Students were recruited from 21 primary and secondary schools located in Lushan County, Sichuan Province, where was hit by the 2013 Lushan Earthquake (Figure 1). The Lushan earthquake, placed at 7.0 on the Richter scale, occurred on April 20, 2013, resulting in 21 missing, 196 deaths, and 11,470 injured [40]. The epicenter located in Lushan County, Sichuan, near the Longmengshan Fault, which has the formation with the potential to cause super strong earthquakes [41].

Data collection was conducted at 1 year after the Lushan earthquake. We used a paper-and-pencil questionnaire and conducted the assessment by well-trained researchers with master’s degree in psychology in each class to supervise participants when they filled the questionnaire. Included participants were the students who consented to participate in the survey. Excluded individuals were those who refused to participate or hadn’t the ability to complete the investigation. In total, 2,917 children and adolescents agreed to participate. Due to that 33 students hadn’t completed the assessments, we finally included 2,887 students in our analysis. The response rate for the assessments was 98.97% (2,887/2,917). Details about sampling strategy showed in Figure 2.

We selected Baoxing County for two reasons: 1) there were 21 primary and secondary schools in this region, satisfying the needs for a large sample; and 2) the directors of the Education Bureau and teachers of these schools paid close attention to the mental health of students. They were willing to sustain this survey. The research protocol was approved by the Ethics Committee of the University of Sichuan and the Education Bureau of Baoxing Country in China and was consistent with the latest version of Helsinki Declaration [IRB approval number: 2017 (103)] and was consistent with the latest vision of Helsinki Declaration. In the present survey, we only attained informed consent from participants and teachers. Because in China, if the local education institution (e.g., the Education Bureau of Baoxing Country) supports a survey as a mental health service to the students, parental consent is not required [42]. It was also made clear to the students that they could choose whether to participate in the survey or not [40]. More detailed information was published in a previously paper [43].

Students completed the following self-report questionnaires in their own classroom: sociodemographic characteristics, Patient Health Questionnaire-9 (PHQ-9), Child and Youth Resilience Measure (CYRM), and two self-constructed questionnaires for trauma-exposure and sleep [44]. Before the investigation was conducted, the purpose and significance of the research were introduced and a psychologist helped interpret some easily confusing items, such as “I solve problems without drugs or alcohol” and “I try to finish what I start.” We provided more detailed information in Supplementary Table 1 (in the onlineonly Data Supplement).

A self-constructed questionnaire included 6 items used to assess trauma exposure severity: 1) Did your immediate family members injured in the earthquake? 2) Did your immediate family members die in the earthquake? 3) Was your house damaged in the disaster? 4) Was there anyone with you after the earthquake? 5) Did the earthquake cause severe property damage to your family? 6) Did you sleep in a tent rather than your own home last night? All answers were coded as binary variables (1 “Yes” or 0 “No”), and we calculated the sum score (range from 0 to 6) to reflect the severity of trauma exposure. The total score of the scale was used to assess the severity of the earthquake trauma exposure.

A self-constructed sleep questionnaire was used to assess insomnia and 3 items were included: 1) difficulty falling asleep, 2) difficulty staying asleep, and 3) early waking. All questions and answers were coded as binary variables (yes or no). If the student’s answers are all “yes,” then this student is considered to have insomnia symptoms.

The self-report symptom scale of the PHQ-9 is a widely used screening tool for depression according to the Diagnostic and Statistical Manual of Mental Disorder-IV (DSM-IV) criteria [45]. The scale consists of 9-items and the items are: Over the last two weeks, how often have you been bothered by any of the following problems? 1) Little interest or pleasure in doing things. 2) Felling down, depressed, or hopeless. 3) Trouble falling or staying asleep, or sleeping too much. 4) Feeling tired of having little energy. 5) Poor appetite or overeating. 6) Feeling bad about yourself—or that you are a failure or having let yourself or your family down. 7) Trouble concentrating on things, such as reading the newspaper or watching television. 8) Moving or speaking so slowly that other people could have noticed. Or the opposite—being so fidgety or restless that you have been moving a lot more than usual. 9) Thought that you would be better off dead, or of hurting yourself in some way. The total scores range from 0 to 27, and can be divided into 4 categories: no depression (≤4), minimal (5-9), mild (10-14), moderate (15-19), and severe (≥20). A cut-off point of 10 has been widely accepted for the diagnostic criteria of probable major depressive disorder (pMDD) in different setting [46]. The Chinese version of PHQ-9 has good reliability and validity [47], and in the present study, the Cronbach’s alpha value was 0.899.

The CYRM-12 is a self-report instrument to measure youth resilience. The specific items are 1) I have people I look up to. 2) Getting an education is important to me. 3) My parent(s)/caregivers(s) know a lot about me. 4) I try to finish what I start. 5) I solve problems without harming myself or others. 6) I know where to go in my community to get help. 7) I feel I belong(ed) to at my school. 8) My family will stand by me during difficult times. 9) My friends stand by me during difficult times. 10) I am treated fairly in my community. 11) I have opportunities to develop skills that will be useful later in life. 12) I enjoy my cultural and family traditions. Items are rated on a 5-point scale from 1=does not describe me at all to 5=describes me a lot. Higher scores suggest higher levels of resilience [44]. In the present study, the Cronbach’s alpha value was 0.907, suggesting good reliability.

Statistical analyses were accomplished by Mplus 7.0 (Linda Muthén & Bengt Muthén, Los Angeles, CA, USA). The number of missing responses on each item of PHQ-9 were 10, 14, 13, 14, 23, 14, 23, 14, 14, 12, and 17 respectively. As for CYRM-12, they were 55, 55, 58, 70, 61, 65, 61, 54, 61, 63, 66, and 63, respectively. Missing values were estimated with maximum likelihood procedures. Considering the difference in scale range between PHQ-9 and CYRM-12, all items’ scores were converted into standardized Z-scores. The 3-step LPA (R3STEP) approach was conducted via maximum likelihood estimation [48]. In the first step, the LPA model was conducted based on the PHQ-9 and CYRM-12 scale scores. In the second step, latent class membership was calculated according to the latent class posterior distribution. In the third step, trauma exposure, sex, age, ethnicity, and insomnia were treated as independent variables to predict latent class membership. Odd ratios (ORs) were calculated according to the beta values and standard errors.

The evaluation indexes of LPA model fitting degree are Akaike Information Criteria (AIC) values, Bayesian Information Criteria (BIC) values, adjusted BIC (aBIC), Entropy, Lo-Mendell-Rubin likelihood ratio test (LMR LRT) and Bootstrap-based Likelihood Ratio Test (BLRT) [49]. The smaller values of AIC, BIC and aBIC are, the better fit of the model is. Entropy is mainly used to evaluate the accuracy of classification, ranging from 0-1. The closer it is to 1, the higher accuracy of the classification is. The LMR LRT and BLRT indicators are used to compare the fit differences between the n and n-1 categories. For instance, an LPA model with 4 categories, the p-values for LMR LRT and BLRT are used to compare the differences in fit between the 4 categories and the 3 category models. If the p values of all LMR LRT and BLRT are significant, it means that the four categories of models fit better than the three categories; if the p values of LMR LRT and BLRT are not significant, it means that the three categories of models are better than the four categories. If the p values of BLRT and LMR LRT are inconsistent, the actual number of potential categories and the number of samples included in the category should be combined to determine the final number of potential categories.

In presented study, a great proportion of individuals were in the mild depression/high resilience group, which was in accordance with findings of Wenchuan earthquake’s youth sample [51]. We both indicated that most individuals obtained positive psychological changes following the earthquake. Based on a previous longitudinal research, resilience does indeed buffer the negative influences of stressful life events over time [52]. As the positive psychology emphasizes, we should focus on the positive aspects (e.g., resilience and post-traumatic growth) of traumatic experience, not only on the negative side [53,54]. Our results highlight the potential beneficial influences from the traumatic events.

An interesting phenomenon of this study also sheds light on the controversial relationship between depression and resilience. Negative relationship between those two constructs is shown in the mild depression/high resilience and severe depression/low resilience subgroups. While, as for the severe depression/high resilience subgroup, depression and resilience levels were reflective of each other. It’s presented a positive association between depression and resilience. These findings indicate complicated reactions among youth survivors, and individual’s heterogeneity should be considered.

The higher level of trauma exposure was related to the higher probability of being classified into the latent class with a high level of depression (severe depression/high resilience and severe depression/low resilience). A link had been found between trauma exposure and depression, such that individuals who witness the disaster, death/injuries of family members and damage to one’s home are being more prone to depression [42,55]. Meanwhile, other studies also had found that a higher level of trauma exposure was related to a higher level of resilience. When youth survivors have been subject to a traumatic event (such as an earthquake), the experience has led to a positive adjustment [56].

The role of demographic characteristics was also explored. Compared to their counterparts, male and younger survivors were more likely to enter mild depression/high resilience. Female and male’s post-traumatic reactions are markedly different. For example, greater levels of self-esteem and mastery, more common among the male than the female, are positively associated with problem-focused coping, which could reduce the negative impacts of stress [57]. In contrast, females use more emotion- focused coping strategies [58], which may make them at risk of trapping in depression. In addition, females more frequently express fear and shame than do male [59,60]. Fear and shame can cause survivors to have a negative outlook on their surroundings, namely, low resilience. Low resilience can make negative reappraisals of trauma-related cues and considered the trauma as more threatening [61]. For youth survivors, they may attain more support from family and society, which are resources related to better mental health. As a type of stress buffer, instrumental aid, active assistance or emotional support from parents or similar others could help to alleviate the impacts of traumatic events [62]. The previous study also emphasizes the role of supportive relationships with caregivers at preventing children from post-traumatic psychopathology, which could help children dampening the threat processing [63].

The roles of sleep characteristics were also clarified. The results showed that insomnia individuals were likely to enter the severe depression/low resilience subgroup. Seelig’s study found that poor sleep is an important factor and related to the construct of resilience [64]. Sleep and resilience can share an underlying neural mechanism and therefore affect one another. Sleep impacted levels of resilience, meaning that greater sleep disturbance could reduce resilience and may potentially predispose individuals to psychopathology such as depression [38].

The current study also provided potential clinical implications. First, based on the person-centered method LPA, our study promotes the understanding of different psychological reactions (depression vs. resilience) as well as individual heterogeneity among youth survivors. Most individuals were in the mild depression/high resilience group, suggesting that resilience may buffer the influence of trauma experience on depression. Thus, in future, we can provide many types of online psychological services for youth survivors to promote their psychological resilience. Second, in our study, we found some important moderating factors (e.g. sleep disturbance) that can influence the relationship between resilience and depression, and indicated more attention should be paid to older insomnia female individuals. As posited by previous research, the presentation of mental disorders are influenced by culture and social milieu [65], and Chinese population is more prone to report more somatic symptoms instead of depression symptoms [66]. Considering sleep disturbance is a relative modifiable factor and more accept mental disorder among Chinese general population. Thus, we can intervene on sleep symptoms to mitigate the outcomes of depressive symptoms [67,68]. Third, it should be aware that our study was conducted at 1-year after the earthquake, which assessed the long-term psychological effect of earthquake exposure. In consistent with Sharma’s study on the Nepal Earthquake [69], we both found a considerable prevalence of depression among youth survivors even after one year of the event. Thus, there is a need for early intervention for youth survivors and we should also pay more attention to the longlasting effects of traumatic event.

Several limitations of our study should be mentioned. First, the findings of our study are based on the self-report scales (e.g., PHQ-9) reported rather than the clinic structure interviewed. And the design of our study was just a cross-sectional study. High quality prospective studies are needed. Second, our research targets on the sample of Chinese youth-earthquake survivors after 1-year exposure, which limits the generalization of the findings. Further studies should examine the profile of depression and resilience in various types of traumatic events among different time points to test the robustness of the model. Third, we only focused on depression in this paper. Other important mental health outcomes (e.g., PTSD, generalized anxiety disorder) are relevant to the trauma events and should not be ignored in further study. Forth, whether resilience is a personality trait or dynamic process is controversial in different studies. In future research, we can evaluate positive psychological through the post-traumatic growth (PTG).