INTRODUCTION
Good sleep is a key determinant of psychological well-being and academic performance in emerging adulthood, and the most recent evidence suggests that poor sleep quality in university students is not an isolated phenomenon but rather highly prevalent. A systematic review and meta-analysis of African students estimated a pooled prevalence of 63.31% for poor sleep quality using the Pittsburgh Sleep Quality Index (PSQI), with consistency across subregions (I2≈97%) [1]. Similarly, a global meta-analysis of 48 studies estimated the prevalence of insomnia symptoms in undergraduate students at 46.9%, with high heterogeneity and differences by continent and instrument [2]. Recent studies conducted in non-English-speaking countries have reported similar findings, with 50%–70% of university students experiencing poor sleep quality [3,4].
Beyond frequency, the links between sleep quality and negative affect are robust. In a series of repeated studies (2020–2023) with university students, sleep quality (PSQI) was moderately to strongly associated with anxiety and depression, with a marked concern about sleep onset (latency) as a key component [5]. Multicenter results in medical students reinforce the convergence between sleep disturbances and greater symptoms of depression, anxiety, and stress, with a worse profile in women [6].
Beyond psychological and behavioral factors, sleep quality is tightly regulated by the circadian timing system, which coordinates the sleep–wake cycle through endogenous biological clocks synchronized primarily by the light–dark cycle. In university students, lifestyle patterns characterized by irregular schedules, nighttime academic demands, social jetlag, and excessive evening light exposure can induce circadian misalignment, defined as a mismatch between internal circadian rhythms and externally imposed sleep–wake schedules. This misalignment has been consistently associated with shorter sleep duration, greater sleep latency, lower subjective sleep quality, and increased daytime dysfunction. Evidence indicates that delayed chronotype—more prevalent in late adolescence and early adulthood—confers higher vulnerability to poor sleep quality, anxiety, and stress when academic schedules are incompatible with delayed circadian phase, a pattern observed in both international samples and Latin American contexts [7,8]. Reviews in college populations can be interpreted within circadian frameworks, wherein circadian dysfunction operates as an upstream mechanism linking sleep disturbances with emotional dysregulation, reframing sleep quality as a marker of circadian health [9,10].
Suicidal ideation and its relationship to sleep demand priority attention [11]. A longitudinal study showed that poor sleep quality predicted increases in suicidal ideation one month later, partially mediated by negative affect and with plausible neurofunctional substrates [12]. In Latin America, an analysis of university students in Honduras found that better sleep and higher self-esteem were associated with lower suicidal ideation, after adjusting for social media use and barriers to help [13]. These results align with studies in the United States that link sleep continuity, quality, and disorders with suicidal ideation and attempts, positioning sleep as a modifiable risk marker [14].
Sex/gender differences add nuance to the phenomenon. Recent data report that, with similar sleep durations, women exhibit worse subjective sleep quality, greater daytime sleepiness, and more stress, suggesting different support needs [15]. Likewise, when analyzing academic performance, distinct patterns emerge: in men, sleep (particularly quality and quantity) predicts performance [7,8]; in women, stress, anxiety, and depression do so [16]. In Latin America, high levels of academic stress and sleep problems have been documented among health and physical education/sports students, with variations by country and sex [17].
Since its inception, the PSQI has been the gold standard for assessing multiple dimensions of sleep, but its factor structure can vary depending on the population and online modality, with implications for the interpretation of scores and cut-off points. In university students from low-income regions, the online version showed adequate reliability and structural validity with a better one-factor fit [18]. In medical students, however, a three-factor solution (efficiency, perceived quality, and daytime disturbances) was supported [19]. In Colombia, recent evidence from nursing students confirms improved model fit and acceptable temporal stability, highlighting the value of local psychometric analyses [20], as is done in the present study.
Regarding intervention, the US Department of Veterans Affairs and Department of Defense (VA/DoD) [21] guidelines recommend cognitive behavioral therapy for insomnia (CBT-I) as a first-line treatment, advising against sleep hygiene as a sole treatment and prioritizing behavioral options over pharmacotherapy whenever possible [21]. In the university setting, there is solid evidence of the effectiveness of digital CBT-I with personalized feedback applications in reducing the severity of insomnia and improving sleep parameters, with effects sustained at 3–4 months [22]. A specific meta-analysis in university students showed moderate effects of psychological interventions on sleep quality, with CBT-I showing an advantage over mindfulness interventions [23]. Moreover, in young people with insomnia and subclinical depression, the CBT-I app reduced the risk of major depressive episodes, indicating a relevant preventive potential for the campus [24].
In this context, the present study focuses on first-semester students enrolled in health and business programs to characterize sleep quality and its relationships with anxiety, stress, depression, insomnia, life satisfaction, resilience, and suicidal risk; to compare these variables according to PSQI categories and sex; and to model PSQI variance using parsimonious and robust regression approaches. By providing region-specific evidence from the Colombian Caribbean—an area under-represented in international sleep research—this study extends prior literature by contextualizing established associations within a Latin American public university setting, rather than introducing new constructs. This contribution addresses regional research gaps and aligns with recommendations to incorporate sleep as a cross-cutting dimension in student health pathways [17,21].
METHODS
Study design
This was an observational, analytical, and cross-sectional study with a correlational-comparative approach. A total of 297 students were analyzed, and comparisons were performed according to sex and PSQI-defined sleep quality categories. Multiple regression models were also used to explain the variation in the overall PSQI score.
Participants
The target population consisted of first- and second-semester university students from the health and business programs at a public university in the Colombian Caribbean. The final sample comprised 297 participants, aged between 15 and 47 years. The sample included 209 females (70.37%) and 88 males (29.63%).
For the analyses, 36 numerical variables were included with less than 30% missing data and ≥5 distinct values, excluding personal identifiers and variables with low variability. The exclusion criteria were: 1) failure to provide informed consent and 2) being under the influence of drugs or medications that would alter the nervous system during the study.
The final sample size (n=297) was determined by available recruitment during the study period and is comparable to sample sizes commonly reported in cross-sectional studies examining sleep quality and emotional symptoms in university populations. This sample size was considered adequate to estimate prevalence parameters with acceptable precision and to detect moderate associations between sleep quality and psychological variables using nonparametric tests and multiple regression models with robust standard errors.
Measurements
Depression Anxiety and Stress Scales-21
The Depression Anxiety and Stress Scales-21 (DASS-21) is a shortened version (21 items) of a 42-item self-report instrument designed to measure three related negative emotional states: depression, anxiety, and stress [25-27]. In the present study, it has meritorious sampling adequacy (Kaiser–Meyer–Olkin [KMO]≈0.946), adequate Bartlett’s sphericity (χ2(210)=4,031.73; p<0.001), and good consistency for total score (α=0.949; α=0.870 for stress, α=0.865 for anxiety, and α=0.848 for depression). The one-factor solution explains approximately 62.2% of the variance.
Athens Insomnia Scale
The 8-item Athens Insomnia Scale (AIS) is based on the International Classification of Diseases, 11th Revision (ICD-11) criteria, and its Spanish version has shown internal homogeneity, reliability, and clinical validity [28]. Recent studies confirm good fit and subgroup invariance for the AIS in the adult population [29]. In the Colombian Caribbean, the AIS has been used with university students to estimate insomnia and its relationship with academic performance, demonstrating local relevance [30]. In the present study, it has meritorious sampling adequacy (KMO≈0.82), clear Bartlett’s sphericity (χ2(28)=797.81; p<0.001), and good consistency (α≈?0.82). The one-factor solution explains approximately 45.43% of the variance, consistent with its use as a screening tool for insomnia.
Okasha Suicidality Scale
The Okasha Suicidality Scale (OSS; 4 Likert-type items) briefly assesses suicide risk. In students from Santa Marta, Colombia, a one-dimensional structure and high internal consistency (α and ω≈0.85) were confirmed, along with convergent and divergent validity [31]. In the present study, the sample adequacy was commendable (KMO=0.80), and Bartlett’s test of sphericity was significant (χ2(6)=795.27, p<0.001), indicating that the matrix is factorable. The internal consistency of the total score was high (α=0.89). A one-factor model explained 76.61% of the total variance.
Pittsburgh Sleep Quality Index
The PSQI includes 19 items (+5 optional) grouped into seven components and produces a global score (0–21) useful for identifying sleep disorders [32]. The Colombian validation documented internal consistency (α≈0.78) and clinical validity through differences between groups (subjective disturbance, use of hypnotics, sleep-onset insomnia), and its use has spread among university students in the country [32,33]. In the present study, at the level of standardized components (quality, latency, duration, efficiency, disturbances, medication, daytime dysfunction), the following was observed (α=0.612; KMO=0.729; Bartlett’s test of sphericity: χ2(21)=268.62; p<0.001; explained variance=33.41%). This pattern reflects the multidimensionality of the PSQI.
Satisfaction with Life Scale
The Satisfaction with Life Scale (SWLS) consists of 5 items with a 7-point Likert scale; higher scores indicate greater life satisfaction [34]. It exhibits a unidimensional structure and high internal consistency (α≈0.88) in the general Spanish population, and good psychometric performance in university students in Bogotá (α≈0.84) using local norms [34,35]. In the present study, the sample adequacy was commendable (KMO=0.84) and Bartlett’s test of sphericity was significant (χ2(10)=574.91, p<0.001), indicating that the matrix is factorable. The internal consistency of the total score was adequate (α=0.83). A one-factor model explained 61.58% of the total variance.
Connor-Davidson Resilience Scale
The Connor-Davidson Resilience Scale (CD-RISC-10) assesses resilience with 10 items (5-point Likert scale) and has shown a unifactorial structure and good metrics (α≈0.85; acceptable temporal stability) in young Spanish-speaking adults [36]. Additional results in Hispanic samples (employed and unemployed individuals) confirm its validity and reliability [37,38]. For the present study, excellent sample adequacy (KMO=0.91) is reported, and Bartlett’s test of sphericity was significant (χ2(45)=1012.15, p<0.001), which supports the factorization of the matrix. The internal consistency of the total score was adequate (α=0.86). A one-factor model explained 45.95% of the total variance.
Statistical analysis
The proportion of missing data per variable did not exceed 30%, consistent with the predefined inclusion criterion (≥70% observed values). After applying these criteria, the analytical dataset retained 297 complete cases for correlational, group comparison, and regression analyses. Given the low overall proportion of missingness and the absence of systematic patterns across key study variables, complete-case analyses were performed without applying data imputation procedures.
Although missing data for individual variables did not exceed 30%, the analytic sample consisted of 297 complete cases with no missing values on the variables included in the main analyses. Participants with missing data on one or more focal variables were therefore excluded from the regression analyses.
Regarding the preparation and assumptions, the normality of the distributions was assessed using the Kolmogorov–Smirnov (K–S) test, revealing non-normality in most variables (p<0.001). Homogeneity of variances was tested using Levene’s test. Based on the assumptions, pooled/Welch t-tests or Mann–Whitney U tests were applied. p-values were adjusted using Holm’s method for multiple comparisons. Effect sizes were reported (Hedges’ g for parametric tests and biserial rank-sum r for nonparametric tests). Additionally, Kruskal–Wallis tests were performed by sex (k=2) and by PSQI categories (k=3), with ε2 as the effect size. Spearman’s rank correlation coefficients (ρ) were estimated between focal variables, with 95% confidence intervals and Holm’s method for p-value adjustment across the entire set of tests. To explain the overall PSQI, ordinary least squares (OLS) and weighted least squares (WLS) regressions (weighted by sex) with robust HC3 (heteroscedasticity-consistent) standard errors were fitted. Collinearity was assessed using the variance inflation factor (VIF), and a parsimonious model was obtained by removing variables with high VIF (e.g., resilience), retaining insomnia, stress, anxiety, depression, suicidal risk, and life satisfaction, as well as sex (female dummy), as predictors. The models explained approximately 58%–59% of the PSQI variance (R2 and adjusted R2). Two-tailed tests with α=0.05 are reported; test statistics, adjusted p-value, 95% confidence intervals (where applicable), and effect sizes are reported according to the type of test.
Multicollinearity among predictors in the regression models was assessed using the VIF. Variables showing high collinearity (VIF>10) were considered problematic for model interpretability. Based on this criterion, resilience was excluded from the parsimonious model due to excessive collinearity with other emotional variables. The remaining predictors showed acceptable VIF values, supporting the stability and interpretability of the regression coefficients. In the final model, VIF values ranged approximately from 1.4 to 3.6. Statistical analyses were performed using the IBM SPSS Statistics version 28.0 (IBM Corp.).
Ethical considerations
The protocol was reviewed and approved by the Research Ethics Committee of the University of Magdalena and was formally established in the Research Vice-Rector’s Office Commencement Act 07162024. Participants aged 15 to 17 years were included because they are formally enrolled university students in the first semester of their academic programs and therefore constitute part of the target population of interest. In accordance with national ethical regulations and institutional requirements, participation of minors was contingent upon informed assent provided by the student and written informed consent obtained from a parent or legal guardian, with institutional authorization. The suicide risk management protocol for minors was reinforced, including immediate detection, clinical notification, support, and priority referral to university and/or external health services, while preserving confidentiality in accordance with applicable regulations. The study was conducted in compliance with Resolution 8430/1993 of the Colombian Ministry of Health and Law 1581/2012 on data protection. The overall risk classification was minimal, with specific mitigation measures for the suicidality component, such as providing support information and clearly defined care pathways upon completion of the questionnaire.
RESULTS
For anxiety, slightly more than two-fifths of the sample fell within the normal range (41.08%), followed by extreme (21.55%) and moderate (20.54%); mild (11.45%) and severe (5.39%) levels were less frequent. This distribution suggests a high central column at the upper end (extreme/moderate) coexisting with an equally large group in the normal range, which anticipates heterogeneity in the severity of anxious distress among all participants.
Emotional symptom indicators (anxiety, stress, and depression) showed heterogeneous distributions, with notable interindividual variability ranging from normal to clinically elevated levels. Anxiety presented higher concentrations at moderate and extreme levels, whereas stress and depression were predominantly within normal ranges but with meaningful upper tails, supporting subsequent analyses by sleep quality and sex (Table 1).
Most participants did not meet the criteria for clinical insomnia, although mild symptoms were common. In contrast, poor sleep quality was prevalent, indicating that sleep difficulties in this population extend beyond insomnia and affect broader aspects of perceived sleep and daytime functioning. Suicide Risk was largely low, yet the presence of medium and high risk cases remains clinically relevant for prevention efforts (Table 1).
Life satisfaction showed a general tendency toward dissatisfaction, consistent with the observed sleep difficulties and emotional symptoms, while resilience levels were predominantly moderate to high. This coexistence suggests that resilience may function as a protective resource for some individuals. As expected, most variables exhibited non-normal distributions, justifying the use of non-parametric analyses (Table 1).
The correlational pattern was robust and aligned with theoretical expectations. Strong associations were observed among negative affect indicators and between insomnia and global sleep quality, while life satisfaction and resilience showed protective associations of moderate magnitude. Overall, correlations ≥0.60 predominate between sleep and negative affect, and among emotional components themselves. The associations with life satisfaction and resilience show the expected direction (protection), with a magnitude of moderate to high (Table 2).
Sex differences were detected in anxiety, stress, insomnia, suicidal risk, sleep quality, and resilience; Depression and Life Satisfaction did not reach significance. In practical terms, the ε2 effects by sex are small (resilience, close to moderate), suggesting statistically detectable but moderate differences in magnitude (Table 3).
In females, differences were observed in all variables, with particularly high values for sleep quality and insomnia, followed by anxiety and stress. In males, differences appeared in sleep quality and insomnia; the remaining variables were close to significance. This pattern indicates the variation in sleep quality levels is more pronounced in females than in males for most variables (Table 4).
All variables showed differences between PSQI categories. Effect sizes (ε2) were moderate to large: PSQI, insomnia, anxiety, stress, and depression; suicidal risk, life satisfaction, and resilience also showed non-trivial effects. This reinforces the clinical importance of sleep quality in the variation of emotional symptoms and well-being in the sample (Table 4).
The OLS multiple regression model explained a substantial proportion of the variance, revealing a positive association between high PSQI scores and insomnia, stress, and anxiety. Depression showed a marginal negative coefficient, likely due to collinearity suppression effects, while sex was not significant. The results of the parsimonious model were comparable (Table 5). The final model, which included anxiety, stress, depression, insomnia, suicide risk, life satisfaction, and sex, showed a high overall fit. Insomnia, stress, and anxiety remained positively associated with lower PSQI scores (Table 6).
Robust and consistent effects were verified: insomnia (strongest effect), stress, and anxiety are positively associated with poorer sleep quality (PSQI), with robust significance in OLS (HC3) and a qualitatively similar pattern under sex-weighted WLS. Regarding collinearity and suppression, VIF diagnostics were examined for all predictors included in the regression models. In the full model, resilience exhibited VIF values exceeding the conventional threshold (VIF>10), indicating problematic multicollinearity and justifying its exclusion from the parsimonious specification. In the final parsimonious model, VIF values for the retained predictors (insomnia, stress, anxiety, depression, suicidal risk, life satisfaction, and sex) fell within acceptable ranges (approximately VIF≈1.4–3.6), supporting the stability and interpretability of the regression coefficients. The marginal negative coefficient observed for depression (p=0.049) is consistent with a suppression effect arising from shared variance among negative affect indicators.
The model fit is high (adjusted R2≈0.58 in OLS) in both the full and parsimonious models, and the WLS shows similar R2 and adjusted R2 values, reinforcing the robustness of the findings under sex weighting. On the other hand, sex: it does not show a significant effect in the complete model, even with weighting in WLS.
DISCUSSION
From an incremental perspective, the present findings complement existing literature by jointly examining sleep quality, insomnia symptoms, negative affect, life satisfaction, resilience, and suicide risk within a single analytical framework. While the observed associations are consistent with previous studies, their documentation in a regional Colombian context provides locally relevant evidence that may inform future longitudinal or interventional research in similar university populations.
This study identified a high prevalence of sleep problems, with approximately two-thirds of participants classified as having poor sleep quality and nearly one-third reporting mild insomnia, a distribution that is consistent with recent descriptions of sleep disturbances in international university samples [39]. These findings characterize a substantial burden of sleep-related complaints in this population rather than indicating specific etiological mechanisms.
From a circadian perspective, the strong associations observed between poorer sleep quality (PSQI), insomnia, stress, and anxiety are conceptually consistent with circadian frameworks that describe sleep disturbances in university students as embedded within broader sleep–wake regulation processes, rather than as isolated nocturnal complaints. In the present study, the convergence between insomnia symptoms and PSQI scores may be understood as reflecting shared behavioral and timing-related features commonly discussed in circadian and sleep–wake models, such as delayed sleep habits and irregular schedules [9,10,21]. However, as no direct circadian markers were measured, these interpretations remain theoretical and associative rather than mechanistic. In this sense, sleep quality can be viewed as a descriptive and transdiagnostic indicator that co-occurs with emotional burden and contextual demands typical of the university environment, supporting its relevance for screening and preventive considerations without implying specific biological pathways or causal directionality.
The associations observed between sleep quality and negative affect were consistent and of moderate to high magnitude: insomnia scores showed strong convergence with the PSQI, and poorer sleep quality was associated with higher levels of anxiety and stress. These patterns align with contemporary evidence reporting stable associations between PSQI scores and emotional symptoms in university students [5,40]. Furthermore, comparisons across PSQI categories revealed moderate to large effect sizes for anxiety, stress, depression, insomnia, life satisfaction, and resilience, supporting the interpretation of sleep quality as a transdiagnostic correlate that systematically co-occurs with emotional and well-being indicators. These patterns should be understood as associative and context-dependent, rather than indicative of directional or causal effects among the studied variables [41].
In regression analyses, approximately 58%–59% of the variance in PSQI scores was statistically explained by a parsimonious set of emotional and behavioral variables, with insomnia, stress, and anxiety showing the strongest associations. The marginal negative coefficient observed for depression is compatible with suppression effects arising from collinearity among negative affect constructs, a phenomenon commonly reported in multivariate models of emotional symptoms [42]. Additionally, small to moderate sex differences were observed, with poorer sleep quality profiles more pronounced in females across PSQI categories. These differences are consistent with prior reports of sex-related variation in subjective sleep quality and stress, even when sleep duration is similar [15,16].
Although the prevalence of moderate or high suicide risk was low, its association with poorer sleep quality and elevated negative affect highlights the relevance of considering sleep indicators within broader psychosocial screening frameworks on campus. This observation is in line with previous studies reporting associations between sleep disturbances and suicidal ideation in university populations [12,13], without implying a causal relationship.
Recommendations
Universities may consider implementing periodic screening strategies that integrate sleep quality measures (e.g., PSQI) alongside brief indicators of anxiety, stress, and depression, framed as a descriptive and preventive approach rather than a diagnostic pathway [41,43]. These efforts can be complemented by the dissemination of evidence-informed information about sleep regulation and common sleep difficulties, as well as the incorporation of clear referral pathways within university settings [23,44].
In line with current guidelines and research evidence, institutions may prioritize non-pharmacological approaches to sleep problems, particularly CBT-I, while avoiding the prescriptive application of clinical interventions in non-clinical academic contexts [21,23,44]. In this regard, digital and app-based tools with personalized feedback may serve as scalable options for health promotion and self-management support, with demonstrated effectiveness in improving sleep outcomes and reducing insomnia severity among university students [22].
Finally, prevention and intervention strategies should adopt a context-sensitive perspective, taking into account sex differences, patterns of stress and daytime functioning, and fluctuations associated with academic demands [15,16,45]. Although causal relationships cannot be inferred, integrating sleep-related indicators into broader awareness and referral frameworks for psychological distress and suicidal ideation may support early identification and prevention efforts in university populations [12,13,45,46].
Limitations
This study has several limitations that should be considered. First, the internal consistency of the PSQI components was modest, which may attenuate effect sizes when used as the dependent variable. In addition, the conceptual overlap between the Athens Insomnia Scale and the PSQI may have contributed to inflated associations, particularly in relation to insomnia.
The cross-sectional design prevents causal inference, and the exclusive use of self-report measures may introduce recall and desirability biases. Furthermore, variability in PSQI structure and cut-off values across populations suggests that the categorization used should be interpreted with caution and supported by local psychometric evidence.
Multicollinearity among emotional variables may also have influenced regression estimates, as suggested by the suppression effect observed for depression. Finally, the use of a single-institution sample limits generalizability, and the timing of data collection in relation to academic demands may have affected stress, anxiety, and sleep patterns.
Conclusion
This study highlights the high prevalence of poor sleep quality among university students and its consistent association with insomnia, stress, and anxiety. The findings support the role of sleep quality as a relevant indicator of emotional distress and well-being in this population. Although causal relationships cannot be inferred, the results underscore the importance of incorporating sleep assessment into broader mental health screening and prevention strategies in university settings. Future research should employ longitudinal designs and objective measures to further clarify the temporal and contextual dynamics of sleep and emotional functioning.
PDF Links
PubReader
ePub Link
XML Download
Full text via DOI
Download Citation
