Abstract

Background

Adolescent health-risk behaviors are prevalent and tend to co-occur. This study aimed to identify patterns of health-risk behaviors among Chinese adolescents during the COVID-19 pandemic and explore the effects of individual and social factors on health-risk patterns.

Methods

This cross-sectional study investigated 1607 adolescents from four high schools in 2021 through stratified cluster random sampling. Latent class analysis was conducted to identify patterns of health-risk behaviors and logistic regression was used to examine the risk and protective factors of latent class membership.

Results

Four latent classes were identified: 鈥淟ow risk鈥� (81.6%), 鈥淧roblematic Internet use鈥� (7.8%), 鈥淎lcohol use鈥� (8.5%), and 鈥淗igh risk鈥� (2.1%). Relative to the 鈥淟ow risk鈥�, adolescents with higher levels of sensation seeking, deviant peer affiliation, and childhood abuse were more likely to be assigned to the 鈥淧roblematic Internet use鈥� class, while those with high degrees of parental monitoring and school connectedness were less likely to be in the 鈥淧roblematic Internet use鈥� class. Those with higher levels of sensation seeking and deviant peer affiliation, lower scores of parental monitoring and school connectedness were more likely to be assigned to the 鈥淎lcohol use鈥� class, compared to the 鈥淟ow risk鈥�. Students in the 鈥淗igh risk鈥� class were more likely to report higher levels of sensation seeking, deviant peer affiliation, and childhood abuse, but lower degrees of parental monitoring and school connectedness than the 鈥淟ow risk鈥� class.

Conclusions

This study identified patterns of multiple risk behaviors among Chinese high school students during the COVID-19 pandemic and found that multi-level individual and social factors affected latent classes of adolescent health-risk behaviors. These findings provide clues for designing effective interventions to reduce health-risk behaviors among adolescents.

Introduction

Health-risk behaviors in adolescence

Adolescence is a critical developmental stage characterized by heightened vulnerability to engaging in health-risk behaviors including bullying, smoking, alcohol consumption, and early sexual activity [1, 2]. Health-risk behaviors are prevalent among Chinese adolescents. For example, a study with a large sample size (n鈥�=鈥�27019) conducted in 2022, in Zhejiang Province, China documented that 3.9% and 16.0% of adolescents smoked cigarettes and consumed alcohol in the past 30 days, and 13.7% of adolescents reported they had been involved in a physical fight within the past 12 months [3]. A cross-sectional survey found that 29.9% of Chinese adolescents met the criteria for possible problematic Internet use [4]. Also, a multicenter survey revealed that 9.75%, 10.57%, 15.17% of Chinese adolescents engaged in skipping school, running away from home, and fighting, respectively [5]. Further, previous studies have revealed that health-risk behaviors can lead to elevated risks of suicidality [6], anxiety and depression [7].

The COVID-19 pandemic introduced unprecedented disruptions to the lives of adolescents, as in-person schooling was replaced with online learning. This significant shift may be considered a traumatic event, potentially amplifying health-related risk behaviors, such as tobacco use, alcohol consumption, and increased electronic screen time [8, 9]. Therefore, it is essential to investigate adolescents鈥� health-risk behaviors during the COVID-19 pandemic, and elucidating the key determinants is informative for developing prevention and intervention strategies.

Patterns of adolescent health-risk behaviors

Previous studies typically combine health-risk behaviors into a composite score assuming equal weights or explore health-risk behaviors separately ignoring the co-occurrence and heterogeneity based on variable-centered approaches [10, 11]. Problem Behavior Theory developed by Jessor [12] noted that co-occurring risk behaviors were prevalent in adolescence. Latent class analysis (LCA), a person-centered method, has an advantage in exploring the clustering of risk behaviors [13]. To date, several studies have discovered distinctive subgroups of health-risk behaviors based on LCA. For instance, a survey investigated health-risk behaviors among Thai secondary school students and found three latent classes, including the low-risk (88%), moderate-risk (11%), and high-risk classes (0.6%) [14]. Based on UK Millennium Cohort Study (n鈥�=鈥�17223), Picoito and colleagues also explored adolescent substance use and antisocial behavior (e.g., smoking, drinking, cannabis use, physical fighting, shoplifting, vandalism, and graffiti), and identified four latent classes, including the 鈥渘ormative鈥� (71.8%), 鈥渁lcohol and physical fighting鈥� (15.3%), 鈥渁lcohol and tobacco鈥� (9.9%) and 鈥淧oly-substance use and antisocial behaviors鈥� classes (3.0%) [15]. Although the aforementioned studies have identified patterns of health-risk behaviors among adolescents, they ignore other prevalent health-risk behaviors in modern society, such as problem Internet use. Thus, the current study offers a more comprehensive understanding of patterns of adolescents鈥� multiple risk behaviors in Chinese context during the COVID-19 pandemic.

Impacts of multi-level factors on adolescent health-risk behaviors

According to the Ecological Theory [16], individual factors and multiple social contexts play crucial roles in individuals鈥� health. At the individual level, sensation seeking refers to a personality trait characterized by a willingness to take risks in response to challenges, such as novel and complex experiences and strong emotions [17]. According to the Sensation Seeking Theory [18], high-sensation-seeking individuals tend to enjoy novel experiences and taking risks. A prior study found that urban and male adolescents as well as adolescents in non-intact families reported more health-risk behaviors [19]. In addition, a systematic review revealed that low parental education was a risk factor of drug abuse among adolescents [20]. Also, the prevalence of adolescent health-risk behaviors differed by age and grade level in school [21]. Therefore, guided by previous research, some socio-demographic variables concerning sex, grade, living places, age, maternal education, paternal education and family structure were adjusted in the current study.

Further, adolescents鈥� behaviors are embedded within social contexts, such as families, peers, and schools [22]. At the family level, childhood abuse is a crucial contributor to the prevalence of health-risk behaviors in adolescents [23]. Household dysfunction, a significant indicator of adverse childhood experiences, includes household mental illness, household substance abuse, household poverty, etc [24]. For example, adolescents exposed to parental mental illness and/or substance use disorder confront an elevated vulnerability to report substance use disorder [25]. Parental monitoring refers to parental awareness, watchfulness, and supervision of adolescent activities in multiple domains (i.e., friends, school, and behaviors at home), and communication to the adolescent that the parent is concerned about, and aware of those activities [26]. Parental monitoring can reduce adolescent sexual behaviors, substance use, and violence [27].

Adolescence as the stage when a second process of separation individuation takes place, during which they tend to individuate away from their family while becoming more susceptible to peer influence [28]. Adolescents prefer selecting peers who are similar to themselves in important ways, due to homophily selection [29]. Both cross-sectional and longitudinal studies have identified deviant peer affiliation as a robust predictor of adolescent problem behaviors [30, 31]. At the school level, the connection to school environment is a particularly protective factor against health-risk behaviors [32]. Prior research explored two dimensions of school connectedness and found that teacher support protected against the initiation of health-risk behaviors [33].

While some literature has demonstrated the association between individual, family, peer, school factors and adolescents鈥� involvement in risk behaviors, the distinct effects of multiple contexts on patterns of health-risk behaviors remain less well understood. Therefore, this study aims to address two research questions: (1) Are there distinct patterns or classes of adolescent health-risk behaviors? (2) How do individual, peer, family, and school factors simultaneously affect those latent classes?

Methods

Study design and participants

The following formula is used to calculate the sample size.

$$\:\text{n}={\text{Z}}_{1-{\upalpha\:}/2}^{2}\times\:\left[\text{P}\times\:\left(1-\text{P}\right)\right]/{\text{d}}^{2}$$

伪鈥�=鈥�0.05, $\:{\text{z}}_{1-{\upalpha\:}/2}$=1.96, $\:\text{d}=0.1\text{p}$, p is expected prevalence of health-risk behaviors among Chinese adolescents. Based on a survey conducted in eight provinces in 2021, 22.2% of Chinese adolescents engaged in high-risk behaviors [34]. The formulae yielded an initial sample size of 1346 students.

Four senior high schools were chosen to recruit participants in Shangqiu city in central China from May to June 2021. With the assistance of teachers, four or five classes were selected randomly from Grades 10 and 11 in each school. This study did not investigate adolescents from grade 12, as Chinese national college entrance exam is typically held on June 7 and 8. All students in the selected classes were invited to complete paper-based questionnaires. To gain rapport with participants, an orientation regarding data anonymity, objectives, and significance of the study was conducted. A total of 1650 questionnaires from 33 classes were collected. Participants with missing information on key variables (n鈥�=鈥�43) were excluded, and 1607 respondents were eligible for final analysis.

Measures

Health-risk behavior

Ten dichotomous questions (Yes or No) were used to measure whether adolescents engaged in following behaviors in the last 6 months, including skipping school, carrying weapons, engaging in fights, smoking cigarettes, drinking alcohol, cheating in tests, problematic Internet use, running away from home, vandalism, and sexual behavior, which were treated as observed indicator variables to identify latent classes. Response of 鈥淵es鈥� means that the student had engaged in the behavior one or more times in the in the last 6 months.

Sensation seeking

Sensation seeking was measured using the 8-item Brief Sensation Seeking Scale for Chinese (BSSS-C) (e.g., 鈥淚 get restless if I do the same thing for a long time鈥�) [35], which had adequate reliability and validity in Chinese adults (Cronbach鈥檚 伪 鈥�=鈥�0.90, Comparative Fit Index鈥�=鈥�0.98, Standardized Root Mean of Residuals鈥�=鈥�0.03). Each item was assessed on a 5-point Likert scale (1鈥�=鈥塻trongly disagree, 2鈥�=鈥塪isagree, 3鈥�=鈥塶either agree nor disagree, 4鈥�=鈥塧gree, 5鈥�=鈥塻trongly agree). The higher composite score demonstrated higher sensation seeking. In the current study, the internal consistency was 0.71.

Deviant peer affiliation

15 items adapted from the National Youth Survey were used to estimate the affiliation with deviant peers in the last year [36], which has been validated in Chinese adolescents (Cronbach鈥檚 伪 was over 0.85) [37]. Sample items included 鈥淗ow many of your friends engaged in a fight in the last year?鈥� 鈥淗ow many of your friends smoked cigarettes in the last year?鈥� Responses were rated on a 3-point Likert scale (1鈥�=鈥塶one, 2鈥�=鈥塻ome, 3鈥�=鈥塵ost). A composite score was calculated, and higher scores suggested more deviant friends. In the current study, the internal consistency was 0.86.

Parental monitoring

The 8-item Parental Monitoring Scale was used to assess parental monitoring (e.g., 鈥淚f I am going to be home late, I tell my parents/guardian鈥�) [38], with a good internal consistency in a previous study (Cronbach鈥檚 伪鈥�=鈥�0.86) [1]. Responses ranged from 1 (Never) to 5 (Always). Higher composite scores represented more stringent parental monitoring. In the present study, the internal consistency was 0.85.

School connectedness

Based on previous tools [39, 40], 10 items were used to assess school connectedness, including teacher support (3 items), school belonging (3 items), and classmate support (4 items), which has been validated in Chinese high school students [41]. Sample items included 鈥淭he teachers at this school treat students fairly?鈥� Items were scored on a 5-point Likert scale ranging from 鈥渟trongly agree鈥� (1) to 鈥渟trongly disagree鈥� (5). Higher scores reflected greater school connectedness. In the current study, the internal consistency was 0.86.

Childhood abuse

Childhood abuse was assessed based on the Childhood Trauma Questionnaire-Short Form (CTQ-SF) [42], and it showed good reliability and validity in the Chinese adolescents (Comparative Fit Index鈥�=鈥�0.91, Tucker-Lewis Index鈥�=鈥�0.90, Root Mean Square Error of Approximation鈥�=鈥�0.06, Cronbach鈥檚 伪鈥�=鈥�0.87) [43]. This study used 15 items to measure emotional abuse (e.g., 鈥渃alled stupid, lazy or ugly by family鈥�), sexual abuse (e.g., 鈥渨as molested鈥�), and physical abuse (e.g., 鈥渨as hit hard by family鈥�). The response options ranged from 1 (never) to 5 (always). The Cronbach鈥檚 伪 was 0.78 in the current study.

Household dysfunction

Questions about household dysfunction came from Centers for Disease Control and Kaiser ACE Study [44]. Household dysfunction was evaluated by the endorsement of the following six experiences (Yes or No) during childhood: lived with anyone who had a problem with alcohol, drug or gambling, parental separation or divorce, witnessed domestic violence, mental illness in household, incarcerated household members, and family financial difficulties. Adolescents were defined as exposed to household dysfunction if they responded 鈥淵es鈥� to any item.

Other factors were also collected including sex (male or female), age (years), grade (10 or 11), living places (urban or rural), maternal and paternal education (elementary school or below, junior high school, senior high school, college or above), and family structure (intact family or non-intact family).

Statistical analysis

First, the categorical variables were summarized as percentages, and continuous variables were expressed as means and standard deviations. Next, LCA with two through five latent classes were conducted using observed dichotomous variables (10 health-risk behaviors listed in Table听1). Model fit was assessed using a combination of Akaike information criterion (AIC), Bayesian information criterion (BIC), sample-size adjusted BIC (aBIC), Entropy, the Lo-Mendell-Rubin likelihood ratio (LMRLR), the adjusted Lo-Mendell-Rubin likelihood ratio (aLMRLR), the Bootstrap Likelihood Ratio Test (BLRT), class size, as well as interpretability [45]. Lower AIC, BIC, aBIC indicate superior fit, and small changes of AIC, BIC, aBIC with increased classes can also be considered as a criterion. Entropy ranged from 0 to 1, with values closer to 1 suggesting greater classification accuracy [46]. Entropy of 0.8 indicates approximately 90% correct group assignment [47], and entropy values above 0.8 are considered acceptable. There were no widely accepted standards for determining the number of samples in each class. Typically, researchers recommended that each class should comprise at least 50 cases and represent no less than 5% of the overall sample [48]. Nonetheless, some studies have incorporated classes with sizes below this 5% threshold or fewer than 50 cases [49]. When determining if a class size is too small, it鈥檚 crucial to examine whether the model fit statistics support the selected model and whether the small class is conceptually meaningful. Additionally, researchers must take into account the total sample size when evaluating the appropriateness of each class size [50]. Significant p-values of LMRLR, aLMRLR, and BLRT indicated that K classes was better than K-1 classes. After we identified the optimal model, students were assigned to each latent class based on the highest posterior probability. Finally, multinomial logistic regression was used to examine the risk and protective factors of latent class membership. LCA was carried out in Mplus 8.0. Other analyses were performed in SPSS 26.0. A two-sided p鈥�<鈥�0.05 was considered statistically significant.

Table 1 Descriptive statistics (n鈥�=鈥�1607)

Results

Descriptive statistics

Table听1 describes the key study variables. The average age was 16.3 years, 53.1% of the sample were male, half of participants came from rural (50.0%), and 46.4% were in grade 10. The majority were from intact families (93.0%), and 27.8% of students experienced household dysfunction during childhood. Problematic Internet use (23.9%) was the most common health-risk behavior in the last 6 months, while vandalism (1.9%) and sexual behavior (2.4%) were the least prevalent.

Patterns of adolescent health-risk behaviors

As shown in Table听2, the improvement in AIC was minimal after the 4-class model. The 3-class and 4-class models had the lowest BIC and aBIC, respectively. LMRLR and aLMRLR in the 5-class model with non-significant p-value, indicating that the 4-class model was more favorable than 5-class model. The entropy value of 4-class model was 0.819, indicating an acceptable level of class separation. While the smallest class consisted of 2.1% of adolescents in the 4-class model, the four identified classes represented empirically significant patterns of adolescent risk behaviors, each with plausible interpretations. Hence, based on a combination of model fit indices, the 4-class solution was selected as the optimal model.

Table 2 Model selection statistics of latent class analysis (n鈥�=鈥�1607)

Four latent classes were identified: 鈥淟ow risk鈥� (81.6%), 鈥淧roblematic Internet use鈥� (7.8%), 鈥淎lcohol use鈥� (8.5%), and 鈥淗igh risk鈥� (2.1%). The 鈥淟ow risk鈥� class was the largest one, marked by low probabilities of all health-risk behaviors. Individuals in the 鈥淧roblematic Internet use鈥� class had the highest probability of problematic Internet use (89.7%), but relatively low probabilities of other health-risk behaviors. Adolescents in the 鈥淎lcohol use鈥� class were marked by a high probability of drinking alcohol (80.9%), but relatively low endorsement of other health-risk behaviors. Members of the 鈥淗igh risk鈥� subgroup reported relatively high probabilities of involvement in seven of ten health-risk behaviors (>鈥�50%), and probabilities of carrying weapons, vandalism and sexual behavior were higher than other classes (Table听3; Fig.听1).

Table 3 Four-latent-class model of health-risk behavior (n鈥�=鈥�1607)

Fig. 1

Patterns of health risk behaviors among Chinese adolescents

The risk and protective factors of latent class membership

As displayed in Table听4, in the logistic regression, the 鈥淟ow risk鈥� class was designated as the reference group. Relative to the 鈥淟ow risk鈥�, adolescents with higher levels of sensation seeking, deviant peer affiliation, and childhood abuse were more likely to be assigned to the 鈥淧roblematic Internet use鈥� class, while those with high degrees of parental monitoring and school connectedness were less likely to be in the 鈥淧roblematic Internet use鈥� class. Likely, those with higher levels of sensation seeking and deviant peer affiliation, lower scores of parental monitoring and school connectedness were more likely to be assigned to the 鈥淎lcohol use鈥� class, compared to the 鈥淟ow risk鈥�. Additionally, males were associated with increased odds of being in the 鈥淎lcohol use鈥� and 鈥淗igh risk鈥� classes, relative to the 鈥淟ow risk鈥�. Moreover, students in the 鈥淗igh risk鈥� class were more prone to report higher levels of sensation seeking, deviant peer affiliation, and childhood abuse, but lower degrees of parental monitoring and school connectedness than the 鈥淟ow risk鈥� class.

Table 4 Adjusted odds ratios for factors predicting latent class membership (Reference鈥�=鈥塋ow risk, n鈥�=鈥�1599)

Discussion

This study explored patterns of health-risk behaviors and contributed to the existing literature by analyzing the multi-level determinants of these patterns among Chinese adolescents through LCA. The current research identified four distinct subgroups: Low Risk (81.6%), Problematic Internet use (7.8%), Alcohol Use (8.5%), High Risk (2.1%) and found that individual, peer, family, and school factors significantly predicted latent patterns. This survey offers a valuable foundation for understanding the types and influencing factors of adolescents鈥� multiple risk behaviors, serving as a practical basis for future research and intervention development.

Congruent with previous research [51, 52], the findings suggested that adolescent health-risk behaviors were multiple and co-occurring. Prior surveys typically found three or four latent classes, commonly one class with relatively low odds of health-risk behaviors, and another one with high endorsement of all health-risk behaviors [14, 53]. Compared to previous literature, the current study identified a novel class (Problematic Internet use) marked by a high likelihood of problematic Internet use. Due to low self-control or self-discipline of adolescents, they may be particularly vulnerable to problematic Internet use in times of COVID-19 [54]. The lockdown measures and consequent lack of social interaction may increase the opportunities for prolonged and intensified using of the Internet, and the surge in adolescent problematic Internet use is a growing concern [55]. Problematic Internet use was the most prominent health-risk behavior in this study (23.9%), indicating that developing effective prevention of problematic Internet use is urgent. Therefore, moderate usage of the Internet should be promoted for adolescents. Also, parents need to improve the communication and monitoring of their child鈥檚 Internet behaviors.

Moreover, a vast number of factors can predict patterns of health-risk behaviors among Chinese adolescents at an individual, peer, family, and school level, which supports the view of the ecological framework that risk and protective factors interact at the micro, mezzo, and macro levels [16]. At the individual level, boys are more likely to belong to 鈥淎lcohol use鈥� and 鈥淗igh risk鈥� compared with girls, which is consistent with previous studies [1]. In line with traditional Chinese social and gender roles, risk-taking among females is viewed less favorably than among males. Besides, high-sensation-seeking adolescents are more likely to belong to the 鈥淧roblematic Internet use鈥�, 鈥淎lcohol use鈥� and 鈥淗igh risk鈥� groups. The positive association between sensation seeking and health-risk behaviors in Chinese context has been well documented [13]. Some scholars hold the view that the occurrence of risk behavior in adolescence was related to sensation seeking, convention challenging, and maturity demonstration, which will be manifested in higher rates of impulsive and health-risk behaviors [12, 18].

Consistent with empirical studies [56, 57], the finding suggested that deviant peer affiliation was associated with an increased adolescents鈥� engagement in health-risk behaviors. Adolescents spend increasing amounts of time with peers, and adolescence is a developmental period characterized by the desire to behave in ways of their friends [58]. Cambron and colleagues found that poor peer interactions independently predicted smoking and drinking behaviors in early adolescence [59]. Also, the Group Socialization Theory emphasized the external environment, especially peer groups, played a key role in environmental adaptation and behavior formation for children [60]. Understanding how deviant peer affiliation influence adolescent behaviors has vital implications for reducing their behavioral problems.

In terms of family level, the results revealed that childhood abuse and parental monitoring acted as significant predictors of adolescents鈥� behaviors. As noted by a study in 34 Quebec high schools, sexual abuse significantly predicted health-risk behaviors, including alcohol abuse, cannabis abuse, and delinquency [61]. Childhood abuse is related to widespread abnormalities in brain structure and function, which further leads to an increased susceptibility to a variety of health-risk behaviors [62]. Parental substance abuse, parental separation/divorce (indicators of household dysfunction) were associated with increased odds of smoking in U.S. adults after controlling for important confounders [63]. Parental monitoring was a vital contributor to patterns of adolescents鈥� multiple risk behaviors in The Bahamas [1]. Parental knowledge of students鈥� whereabouts, companions, and activities can prevent opportunities for involvement in risk behaviors or spending time with peers who might promote such behaviors [64]. These findings highlight the significance of reducing childhood abuse and fostering parental monitoring to support healthy adolescent development.

Adolescents in the problematic Internet use, alcohol use and high risk classes reported low levels of school connectedness than those in the low risk class. A survey documented that school connectedness reduced involvement in problematic Internet use among Chinese students [65]. Data from the 2021 nationally representative Youth Risk Behavior Survey showed that school connectedness was negatively related to risk behaviors among U.S. high school students [66]. A longitudinal study also found the protective effects of school connectedness on alcohol use among adolescents [67]. This study emphasize school connectedness will protect against multiple adolescent health risks, and school-based strategies should promote safe and supportive environments for students.

During the time of our data collection, the COVID-19 pandemic caused widespread disruptions to school operations and increased stress and trauma for some adolescents. These results have crucial implications for prevention and intervention of adolescent risk behaviors in the context of a pandemic and increased adversity. First, fostering collaboration among individuals, peers, families, and schools is essential for reducing adolescent risk behaviors. Comprehensive, multi-component interventions are likely to be more effective than single-component interventions. Second, targeted interventions and professional support should be prioritized for vulnerable adolescents, particularly those exposed to factors such as childhood abuse, etc.

Limitations

The following limitations should be noted in this study. First, all adolescents were from a single city in China and the sample size was relatively small, which may limit the generalizability of these findings. Second, as the questionnaire included sensitive topics, peer, family, and school variables were assessed from the adolescents鈥� perspectives. This approach may introduce social desirability bias, leading to potential underestimate of risk behaviors. Third, the cross-sectional design of this survey did not allow for causal inferences among the variables. Fourth, this study was conducted in only four senior high schools focusing on grades 10 and 11, which may introduce selection bias. Fifth, this study was conducted in 2021 during the COVID-19 pandemic, thus, the results and implications may differ from findings obtained before and after the pandemic. Nevertheless, these findings, despite being based on older data, provide valuable insights for developing prevention and intervention strategies for adolescents in the context of a pandemic and increased adversity. Therefore, to untangle these limitations, future studies should employ longitudinal designs using nationally representative samples, and data from parents, peers, and teachers should be added.

Conclusions

This study identified patterns of multiple health-risk behaviors among Chinese high school students and found that multi-level individual and social factors affected latent classes of adolescent health-risk behaviors. These findings may provide clues for designing effective interventions to reduce health-risk behaviors among adolescents.