This study used a cross-sectional design. An online questionnaire was administered to college students in China via the Wenjuanxing platform, from September 2024 to January 2025, with questionnaires distributed through official university channels. Using a stratified-cluster sampling method, the survey covered 70 universities across 18 provincial administrative regions, including 15 medical colleges and 55 non-medical universities. The specific distribution of regions and the number of universities are shown in Table 1. Notably, this survey was conducted during the post-pandemic normalization of COVID-19 prevention and control. Although potential residual effects of previous pandemic experiences (e.g., long-term impacts on sleep patterns or stress responses) on participants could not be entirely ruled out, the nationwide multi-center sampling strategy—covering diverse regions and educational institutions—minimized biases in comparing group differences between medical and non-medical students. Thus, the core analysis of inter-group disparities remained robust despite these contextual factors, which were acknowledged but not the focus of the present study. All participants’ data were de-identified before retrieval. Privacy information was protected through encrypted storage and restricted access only to authorized members of the research team. The research protocol was approved by the Ethics Review Committee of Chongqing Medical and Pharmaceutical College (Approval No.: KYLLSC20240922005) and strictly adhered to the ethical guidelines of the Helsinki Declaration.

All participants provided online informed consent prior to answering the questionnaire, which acknowledged potential risks and confidentiality measures of the study. The research team ensured data quality through IP-address restrictions and by monitoring response duration (estimated 10–20 minutes). A total of 12,090 initial questionnaires were collected. After excluding 118 invalid questionnaires due to severely unreasonable response time (5 minutes), missing key information (e.g., sex, grade, age) or variables (e.g., incomplete subscales of Depression Anxiety Stress Scales-21 (DASS-21)), and logical contradictions (e.g., conflict between “17 years old” and “senior year”; coexistence of “daily sleep duration $>$16 hours” and “no daytime dysfunction”), a total of 11,972 valid samples were finally included, with an effective rate of 99.02%. Among the samples, males accounted for 34.56% (n = 4137) and females accounted for 65.44% (n = 7835). The samples were predominantly aged 18–22 (mean = 19.8 years, SD = 1.3).

In addition to basic information (age, sex, university, grade, major), standard survey tools were used to measure core variables. Mental health was assessed using the DASS-21, a 21-item scale with three subscales (depression, anxiety, stress; 7 items each). Responses use a 4-point Likert scale (0 = never to 3 = always), with subscale scores $\le$9 (depression), $\le$7 (anxiety), and $\le$14 (stress) indicating normal levels; scores $\ge$10, $\ge$8, or $\ge$15 suggest mild or higher symptoms (Lovibond and Lovibond, 1995). The Chinese version has been validated for Chinese college students (Wang et al, 2016), with a Cronbach’s $\alpha$ of 0.947 in this study.

Sleep quality was evaluated using the Pittsburgh Sleep Quality Index (PSQI). The original PSQI scale was developed by (Buysse et al, 1989) and includes 18 self-rated and 5 observer-rated items, assessing seven components (subjective sleep quality, sleep latency, sleep duration, habitual sleep efficiency, sleep disturbances, hypnotic use, daytime dysfunction) over the previous month. Scores range from 0 to 21, with $>$5 indicating poor sleep quality. These grade-related differences indicate distinct mental health and sleep patterns across academic years, yet grade itself explains only a small proportion of variance (small effect sizes), suggesting that numerous other factors influenced college students’ sleep and psychological well-being (Buysse et al, 1989). The scale showed good internal consistency in this study (Cronbach’s $\alpha$ = 0.81).

Statistical analyses were performed using SPSS 26.0 (IBM Corp., Armonk, NY, USA) and R 4.4.2 (R Foundation for Statistical Computing, Vienna, Austria). First, normality tests (Shapiro-Wilk test) were conducted for continuous variables (DASS subscale scores, DASS total score, PSQI total score). Variables meeting normality were described by mean standard deviation, and independent samples t-tests were used. Since most continuous variables did not follow a normal distribution, they were described by median (interquartile range, IQR). Nonparametric tests were applied for group comparisons: the Mann-Whitney U test for two groups and the Kruskal-Wallis H test for multiple groups (by grade). Categorical variables were presented as frequency (%) and analyzed using Pearson’s chi-square test or Fisher’s exact probability test (when theoretical frequency 5), with Cramer’s V calculated as the effect size.

To explore the relationship between sleep quality and mental health, Spearman’s rank correlation analysis was used to assess the association strength between PSQI and DASS scores, as well as between sleep quality (PSQI score) and mental health indicators (DASS subscale scores and total score). All significance tests were two-tailed, and p $\le$ 0.05 was considered statistically significant.

Significant differences in demographic characteristics were observed between medical and non-medical students. As shown in Table 2, chi-square tests indicated statistically significant differences in grade distribution ($\chi$² = 414.021), age distribution ($\chi$² = 261.312), mental health status (depressive symptoms: $\chi$² = 168.659; anxiety symptoms: $\chi$² = 19.455; stress symptoms: $\chi$² = 75.113), and sleep quality (PSQI total score: $\chi$² = 135.813) (all p 0.001). Specifically, grade distribution showed significant disciplinary differences: non-medical students dominated in lower grades (freshmen and sophomores), accounting for 56.76% and 55.27%, respectively, whereas medical students significantly outnumbered non-medical students in higher grades (juniors and seniors), reaching 68.99% and 82.99%, respectively.

In age distribution, medical students significantly dominated the $\le$17-year-old group (60.78%), reflecting the early recruitment of outstanding high school graduates into medical programs. Non-medical students prevailed in the 18–20-year-old groups (57.52%, 56.01%, 50.61%), whereas medical students rebounded to 63.67%–74.67% in the $\ge$21-year-old groups. This U-shaped distribution highly aligned with the three-stage training model of medical education (“preparatory-basic-clinical”). In terms of mental health, non-medical students reported slightly higher rates of depressive symptoms, whereas medical students reported higher stress symptoms; both groups had high anxiety-symptom reporting rates with relatively small differences (see section 3.2 for details).

Mann-Whitney U test results showed that the DASS total score distribution in both groups approximated a pyramid, mostly concentrated at 20–30 points (Table 3 & Fig. 1). The median total DASS score for both groups was 31 (p = 0.094). The two groups were not significantly different, and the effect size was extremely small (Cohen’s d = 0.024) (Table 4), indicating that the overall level of psychological distress was similar in both groups. Medical students did not exhibit significantly higher overall distress than did non-medical students, though subtle differences existed in specific dimensions.

Fig. 1.

Comparison of the distribution of psychological distress levels (total DASS scores) between medical and non-medical students. Note: 1 = Medical students, 0 = Non-medical students, Y axis = n, X axis = DASS score.

Further comparison of DASS-21 subscales revealed (details in Table 2) the following results. Depression: non-medical students reported higher depressive symptoms (59.81% vs. 48.48%), possibly influenced by employment uncertainty. Anxiety: both groups showed high anxiety reporting rates (81.85% vs. 79.77%), with similar severe anxiety proportions (8.86% vs. 7.87%), suggesting universal academic stress. Stress: the prevalence of mild (11.38%) and moderate (2.38%) stress in non-medical students was lower than in medical students (16.21% and 3.38%, respectively); no extremely severe cases were found in either group. Although statistically non-significant, these trends held clinical implications: medical students experienced greater stress but slightly better depression outcomes.

For sleep quality, medical students had a PSQI median of 8 vs. 9 for non-medical students, with a significant difference (p 0.001) and moderate effect size (Cohen’s d = 0.242) (see Table 4). The specifics (see Table 3) were as follows. Good sleep quality: 23.05% (medical) vs.16.15% (non-medical). Moderate-severe sleep disorders: 20.47% (medical) vs. 27.03% (non-medical). Mild sleep problems: similar (55.33% vs. 55.60%). PSQI frequency distributions (Fig. 2) showed bell-shaped curves for both groups, with non-medical students having a higher proportion in the poor sleep ($\ge$12) range. Overall, medical students, despite greater psychological stress, had no worse and even better sleep quality than did non-medical students, a phenomenon termed “stress-sleep decoupling” where stress and sleep did not deteriorate synchronously.

Fig. 2.

Differences in the distribution of sleep quality (total PSQI scores) between medical and non-medical students. Note: 1 = Medical students, 0 = Non-medical students, Y axis = n, X axis = PSQI score.

Using sex as the grouping variable, results of the Mann-Whitney U test (Table 5) showed significant differences in both DASS total scores and PSQI total scores between males and females (all p 0.001). Females had a DASS median of 32 vs. 29 for males (Cohen’s d = 0.171), and a PSQI median of 9 vs. 8 (Cohen’s d = 0.200). Although both represented small effect sizes, the consistency across measures indicated that female students face greater mental-health distress and poorer subjective sleep quality. This finding underscores the need to account for sex differences in mental health and sleep interventions. Targeted support for female students—such as emotional regulation training, work-family-role-balance counseling, and sleep hygiene education—may mitigate these disparities.

As shown in Table 6, Kruskal-Wallis H tests comparing mental health (depression, anxiety, stress) and sleep scores across grades revealed significant grade effects (H values: 53.350–230.515, all p 0.001), though effect sizes were small (Cohen’s f 0.1 for all). This indicates that although grade differences were statistically significant, their practical impact on outcomes was limited. Median scores showed that mental health indicators fluctuated across grades rather than changing monotonically, whereas PSQI did not fully align with mental health trends.

Specifically, in terms of mental health, the stress level of sophomores showed the first significant peak: the median score of the DASS stress subscale for sophomores was 12, which is 1 point higher than that of freshmen. Additionally, 52.8% of sophomores scored $\ge$12, which was significantly higher than the 39.9% of freshmen ($\chi$² = 182.05, p 0.001). From the perspective of interval estimation, the 95% confidence interval for sophomores’ stress scores was 11.8–12.2, compared with 10.9–11.1 for freshmen. This data characteristic is consistent with the “sophomore slump” phenomenon (Webb and Cotton, 2019). By the junior year, the median score remained the same as that of the sophomore year, although the standard deviation further increased to 3.743. This indicates that the differentiation in the distribution of stress within the group persisted, rather than a simple “decline”. It suggests that although some students recovered psychologically after adapting to the sophomore year, others continue to face persistent stress. A notable trough emerged in seniors, with DASS total scores dropping to a median of 28 (vs. 33 for sophomores and 32 for juniors). This may have been due to the relative ease of the senior year, where some students gain valuable respite and adjustment time. However, fifth-year students (final-year undergraduates) experienced a resurgence in all DASS subscales (depression: 12; anxiety: 11; stress: 13—all grade highs), possibly driven by employment pressures and clinical internship responsibilities. Moreover, anxiety showed a “double-peak” distribution in the second and fifth years, reflecting a shift in the sources of anxiety: early-stage stress is situational (e.g., exams, grades), whereas late-stage stress is career-related chronic anxiety. It should be noted that the sample size of fifth-year students is small (n = 33, 0.54%), with limited statistical power, and the results more reflect the potential changing trends of this group.

PSQI also fluctuated by grade but followed a different pattern than mental health indicators. Sophomores and fifth-year students had PSQI medians of 9 (vs. 8 for freshmen, juniors, and seniors), indicating poorer subjective sleep during high-stakes academic periods (sophomore year) and pre-graduation (fifth year). Notably, seniors showed reduced mental stress (DASS) but unchanged PSQI (median = 8), suggesting asynchronous mental health and sleep dynamics. This dissociation implies that sleep may be influenced by independent variables like internship schedules or behavioral habits, beyond psychological stress. In sophomore and fifth-year students, a bidirectional stress-sleep deterioration pattern emerged: academic/employment stress $\to$ sleep deprivation $\to$ daytime cognitive impairment $\to$ heightened stress (Lund et al, 2010).

The statistical results and trends of the aforementioned grade differences indicate that the dynamic impact of grade on college students’ psychology and sleep is characterized by “stage-specific fluctuations and interweaving of multiple factors”. There are statistically significant differences in mental health and sleep characteristics among students of different grades, which provides an important reference for understanding the stage-specific characteristics of the group. However, it should be noted that the effect sizes in some multiple-group comparisons were small, suggesting that this phenomenon may not have reflected strong individual correlations. Moreover, the independent impact of grade was limited; actual differences are also influenced by multiple moderating factors such as living habits, social support, and academic pressure. The clinical significance of the dynamic impact of grade on college students’ psychology and sleep needs to be comprehensively judged by combining the core stressors of each grade and group characteristics.

Spearman’s rank correlation coefficients were used to examine the association between PSQI total scores and DASS subscale scores (Fig. 3). The p-values for all pairwise correlations in this analysis are provided in Supplementary Material. All primary variables showed statistically significant pairwise correlations (p 0.01). The DASS total score exhibited a moderate positive correlation with PSQI ($\rho$ = 0.489, p 0.001), indicating that higher psychological distress was associated with poorer sleep quality—a synchronous worsening trend between stress and sleep problems. Subscale analyses showed moderate positive correlations between PSQI and depression ($\rho$ = 0.42), anxiety ($\rho$ = 0.45), and stress ($\rho$ = 0.44) scores (all p 0.001). Grade showed weak positive correlations with DASS ($\rho$ = 0.087) and PSQI ($\rho$ = 0.041), confirming its limited direct impact on mental health/sleep. Sex correlated positively with both DASS and PSQI ($\rho$ = 0.108, p 0.001), consistent with earlier findings that females reported higher distress and poorer sleep. Age showed negligible correlations with other variables ($|$$\rho$$|$ 0.05), suggesting that age differences had minimal impact among college students. All variables showed perfect self-correlation ($\rho$ = 1.000), and most off-diagonal correlations were significantly non-zero, validating the reliability of the correlation matrix.

Fig. 3.

Spearman correlation heatmap: Associations between mental health indicators, sleep quality, and demographic factors. Supplementary data tables with p-values ​​for association analysis are provided in the Supplementary Material.

Contrary to our expectations and the common understanding in previous studies that “stress and sleep deteriorate synchronously” (Alwhaibi and Al Aloola, 2023; Huang et al, 2024), our study observed that although medical students had significantly higher DASS stress scores than did non-medical students (positive rate of the stress scale: 19.75% vs. 13.93%), their median PSQI total score was slightly lower (8 vs. 9). That is, a “stress-sleep decoupling” occurred in the medical student population: subjective sleep quality did not decrease with increased psychological stress, and some sleep indicators were even better than those of non-medical students. We should emphasize that because these were cross-sectional observational findings, this phenomenon reflected the association pattern at a specific time point, and cannot rule out the influence of unobserved variables (such as socioeconomic factors, technology use, social support scales, comorbidities, exercise, etc.), and requires verification of a causal relationship and mediating effects through longitudinal studies. Despite needing verification of its mechanism and stability, this phenomenon—observed in a large sample of Chinese medical students—suggests a unique stress-sleep association pattern in medical education, offering a new perspective for understanding sleep quality’s protective mechanisms in high-stress environments.

The present study revealed a “peak-trough-peak” fluctuation in mental health and sleep across grades, a non-linear pattern indicating that different academic years entail distinct stressors and adaptation tasks.

This study confirmed that female college students exhibit poorer mental health and sleep quality than do male students, consistent with prior research: women have 1.4 times higher insomnia susceptibility than men (Zhang and Wing, 2006), nearly double the likelihood of poor sleep quality (Madrid-Valero et al, 2017), and higher rates of depression, anxiety, and neuroticism (Sung and Kim, 2020; Kajonius et al, 2025).

This study has some limitations that should be noted when interpreting the results. First, restricted causal inference: the cross-sectional design only reflects the association patterns between stress and sleep (e.g., the “decoupling” phenomenon) at a specific time point, and cannot verify dynamic causal relationships or long-term stability. Subsequent longitudinal studies are needed for further validation. Second, sample representativeness and heterogeneity: medical sub-disciplines (e.g., clinical medicine vs. basic medicine) were not differentiated, which may mask inter-subdiscipline differences; the fifth-year sample accounted for only 0.54%, potentially underestimating the actual stress levels of senior students. Future studies should optimize this issue through targeted sampling. Third, measurement and selection biases: reliance on subjective scales (PSQI, DASS-21) may lead to underestimation of real problems due to medical students’ cognitive tendencies related to health knowledge; survival bias may weaken the assessment of cumulative effects of stress/sleep problems. Subsequent research can be improved by combining objective monitoring (e.g., polysomnography) and longitudinal designs.