This study used publicly available data from the Global School-based Student Health Survey (GSHS) among 29 countries and territories. The GSHS is a large and representative survey of students’ health behaviours and risk factors using a standardized two-stage probability sampling approach for participant selection. The survey was developed by the World Health Organization (WHO) and the US Centers for Disease Control and Prevention (CDC), and other United Nations (UN) allies and country-specific institutions (1, 15). The survey draws content from the CDC Youth Risk Behaviours Survey (YRBS), for which test–retest reliability has been established (16). Information, including the aims, methodology, and sampling procedure of the GSHS, is available at http://www.cdc.gov/gshs/. This study used the most recent data available in GSHS-participating countries and territories. Countries and territories with their most recent data released before 2015 were excluded from the analysis. Also, countries with recent datasets missing information on the variables of interest for the present study were excluded. By applying the inclusion and exclusion criteria, data from adolescents aged 11–18 years in 29 countries and territories were considered recent and eligible to be included in the analysis (see Table 1).

Data collection was conducted during a regular class period with the help of trained enumerators from local social work organizations and other research institutions. The GSHS questionnaire was close-ended, structured and developed in English but translated into the local language in each country. The questionnaire included multiple-choice questions. After informed consent was obtained from the students, parents, and school officials, the questionnaires were administered. The trained enumerators explained the aim of the study and the instructions in detail to the students before the questionnaires were distributed. In addition, the survey ensured that students’ privacy was protected through anonymous and voluntary participation. The study complied with the Declaration of Helsinki, and each country’s Institutional review boards or ethics committees reviewed and approved all GSHS surveys.

The respondent’s substance use was assessed with single items for tobacco, marijuana, cigarette, and alcohol use. The items included “During the past 30 days, how many days did you smoke cigarettes?” (0 days, 1 or 2 days, 3 to 5 days, 6 to 9 days, 10 to 19 days, 20 to 29 days and all 30 days), “During the past 30 days, on how many days did you use any tobacco products other than cigarettes, such as country specific examples?” (0 days, 1 or 2 days, 3 to 5 days, 6 to 9 days, 10 to 19 days, 20 to 29 days and, all 30 days), “During the past 30 days, how many times have you used marijuana?” (1 = 0 times, 2 = 1 or 2 times, 3 = 3–9 times, 4 = 10–19 times and 5 = 20 or more times), and “During the past 30 days, how many days did you have at least one drink containing alcohol?” (0 days, 1 or 2 days, 3 to 5 days, 6 to 9 days, 10 to 19 days, 20 to 29 days and all 30 days). All variables were re-coded on a dichotomous scale (Yes or No, where 0 times/days were considered as No and at least 1 day/time was considered Yes) as in other existing GSHS studies (17–19). We created a composite variable, “substance use,” by assigning a score of “1” each for “Yes” responses to smoking cigarettes, other tobacco, marijuana, and alcohol. A “0” score was assigned if the response was “No”. The score for each participant was added with a possible total score of 0–4. Participants who scored ≥1 were considered substance users.

The primary outcome of this study was a self-reported serious injury. This variable was assessed in response to the question, “During the past 12 months, how many times were you seriously injured?” The GSHS questionnaire defined serious injury as an injury that makes the respondent miss at least one full day of usual activities such as school, sports, or a job or requires treatment by a doctor or nurse. Response options to the question were 1 = 0 times; 2 = 1 time; 3 = 2 or 3 times; 4 = 3 or 4 times; 4 = 4 or 5 times; 5 = 6 or 7 times; 6 = 8 0r 9 times; 7 = 10 or 11 times; 8 = 12 or more times. For analytic purposes, the options were further categorized into 1 = 0 [No] and 2 to 12 or more times = 1 [Yes] for this study. The students whose response option was “0 times” showed that they had not sustained any serious injury, while the remaining response options meant that they had sustained one or more injuries in the 12 months preceding the survey (9, 20, 21). We dichotomized the variable to simplify the analysis by analyzing group differences rather than individual differences (22).

Perpetration of and victimization by interpersonal violence were the mediating variables in this study. Perpetration of interpersonal violence was measured as “During the past 12 months, how many times were you in a physical fight?” and victimization by interpersonal violence was assessed as “During the past 12 months, how many times were you physically attacked?” The two variables had the same response options including 1 = 0 times; 2 = 1 time; 3 = 2 or 3 times; 4 = 4 or 5 times; 5 = 6 or 7 times; 6 = 8 or 9 times; 7 = 10 or 11 times; and 8 = 12 or more times. For simplification and analytic purposes, the responses were further categorized into “No” for those who responded “0 times,” with the remaining responses as “Yes” in both variables Previous studies have used perpetration of and victimization by interpersonal violence to measure interpersonal violence (20, 23, 24).

The “svyset” command in Stata was used for the statistical analyses to adjust the complex sampling design employed by the GSHS survey using the weight, primary sampling unit (PSU), and stratum variables. Descriptive statistics were used to analyze and describe respondents’ characteristics stratified by substance use status. The chi-square test of independence was used to determine the association between substance use and background characteristics. Country-specific prevalence of substance use and injury was summarised using proportions and 95% confidence intervals. To calculate the overall prevalence of injury and substance use, we performed random-effects meta-analysis in stata. We investigated the between-country heterogeneity using the I-square ( I 2 ) statistic and the high heterogeneity ( I 2 >95%) was used to justify the random-effect modelling. Bivariate and multivariable logistic regression models were used to assess the influence of substance use on injuries, adjusting for other risk factors of injury. Odds ratios and 95% confidence intervals were reported. Variables with a value of p of 0.05 or less were considered statistically significant in the multivariable analysis. A path analysis using the structural equation model examined the potential mediation effect of perpetration of and victimization by interpersonal violence on the association between substance use and injury. Direct and indirect effects were tested, and the decomposition of total, direct, and indirect effects were specified (25). A sub-group analysis was also conducted to determine the associations between substance types and forms of injuries. This analysis employed a multinomial logistic regression analysis, with injury types as the outcome variable and substance types (tobacco, cigarette, marijuana, and alcohol) as the explanatory variables. Broken bone/dislocated joint was considered the base outcome, while concussion/head injury, gunshot wound, bad burn, and poison were the other outcomes used for the multinomial regression analysis. The data were analyzed from multiple countries jointly, and a country variable was included in the regression models to account for the joint analysis derived from multiple surveys.

Table 2 shows the background characteristics of the respondents by substance use. In general, the majority of respondents were below the age of 15. Females constituted a higher proportion at 52.0%. The majority of the respondents (78.6%) reported having three or more close friends. A significant percentage, 68.0%, engaged in physical activity, while 79.5% received support from peers. Approximately 45.4% experienced feelings of loneliness. Around one-third of respondents reported being bullied, 9.2% had a suicidal plan, 8.1% had attempted suicide, and roughly 9.6% had contemplated suicide. Aside from hunger, all other background characteristics (age, sex, number of close friends, physical activity, peer support, loneliness, worry, bullying, suicidal plan, suicidal attempt, and suicidal consideration) were significantly associated with substance use.

Figures 1, 2 present the country-specific prevalence of injuries and substance use, respectively. Overall, the injury prevalence was 41.7% (95%CI = 37.3, 46.1) among in-school adolescents and was significantly higher in males than females (47.3% vs. 36.4%, p = 0.001), respectively. The overall pooled prevalence of substance use was 33.6 (95%CI = 28.5, 38.6). Substance use was significantly higher among males than females (37.8% vs. 29.4%, p = 0.001), respectively.

Table 3 shows the influence of substance use on injuries among in-school adolescents. It also presents other risk factors associated with injuries. When unadjusted, the odds of injuries were 94% higher among in-school adolescents who used substances than their counterparts who did not use substances [OR = 1.94, 95%CI = 1.80, 2.09]. When the effect of substance use on injuries was adjusted for all other variables, the odds of injuries were 40% higher among in-school adolescents who used substances [AOR = 1.40, 95%CI = 1.31, 1.49].

Figure 3 presents a path analysis showing how perpetration of and victimization by interpersonal violence potentially mediates the effect of substance use on injuries. It revealed a positive direct effect of substance use on injuries [β = 0.10; 95%CI = 0.09, 0.11; p = 0.001]. Decomposition of the effects revealed a positive indirect effect of substance use on injuries running through perpetration of and victimization by interpersonal violence [β = 0.06; 95%CI = 0.05, 0.07; p = 0.001] with a total positive effect of both perpetration [β = 0.18; 95%CI = 0.16, 0.19; p = 0.001] and victimization on injuries [β = 0.22; 95%CI = 0.21, 0.24; p = 0.001]. This finding indicates a significant mediation effect of perpetration of and victimization by interpersonal violence on the association between substance use and injuries.

Table 4 presents a subgroup analysis showing the association between substance types and forms of injuries using multinomial logistic regression analysis. Compared to adolescents who do not use tobacco, those who used tobacco were 3.98 times more likely to sustain gunshot wound [OR = 3.98, 95%CI = 2.05, 7.75] rather than broken bone/dislocated joint. Also, compared to those who did not use cigarettes, adolescents who smoked cigarettes had 45% lower odds of sustaining a cut/stab wound rather than a broken bone/dislocated joint [OR = 0.55, 95%CI = 0.44, 0.68]. Moreover, compared to non-smokers of marijuana, adolescents who smoked marijuana had 2.81 higher odds of sustaining a gunshot wound [OR = 2.81, 95%CI = 1.29, 6.10] rather than broken bone/dislocated joint. Also, compared to non-users of alcohol, alcohol users were 25% less likely to sustain cut/stab wound [OR = 0.75, 95%CI = 0.62, 0.91], 53% more likely to sustain concussion/head injury [OR = 1.53, 95%CI = 1.19, 1.97], and two and half times more likely to sustain gunshot wound [OR = 2.50, 95%CI = 1.52, 4.12] rather than broken bone/dislocated joint.