0 The Dynamic Association between Sleep Quality, Suicide Risk, and Perceived Social Support: Based on Social Media Data OriginalPaper 1 1 2018 127 Binyu Wang binyu_wang@mails.ccnu.edu.cn Xiayu Du duxiayu@mails.ccnu.edu.cn Tingshao Zhu tszhu@psych.ac.cn Zongkui Zhou zhouzk@ccnu.edu.cn Xingyun Liu +86- 15201683001 liuxingyun@ccnu.edu.cn Zhihong Ren +86- 13627131550 ren@ccnu.edu.cn Key Laboratory of Adolescent Cyberpsychology and Behavior (CCNU) Ministry of Education 430079 Wuhan China Key Laboratory of Human Development and Mental Health of Hubei Province, School of Psychology Central China Normal University 430079 Wuhan China CAS Key Laboratory of Behavioral Science, Institute of Psychology 100101 Beijing China Department of Psychology University of Chinese Academy of Sciences 100049 Beijing China School of Psychology Liaoning Normal University 116029 Dalian China School of Psychology Central China Normal University Luoyu Road No.152, Hongshan District 430079 Wuhan China Binyu Wang1,2, Xiayu Du1,2, Tingshao Zhu3,4, Zongkui Zhou1,2, Xingyun Liu1,2, Zhihong Ren1,2,5 1Key Laboratory of Adolescent Cyberpsychology and Behavior (CCNU), Ministry of Education, Wuhan 430079, China 2Key Laboratory of Human Development and Mental Health of Hubei Province, School of Psychology, Central China Normal University, Wuhan 430079, China 3CAS Key Laboratory of Behavioral Science, Institute of Psychology, Beijing 100101, China 4Department of Psychology, University of Chinese Academy of Sciences, Beijing 100049, China 5School of Psychology, Liaoning Normal University, Dalian 116029, China Author Email Binyu Wang: binyu_wang@mails.ccnu.edu.cn; Xiayu Du: duxiayu@mails.ccnu.edu.cn; Tingshao Zhu: tszhu@psych.ac.cn; Zongkui Zhou: zhouzk@ccnu.edu.cn; Xingyun Liu: liuxingyun@ccnu.edu.cn; Zhihong Ren: ren@ccnu.edu.cn. Author ORCID Binyu Wang: 0009-0000-8331-4555; Xiayu Du: 0000-0002-1870-9216; Tingshao Zhu: 0000-0003-0020-3812; Zongkui Zhou: 0000-0002-0958-6568; Xingyun Liu: 0000-0001-7118-0465; Zhihong Ren: 0000-0002-3753-6242. Correspondence Xingyun Liu, School of Psychology, Central China Normal University, Luoyu Road No.152, Hongshan District, Wuhan, 430079, China; Email: liuxingyun@ccnu.edu.cn; Tel: +86-15201683001. Zhihong Ren, School of Psychology, Central China Normal University, Luoyu Road No.152, Hongshan District, Wuhan, 430079, China; Email: ren@ccnu.edu.cn; Tel: +86-13627131550. Author ContributionBinyu Wang and Xiayu Du contributed equally to this work and should be considered as co-first authors.Binyu Wang: Conceptualization, Formal analysis, Software, Writing – original draft; Xiayu Du: Conceptualization, Writing – review & editing; Tingshao Zhu: Data curation; Zongkui Zhou: Supervision; Xingyun Liu: Data Curation, Writing – review & editing; Zhihong Ren: Supervision, Writing – review & editing, Funding Acquisition; Binyu Wang: Conceptualization, Formal analysis, Software, Writing – original draft; Xiayu Du: Conceptualization, Writing – review & editing; Tingshao Zhu: Data curation; Zongkui Zhou: Supervision; Xingyun Liu: Data Curation, Writing – review & editing; Zhihong Ren: Supervision, Writing – review & editing, Funding Acquisition; Funding This research was supported by the Major Program of the National Social Science Foundation of China (grant No. 22&ZD187). Abstract Purpose Suicide risk and sleep problems are significant public health concerns, with perceived social support acting as a key social determinant of mental health. This study aimed to investigate the dynamic, reciprocal relationships among public sleep quality (SQ), suicide risk (SR), and perceived social support (SS) from a macro-level, epidemiological perspective to understand their systemic interplay over time. Methods We constructed a provincial-level annual panel dataset using public data from Weibo, a major Chinese social media platform, spanning 14 years (2010–2023) across 31 provinces. The annual frequencies of keywords related to SQ, SR, and SS were calculated to serve as population-level indicators. Dynamic Structural Equation Modeling (DSEM) was employed to analyze the autoregressive, cross-lagged, and contemporaneous effects among these variables. Results All three variables demonstrated significant temporal continuity. Longitudinally, higher prior SR predicted subsequent declines in both SQ and SS. Paradoxically, however, better prior SQ (i.e., lower scores on the measure) predicted higher subsequent SR, while higher prior SS predicted poorer subsequent SQ. Concurrently, poorer SQ and lower SS were associated with higher SR. Conclusions Utilizing large-scale social media data, this study uncovers the complex feedback loops governing the population-level dynamics of sleep quality, perceived social support, and suicide risk. The findings highlight the necessity of a systemic public health approach for suicide prevention that considers these complex, time-dependent, and sometimes paradoxical interactions, moving beyond simplistic linear risk models. Keywords: Sleep quality suicide risk perceived social support social media dynamic structural equation model The Dynamic Association between Sleep Quality, Suicide Risk, and Perceived Social Support: Based on Social Media Data 1 Introduction As a global public health challenge causing over 700,000 deaths annually and imposing substantial societal burdens [1], suicide risk (SR) may potentially be mitigated by improving sleep quality (SQ) and perceived social support (SS) [2, 3]. The negative link between poor sleep quality and elevated suicide risk is well-established. For instance, a ten-year longitudinal study established that sleep disturbance is a risk factor for suicide, independent of depressive mood [4]. Multiple reviews corroborate this finding, documenting that both objective and subjective sleep disturbances (such as insomnia, nightmares, and poor self-reported sleep quality) can elevate the risk for suicidal ideation, suicide attempts, and death by suicide [2, 5–6]. Mechanistically, anxiety and depressive symptoms serially mediate the pathway from poor sleep to suicide risk [7]. However, the effect of suicide risk on sleep quality may be mediated by presleep arousal (PSA), which is defined as a state of excessive arousal at bedtime that is counterproductive to sleep onset and maintenance [8]. Specifically, one study concluded that this association is mediated by presleep cognitive arousal (PSA-C): the degree of mental alertness during attempts to fall asleep that often manifests as rumination and worry, factors strongly associated with suicide risk [9]. Similarly, perceived social support is widely recognized as a crucial protective factor against suicide risk. Perceived social support refers to an individual's subjective appraisal of the aid available from various sources, such as family, friends, colleagues, or the community. This support can be emotional, informational, or instrumental (i.e., tangible assistance) [10]. Indeed, the perception of social support seems to be a key factor in mitigating suicidal ideation [11]. Subsequent research has demonstrated that perceived social support moderates the relationship between impulsivity and suicide risk, acting as a buffer for individuals with higher levels of impulsivity [12]. In a similar vein, other studies have found that perceived social support buffers against suicidal ideation by enhancing self-esteem and the utilization of social support [13]. Furthermore, studies using representative samples from the United States and the United Kingdom have associated perceived social support with a reduced likelihood of lifetime suicide attempts, suggesting its role as a cross-cultural protective factor [14]. These findings are echoed in research from China, where psychological autopsy studies of rural youth who died by suicide identified the absence of perceived social support as a critical risk factor, particularly for individuals with psychiatric disorders [15]. Beyond empirical evidence, relevant theories also support this relationship. For instance, individuals who perceive social support are likely to experience a greater sense of belonging. According to the Interpersonal Theory of Suicide, a thwarted sense of belonging increases suicide risk, whereas an established sense of belonging mitigates it [16]. However, the relationship is bidirectional, as suicide risk can also influence perceived social support. For example, according to cognitive theory, individuals with depression and suicide risk exhibit the negative cognitive triad: negative views of the self, the world (including others), and the future [17]. This negative cognitive filter may impair their ability to objectively appraise genuine acts of support, leading to social isolation and a subsequent decline in perceived social support. Furthermore, the stigma surrounding mental illness creates a significant barrier to accessing formal support systems such as mental health services. This barrier to help-seeking arises from both external, anticipated social pressures (public stigma) and internalized feelings of shame and self-devaluation (self-stigma) [18]. This stigma-induced barrier can sever crucial avenues of social support, thereby diminishing an individual's level of perceived social support. A reciprocal relationship is also evident between sleep quality and perceived social support. From a workplace perspective, for instance, improvements in network and emotional support have been shown to lower the risk of sleep disturbances [19]. Similarly, within intimate relationships, supportive partnerships are positively associated with sleep quality, whereas relationships characterized by rejection are predictive of poorer sleep outcomes [20]. From a broader societal perspective, perceived social support exhibits an inverse association with both validated questionnaire scores (e.g., the PSQI) and diary-assessed sleep quality metrics [21]. This association is robust; a meta-analysis demonstrated that higher levels of perceived social support are linked to improved sleep quality, independent of variations in the operationalization of support, study design, or chronic conditions [22]. In the reverse direction, sleep quality also affects the perception of social support. Research has found that sleep-deprived individuals not only report increased feelings of loneliness but also tend to maintain greater social distance from others. Moreover, they are perceived by healthy observers as less socially attractive, thereby reducing the observers' willingness to engage in social interaction [23]. This social withdrawal can ultimately reduce perceived social support. While these dyadic relationships are well-documented, a critical gap remains in understanding how these three factors (sleep quality, suicide risk, and perceived social support) interact dynamically as a system over time, particularly at a collective, societal level. Large-scale social media data offer a unique opportunity to investigate these population-level dynamics. Weibo, a major Chinese social media platform, is a rich data source where users often freely discuss personal topics like sleep and perceived social support [24, 25]. Additionally, Dynamic Structural Equation Modeling (DSEM) provides advanced methods to examine these complex longitudinal relationships. This study applies DSEM to a 14-year Weibo word frequency dataset. We aim to elucidate the annual dynamic relationships among public SQ, SR, and SS in China (2010–2023). Based on the literature, we formulated three hypotheses within a dynamic systems framework: Hypothesis 1 Autoregressive Effects. We hypothesized that each variable would exhibit a significant positive autoregressive effect, reflecting its temporal continuity. That is, the score of each variable at time t-1 (​) will be positively associated with its score at time t (​) (where SX refers to any of these three variables). Hypothesis 2 Cross-Lagged Effects. We posited a series of cross-lagged effects, anticipating that each variable would influence the others over time. Specifically, ​ will be positively associated with and negatively associated with ​; ​ ​will be positively associated with both ​ and ; and ​​ will be negatively associated with both and ​. Hypothesis 3 Contemporaneous Effects. We hypothesized contemporaneous associations between the variables, reflecting within-year effects on SR. Specifically, will be positively associated with ​, while will be negatively associated with . 2 Method 2.1 Data Source and Processing Text data were sourced via the public Weibo Application Programming Interface (API) from a database covering over 1.16 million active users on Weibo (Sina Weibo), a leading Chinese microblogging platform [26]. The dataset spanned January 1, 2010, to December 31, 2023, consisting of posts from all active users. Following previous research [26, 28], "active users" were defined as those with > 500 posts since registration and recent activity. To ensure personal perspectives, only posts from private users were retained, excluding institutional accounts (e.g., commercial entities, celebrities, government agencies). These textual data were then processed using the TextMind system, developed by the Computational Cyberpsychology Lab at the Institute of Psychology, Chinese Academy of Sciences [29]. The system performs an integrated process, from automatic Chinese word segmentation to the calculation of keyword frequencies based on specific dictionaries [26, 30]. For this study, data were aggregated to the provincial level; scores for SQ, SR, and SS were operationalized as the annual frequency of specific keywords within each of the 31 provinces. The research protocol was approved by the Institutional Review Board of the College of Life Sciences at Central China Normal University (Approval No. CCNU-IRB-202504006b). 2.2 Dictionary-Based Measures Sleep Quality (SQ). SQ was measured using a 92-keyword Sleep Quality Dictionary, which was developed by referencing established domestic and international sleep quality questionnaires and has demonstrated good validity [31]. Higher scores on this measure indicate poorer sleep quality. Suicide Risk (SR). SR was assessed using the Chinese Suicide Dictionary, which was specifically designed to detect suicide risk on social media. The dictionary comprises 2,168 words across 13 distinct categories (e.g., suicidal ideation, suicidal behaviors) and has been shown to yield more accurate estimations of suicide risk at both post and user levels compared to general-purpose dictionaries [32]. A higher score represents a greater level of suicide risk. Perceived Social Support (SS). SS was quantified using the 1,312-word Perceived Social Support Dictionary. It incorporates a dynamic corpus-updating mechanism to track the rapid evolution of internet language and has demonstrated good reliability [33]. A higher score corresponds to a greater level of perceived social support. 2.3 Data Analysis We used Dynamic Structural Equation Modeling (DSEM) in Mplus version 8.3 [34] to examine the dynamic relationships among SQ, SR, and SS. This approach tested our hypotheses on autoregressive, cross-lagged, and contemporaneous relationships. The DSEM was specified as a two-level model to account for the nested data structure (i.e., annual observations nested within provinces). Level 1 (within-province) modeled the time-series dynamics, estimating the autoregressive, cross-lagged, and contemporaneous effects for SQ, SR, and SS, as well as their within-province residuals. Level 2 (between-province) modeled the variability across provinces by treating the Level-1 parameters as random effects. DSEM relies on Bayesian analysis via Markov Chain Monte Carlo (MCMC) estimation algorithms, effective for complex models and improving convergence [35]. These algorithms generate a posterior distribution for each parameter. Statistical significance is determined by examining the 95% Credible Interval (CI); if the interval (delineated by the 2.5th and 97.5th percentiles of the posterior distribution) does not contain zero, the estimate is considered significant. Model convergence was assessed using the Potential Scale Reduction (PSR) factor, which compares within-chain to between-chain variance; a value approaching 1.0 indicates convergence [36]. A PSR > 1.05 suggests convergence issues. Our final model (27,000 iterations) achieved adequate convergence with a PSR of 1.038. Note where W is the within-chain variance and B is the between-chain variance. We used noninformative priors, the default setting in Mplus, which is appropriate for the exploratory nature of this study, as they assume all values are equally likely. All variables were latent, province-mean centered to disentangle within-province from between-province variance. This latent centering approach is superior to observed mean centering as it reduces bias in the estimation of autoregressive and cross-lagged effects. Finally, the TINTERVAL = 1 command was used to specify the 1-year interval between each measurement point. As shown in Fig. 1 (for visual clarity, correlations among the random effects are omitted from the diagram), the model specifies parameters at two levels: At Level 1 (Within-Province), the model captures the time-series dynamics within each province (i). These dynamics are composed of several key parameters, including the province-specific intercepts (, , and ​), which quantify the mean level of each variable for a given province, reflecting its long-term equilibrium across the study period; the autoregressive paths (, , and ), which capture the influence of a variable in the previous year on itself in the current year; the cross-lagged paths ( through ​), which capture the influence of one variable in the previous year on a different variable in the current year; and the contemporaneous paths ( and ​), representing the within-year associations of SQ and SS with SR at time t. At Level 2 (Between-Province), the model estimates the population-level means of the Level 1 parameters, which represent the average dynamic effects across all 31 provinces. These are denoted by gamma coefficients (), including the population means for the intercepts (, , and ​), autoregressive coefficients (, , and ​), cross-lagged coefficients ( through ​), and contemporaneous coefficients ( and ​).

Fig. 1 Path Diagram of the Dynamic Structural Equation Model (DSEM)

3 Result 3.1 Descriptive statistics Table 1 presents descriptive statistics and correlations for the 434 total observations (14 years × 31 provinces). No data were missing for SQ, SR, or SS. Skewness and kurtosis values confirmed the variables' distributions were appropriate for DSEM analysis. Only the correlation between SR and SS was statistically significant, though small (r = -0.042). SQ SR SS Mean 0.0119 0.0461 0.000104 S.D. 0.00128 0.00631 0.000025 Min 0.00833 0.0322 0.0000457 Max 0.0168 0.0616 0.000192 Skewness 0.747 0.468 -0.080 Kurtosis -0.069 -0.664 -0.471 SQ - SR 0.031 - SS -0.025 -0.042* - Note. * < 0.05

Table 1 Descriptive Statistics and Bivariate Correlations Among Variables.

3.2 DSEM results DSEM results are summarized in Fig. 2 and detailed in Table 2; all variables were standardized (z-scored) before analysis. The population-level mean intercepts were not significant for SQ, but were significant for both SR (Median = -0.875, 95% CI [-1.151, -0.649]) and SS (Median = 0.926, 95% CI [0.704, 1.179]). Significant random intercept variances for SQ (0.015), SR (0.002), and SS (0.002) indicated substantial between-province heterogeneity.

Fig. 2 Path Diagram of the DSEM results

Intercept Effect Notation Posterior median 95% credible interval Int(SQ, ) -0.068 [-0.361,0.215] Int(SR, ) -0.875 [-1.151,-0.649] Int(SS, ) 0.926 [0.704,1.179] Ln(Var[SQ], ) -1.652 [-1.898,-1.406] Ln(Var[SR], ) -2.859 [-3.137,-2.587] Ln(Var[SS], ) -1.259 [-1.510,-1.015] Path coefficient Predictor Outcome Notation Posterior median 95% credible interval SQ at t-1 SQ at t() 0.768 [0.687,0.858] SR at t-1 SR at t() 0.699 [0.635,0.761] SS at t-1 SS at t() 0.130 [0.045,0.210] SR at t-1 SQ at t() 0.099 [0.026,0.207] SS at t-1 SQ at t() 0.154 [0.093,0.236] SQ at t-1 SR at t() -0.121 [-0.197,-0.035] SS at t-1 SR at t() 0.024 [-0.029,0.076] SQ at t-1 SS at t() 0.000 [-0.085,0.089] SR at t-1 SS at t() -0.626 [-0.724,-0.530] SQ at t SR at t() 0.263 [0.200,0.338] SS at t SR at t () -0.170 [-0.256,-0.097] Between-level residual variance Effect Notation Posterior median Var() 0.015 Var() 0.002 Var() 0.002 Var() 0.003 Var() 0.001 Var() 0.002 Var() 0.002 Var() 0.002 Var() 0.002 Var() 0.001 Var() 0.014 Var() 0.002 Var() 0.001 Var() 0.007 Var() 0.275 Var() 0.371 Var() 0.281 Note. Boldface indicates that the 95% Credible Interval does not contain zero.

Table 2 DSEM Results for Path Coefficients and Variance Components.

First, all three variables demonstrated significant positive autoregressive effects, indicating temporal continuity. The standardized coefficients were significant for SQ ( = 0.768, 95% CI [0.687, 0.858]), SR ( = 0.699, 95% CI [0.635, 0.761]), and SS ( = 0.130, 95% CI [0.045, 0.210]). All coefficients were between − 1 and 1, confirming time series stationarity [37]. Next, we examined the cross-lagged effects between the variables. Four of the six possible pathways were found to be significant: Higher prior suicide risk () predicted poorer subsequent sleep quality (, = 0.099, 95% CI [0.026, 0.207]) (i.e., a higher score). Higher prior perceived social support () also predicted poorer subsequent sleep quality (, = 0.154, 95% CI [0.093, 0.236]) (i.e., a higher score). Higher prior sleep quality () (i.e., a lower score) predicted large subsequent suicide risk (, = -0.121, 95% CI [-0.197, -0.035]). Higher prior suicide risk () predicted a large decrease in subsequent perceived social support () ( = -0.626, 95% CI [-0.724, -0.530]). The effect of prior suicide risk on subsequent social support was notably large in magnitude, whereas the other cross-lagged effects were relatively small. Contemporaneously, both sleep quality and social support significantly predicted suicide risk. Specifically, within the same year, poorer sleep quality () was associated with higher suicide risk (, = 0.263, 95% CI [0.200, 0.338]), while higher perceived social support () was associated with lower suicide risk (, = -0.170, 95% CI [-0.256, -0.097]). The variances for both contemporaneous effects were also significant, indicating notable between-province variability in the strength of these within-year relationships. Finally, the model accounted for a substantial portion of the variance in all three variables, demonstrating considerable explanatory power. The values were 62.8% (95% CI [54.4%, 74.0%]) for SQ, 80.8% (95% CI [76.3%, 84.8%]) for SR, and 55.7% (95% CI [47.8%, 63.1%]) for SS. 4 Discussion This study used DSEM and a large-scale Weibo dataset to investigate the dynamic relationships among sleep quality, suicide risk, and perceived social support. While some findings were expected, several key cross-lagged effects emerged as counterintuitive. These paradoxical results challenge conventional assumptions and require deeper interpretation. 4.1 Continuity of Sleep Quality, Suicide Risk, and Perceived Social Support As hypothesized, sleep quality showed strong continuity. This finding resonates with research by Pierson-Bartel and Ujma [38], who found moderate to high correlations between objective measures of sleep quality (such as sleep macrostructure and quantitative EEG indicators as measured by mobile EEG devices) and subjective measures (such as self-reports). This consistency reinforces the stability of sleep quality. Likewise, suicide risk showed high continuity, as hypothesized. This finding echoes the work of Preti and Lentini [39], who demonstrated that historical suicide time-series data can predict future trends, thus corroborating the strong autoregressive effect observed in our study. The high autoregressive coefficient suggests a societal-level "memory effect," where major events have lingering impacts on suicide risk. Its effects may dissipate over months or years. One potential mechanism for this societal memory may be related to the phenomenon of suicide contagion. In their systematic review, Cheng et al. [40] outlined four mechanisms for the formation of suicide clusters: transmission, imitation, contextual influences, and group affiliation. Among these, imitation is considered the mechanism with the most heuristic utility. In contrast to the stability of the other two variables, perceived social support showed only weak positive continuity, a finding that was also consistent with our hypothesis. This suggests that perceived social support is a more volatile construct, with year-to-year fluctuations that are difficult to predict at the regional level. This phenomenon can be explained through the lens of the Conservation of Resources (COR) theory, which conceptualizes social support as a valuable resource. COR theory posits that individuals are fundamentally motivated to obtain, retain, protect, and accumulate valued resources [41]. The theory explains resource dynamics through "loss spirals" and "gain spirals," wherein the loss or gain of one resource can trigger a cycle of further gains or losses. Because this process is sensitive to cultural and social contexts, the interplay of these resource spirals with societal events makes perceived social support difficult to maintain and, consequently, hard to predict. This macro-level instability mirrors findings at the micro-level; for instance, a daily diary study by Coppersmith et al. [42] found high day-to-day variability in perceived social support within individuals. 4.2 The relationship between sleep quality and suicide risk As hypothesized, higher prior suicide risk predicted poorer subsequent sleep quality. This is likely because suicide risk is often comorbid with psychiatric disorders like depression, anxiety, or PTSD. The core symptoms of these conditions, including rumination and psychomotor agitation, can directly disrupt sleep architecture, leading to difficulty initiating sleep, sleep fragmentation, or early morning awakening. The most striking finding, however, was the paradoxical effect of sleep on later suicide risk. Contrary to our hypothesis, better prior sleep quality predicted an increase in subsequent suicide risk. This phenomenon is conceptually analogous to "paradoxical suicidality," a term describing the emergence or worsening of suicidal ideation or behaviors in patients initiating antidepressant treatment, despite these medications being intended to alleviate, rather than exacerbate, such risk. Explanations for this effect include the possibility that medication increases energy and agitation before mood improves, providing the capacity to act on pre-existing suicidal impulses [43] or that it enhances the cognitive capacity for suicidal ideation [44]. Applying this logic, we propose that the paradoxical effect in our study operates via a similar mechanism: improved sleep may restore the physical energy and cognitive resources required for an at-risk individual to plan and act on suicidal impulses, thus leading to an observable increase in suicide risk. At the contemporaneous level, and consistent with our hypothesis, poorer sleep quality was associated with higher suicide risk within the same year. This macro-level, within-year association is consistent with micro-level, high-frequency findings from Ecological Momentary Assessment (EMA) studies, which show that poor sleep can significantly predict suicidal ideation on the following day [45]. This finding underscores the importance of addressing immediate sleep problems as a potential strategy for mitigating suicide risk. 4.3 The relationship between suicide risk and perceived social support Our analysis of the lagged relationships between perceived social support and suicide risk yielded a noteworthy asymmetry. Contrary to our hypothesis, prior perceived social support did not significantly predict a change in subsequent suicide risk. Conversely, and also contrary to our hypothesis, a strong negative cross-lagged relationship emerged in the opposite direction: higher prior suicide risk was a powerful predictor of a decline in subsequent perceived social support. This erosion of social support can be understood from the perspectives of both the support recipient and the provider. From the recipient's perspective, psychological mechanisms may lead to social withdrawal. For instance, Joiner's Interpersonal Theory of Suicide posits that perceived burdensomeness and thwarted belongingness are significant correlates of suicide risk, a finding robustly validated by meta-analysis [46]. Individuals with high suicide risk may internalize a sense of being a burden, which, combined with feelings of thwarted belongingness, can foster social withdrawal and thus reduce perceived support. Moreover, individuals at high risk for suicide may exhibit an attentional bias toward negative stimuli [47], causing them to overlook supportive cues from their environment and further diminishing their perception of available support. From the provider's perspective, social-psychological phenomena can inhibit supportive action. The diffusion of responsibility, for example, may lead potential supporters to assume others will intervene in a crisis, thereby reducing their own supportive expressions [48]. Furthermore, a provider's self-efficacy is crucial. Research shows a significant positive correlation between one's self-efficacy and their willingness to intervene in a suicide crisis [49]. A lack of knowledge or skills can lead supporters to feel a low sense of self-efficacy and anticipate that their intervention would be futile, prompting them to refrain from offering support. In contrast to the lagged findings, the contemporaneous relationship was consistent with our hypothesis: higher perceived social support was associated with lower concurrent suicide risk. This macro-level, within-year association mirrors findings from micro-level daily diary studies, which have also found a significant negative correlation between perceived social support and same-day suicidal ideation [42]. This finding reinforces the role of perceived social support as both a crucial protective factor and a viable target for intervention. Indeed, higher levels of perceived social support have been linked to a reduced likelihood of lifetime suicide attempts in representative national samples [13]. As an intervention target, a meta-analysis by Hou et al. [3] concluded that while social support interventions may not always reduce suicide attempts, they are a valuable component of broader suicide prevention strategies. 4.4 The relationship between sleep quality and perceived social support The dynamic between sleep and social support was also unexpectedly asymmetric. While prior sleep quality did not significantly predict subsequent social support, higher prior social support paradoxically predicted a decline in subsequent sleep quality, contrary to our hypothesis. This result runs counter to the general consensus from meta-analytic evidence, which indicates that higher social support is typically associated with better sleep quality [22]. However, our paradoxical finding aligns with a smaller body of literature that has reported similar negative associations [e.g., 50, 51]. A compelling explanation for this paradox, suggested by Kent et al. [22], lies in the crucial distinction between perceived and received social support. While the perception of available support is beneficial, the act of receiving support can sometimes be harmful. When received support is either mismatched with an individual's needs or exceeds their coping abilities, it can become a stressor that creates a psychological burden and disrupts sleep [52]. Our findings, situated in the collectivist context of China, may be particularly illuminated by this distinction. Here, a cultural emphasis on interpersonal harmony may compel individuals to accept unwanted or unhelpful support to avoid relational friction. This compelled acceptance, rather than genuine utilization of support, could increase psychological burden and, in turn, disrupt sleep. 4.5 Implications for Research and Practice This study holds several key research and practical implications. Methodologically, it demonstrates the utility of social media data for population-level mental health monitoring. Theoretically, it provides ecological validation for the Interpersonal Theory of Suicide. Practically, it offers a new macro-level perspective on SR dynamics, informing public health policy. At the individual level, long-term sleep management, monitored via wearables or diaries, is a critical intervention point. When poor sleep is detected, interventions like CBT-I can be used to improve sleep and potentially reduce downstream suicide risk. Crucially, the paradoxical finding that improved sleep can precede higher suicide risk underscores the need for enhanced monitoring during this apparent recovery phase. At the societal level, culturally tailored programs are needed to disseminate knowledge of suicide intervention, foster responsibility, promote effective support, and enhance public awareness. Given the societal "memory effect," government agencies should consider longitudinal models to monitor suicide risk after major events. Concurrently, suicide-related information must be carefully managed to prevent contagion. 4.6 Limitations and future directions This study has several limitations. First, our Weibo data are susceptible to demographic biases, likely skewed towards younger, urban, and educated users. This limits generalizability to rural, elderly, and non-Internet-using groups. Additionally, our self-report variables are subject to social desirability, recall, and expressive biases. Second, word frequency analysis remains subject to expressive biases. Social stigma, for instance, might lead to under-reporting due to users avoiding sensitive terms, potentially resulting in an underestimation of true risk. Conversely, the use of metaphor, sarcasm, and evolving slang presents misclassification risks, as static dictionaries might fail to capture the true sentiment of nuanced expressions. Additionally, the tendency for exaggerated expressions of emotion and behavior on social media could introduce noise into the frequency counts, potentially resulting in an overestimation of the true risk. Third, methodological limitations exist within our model. DSEM assumes linear relationships and fixed time lags, while actual psychological processes might be non-linear or involve heterogeneous delays (e.g., effects manifesting over weeks versus years). Moreover, using annual data precludes capturing short-term dynamics, and excluding potential covariates limits the model's explanatory power. Finally, limited generalizability may exist outside China's unique socio-cultural context (e.g., mental health stigma, collectivist pressures, censorship), potentially limiting direct extrapolation to individualistic cultures. Indeed, measuring these constructs within a collectivist context may require more culturally-attuned instruments to capture unique influencing factors. Despite these limitations, this study holds significant value. Its application of DSEM to large-scale, longitudinal social media data yielded novel population-level insights into the interplay of sleep quality, perceived social support, and suicide risk. Building upon this foundation, future research should aim to utilize more granular data (e.g., monthly) for analyzing finer-grained dynamics; incorporate validated scales as covariates for specific populations; and test the generalizability of these in findings within individualistic cultures, exploring culture's moderating role. 5 Conclusion Using a 14-year provincial Weibo dataset, this study applied DSEM to examine the dynamics of public sleep quality, suicide risk, and perceived social support in China. Our findings revealed a system of complex and reciprocal influences. Specifically, the results revealed that all three variables demonstrated significant temporal continuity. The cross-lagged analysis uncovered several key predictive relationships. As expected, higher prior suicide risk predicted a subsequent decline in sleep quality. However, three findings were contrary to our hypotheses: higher prior suicide risk also predicted a decline in subsequent perceived social support; poorer prior sleep quality predicted a decrease in future suicide risk; and higher prior social support predicted a decline in future sleep quality. At the contemporaneous level, both poor sleep quality and low perceived social support were strongly associated with higher suicide risk within the same year. Taken together, these findings reveal that the relationships among sleep quality, perceived social support, and suicide risk are not static or unidirectional. Rather, they constitute a dynamic system with nuanced, time-dependent feedback mechanisms. This underscores the necessity for public health interventions to move beyond simple, linear models and instead account for these complex interactions to effectively mitigate both present and future suicide risk. Statements and Declarations Funding This research was supported by the Major Program of the National Social Science Foundation of China (grant No. 22&ZD187). Competing interests The authors declare that they have no conflict of interest. Ethics approval All procedures performed in this study involving human data were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards. Specifically, this study strictly adheres to the principles of the Declaration regarding the protection of research participants' privacy and the confidentiality of their personal information (Article 24). As the research is based on publicly available, anonymized social media data, it did not involve the direct recruitment of human participants. The research protocol was reviewed and approved by the Institutional Review Board of the College of Life Sciences at Central China Normal University (Approval No. CCNU-IRB-202504006b). Consent to participate Informed consent to participate was not required as this study is based on publicly available, anonymized social media data and does not contain clinical studies or individual patient data. Data AvailabilityData sets generated during the current study are available from the corresponding author on reasonable request. Authors' contributions Binyu Wang and Xiayu Du contributed equally to this work and should be considered as co-first authors. Binyu Wang: Conceptualization, Formal analysis, Software, Writing – original draft; Xiayu Du: Conceptualization, Writing – review & editing; Tingshao Zhu: Data curation; Zongkui Zhou: Supervision; Xingyun Liu: Data Curation, Writing – review & editing; Zhihong Ren: Supervision, Writing – review & editing, Funding Acquisition; Reference 1. World Health Organization (2021) Live life: An implementation guide for suicide prevention in countries, 1st ed. World Health Organization, Geneva 2. Bernert RA, Kim JS, Iwata NG, Perlis ML (2015) Sleep disturbances as an evidence-based suicide risk factor. Curr Psychiatry Rep 17:15. https://doi.org/10.1007/s11920-015-0554-4 3. Hou X, Wang J, Guo J, et al (2022) Methods and efficacy of social support interventions in preventing suicide: a systematic review and meta-analysis. Evid Based Mental Health 25:29–35. https://doi.org/10.1136/ebmental-2021-300318 4. Bernert RA, Turvey CL, Conwell Y, Joiner TE (2014) Association of poor subjective sleep quality with risk for death by suicide during a 10-year period: a longitudinal, population-based study of late life. JAMA Psychiatry 71:1129. https://doi.org/10.1001/jamapsychiatry.2014.1126 5. Harris LM, Huang X, Linthicum KP, et al (2020) Sleep disturbances as risk factors for suicidal thoughts and behaviours: a meta-analysis of longitudinal studies. Sci Rep 10:13888. https://doi.org/10.1038/s41598-020-70866-6 6. Liu RT, Steele SJ, Hamilton JL, et al (2020) Sleep and suicide: a systematic review and meta-analysis of longitudinal studies. Clin Psychol Rev 81:101895. https://doi.org/10.1016/j.cpr.2020.101895 7. Yiyue Y, Kaiqi G, Rujie W, et al (2023) Effects of sleep quality on suicide risk in COVID-19 patients: The chain mediating of anxiety and depressive symptoms. Heliyon 9:e15051. https://doi.org/10.1016/j.heliyon.2023.e15051 8. Nicassio PM, Mendlowitz DR, Fussell JJ, Petras L (1985) The phenomenology of the pre-sleep state: The development of the pre-sleep arousal scale. Behav Res Ther 23:263–271. https://doi.org/10.1016/0005-7967(85)90004-x 9. Grove JL, Carlson SE, Parkhurst KA, Smith TW (2022) The role of pre-sleep arousal in the connection between insomnia and suicide risk. Death Stud 46:2523–2529. https://doi.org/10.1080/07481187.2021.1964109 10. Cobb S (1976) Social support as a moderator of life stress. Psychosom Med 38:300–314. https://doi.org/10.1097/00006842-197609000-00003 11. Chioqueta AP, Stiles TC (2007) The relationship between psychological buffers, hopelessness, and suicidal ideation: Identification of protective factors. Crisis 28:67–73. https://doi.org/10.1027/0227-5910.28.2.67 12. Kleiman EM, Riskind JH, Schaefer KE, Weingarden H (2012) The moderating role of social support on the relationship between impulsivity and suicide risk. Crisis 33:273–279. https://doi.org/10.1027/0227-5910/a000136 13. Kleiman EM, Liu RT (2013) Social support as a protective factor in suicide: Findings from two nationally representative samples. J Affect Disord 150:540–545. https://doi.org/10.1016/j.jad.2013.01.033 14. Kleiman EM, Riskind JH (2013) Utilized social support and self-esteem mediate the relationship between perceived social support and suicide ideation: a test of a multiple mediator model. Crisis 34:42–49. https://doi.org/10.1027/0227-5910/a000159 15. Zhang J, Xiao S, Zhou L (2010) Mental disorders and suicide among young rural chinese: a case-control psychological autopsy study. Am J Psychiatry 167:773–781. https://doi.org/10.1176/appi.ajp.2010.09101476 16. Joiner T (2009) Why people die by suicide. Harvard University Press, Cambridge. 17. Beck AT (1976) Cognitive therapy and the emotional disorders. International Universities Press, New York 18. Corrigan P (2004) How stigma interferes with mental health care. Am Psychol 59:614–625. https://doi.org/10.1037/0003-066x.59.7.614 19. Nordin M, Westerholm P, Alfredsson L, Akerstedt T (2012) Social support and sleep. Longitudinal relationships from the WOLF-Study. Psych 03:1223–1230. https://doi.org/10.4236/psych.2012.312a181 20. Kent RG, Uchino BN, Cribbet MR, et al (2015) Social relationships and sleep quality. Ann Behav Med 49:912–917. https://doi.org/10.1007/s12160-015-9711-6 21. Chung J (2017) Social support, social strain, sleep quality, and actigraphic sleep characteristics: Evidence from a national survey of US adults. Sleep Health 3:22–27. https://doi.org/10.1016/j.sleh.2016.10.003 22. Kent De Grey RG, Uchino BN, Trettevik R, et al (2018) Social support and sleep: A meta-analysis. Health Psychol 37:787–798. https://doi.org/10.1037/hea0000628 23. Ben Simon E, Walker MP (2018) Sleep loss causes social withdrawal and loneliness. Nat Commun 9:3146. https://doi.org/10.1038/s41467-018-05377-0 24. Tian X, Yu G, He F (2016) An analysis of sleep complaints on Sina Weibo. Comput Hum Behav 62:230–235. https://doi.org/10.1016/j.chb.2016.04.014 25. Shi J, Chen L (2014) Social support on Weibo for people living with HIV/AIDS in China: a quantitative content analysis. Chin J Commun 7:285–298. https://doi.org/10.1080/17544750.2014.926954 26. Li L, Li A, Hao B, et al (2014) Predicting active users’ personality based on micro-blogging behaviors. PLoS ONE 9:e84997. https://doi.org/10.1371/journal.pone.0084997 27. Weibo Corporation (2024) Announcement of the Fourth Quarter and Fiscal Year 2023 Financial Results and Dividend. Hong Kong Exchanges and Clearing Limited. https://www.hkexnews.hk. Accessed 6 June 2025 28. Li H, Han Y, Xiao Y, et al (2021) Suicidal ideation risk and socio-cultural factors in china: a longitudinal study on social media from 2010 to 2018. IJERPH 18:1098. https://doi.org/10.3390/ijerph18031098 29. Gao R, Hao B, Bai S, et al (2013) Improving user profile with personality traits predicted from social media content. In: Proceedings of the 7th ACM conference on Recommender systems. ACM, Hong Kong China, pp 355–358 30. Gao R, Hao B, Li H, et al (2013) Developing simplified Chinese psychological linguistic analysis dictionary for microblog. In: Imamura K, Usui S, Shirao T, et al (eds) Brain and Health Informatics. Springer International Publishing, Cham, pp 359–368 31. Lin J, Sun A, Chen J, et al (2023) Sleep quality and life satisfaction: based on big data of weibo in different regions. ChinaXiv [preprint] https://doi.org/10.12074/202307.00067V1 32. Lv M, Li A, Liu T, Zhu T (2015) Creating a Chinese suicide dictionary for identifying suicide risk on social media. PeerJ 3:e1455. https://doi.org/10.7717/peerj.1455 33. Isak G, Du X, Wang B, et al (2025) An analysis of the status and influencing factors of perceived social support among adolescents: based on weibo big data. ChinaXiv [preprint] https://doi.org/10.12074/202507.00050 34. Muthen LK, Muthen B (2017) Mplus user’s guide: Statistical analysis with latent variables, Eighth edition. Muthén & Muthén, Los Angeles, CA 35. Muthén B, Asparouhov T (2012) Bayesian structural equation modeling: A more flexible representation of substantive theory. Psychol Methods 17:313–335. https://doi.org/10.1037/a0026802 36. Asparouhov T, Muthén B (2010) Bayesian analysis using mplus: Technical implementation 37. McNeish D, Hamaker EL (2020) A primer on two-level dynamic structural equation models for intensive longitudinal data in Mplus. Psychol Methods 25:610–635. https://doi.org/10.1037/met0000250 38. Pierson-Bartel R, Ujma PP (2024) Objective sleep quality predicts subjective sleep ratings. Sci Rep 14:5943. https://doi.org/10.1038/s41598-024-56668-0 39. Preti A, Lentini G (2016) Forecast models for suicide: Time-series analysis with data from Italy. Chronobiol Int 33:1235–1246. https://doi.org/10.1080/07420528.2016.1211669 40. Cheng Q, Li H, Silenzio V, Caine ED (2014) Suicide contagion: a systematic review of definitions and research utility. PLoS ONE 9:e108724. https://doi.org/10.1371/journal.pone.0108724 41. Halbesleben JRB, Neveu J-P, Paustian-Underdahl SC, Westman M (2014) Getting to the “COR”: Understanding the role of resources in conservation of resources theory. J Manag 40:1334–1364. https://doi.org/10.1177/0149206314527130 42. Coppersmith DDL, Kleiman EM, Glenn CR, et al (2019) The dynamics of social support among suicide attempters: a smartphone-based daily diary study. Behav Res Ther 120:103348. https://doi.org/10.1016/j.brat.2018.11.016 43. Friedman RA, Leon AC (2007) Expanding the black box — depression, antidepressants, and the risk of suicide. N Engl J Med 356:2343–2346. https://doi.org/10.1056/nejmp078015 44. Pierson EE (2009) Antidepressants and suicidal ideation in adolescence: a paradoxical effect. Psychol Sch 46:910–914. https://doi.org/10.1002/pits.20432 45. Littlewood DL, Kyle SD, Carter L-A, et al (2019) Short sleep duration and poor sleep quality predict next-day suicidal ideation: an ecological momentary assessment study. Psychol Med 49:403–411. https://doi.org/10.1017/s0033291718001009 46. Chu C, Buchman-Schmitt JM, Stanley IH, et al (2017) The interpersonal theory of suicide: a systematic review and meta-analysis of a decade of cross-national research. Psychol Bull 143:1313–1345. https://doi.org/10.1037/bul0000123 47. Cha CB, Najmi S, Park JM, et al (2010) Attentional bias toward suicide-related stimuli predicts suicidal behavior. J Abnorm Psychol 119:616–622. https://doi.org/10.1037/a0019710 48. Darley JM, Latane B (1968) Bystander intervention in emergencies: Diffusion of responsibility. J Pers Soc Psychol 8:377–383. https://doi.org/10.1037/h0025589 49. Mansoor M, Ahmad KB (2022) Effectiveness of suicide prevention skills training on the willingness to intervene and the self-efficacy among young adult students. Pak J Clin Psychol 21:3–16 50. Brummett BH, Babyak MA, Siegler IC, et al (2006) Associations among perceptions of social support, negative affect, and quality of sleep in caregivers and noncaregivers. Health Psychol 25:220–225. https://doi.org/10.1037/0278-6133.25.2.220 51. Tomfohr LM, Edwards KM, Madsen JW, Mills PJ (2015) Social support moderates the relationship between sleep and inflammation in a population at high risk for developing cardiovascular disease. Psychophysiology 52:1689–1697. https://doi.org/10.1111/psyp.12549 52. Cohen S, Wills TA (1985) Stress, social support, and the buffering hypothesis. Psychol Bull 98:310–357. https://doi.org/10.1037/0033-2909.98.2.310 YesAbstractAbstract Purpose: Suicide risk and sleep problems are significant public health concerns, with perceived social support acting as a key social determinant of mental health. This study aimed to investigate the dynamic, reciprocal relationships among public sleep quality (SQ), suicide risk (SR), and perceived social support (SS) from a macro-level, epidemiological perspective to understand their systemic interplay over time. Methods: We constructed a provincial-level annual panel dataset using public data from Weibo, a major Chinese social media platform, spanning 14 years (2010–2023) across 31 provinces. The annual frequencies of keywords related to SQ, SR, and SS were calculated to serve as population-level indicators. Dynamic Structural Equation Modeling (DSEM) was employed to analyze the autoregressive, cross-lagged, and contemporaneous effects among these variables. Results: All three variables demonstrated significant temporal continuity. Longitudinally, higher prior SR predicted subsequent declines in both SQ and SS. Paradoxically, however, better prior SQ (i.e., lower scores on the measure) predicted higher subsequent SR, while higher prior SS predicted poorer subsequent SQ. Concurrently, poorer SQ and lower SS were associated with higher SR. Conclusions: Utilizing large-scale social media data, this study uncovers the complex feedback loops governing the population-level dynamics of sleep quality, perceived social support, and suicide risk. The findings highlight the necessity of a systemic public health approach for suicide prevention that considers these complex, time-dependent, and sometimes paradoxical interactions, moving beyond simplistic linear risk models. Keywords: Sleep quality; suicide risk; perceived social support; social media; dynamic structural equation model.