How rural residence and gender associate with Mental Health: a cross-sectional study among Chinese early adolescents

Background: Mental health disorders are the leading cause of disability in youth globally. China has the world’s second largest pediatric population with growing urban-rural inequalities. Objectives: 1) to examine the prevalence and gender differences in drug use, depressive symptoms, social support and externalizing behaviors in adolescents from an underserved rural community; 2) to compare results to a known urban sample Methods: A cross-sectional study was conducted among children 12-14 years-old from rural Guizhou (N=76) using the 2003 Global School-based Student Health Survey. Prevalence and gender differences of 19 mental health outcomes were calculated and compared to those from a public urban Beijing sample (N=1,629). Associations of outcomes with gender and rural residence were assessed using multivariate logistic regression models in a combined analysis. Results: The prevalence of 9 out of the 19 investigated outcomes was signicantly higher in the rural sample. In rural Guizhou, female gender was associated only with lower prevalence of physical ghts. In the combined analysis, rural residence was associated with higher risks of lifetime troubles due to drinking, loneliness, insomnia, hopelessness, injuries and absenteeism. Girls had lower risks of recent drinking, inebriation, troubles due to drinking, recent smoking, ghts, injures and bullying victimization compared to boys. Conclusion: Higher risks of excessive drinking, depressive symptoms, poor social support, and externalizing behaviors were found in rural adolescents compared to their urban peers. Girls had lower risks of drug use, poor peer support and externalizing behaviors. These ndings call for targeted interventions informed by contextual and gender specications.

patterns of higher prevalence of substance use (She et al., 2016), anxiety (Liu et al., 2018), lower selfesteem and social support (Zhang et al., 2017) in rural youth. The causes are likely multifactorial. Rural children reportedly have lower rates of preschool and kindergarten enrollment, poorer social skills and less caregiver involvement in their education (Chen et al., 2015). In addition, more than 20 million have been impacted by the family disruption due to parental out-migration in the last few decades (Gao et al., 2010), which increased risks in educational and health problems (Meng & Yamauchi, 2017). Rural China is also known for traditional gender roles-girls are often assigned more domestic responsibilities and receive fewer resources (Tian et al., 2018). Historically this led to higher female child mortality (Ren, 1995) and poorer health (Li et al., 2004). Recently the gender gap seems to be closing for all health outcomes except for mental health . This phenomenon is poorly understood.
Guizhou (Appendix 1) is the poorest province in China based on per capita income (National Bureau of Statistics of China, 2014). Zhijin county is a mountainous area of Guizhou particularly isolated due to di cult transportation (at least 6 hours of cliff-side bus ride from the nearest station at the time of data collection). In addition to limited economic and healthcare resources, more than one third of the rural children have been "left behind" by parents due to out-migration (All China Women's Federation, 2013).
These children suffer from long-term separation from parents during formative years and often live with overburdened relatives or grandparents (Zhao et al., 2018). This phenomenon of "left-behind-children" is common among LMICs and has been associated with higher rates of depression, anxiety, suicidal ideation, conduct disorder and substance use (Fellmeth et al., 2018).
Objectives 1) to examine the prevalence and gender differences in drug use, depressive symptoms, social support and externalizing behaviors in early adolescents in an underserved rural county in China; 2) to compare outcomes to those in a public urban sample; 3) to examine the associations of gender and residence with mental health outcomes in all subjects Methods This study used cross-sectional survey data collected by the authors in grades 4-6 (ages [11][12][13][14] at two underserved schools in rural Zhijin County (Guizhou province, China). A total of 100 students completed the Chinese version of the 2003 Global School Health Survey, or GSHS (WHO, 2003). The sample size satis ed our power calculation using an α of 0.05 and a hypothesized between-group difference of 10%.
The data collection was twofold. First, two local schools participating in a community library project were contacted and both agreed to enroll in the study. After obtaining consent from the principals, we randomly selected 2 classrooms from each school (cluster randomization) with fty students each. All guardians of the 100 students gave verbal assesnt prior to student participation. The urban Beijing sample included in the rural-urban analysis was previously published by the Chinese Center for Disease Control (CCDC) in collaboration with the World Health Organization and the US CDC (USCDC, 2016). The same 2003 GSHS was used in a 2-stage cluster sample selection process. First, a list of all middle Page 4/20 schools in Beijing was gathered and schools were selected with probability proportional to enrollment size. Next, classes were randomly selected as clusters and all students in these classes were eligible to participate. The school response rate in Beijing was 100% and the student response rate was 99%. A total of 2,348 students ages 12-16 were included in the Beijing sample. For this study students ages 12-14 were chosen to represent early adolescents. Our analysis included 76 Zhijin students and 1,629 Beijing students.
For the survey, the 2003 GSHS is a widely used tool validated and applied by the WHO in many countries in conjunction with local governments (WHO, 2003). It covered 68 core questions in 8 categories: "Hygiene," "Drug Use," "Mental Health," "Nutrition," "Activity," "Protective Factors," "Tobacco," and "Safety." Table 1 lists the 19 survey questions from the Chinese GSHS used to create all the binary categorical outcome variables in our study. We grouped them into four domains for conceptualization: "drug use", "depression", "social support" and "externalizing behaviors".

Hopelessness
In the past 12 months, did you ever feel so sad or hopeless almost every day for 2 weeks or more in a row that you stopped doing your usual activities?

Suicidal ideation
In the last 12 months, how many times have you seriously considered suicide?

Suicidal plan
In the last 12 months, how many times have you seriously planned a suicide?

Close friends
How many close friends do you have?

Helpful schoolmates
In the past 30 days, how often did you nd most schoolmates kind or helpful?

Parental supervision
In the past 30 days, how often did your guardians really knew what you were doing with your free time?

Parental understanding
In the past 30 days, how often did your guardians understand your problems or worries?

Fighting
In the past 12 months, how many times were you involved in a physical ght?

Injuries
In the past 12 months, how many times were you seriously injured?

Bullying
In the past 30 days, on how many days were you bullied?

Absence
In the past 30 days, on how many days were you absent without permission?
For the analysis, we used descriptvie statistics to calculate the prevalence of each outcome variable in the rural sample. Then we compared the results with those from the urban sample using the χ 2 test. Next, we calculated the prevalence of each outcome variable in the rural sample by gender. The χ 2 test was again used to assess gender differences by statistical signi cance. The same process was applied to examine gender differences in the urban sample. Finally, multivariate logistic regression models were applied to a combined rural-urban sample (N = 1,705) accounting for the effects of clustered sampling to examine the associations between each mental health outcome, gender and residence adjusting for age, grade, height and weight. In sensitivity analysis, interaction terms between gender and residency were added to test potential effect modi cation on the outcomes. All analyses were performed using 95% con dence intervals. A two-sided P value of < 0.05 was used to de ne statistical signi cance. Data analysis was completed using STATA version 15.0 (StataCorp, College Station, TX). Table 2 summaries the background characteristics of rural and urban students.    Table 3 presents the associations of mental health outcomes with female gender and rural residence in the combined rural-uran sample (N = 1,705), adjusting for age, grade, height and weight. We also performed a sensitivity analysis that showed no statistically signifcance in any of the interaction terms (gender*residence) in the multivariate logistic regression models-therefore omitted from the table. Girls were less likely to have drunk alcohol in the past month (OR 0.65, p = 0.01), ever gotten drunk (OR 0.13, p = 0.02), ever gotten into trouble because of drinking (OR 0.40, p < 0.001), smoked in the past month (OR 0.13, p < 0.001), or experienced passive smoking in the past week (OR 0.74, p = 0.03. They were less likely to perceive their schoolmates as rarely or never helpful in the past month (OR 0.47, p < 0.001). They were less likely to have gotten into physical ghts (OR 0.17, p < 0.001) or gotten injured (OR 0.65, p = 0.01) in the past year, or suffered from bullying in the past month (OR 0.68, p = 0.01). Rural students were more likely to have ever gotten into trouble due to alcohol (OR5.09, p = 0.001). They were more likely to report frequent loneliness (OR 2.08, p = 0.03), frequent insomnia (OR 2.07, p = 0.04) and hopelessness for 2 weeks straight (OR 2.56, p = 0.04) in the past year. They were more likely to perceive fellow schoolmates as rarely or never helpful (OR 2.90, p = 0.001), and their guardians as rarely or never understanding of their problems (OR 4.12, p < 0.001) in the past month. Finally, they were much more likely to have been seriously injured in the past year (OR 5.00, p < 0.001) and to have missed school unexcused in the past month (OR 5.21, p < 0.001).

Discussion
This study used a unique rural sample from the poorest province in China to study the prevalence of mental health outcomes in underserved early adolescents, and to examine the associations between mental health, gender and rural residence by comparing this sample to a public urban sample. Major ndings include: (1) 9/19 investigated mental health outcomes were found to be more prevalent in the rural sample compared to the urban sample; (2) rural residence was associated with signi cantly higher odds for 7/19 outcomes after adjusting for age, grade, weight and height; (3) girls were overall less likely to report drug use, poor peer support and externalizing behaviors but this gender difference was mostly driven by the urban sample.
In terms of drug use, the prevalence of alcohol use in the past month was similar in the rural (13.3%) and the urban samples (10.3%). However, the prevalence of having ever gotten into trouble due to alcohol was signi cantly higher in rural Zhijin (14.9% vs. 3.7%, p < 0.001). In the combined analysis, rural residence was also signi cantly associated with this outcome (OR 5.09, p = 0.001). These ndings highlighted excessive drinking reaching harm as more likely in rural adolescents. Consequences of youth excessive drinking on a single occasion have been documented to range from poisoning to motor vehicle crash deaths, drownings and falls, suicides and burns to later dependency and injuries globally (Jernigan, 2001). Our ndings were consistent with previous literature in many Chinese adolescents start drinking alcohol before 6th grade (Li et al., 1996). Therefore, alcohol use should be a priority target for mental health interventions, ideally starting before 6th grade and especially in rural China.
In terms of depressive symptoms, an alarming 32% of Zhijin students have felt "sad and hopeless almost every day for 2 weeks straight in the past year" (vs. 18.2% in Beijing, p = 0.003). In the combined analysis, rural residence was also signi cantly associated with this outcome (OR 2.56, p = 0.04). Though suicidal thought and planning were not signi cantly different between samples, hopelessness may be an early red ag for later depression (Mac Giollabhui et al., 2018). The combined analysis also showed that rural residence was associated with frequent insomnia (OR 2.07, p = 0.01). Previous studies have shown insomnia as a predictor of depression in Chinese teenagers (Luo et al., 2014). Future interventions against adolescent depression could use insomnia as a measurable target in early adolescents especially in rural China, before clinical depression was evident.
Under "social support", 38.2% of Zhijin students (vs. 17.9% in Beijing, p < 0.001) rarely or never found their schoolmates helpful. 46.1% of them has little or no supervision on their free time (vs. 30.1% in Beijing, p = 0.003) and more than 63.2% felt not understood by their guardians (vs. 34.0% in Beijing, p < 0.001). In the combined analysis, rural residence was associated both with poor peer support (OR 2.90, p = 0.001) and poor parental understanding (OR 4.12, p < 0.001). Poor parental involvement and inadequate parenting practices have been shown to be modi able mediators between socioeconomic status and child mental health, with a few published family-level programs that improved children's cognitive and socioemotional skills by optimizing parenting skills and involvement (Verhulst et al., 2020). On the other hand, peer relationships have been found as an effective buffer between traumatic life events and depression in Chinese teens (Greenberger et al., 2000), making it an important target for school-level interventions.
Under "externalizing behaviors", the prevalence of being bullied in the past month was 47.4% in rural Zhijin (vs. 20.3% in Beijing, p < 0.001). The high rural prevalence was concerning for being much higher than the 13.3% previous estimate in China (Eslea et al., 2003), the 35.5% in neighboring LMICs such as the Philippines (Rudatsikira, 2008), and almost four times the 12% average in developed Western Europe (WHO, 2016). Bullying victimization in early teenage years have been shown to be associated with anxiety and depression (Bond et al., 2001), and with teenage suicidal ideation in a Chinse study (Liu et al., 2017).
Given our ndings in early adolescents, interventions targeting bullying should start at or before 12 yearsold. The prevalence of serious injuries was also much higher in the rural sample (46.7% in Zhijin vs. 16.2% in Beijing, p < 0.001), con rmed by an OR of 5.0 (p < 0.001) associated with rural residence in combined analysis. One explanation is the risk of farm-work-related injuries in rural youth, with known associations to sleep disturbances and school-related stress (Postel et al., 2009). Other researchers suggested maltreatment by guardians and involvement in violent episodes as major risk factors (Shi et al., 2014). Injuries, in turn, can lead to unexplained school absenteeism, con rmed by our nding of OR 5.21 for missing school (p < 0.001) associated with rural residence. High absenteeism is likely to also involve anxiety, transport, bullying and di culties with schoolwork (Melvin et al., 2019).
In terms of gender differences, combined analysis showed that girls were overall less likely to have drunk recently, to ever get drunk, to have drinking-related troubles, to smoke and to experience passive smoking.
However, separate gender analysis by sample showed that most gender differences were noted in the urban sample alone, likely due to the relatively small size of our rural sample. The combined analysis ndings were consistent with the global male predominance in smoking (WHO, 2010) and drinking (Wilsnack et al., 2009), as well as parallel ndings in Chinese adolescents (Yue et al., 2016). Being a girl was overall associated with better perceived social support from their peers but not from their guardians.
This was interesting because Chinese families are known to favor male children in resource-allocation (Tian et al., 2018). Perceived peer support is thus especially important for girls as it is a protective factor associated with increased prosocial behavior, better motivation and academic performance in school (Wentzel et al., 2004). Finally, girls had lower odds of suffering from physical ghts, serious injuries and bullying. These ndings were consistent with the literature on male predominance of externalizing behaviors in children (Chaplin & Aldao, 2013). Overall, gender differences found in the combined analysis raise questions regarding the necessity to adopt different strategies and outcome measures by gender when designing future interventions.
This study has several strengths. First, its unique, underserved rural population from Guizhou has rarely been studied despite its needs and limited resources. Second, the combined analysis highlights the ruralurban inequality in adolescent mental health in China. Third, the GSHS survey covers a variety of mental health topics and has a wide international data base. Our ndings help to add to the knowledge in adolescent mental health especially in LMICs. This study has a few important limitations. First, a crosssectional study is unable to con rm causality or temporality. Second, the speci c rural Guizhou sample and urban Beijing sample may not be generalizable to other contexts. They are examples of health inequity in China due to social determinants of health. Third, the rural and urban samples were collected in different studies, although this was done using the same GSHS tool in anonymous data collection.
There is no reason to believe that the survey procedure may induce any difference. The comparison of the prevalence between samples may be affected by age distribution difference between your rural and urban sample. However, we were able to adjust the difference when estimating the ORs. Finally, the lack of information on internalized mental health measures and broader social-economic factors limited our ability to draw conclusions on important mental health outcomes such as anxiety, self-esteem and selfe cacy and the associations between outcomes and speci c social determinants of health. More standardized, comprehensive mental health data in children and adolescents in LMICs are needed to understand the evolution of mental health in adolescents and their determinants.

Conclusion
This study is an examination of the prevalence of common mental health outcomes and their associations with gender and rural residence among early Chinese adolescents, using an underserved rural sample and a known public urban sample surveyed with the 2003 GSHS. Higher prevalence of excessive drinking, hopelessness, poor social support, and externalizing behaviors was found in rural adolescents compared to their urban peers. Girls had lower risks of drug use, poor peer support and externalizing behaviors. These ndings call for targeted interventions informed by contextual and gender speci cations.
Declarations *Ethics approval and consent to participate This study has been determined to have "exempt" status from the Stanford University IRB due to its lack of identi able data and minimal risk with secondary data analysis from a previously collected sample and a public sample (please see attachment under supplementary les).

*Consent for publication
It was not possible to reach any individuals for consent due to the de-identi ed nature of our data.
The author(s) declared no potential con icts of interest with respect to the research, authorship, and/or publication of this article.
This study was approved by the IRB at Stanford University School of Medicine and at the Zhijin County Medical O ce.
All participants and authors have provided either written consent or verbal assent for publication. Data and mateiral are available for review upon request. This study did not receive any o cial research funding.
Dr. She contributed to the conception and design of the study, contributed to acquisition, analysis, and interpretation of data, drafted the manuscript, critically revised the manuscript for important intellectual content and gave nal approval for publication. Dr. Zhao substantially contributed to acquisition, analysis, and interpretation of data, critically revised the manuscript for important intellectual content and gave nal approval for publication. Dr. Li substantially contributed to the conception and design of the study, contributed to analysis and interpretation of data, critically revised the manuscript for important intellectual content and gave nal approval for publication.
We gratefully acknowledge the contribution and generosity of the teachers, principals and families who