Introduction
In the newly released World Mental Health Report by the World Health Organization (WHO), mental health is defined as a state of well-being that enables individuals to cope with the stresses of life, realize their abilities, learn and work well, and contribute to their communities.1 The Global Burden of Disease Study (GBD) 2019 indicates that mental disorders account for 4.9% of the global total disease burden, remaining one of the top ten leading causes of burden worldwide.2 The lifetime prevalence of any type of mental disorder among Chinese adults stands at 16.6%,3 and the age-standardized prevalence of mental disorders among Chinese children and adolescents is 8.9%.4 Mental disorders and comorbid mental disorders are also common among college students,5–7 especially freshmen.8–11 The prevalence rates of depressive and anxiety symptoms among college students are 33.6% and 39.0% respectively.12 College years represent a critical developmental period for students transitioning from late adolescence to early adulthood,13 but the high incidence of psychological problems has been associated with poorer college students’ academic performance, physical health, social relationships, and quality of life.14–16
Alcohol consumption, even alcohol abuse, is quite prevalent among college students and has become a public health concern that cannot be ignored.17 A study showed that the alcohol consumption rate among male and female college students in China was 66.8% and 31.7%,18 and there was a trend of high-risk drinking among Chinese college students.19 Although existing research has confirmed that alcohol use is closely linked to mental health problems, such as poorer academic performance,20 social anxiety,21 an immature state of self-identity,22 and lower self-esteem,23 the specific pathways through which drinking is associated with psychological well-being remain unclear, thus highlighting the need for further research. In current college students’ mental health assessment, consideration of alcohol consumption remains insufficient and has not been fully integrated into the comprehensive assessment system.
A health-promoting lifestyle (HPL) can be regarded as a positive lifestyle, encompassing a variety of health-enhancing behaviors such as self-actualization, health responsibility, physical activity, nutrition, interpersonal support, and stress management.24 A study showed that participants with a healthy lifestyle pattern had a 69% lower risk of developing new-onset mental disorders, and healthy lifestyles were closely associated with reduced mortality rates and extended life expectancy.25 Therefore, adhering to HPL is a crucial protective factor for maintaining and improving mental health. Although existing studies have fully confirmed the direct positive association between HPL and mental health, research on whether HPL may play a mediating role between alcohol consumption and mental health is relatively scarce.
An interaction also exists between sleep quality and mental disorders.26 Disruptions in sleep quality, continuity and duration have been associated with the occurrence or worsening of mental symptoms in vulnerable individuals; conversely, mental symptoms can repeatedly intensify poor sleep and disrupt clock-controlled processes.27 A study involving adolescents aged 16–18 years revealed that individuals who met the criteria for insomnia and had a sleep duration of <6 hours had an over eightfold increased risk of developing depression.26 At the same time, the negative association between alcohol consumption on sleep quality has also been clearly confirmed, even low levels of alcohol consumption may be associated with poorer sleep health.28 Current research on whether sleep quality plays a mediating role in the association between alcohol consumption and psychological symptoms remains insufficient.
HPL, sleep quality, and psychological symptoms are closely interrelated.26,27,29,30 The present study focused on the HPL → sleep quality sequence, because HPL was placed earlier in the theoretical model, whereas sleep quality was placed subsequently in the model. Intervention studies have shown that lifestyle modifications can improve sleep quality.31,32 Given the cross-sectional design, this sequence should be regarded as a theoretically guided framework rather than an empirically established temporal ordering.
Although the associations between drinking, lifestyle, sleep quality, and mental health symptoms have been established, most prior studies have examined these factors in isolation or through single-mediator models, whereas the sequential pathway of drinking status → health-promoting lifestyle → sleep quality → psychological symptoms has been rarely explored. The potential sequential pathways have not been empirically tested. First-year medical students may be particularly vulnerable to these interacting factors, as their physical activity and sleep quality tend to deteriorate.33,34 Based on existing research, this study aims to explore the mediating roles of HPL and sleep quality in the association between drinking status and mental health among college students, and further examines whether these factors operate in parallel or form a chain-mediated pathway.
Two hypotheses are proposed: first, HPL and sleep quality each have significant mediating associations in this association; second, HPL and sleep quality form a chain mediating association. This study is expected to provide a new perspective for understanding the associations between alcohol consumption and college students’ mental health and offer a theoretical basis for formulating targeted mental health interventions for this group.
Materials and Methods
Study Design and Participants
A cross-sectional survey was conducted by a medical school in Shanghai, China, and data were collected in October 2024. The study was conducted via an online questionnaire. If the students agreed to participate in the study, they could proceed with filling out the survey.
After reviewing the electronic informed consent form, students were required to either scan a QR code or click on the survey link via their mobile phones/computers to complete the questionnaire. The final analytical sample comprised 2272 first-year students, with no missing data across all variables.
Quality-control procedures were implemented through the Wenjuanxing platform. All items were mandatory and participants could not submit a questionnaire with missing answers. The platform was set to allow only one submission per IP address/device and to record completion time.
After data collection, the dataset was screened for duplicate responses, unusually short completion times, invariant responding, and logical inconsistencies.
The study was performed in accordance with the Declaration of Helsinki and approved by the Ethics Committee of Shanghai University of Medicine and Health Sciences (2024-TG2-01-640102197110170019) on October 8, 2024. All procedures were performed in accordance with ethical standards. Online informed consent was obtained from each participant.
Measurements
Mental Health
Mental health was assessed using the Symptom Checklist-90 (SCL-90), a self-reported multidimensional instrument developed by Derogatis,35 with its Chinese version revised by Wang Zhengyu.36 This scale comprises a total of 90 items, each rated on a 5-point Likert scale, including 9 dimensions (somatization, obsessive-compulsive, interpersonal sensitivity, depression, anxiety, hostility, phobia anxiety, paranoid ideation, psychoticism) and one other factor. Collectively, these nine symptom dimensions and one additional factor form the overall mental health score. When the total score of SCL-90 is treated as a continuous variable, higher total scores are indicative of more psychological symptoms.35 When the total score is treated as a binary variable, a positive case was defined as either a total SCL-90 score > 160 or more than 43 positive items (defined as items with score ≥ 2).37,38 The scores of its nine symptom dimensions plus the additional factor were extracted individually for subsequent analysis of the potential associations between drinking and specific psychological symptoms.
It should be noted that the SCL-90 is a self-report screening tool that captures the severity of psychological symptoms over the past week rather than providing a clinical diagnosis of mental disorders. In the present study, the Cronbach’s α for the SCL-90 was 0.978, and validity is presented in Table S1.
Drinking Status
In this study, abstainers were defined as those who had never or almost never consumed alcohol in the past 12 months and had not drunk weekly in the past either.39 Participants who did not meet the criteria for abstainers were classified as drinkers. This group included former weekly drinkers, reduced-intake drinkers, occasional drinkers, and weekly drinkers. Due to the lack of detailed information on drinking quantity, frequency, binge drinking, or alcohol-related problems in the questionnaire, a binary classification was adopted.
Health-Promoting Lifestyle
Health-promoting lifestyle was measured using the Health-Promoting Lifestyle Profile-II (HPLP-II), originally developed by Walker et al.24 The present study used the 40-item Chinese version revised for the Chinese population by Cao Wenjun et al.40 The HPLP-II was originally developed as the Health-Promoting Lifestyles Profile (HPLP) within a wellness framework.24 This measure consists of 40 items across 6 dimensions: interpersonal relationships, health responsibility, stress management, nutrition, physical activity, and spiritual growth. The HPLP-II is scored on a 4-point Likert scale ranging from 40 to 160, with higher scores indicating higher levels of health-promoting behaviors.24 In the present study, the Cronbach’s α for the HPLP-II was 0.968, and validity is presented in Table S1.
Sleep Quality
Sleep quality was evaluated using the Pittsburgh Sleep Quality Index (PSQI).41 This questionnaire has been widely used in China and has demonstrated good reliability and validity.42,43 The PSQI comprises 19 self-rated items. According to the standard scoring procedure, these items are combined to form seven component scores (sleep quality, sleep latency, sleep duration, sleep efficiency, sleep disturbances, use of sleeping medication, and daytime dysfunction), each scored from 0 to 3. The seven component scores are summed to obtain a global score ranging from 0 to 21.
A higher PSQI score indicates poorer sleep quality and vice versa.41 In the present study, the Cronbach’s α for the PSQI was 0.672, and validity is presented in Table S1.
Socio-Demographic Data
In this study, social demographic data were collected, encompassing age, gender, nation, household registration, student source location, monthly household income, father’s educational background, mother’s educational background, parental marital status, body mass index (BMI), and smoking.
Data Analysis
R version 4.5.0 was used to analyze data. The normality of the distribution of continuous variables, including age, BMI, HPLP-II, PSQI, and SCL-90, was assessed using the Shapiro–Wilk test. Descriptive statistics were used to summarize the data: due to non-normal distributions, continuous variables were expressed as medians with interquartile ranges (IQR), and categorical variables were presented as counts and percentages.
Differences between participants with and without elevated psychological symptoms were analyzed using Mann–Whitney U-tests for continuous variables and chi-squared tests for categorical variables. The correlation coefficients among variables were calculated using Spearman correlation analysis.
To further examine the mediating roles of HPLP-II and PSQI in the association between drinking status and SCL-90, we performed structural equation modeling (SEM) with bootstrap resampling using 10,000 bootstrap samples. The models were estimated using maximum likelihood (ML), with standard errors based on 10,000 bootstrap samples. In these models, both HPLP-II and SCL-90 were treated as latent variables, with HPLP-II measured by its six dimensions and SCL-90 measured by its nine symptom dimensions plus the additional factor.
The total SCL-90 score and its nine symptom dimensions plus the additional factor were analyzed as continuous outcomes in separate models, with all covariates (age, gender, smoking status, BMI, nation, rural/urban residence, student source location, monthly household income, father’s education, mother’s education, and marital status dummies) incorporated as observed exogenous variables directly regressing on each endogenous variable in the structural model. Both direct and indirect pathways were tested. Model fit indices were assessed, including the chi-square/df ratio (χ2/df), comparative fit index (CFI), Tucker-Lewis index (TLI), root mean squared error of approximation (RMSEA), and standardized root mean square residual (SRMR).
As a complementary analysis, a Bayesian SEM was also estimated in which the SCL-90 outcome was dichotomized into screen-positive versus screen-negative categories. This binary parameterization was chosen to improve clinical interpretability and to complement the continuous frequentist model by estimating associations with a clinically relevant threshold rather than assuming a linear relationship. Given that dichotomization may reduce statistical power and discard information, this analysis was interpreted as complementary rather than as a direct validation of the frequentist model.
The Bayesian SEM was constructed using the blavaan package (version 0.5.8). Regression coefficients were assigned Normal (0, 0.5) priors, thresholds Normal (0, 2) priors, and all other parameters received default priors. Four Markov chain Monte Carlo (MCMC) chains were run with 2000 warm-up and 10,000 retained iterations per chain, yielding 40,000 posterior samples.
In the Bayesian model, HPLP-II was modeled as a latent variable measured by its six dimensions, whereas SCL-90 was entered as a binary observed outcome.
Three sensitivity analyses were conducted within the primary frequentist bootstrap SEM framework. First, because three SCL-90 items in the “other” category (items 44, 64, and 66) are sleep-related and may overlap in content with the PSQI, potentially inflating the PSQI-SCL-90 association, SCL-90 scores were recalculated after excluding these items. Second, the analysis was repeated using only these nine dimensions.
Third, because the cross-sectional design cannot empirically establish the temporal ordering of HPLP-II and PSQI, and the HPLP-II → PSQI sequence was theory-driven, an alternative model with PSQI preceding HPLP-II was estimated. All three sensitivity analyses were restricted to the primary frequentist bootstrap SEM framework.
Results
Sample Characteristics
A total of 2272 first-year students were enrolled in this study, including 1422 females (62.59%) and 850 males (37.41%). The demographic baseline characteristics stratified by the SCL-90 score are presented in Table 1, including 1915 (84.29%) participants in the SCL-90 screen-negative group and 357 (15.71%) in the SCL-90 screen-positive group. Drinking was prevalent (n=1005, 44.23%), with a significantly higher proportion in the SCL-90 screen-positive group (n=190, 53.22%) compared to the SCL-90 screen-negative group (n=815, 42.56%) (p < 0.001).
The median HPLP-II score in the SCL-90 screen-positive group (91.0, IQR: 82.0, 100.0) was significantly lower than that in the SCL-90 screen-negative group (109.0, IQR: 94.0, 128.0) (p < 0.001). The median PSQI score in the SCL-90 screen-positive group (7.0, IQR: 5.0, 9.0) was significantly higher than that in the SCL-90 screen-negative group (3.0, IQR: 2.0, 5.0) (p < 0.001). The median SCL-90 score in the SCL-90 screen-positive group (173.0, IQR: 158.0, 195.0) was significantly higher than that in the SCL-90 screen-negative group (103.0, IQR: 93.0, 118.0) (p < 0.001).Table 1Participants’ Demographic InformationDemographicsTotal (N = 2272)SCL-90 Screen-Negative (N = 1915)SCL-90 Screen-Positive (N = 357)p valueAge18.0 (18.0, 19.0)18.0 (18.0, 19.0)18.0 (18.0, 18.0)<0.001Gender0.280 Female1422 (62.59%)1189 (62.09%)233 (65.27%) Male850 (37.41%)726 (37.91%)124 (34.73%)Nation0.882 Han2095 (92.21%)1767 (92.27%)328 (91.88%) Other177 (7.79%)148 (7.73%)29 (8.12%)Household registration0.011 Rural623 (27.42%)505 (26.37%)118 (33.05%) Urban1649 (72.58%)1410 (73.63%)239 (66.95%)Student source location0.012 Shanghai1255 (55.24%)1080 (56.40%)175 (49.02%) Non-Shanghai1017 (44.76%)835 (43.60%)182 (50.98%)Monthly household income0.270 <150051 (2.24%)40 (2.09%)11 (3.08%) 1500–3000139 (6.12%)115 (6.01%)24 (6.72%) 3001–5000309 (13.60%)255 (13.32%)54 (15.13%) 5001–8000492 (21.65%)407 (21.25%)85 (23.81%) >80001281 (56.38%)1098 (57.34%)183 (51.26%)Father’s educational background0.881 Primary or below143 (6.29%)123 (6.42%)20 (5.60%) Junior high461 (20.29%)395 (20.63%)66 (18.49%) High school or secondary578 (25.44%)480 (25.07%)98 (27.45%) College482 (21.21%)404 (21.10%)78 (21.85%) Bachelor505 (22.23%)426 (22.25%)79 (22.13%) Master or above103 (4.53%)87 (4.54%)16 (4.48%)Mother’s educational background0.425 Primary or below213 (9.38%)185 (9.66%)28 (7.84%) Junior high481 (21.17%)409 (21.36%)72 (20.17%) High school or secondary521 (22.93%)426 (22.25%)95 (26.61%) College486 (21.39%)407 (21.25%)79 (22.13%) Bachelor508 (22.36%)436 (22.77%)72 (20.17%) Master or above63 (2.77%)52 (2.72%)11 (3.08%)Parental marital status0.005 Married1930 (84.95%)1646 (85.95%)284 (79.55%) Separated37 (1.63%)31 (1.62%)6 (1.68%) Divorced193 (8.49%)148 (7.73%)45 (12.61%) Remarried68 (2.99%)51 (2.66%)17 (4.76%) Widowed44 (1.94%)39 (2.04%)5 (1.40%)BMI21.2 (19.1, 24.0)21.2 (19.1, 23.9)21.3 (19.1, 24.7)0.497Smoking0.738 Yes265 (11.66%)221 (11.54%)44 (12.32%) No2007 (88.34%)1694 (88.46%)313 (87.68%)Drinking<0.001 Yes1005 (44.23%)815 (42.56%)190 (53.22%) No1267 (55.77%)1100 (57.44%)167 (46.78%)HPLP-II105.0 (91.0, 124.0)109.0 (94.0, 128.0)91.0 (82.0, 100.0)<0.001PSQI4.0 (2.0, 6.0)3.0 (2.0, 5.0)7.0 (5.0, 9.0)<0.001SCL-90107.0 (95.0, 131.0)103.0 (93.0, 118.0)173.0 (158.0, 195.0)<0.001Abbreviations: BMI, Body Mass Index; HPLP-II, Health-Promoting Lifestyle Profile-II; PSQI, Pittsburgh Sleep Quality Index; SCL-90, Symptom Checklist-90.
Correlation Analysis of Variables
As presented in Figure 1, drinking status, HPLP-II, PSQI, and SCL-90 exhibited significant correlations with one another (all p values < 0.001). Specifically, drinking status was significantly associated with a lower HPLP-II score (r = −0.10), higher PSQI score (r = 0.12), and higher SCL-90 score (r = 0.11) (all p values <0.001). The SCL-90 score demonstrated a positive correlation with the PSQI score (r = 0.60, p < 0.001) and a negative correlation with the HPLP-II score (r = −0.52, p < 0.001).
The HPLP-II score showed a negative correlation with the PSQI score (r = −0.45, p < 0.001). Figure 1Spearman correlation coefficient heatmap.A Spearman correlation coefficient heatmap with identical row and column labels: Age, Gender, Nation, Household Registration, Student Source Location, Monthly Household Income, Fathers Educational Background, Mothers Educational Background, Parental Marital Status, Body Mass Index, Smoking, Drinking, Health Promoting Lifestyle Profile Two, Pittsburgh Sleep Quality Index, Symptom Checklist 90. Cells show correlation coefficients and some cells include one to three asterisks.
A vertical legend at right is labeled from minus one at the bottom through zero to one at the top. Diagonal cells read one point zero zero.A correlation heatmap for Age to Symptom Checklist 90, with matching row and column labels and cell numbers.Notes: * p < 0.05, ** p < 0.01, *** p < 0.001.Abbreviations: BMI, Body Mass Index; HPLP-II, Health-Promoting Lifestyle Profile-II; PSQI, Pittsburgh Sleep Quality Index; SCL-90, Symptom Checklist-90.
Frequentist Structural Equation Modeling with Continuous Variables
A frequentist SEM was built, with the continuous latent variable of SCL-90 as the dependent variable. The model exhibited moderate fit (χ2 = 3048.099, df = 327, χ2/df = 9.321, p < 0.001, CFI = 0.923, TLI = 0.904, RMSEA = 0.061, SRMR = 0.025). A post hoc power analysis for SEM was performed.
With N = 2272, df = 327, α = 0.05, and the observed RMSEA = 0.061, the sample yielded power > 0.99. In the measurement model, all standardized factor loadings were significant (p < 0.001), ranging from 0.67 to 0.84 for HPLP-II and 0.78 to 0.92 for SCL-90. Composite reliability (CR) was 0.901 for HPLP-II and 0.963 for SCL-90, and average variance extracted (AVE) was 0.608 and 0.728, respectively, supporting reliability and convergent validity.
Full loadings are provided in Table S2. The SEM showed that drinking status was directly associated with HPLP-II (β = −0.115, 95% CI [−0.158, −0.072], p < 0.001) and PSQI (β = 0.054, 95% CI [0.014, 0.095], p = 0.009). HPLP-II was negatively associated with SCL-90 (β = −0.186, 95% CI [−0.248, −0.110], p < 0.001) and PSQI (β = −0.446, 95% CI [−0.483, −0.404], p < 0.001), while PSQI was positively associated with SCL-90 (β = 0.457, 95% CI [0.409, 0.510], p < 0.001).
More detailed information is presented in Figure 2. The bootstrap method was adopted to examine the mediation associations, with the results presented in Table 2. When the score of HPLP-II served as a single mediator, the indirect effect value was 0.021 (95% CI [0.011, 0.034], p < 0.001), indicating a significant mediating role of HPLP-II (drinking status → HPLP-II → SCL-90).
When the score of PSQI functioned as a single mediator, the indirect effect value was 0.025 (95% CI [0.007, 0.044], p = 0.010), indicating a significant mediating role of PSQI (drinking status → PSQI → SCL-90). When the HPLP-II and PSQI formed a chain mediation, the indirect effect value was 0.023 (95% CI [0.015, 0.033], p < 0.001), confirming the existence of a significant chain mediation association (drinking status → HPLP-II → PSQI → SCL-90). The total indirect effect across all mediating pathways was 0.070 (95% CI [0.045, 0.095], p < 0.001), while the total effect of drinking status on SCL-90 was 0.087 (95% CI [0.046, 0.129], p < 0.001).
The direct effect was not significant (β = 0.018, 95% CI [−0.017, 0.054], p = 0.318), suggesting that the association between drinking status and SCL-90 was largely accounted for by the indirect pathways.Table 2Direct and Indirect Effects of the Frequentist Structural Equation ModelParameterEstimateStd.allSEZ valuep valueCI.lowerCI.upperDirect effectsDrinking status → SCL-900.0280.0180.0180.9990.318−0.0170.054Indirect effectsDrinking status → HPLP-II → PSQI → SCL-900.0370.0230.0055.030<0.0010.0150.033Drinking status → PSQI → SCL-900.0390.0250.0102.5910.0100.0070.044Drinking status → HPLP-II → SCL-900.0340.0210.0063.588<0.0010.0110.034Total indirect effects0.1100.0700.0135.525<0.0010.0450.095Total effects0.1380.0870.0214.151<0.0010.0460.129Note: SE, Z, p, and the 95% CI all correspond to the standardized estimates.Abbreviations: HPLP-II, Health-Promoting Lifestyle Profile-II; PSQI, Pittsburgh Sleep Quality Index; SCL-90, Symptom Checklist-90; Std.all, standardized estimate; SE, standard error; CI.lower, lower bound of 95% confidence interval; CI.upper, upper bound of 95% confidence interval.Figure 2The frequentist structural equation model of the study.A path diagram with four main rectangles arranged left to right: Drinking at lower left, HPLP dash II at left center, PSQI at center and SCL dash 90 at right center. From Drinking, an arrow goes up right to HPLP dash II labeled, beta equals minus 0.115, 95 percent C I left parenthesis minus 0.158 comma minus 0.072 right parenthesis, S E equals 0.022, P less than 0.001. From Drinking, an arrow goes up right to PSQI labeled, beta equals 0.054, 95 percent C I left parenthesis 0.014 comma 0.095 right parenthesis, S E equals 0.021, P equals 0.009.
From Drinking, a horizontal arrow goes right to SCL dash 90 labeled, beta equals 0.018, 95 percent C I left parenthesis minus 0.017 comma 0.054 right parenthesis, S E equals 0.018, P equals 0.318. From HPLP dash II, an arrow goes right to PSQI labeled, beta equals minus 0.446, 95 percent C I left parenthesis minus 0.483 comma minus 0.404 right parenthesis, S E equals 0.020, P less than 0.001.A path diagram of a structural equation model linking Drinking, HPLP dash II, PSQI and SCL dash 90.Abbreviations: HPLP-II, Health-Promoting Lifestyle Profile-II; PSQI, Pittsburgh Sleep Quality Index; SCL-90, Symptom Checklist-90.
To further investigate how drinking is associated with specific psychological symptoms, we conducted separate SEM analyses using each of the nine symptom dimensions plus the additional factor as exploratory analyses. The chain mediation associations of HPLP-II and PSQI were directionally consistent across the nine symptom dimensions plus the additional factor (all p values < 0.001). To account for multiple testing across these dimension-specific analyses, false discovery rate (FDR) correction was applied (Table S3).
The results remained consistent after FDR correction. Notably, when the outcome variable was the dimension of somatization, the direct effect of drinking status on this dimension was significant (β = 0.058, 95% CI [0.021, 0.095], p = 0.002), whereas the direct effects for the other nine dimensions were not significant. Detailed model parameters and path diagrams are provided in the Supplementary (Figures S1–S10 and Table S3).
Bayesian Structural Equation Modeling with a Binary Variable
A Bayesian binary SEM was constructed, with the binary indicator of SCL-90 as the dependent variable. As shown in Figure 3, the posterior mean for the path from drinking status to HPLP-II was −0.115, with a 95% credible interval (CI) of (−0.157, −0.070), indicating that drinking status was negatively associated with health-promoting lifestyle and positively associated with PSQI (mean = 0.054, 95% CI [0.015, 0.093]). HPLP-II was negatively associated with PSQI (mean = −0.444, 95% CI [−0.481, −0.406]), and HPLP-II and PSQI were negatively (mean = −0.332, 95% CI [−0.398, −0.259]) and positively (mean = 0.397, 95% CI [0.343, 0.448]) associated with SCL-90, respectively.
As shown in Table 3, the indirect effect values of the three mediating pathways (drinking → HPLP-II → SCL-90, drinking → PSQI → SCL-90, drinking → HPLP-II → PSQI → SCL-90) were 0.038 (95% CI [0.022, 0.055]), 0.021 (95% CI [0.006, 0.038]), and 0.020 (95% CI [0.012, 0.029]), respectively. The total indirect effect was 0.079 (95% CI [0.053, 0.106]), and the total effect was 0.128 (95% CI [0.073, 0.214]). All R-hat values were < 1.001, and all ESS values were > 4000, indicating good convergence.Table 3Direct and Indirect Effects of the Bayesian Binary Structural Equation ModelParameterStandardized EstimateCI.lowerCI.upperDirect effectsDrinking status → SCL-900.049−0.0080.104Indirect effectsDrinking status → HPLP-II → PSQI → SCL-900.0200.0120.029Drinking status → PSQI → SCL-900.0210.0060.038Drinking status → HPLP-II → SCL-900.0380.0220.055Total indirect effects0.0790.0530.106Total effects0.1280.0730.214Abbreviations: HPLP-II, Health-Promoting Lifestyle Profile-II; PSQI, Pittsburgh Sleep Quality Index; SCL-90, Symptom Checklist-90; CI.lower, Lower bound of the 95% credible interval; CI.upper, Upper bound of the 95% credible interval.Figure 3The Bayesian binary structural equation model of the study.Six labeled rectangles along the top connect by separate diagonal lines down to a rectangle near the upper middle labeled, HPLP dash 2.
From left to right the top rectangles read, interpersonal relations, stress management, health responsibility, nutrition, physical activity and spiritual growth, with lambda labels on the connecting lines reading, lambda equals 0.675, lambda equals 0.796, lambda equals 0.825, lambda equals 0.769, lambda equals 0.834 and lambda equals 0.774. To the right of HPLP dash 2 is a rectangle labeled, PSQI, connected by a right pointing horizontal arrow labeled, beta equals minus 0.444. At the bottom left, a rectangle labeled, Drinking, sends an up right arrow to HPLP dash 2 labeled, beta equals minus 0.115 and a long diagonal arrow to PSQI labeled, beta equals 0.054.
At the bottom right, a rectangle labeled, SCL dash 90, receives a diagonal arrow from HPLP dash 2 labeled, beta equals minus 0.332 and a diagonal arrow from PSQI labeled, beta equals 0.397.A schematic of a Bayesian binary structural equation model linking Drinking, HPLP dash 2, PSQI and SCL dash 90.Abbreviations: HPLP-II, Health-Promoting Lifestyle Profile-II; PSQI, Pittsburgh Sleep Quality Index; SCL-90, Symptom Checklist-90.
Sensitivity Analysis
The results of the sensitivity analyses were consistent with the main findings. When the three sleep-related SCL-90 items or the items included in the “other” dimension were excluded from the composite scores, the estimated associations remained consistent with the primary analyses (Tables S4 and S5). In the alternative model in which PSQI preceded HPLP-II, the pattern of associations was likewise consistent with the main findings (Table S6).
Discussion
In this study, a cross-sectional survey was conducted among 2272 freshmen from a medical college in Shanghai, aiming to explore the association between drinking status and mental health in college students and analyze the mediating roles and chain mediation pathway of HPL and sleep quality in this association. The frequentist SEM indicated that HPL and sleep quality were associated with mental health through the proposed mediating pathways, and directionally consistent associations were observed in the complementary Bayesian analysis. Furthermore, these factors formed a significant chain mediation pathway, through which drinking status was associated with mental health symptoms.
The associations between drinking status, HPL, sleep quality and mental health have been substantiated in past studies. However, existing research has mostly concentrated on direct associations among variables or single mediating pathways. Explorations into the specific complex pathways through which alcohol consumption is associated with mental health remain scarce, particularly whether HPL and sleep quality act synergistically in this process.
To our knowledge, this is the first study to systematically examine the chain mediating association of HPL and sleep quality between drinking status and mental health symptoms, thus providing a basis for understanding the multi-pathway associations linking alcohol consumption to college students’ mental health.
Our findings were consistent with the hypothesized mediating roles of HPL and sleep quality and suggested a significant chain mediation association. This result is consistent with previous research. Drinking often coexists with unhealthy behaviors, such as poor diet,44 and physical inactivity,45 which are collectively associated with depressive symptoms.46 Conversely, interventions targeting these behaviors, including physical activity, have demonstrated potential for improving mental well-being.47,48 Sleep quality functioned as another critical mediator.
Research has confirmed that even low-dose alcohol can significantly disrupt sleep architecture,49,50 reducing REM sleep which is crucial for learning,51 emotional memory processing and regulation.52 Furthermore, sleep and mental health share common neurobiological53 and genetic substrates.54,55 Sleep deprivation disrupts key neurotransmitter systems, providing a classic pathophysiological basis for mental disorders.51
This study identified a significant chain mediating association of HPL and sleep quality between drinking status and mental health symptoms. Similarly, current research has indicated that unhealthy lifestyle factors are significantly associated with poor sleep quality.29,56 Several biological pathways have been proposed to explain this association. For example, unhealthy lifestyles may contribute to oxidative stress and inflammatory responses, which have been linked to impaired sleep quality.30,57 Additionally, lifestyle factors such as irregular schedules and nighttime blue light exposure disrupt circadian rhythms.32,58 Lifestyle can also alter the gut microbiota composition and function, influencing the production of sleep-related metabolites.59–61 Conversely, interventions targeting lifestyle factors such as physical activity and diet have shown significant efficacy in improving sleep quality.31,62 Thus, the intrinsic connection between lifestyle and sleep quality provides a theoretical basis for the proposed chain mediating pathway.
Notably, the chain mediation associations were directionally consistent across all dimensions of the SCL-90. An exploratory direct association was also observed between drinking status and somatization, beyond the indirect pathways involving HPL and sleep quality. This suggests that there may be a residual direct association between alcohol use and the physical manifestations of mental health symptoms, possibly involving biological pathways such as oxidative stress, which has been linked to neuronal damage,63 systemic inflammatory responses,64 or alcohol-induced hyperalgesia.65 Therefore, for students presenting with predominant somatic symptoms, intervention strategies may consider alcohol reduction as one component.
This study has several strengths. In terms of research perspective, it extends beyond single mediation studies through a shift from examining isolated factors to examining the chain mediating pathway. In data analysis, potential confounding factors such as age, gender, family socioeconomic status, BMI, and smoking status were fully considered and adjusted for in the model.
In addition, both frequentist and Bayesian SEM were used. The SCL-90 outcome was modeled as continuous in the frequentist model and binary in the Bayesian model. The two approaches provide complementary evidence.
However, this study also has limitations. First, as a cross-sectional study, this research can only show associations between variables rather than establish causality. The temporal ordering assumed in the serial mediation model was theory-driven rather than empirically demonstrated.
Alternative orderings among lifestyle, sleep quality, and psychological symptoms are also plausible and cannot be ruled out. Reverse causation also cannot be excluded. In addition, all variables were self-reported within a single online survey, so the observed associations may be affected by common-method bias.
Although we adjusted for a range of sociodemographic and behavioral covariates, unmeasured or residual confounding cannot be ruled out. Second, participants were divided into drinkers and non-drinkers based on their drinking status. The binary classification combined former, reduced-intake, occasional, and weekly drinkers into a single group and did not quantify alcohol volume, type, frequency, binge drinking, or alcohol-related problems.
This may introduce exposure misclassification and limit the assessment of dose-response relationships. In particular, former drinkers may differ from current drinkers in health status and reasons for cessation, which could further bias the estimated association between drinking and mental health symptoms. Third, the study sample was only recruited from first-year students at a single medical college in Shanghai, resulting in high homogeneity of the sample, which limits the generalizability of the findings.
Future studies should include more diverse and representative populations to validate the universality. Fourth, the effect sizes of drinking status on the mediators and outcomes were relatively modest. With a sample of 2272 participants, even relatively small effects can readily achieve statistical significance.
At the same time, the binary drinking classification may have further attenuated these associations. The effect sizes should therefore be interpreted cautiously. Future longitudinal studies adopting validated alcohol-use measures are required to examine whether these relatively weak associations can translate into clinically meaningful intervention effects.
Conclusion
In conclusion, this cross-sectional study found that health-promoting lifestyle and sleep quality may be involved in the association between drinking status and mental health symptoms among first-year medical students, and that these two factors formed a significant chain mediating association. These findings suggest that health-promoting lifestyle and sleep quality may be potential targets for mental health promotion in colleges, including strengthening publicity and education on the hazards of alcohol, guiding the adoption of healthy lifestyles, and integrating sleep health into the mental health work system. Longitudinal or intervention studies are needed to confirm the direction and practical significance of the proposed pathway.