testing could not be conducted. Together, these issues may increase between-study variance and widen CIs; non-significant pooled estimates should therefore be viewed as potentially underpowered rather than evidence of no association. Accordingly, non-significant pooled estimates should be interpreted with caution, as limited statistical power and measurement non-equivalence may bias effects toward the null. Accordingly, it is expected that future investigations will expand the scope of research, subdivide measurement tools, ensure national coverage, and discuss causality by domain, thereby providing more accurate insights. From a service-planning perspective, clearly reporting instrument/cut-off choices and the academic-calendar timing of surveys in future studies will facilitate cross-site comparability and help universities allocate counselling resources more efficiently.
Conclusion
Based on the findings of this study, prevention and counselling measures for anxiety among college students should focus on objectively and comprehensively understanding the characteristics of their specialities, improving speciality satisfaction, fostering an optimistic outlook on employment prospects, reducing stigma toward students from single-parent families through non-discriminatory support, and providing emotional support for graduates. At the same time, it is of great importance to encourage students to actively participate in various physical activities and to establish counselling spaces that support emotional expression. In this regard, the prevention of anxiety symptoms is preferable to treatment. Only by clearly identifying the causes of anxiety symptoms and understanding their risk and protective factors can effective mental health education be implemented, thereby helping to reduce the occurrence of anxiety symptoms and promote the healthy development of both physical and mental well-being in college students.
PDF Links
PubReader
ePub Link
XML Download
Full text via DOI
Download Citation
Supplement1
Print
