Previous studies have demonstrated the benefits of PA across various age groups and populations. Among preschool children, even a single bout of moderate-to-vigorous PA has been shown to improve sleep quality, enhance cognitive function, and regulate insulin sensitivity, and regular PA is associated with more pronounced improvements in these domains (33). Moreover, individuals across diverse backgrounds, including pregnant and postpartum women, people of all genders and ages, individuals with disabilities, and those seeking to reduce their risk of disease, can benefit from PA interventions (34). A substantial body of epidemiological research has demonstrated that PA can significantly enhance self-esteem and subjective well-being. Adults who engage in regular PA tend to report fewer symptoms of depression and anxiety. These findings support the view that exercise may play a preventive role in the development of mental disorders (35). The theoretical research of PA suggests that it can reduce body fat, which subsequently enhances cardiopulmonary function, improves muscular strength, and promotes skeletal health, ultimately contributing to better mental health (36). We have discussed the psychological definitions of SR and SUR, both of which can lead to serious consequences. However, accumulating evidence suggests that these risks may be mitigated through PA. Based on these findings, we hypothesize that there may be a significant relationship between PA, SR, and SUR among adolescents. Accordingly, we propose the following research question:

The complex mediating roles of sleep duration (SD), healthy diet (HD), and mental health (MH) in the associations between PA, SR, and SUR among adolescents warrant further examination. Although direct significant relationships have been identified among SD, HD, MH, SR and SUR, research specifically focusing on adolescents remains limited. For example, a meta-analysis has identified that insomnia can prospectively predict suicidal thoughts and behaviors (37). Sleep disturbances can also be an important risk factor for SR and suicide attempts among otherwise healthy adolescents (38). A healthy diet, such as the Mediterranean diet, may prospectively reduce depressive symptoms, and depression is one of the key antecedent factors associated with increased suicide risk (39). Moreover, individuals with depression are estimated to have a 20-fold increased risk of suicide compared to the general population. Depression is currently recognized as one of the most well-established and stable predictors of suicide (40). Depression and SUR are highly comorbid and exert bidirectional influences. On one hand, depression may lead to substance misuse through a self-medication mechanism; on the other hand, SUR can exacerbate depressive symptoms and increase the risk of suicide (41). Individuals with SUR exhibit significantly elevated suicide risk, with ORs typically ranging from 2.5 to 7.5. The risk is particularly pronounced among individuals with polysubstance use, as well as those using alcohol and opioids (42). Given the established associations among PA, SR, and SUR, it is worth exploring whether SD, HD, and MH may serve as potential mediators in these relationships. Accordingly, we propose the following research question:

Chain mediation (also called serial mediation) models posit that an independent variable influences the outcome through a sequence of mediators, implying a causal chain among the mediators themselves (43). This model allows researchers to test more complex mechanisms of indirect effects beyond simple mediation (44). In this study, we aim to elucidate the complex mediating mechanisms underlying the relationship among the key variables in the model. In addition, this study seeks to further identify the sequential mediation pathways that explain how these variables are interrelated. We propose the following research question, and present a hypothesized model (Figure 1).

The data for this study were derived from the Youth Risk Behavior Surveillance System (YRBSS), a national survey conducted in the United States. The YRBSS was established in 1991 by the Centers for Disease Control and Prevention (CDC) to investigate critical health-risk behaviors among high school students. For the present analysis, we utilized the most recent cross-sectional dataset, collected in 2023.

Institutional review boards at CDC and ICF, the survey contractor, approved the study protocol for YRBSS. All procedures were conducted in accordance with the relevant ethical guidelines and federal regulations. Data collection was carried out in compliance with applicable U.S. federal law and CDC policy. All participants provided voluntary, anonymous responses, and parental consent procedures were followed prior to data collection. The publicly available dataset can be freely accessed from the official CDC website.

Due to the questionnaire targeting to collect disparate information from adolescents, it is not suitable to construct potential variable. However, a recent test–retest study of most of the 2023 survey questions indicated substantial reliability among the questions (45). Although the official guidelines indicate that the high dataset has high reliability, we conducted additional reliability tests across multiple dimensions. The results demonstrated acceptable reliability, as reported in the following sections.

The 2023 TRBSS dataset included 254,675 respondents. Figure 2 illustrates the sample filtering process. Participants with missing values in key demographic covariates were excluded, namely gender (n = 863), age (n = 907), grade (n = 1,003), and race (n = 3,200), resulting in a remaining sample of 248,702. Subsequently, participants with missing values on core study variables were also excluded. These variables included PA, HD, SUR, SR, SD, and MH. After excluding participants with missing data on any of variables, the final analytic sample comprised 15,734 participants.

Data processing and statistical analyses were conducted using R 4. 5. For mediation analysis, the bootstrap method was adopted based on Hayes’ recommendation. Compared to Sobel test, the bootstrap method estimates the sampling distribution directly without assuming normality, providing greater statistical power and a lower Type I error rate (46). Arguably, we employed the bootstrap method with 5,000 resamples to estimate the 95% confidence intervals of the mediation effects. All mediation analyses and model fit test were also conducted using R 4.5.

Model fit was considered acceptable when all of the following criteria were met: a non-significant chi-square (χ²) test result (47). A comparative Fit Index (CFI) and Tucker-Lewis Index (TLI) greater than 0.95. And a Root Mean Square Error of Approximation (RMSEA) being less than 0.08 (48).

Table 2 presents the covariate variables included in this study. Among the final sample, the gender distribution was nearly balanced, with 7,672 females and 8,062 males. Age was categorized into seven groups, ranging from 12 or younger to 18 or older. The distribution approximated normality, with the majority of participants aged 15 (24.9%), 16 (26.0%), and 17 (25.2), which aligns with the typical adolescent age range.

Participants were primarily in grades 9 through 12, with a relatively even distribution across grade levels and a mean grade level of 2.49. Regarding race, White participants comprised the largest group (48.7%), followed by Hispanic/Latino, Black or African American, and other racial categories. We reported frequencies, percentages, and means with standard deviations to provide a comprehensive overview of sample characteristics. All variables included in the correlation matrix were either continuous or ordinal and were treated as approximately continuous.

Table 3 presents the means, standard deviations, and intercorrelations among the key study variables, including gender, age, grade level, PA, HD, SUR, SR, SD and MH. PA was significantly correlated with HD (r = 0.26, p < 0.001), SUR (r = 0.02, p < 0.05), SR (r = −0.10, p < 0.001), SD (r = 0.16, p < 0.001) and MH (r = 0.03, p < 0.001). These results indicate that higher levels of PA are associated with higher HD, longer SD, better MH and lower SR. Although a weak positive correlation was observed between PA and SUR, this unexpected association warrants further analysis to clarify the complex relationships among these variables. Race was not included in Table 3, as it is a nominal variable lacking a meaningful numeric scale for correlation analysis. These findings support the hypothesized associations and provide empirical justification for conducting subsequent mediation analysis. While several covariates (e.g., gender, age, grade) showed modest correlations with the main variables, these effects were small and are not considered potential confounders.

The YRBSS is designed as a surveillance tool that provides observed behavioral indicators rather than latent constructs, and it does not require factor validation such as AVE or CR (49). However, in order to increase the measuring accuracy, we still tried to construct a latent variable (e.g., mental health, substance use risk) in line with prior SEM research and to enhance construct validity.

The measurement model demonstrated acceptable reliability and convergent validity. Standardized factor loadings ranged from 0.34 to 0.84, with most values above the recommended 0.50 threshold (50). Composite reliability (CR) ranged from 0.675 to 0.795, and AVE ranged from 0.401 to 0.568. Although AVE values for PA and Healthy Diet were slightly below the 0.50 criterion, their CR values remained above 0.60, supporting convergent validity (51). Overall, the results confirmed that the latent constructs were measured reliably (Table 4).

We used maximum likelihood estimation with robust standard errors (MLR) to handle non-normality. All items were retained without parceling to avoid information loss. Model modifications were not applied to maintain theoretical purity. The model demonstrated the overall model acceptable fit: χ² (93) = 1206.60, p < 0.001, RMSEA = 0.0028, 95% CI [0.026, 0.031], CFI = 0.981, TLI = 0.975, SRMR = 0.023. Although the chi-square statistic was significant, which is common in models with large sample sizes and complex structure, alternative fit indices fell within acceptable ranges. The alternative fit indices fell within conventionally accepted thresholds, indicating satisfactory model fit and construct validity.

Table 5 presents the standardized path coefficients, standard errors, 95% confidence intervals, and significant levels for the direct effects among the key study variables. PA demonstrated significant direct associations with SR, MH, SD, SUR, HD, with all p-values less than 0.001. Specifically, PA significantly predicted better MH (β = −0.241, 95% CI [0.036, 0.077]), longer SD (β = 0.177, 95% CI [0.159, 0.194]) and higher HD (β = 0.395, 95% CI [0.375, 0.415]).

We observed that two the standardized direct paths showed counterintuitive directions: PA significantly predicted SR (β = 0.047, 95% CI [0.036, 0.077]) and SUR (β = 0.085, 95% CI [0.064, 0.106]). Although the direct effect of PA on SR was small and positive, the strong indirect pathway through improved mental health (β = −0.241) and its substantial predictive power on suicide risk (β = 0.739) suggest that the total effect is largely driven by the indirect mechanisms. The pattern is commonly observed in chain mediation models, particularly when suppressor effects or competing pathways exist. A similar explanation may apply to the relationship between PA and SR.

MH was positively associated with both SR (β = 0.395, 95% CI [0.724, 0.754], p < 0.001) and SUR (β = 0.348, 95% CI [0.326, 0.371], p < 0.001). In addition, SD significantly predicted both better MH (β = −0.303, 95% CI [−0.321, −0.286], p < 0.001) and lower SUR (β = −0.064, 95% CI [−0.084, −0.045], p < 0.001). HD was positively associated with MH (β = 0.059, 95% CI [0.033, 0.084], p < 0.001). SUR was a significant positive predictor to SR (β = 0.114, 95% CI [0.091, 0.137], p < 0.001). However, the direct path from HD to SR was not statistically significant (β = 0.016, 95% CI [−0.005, 0.037], p = > 0.05). Nevertheless, this path was retained in the mediation analysis, as on indirect effects can still be significant even when the constituent path is not significant (44).

These results suggest that the effect of PA on SR and SUR may operate primarily through indirect pathways (e.g., via MH, SD, HD), providing initial support for the hypothesized mediation structure to be tested in the subsequent analysis.

Table 6 presents the standardized direct, indirect, and total effects from PA to SUR. We observed that the total effect (β = −0.021, 95% CI [−0.037, −0.006], p < 0.001) shown that higher levels of PA can significantly predict lower SUR, supporting H2a.

The mediation analysis revealed several significant pathways. RQ1a (PA → MH → SUR) was supported, indicating that MH significantly mediated the effect of PA on SUR (β = −0.062, 95% CI [−0.070, −0.055], p < 0.001). More importantly, the pathway (PA → SD → SUR) was also supported, demonstrating that SD significantly mediated the relationship between PA and SUR (β = −0.009, 95% CI [−0.012, −0.006], p < 0.001). In addition, a significant chain mediation pathway (PA → SD → MH → SUR) was observed (β = −0.014, 95% CI [−0.016, −0.012], p < 0.001), highlighting a complex multi-step mechanism through which PA influences SUR. These findings suggest that lower levels of PA may initially lead to deteriorated MH or reduced SD, which in turn contribute to increased SUR. The Chain mediation model illustrates that insufficient PA may negatively impact SD, which subsequently impairs MH, ultimately leading to elevated levels of SUR. Together, these results emphasize that both MH and SD are the critical mediating mechanisms linking PA to SUR among adolescents.

Table 7 presents the standardized direct, indirect, and total effects from PA to SR. The total effect (β = −0.024, 95% CI [−0.028, −0.021], p < 0.001) demonstrates that higher levels of PA significantly predict lower levels of SR, thereby supporting H2b.

The mediation and chain mediation analyses revealed several key pathways linking PA to SR. RQ1b was supported, indicating that MH significantly mediated the effect of PA on SR (β = −0.03, 95% CI [−0.033, −0.027], p < 0.001). This suggesting that higher levels of PA may enhance MH, thereby reducing SR. The pathway from PA to SR visa HD was not supported (β = 0.001, 95% CI [0.000, 0.003], p < 0.001), indicating that HD did not serve as a significant mediator. However, the indirect effect via SUR was significant (β = 0.002, 95% CI [0.001, 0.002], p < 0.001), suggesting that PA may reduce SUR, which in turn decreases SR. A significant chain mediation pathway (PA → MH → SUR → SR) was also identified (β = −0.002, 95% CI [−0.002, −0.001], p < 0.001), indicating that higher levels of PA can improve MH, which subsequently lowers SUR, ultimately leading to minimize SR. Another chain pathway (PA → HD → MH → SR) was supported (β = 0.003, 95% CI [0.002, 0.004], p < 0.001), highlighting the combined mediating role of HD and MH. Similarly, the sequential pathway (PA → SD → MH → SR) was also significant (β = −0.007, 95% CI [−0.007, −0.006], p < 0.001), suggesting that PA promotes SD, which enhances MH and consequently reduces SR. In contrast, the pathway (PA → SD → SUR → SR) although statistically significant (p < 0.001), yielded a near-zero effect size (β ≈ 0.000 95% CI [0.000, 0.000], p < 0.001), indicating a lack of practical significance (Figure 3).

We examined the direct and indirect relationships between PA, SR, SUR among adolescents, focusing specifically on the mediating roles of SD, HD, and MH. The results generally supported the proposed mediation and chain mediation hypotheses, though with some notable exceptions.

Consistent with hypotheses, higher levels of PA significantly predicted lower overall levels of both SUR and SR. However, direct pathways from PA to SR and SUR were unexpectedly positive, suggesting the presence of suppressor effects and highlighting the critical role of indirect pathways. Indeed, MH emerged as a central mediator, significantly mediating the relationships between PA and both SUR and SR. Additionally, SD was identified as a significant mediator linking PA to SUR, indicating that adequate sleep contributes to reduced substance use risk.

Chain mediation analyses further underscored the complexity of these relationships. PA influenced SUR through a sequential mediation pathway involving SD and MH, highlighting a stepwise protective mechanism (PA → SD → MH → SUR). Similarly, PA influenced SR through multiple significant sequential mediation pathways, including PA → MH → SUR → SR, and PA → SD → MH → SR, reinforcing the central role of MH as a pivotal mediating factor. The sequential mediation pathway via HD and MH (PA → HD → MH → SR) was statistically significant but presented a minor positive effect. Conversely, the sequential mediation via SD and SUR (PA → SD → SUR → SR) were not supported.

These findings illustrate that the beneficial impacts of PA on SR and SUR among adolescents predominantly operate indirectly through improved MH, increased SD, and partially via HD practices. These mechanisms suggest targeted avenues for intervention strategies designed to enhance adolescent risk.

Hypothesis H2a was supported, as higher levels of PA significantly predicted lower SUR (β = −0.021, p < 0.001). This finding aligns with previous studies suggesting physical activity’s protective effects against substance use, possibly through behavioral regulation and healthy lifestyle promotion. For instance, a longitudinal study of adolescents demonstrated that higher engagement in regular PA was associated with reduced likelihood of substance abuse behaviors, supporting our observed protective relationship (52).

Hypothesis H2b was also confirmed; PA significantly predicted reduced SR (β = −0.024, p < 0.001). Although the direct pathway between PA and SR was unexpectedly small and positive (β = 0.047, p < 0.001), the robust negative indirect effects primarily via MH indicate a suppression effect. Previous research supports this indirect protective pathway, emphasizing that PA reduces depression and anxiety, thereby lowering SR (53).

Regarding mediation hypotheses, H1a was supported as SD significantly mediated the relationship between PA and SUR (β = −0.009, p < 0.001). Adequate sleep has been consistently documented as an essential protective factor against adolescent substance use, likely by promoting emotional regulation and reducing impulsivity (54).

Hypothesis H1b was strongly supported, highlighting MH as a significant mediator in the relationships among PA, SUR, and SR (β ranging from −0.03 to −0.062, p < 0.001). Consistent with earlier findings, our results underline MH as a central explanatory factor linking physical activity to adolescent mental health outcomes, including SUR (55) and SR (56).

Hypothesis H1c regarding HD mediating effect of PA on SR was not supported (β = 0.001, p < 0.001). Although HD was positively associated with improved MH, the direct mediating role of HD alone was negligible. This outcome aligns with studies suggesting dietary effects on mental health are subtle and often operate indirectly via complex biopsychosocial pathways rather than simple, direct mediation (57).

Chain mediation hypotheses (RQ1a) received support. Revealing a significant sequential pathway from PA through SD and MH to SUR (β = −0.014, p < 0.001), suggesting a cumulative protective process through improved sleep and subsequent mental health enhancement. This finding is congruent with existing literature emphasizing multi-step health behavior chains among adolescents (58).

RQ1b was also confirmed, demonstrating significant chain mediation through HD and MH to SR (β = −0.007, p < 0.001). This result emphasizes the complex interplay among behavioral health factors in reducing suicide risk, consistent with the previous theoretical model proposed (59).

RQ1c, involving chain mediation via MH and SUR (β = −0.002, p < 0.001), was statistically significant, suggesting that complex interplay among mental health factors in reducing suicide risk. This finding is congruent with existing literature emphasizing multi-step health behavior chains among adolescents (60).

These interpretations underline the central role of indirect pathways—particularly mental health—as the primary mechanisms through which physical activity influences critical adolescent health outcomes.

The mediation analyses provide robust support for indirect pathways linking PA to SUR and SR, highlighting the critical roles of MH and SD. MH was the most influential mediator, significantly mediating relationships between PA and both SUR and SR. Adolescents with higher PA typically experience improved mental health, consequently reducing SUR and SR, consistent with prior findings emphasizing MH as a key intervention target. SD also significantly mediated the relationship between PA and SUR, reinforcing prior research that adequate sleep contributes to better emotional regulation and reduced substance use among adolescents.

However, the mediating role of HD between PA and SR was negligible, suggesting dietary influences alone may have limited practical significance. Chain mediation analyses further supported multi-step pathways (PA → SD → MH → SUR; PA → MH → SUR → SR), underscoring the interconnectedness of these health behaviors in reducing risk.

The results indicated several important theoretical contributions to the literature on adolescent health behaviors and risk prevention. It enriches the understanding of the mechanisms linking PA to SR and SUR by highlighting the central mediating roles of MH and SD. The identification of MH as a pivotal mediator underscores its significance within adolescent behavioral health models, reinforcing theories that emphasize emotional and psychological regulation as critical pathways through which lifestyle behaviors impact health outcomes.

By explicitly examining sequential mediation (chain mediation) pathways involving SD, MH, and SUR, this study advances current conceptualizations of health behavior interconnectivity among adolescents. It demonstrates empirically that PA exerts its protective effects through complex, multi-step processes, emphasizing an integrated approach to health promotion rather than isolated behavioral interventions.

In conclusion, the complex positive direct associations observed between PA and both SR and SUR contribute to the theoretical discourse on suppressor effects and competing mechanisms in behavioral health research. These findings highlight the necessity of considering indirect and suppressor effects within health behavior theories and interventions, urging related practitioners to adopt a nuanced understanding of how protective behaviors may sometimes interact with contextual factors in complex ways.

The findings provide important practical implications for practitioners, educators, and policymakers aiming to promote adolescent well-being. The identification of MH and SD as key mediators between PA and adolescent risk behaviors underscores specific target areas for intervention programs.

Interventions designed to enhance MH among adolescents through regular PA should be prioritized. The related institutions could implement structured PA integrated with emotional and psychological support to foster comprehensive health benefits. Additionally, given SD protective role against SUR, educational programs emphasizing healthy sleep hygiene and routines may effectively complement existing SUR prevention efforts.

Moreover, recognizing the chain mediation effect (PA → SD → MH → SUR) highlights the value of adopting integrated intervention strategies. Practitioners should consider multi-component programs that simultaneously address PA levels, SD, and MH to optimize reductions in SUR and SR among adolescents.

Despite its contributions, the current study has several limitations. First, the cross-sectional design prevents causal inferences from being established, as it does not allow for the determination of temporal sequence among variables. And family environment factors (e.g., parental supervision, economic status) and social support (e.g., partner relationship) may simultaneously influence PA, SD, MH, and SUR. Longitudinal or experimental designs are required to verify the directionality and causal relationships identified in the model. Second, the generalizability of our findings is limited by the sample composition, which included only U.S. adolescents. The lack of data from developing countries and some minority groups may restrict the universality of the conclusions. Future studies should validate these relationships in more diverse, multi-ethnic, and cross-national samples to enhance external validity. Third, the YRBSS does not collect detailed physical activity dose metrics, such as total weekly energy expenditure (e.g., MET-min/week), which limits our ability to examine potential dose–response or threshold effects, including the possibility that excessive activity could have adverse impacts on sleep or mental health. Finally, employing the identified model as a basis for designing and implementing longitudinal intervention studies could yield more definitive insights into the pathways connecting physical activity and adolescent health outcomes.