Adolescent suicidal behaviours in 32 low- and middle-income countries

Abstract Objective To estimate prevalence of suicidal ideation and suicidal ideation with a plan in each surveyed country and to examine cross-national differences in associated risk factors. Methods We analysed data of students aged 13–17 years who participated in the 2003–2012 Global School-based Health Surveys in 32 countries, of which 29 are low- and middle-income. We used random effects meta-analysis to generate regional and overall pooled estimates. Multivariable logistic regression was used to estimate risk ratios for the associated risk factors. Population attributable fractions were estimated based on adjusted risk ratios and the prevalence of the determinants within each exposure level. Findings Across all countries, the pooled 12-month prevalence of suicide ideation were 16.2% (95% confidence interval, CI: 15.6 to 16.7) among females and 12.2% (95% CI: 11.7 to 12.7) among males and ideation with a plan were 8.3% (95% CI: 7.9 to 8.7) among females and 5.8% (95% CI: 5.5 to 6.1) among males. Suicide ideation in the WHO Region of the Americas was higher in females than males, with an estimated prevalence ratio of 1.70 (95% CI: 1.60 to 1.81), while this ratio was 1.04 (95% CI: 0.98 to 1.10) in the WHO African Region. Factors associated with suicidal ideation in most countries included experiences of bullying and physical violence, loneliness, limited parental support and alcohol and tobacco use. Conclusion The prevalence of adolescent suicidal behaviours varies across countries, yet a consistent set of risk factors of suicidal behaviours emerged across all regions and most countries.


Introduction
Worldwide, suicide accounts for an estimated 6% of all deaths among young people. 1 As the second leading cause of mortality among females and the third leading cause among males aged 10-24 years, youth suicide is a major global public health concern. 1 Low-and middle-income countries are home to more than 90% of the world's children and youth and also account for over 75% of global suicide deaths. 2 However, compared to high-income countries relatively little is known about the epidemiology of adolescent suicide and suicidal behaviours in low-and middle-income countries Suicidal behaviours include ideation (thinking about killing oneself), planning suicide, attempting suicide and suicide itself. 2 Suicidal ideation often emerges in adolescence and is prevalent among this age group, particularly among females. 3 Across 17 European countries, the lifetime prevalence of ideation among students aged 15-16 years ranged from 15% (Armenia) to 31.5% (Hungary), while the lifetime prevalence of suicide attempts ranged from 4.1% to 23.5% in the same two countries, respectively. 4 Across 49 low-and middle-income countries, 15.3% of adolescents aged 13-15 years had seriously considered suicide in the past year. 5 Given that suicide ideation strongly and prospectively relates to suicide attempts and suicide, 6,7 identifying potentially modifiable risk factors is essential for preventing these deaths.
Major risk factors for youth suicidal behaviours include being female, exposure to bullying and violence, alcohol and drug use, mental disorders and weak family and peer relationships. 4,8,9 While much of this evidence comes from Europe and North America, recent research has expanded the knowledge of the determinants of youth suicidal behaviours in several low-and middle-income countries. Many factors associated with youth suicidal behaviours in low-and middle-income countries overlap with established risk factors from high-income countries, including bullying, 10,11 physical and sexual abuse, 10,12,13 mental disorders and depressive symptoms, 10,[13][14][15][16] substance use, 10,14 and weak family and social relationships. 14 However, research in some low-and middle-income countries suggests that gender and common mental health problems contribute less to suicidal behaviours. 14,17,18 While studies from individual countries have provided insights about youth suicidal behaviours, differences in variable definitions and measures, study populations and analytical approaches make it difficult to compare the prevalence of and risk factors for youth suicidal behaviours across different settings.
The Global School-Based Health Survey (GSHS) has been conducted in over 80 countries worldwide and aims to provide comparable data on the health of adolescents aged 13-17 years 19 . GSHS data have been used to show crossnational variation in the prevalence of adolescent suicide ideation, 5 examine suicide ideation in relation to psychosocial distress in seven African countries 20 and study adolescent suicidal behaviours in individual low-and middle-income countries. 10,15,18,21,22 In the present study, we estimate the crossnational prevalence of suicidal ideation and ideation with planning and examine correlates of these outcomes in a large and diverse sample of countries.

Data source
GSHS is a self-administered, school-based survey developed by the World Health Organization (WHO) and the United States Centers for Diseases Control and Prevention, in collaboration with the United Nations Children's Emergency Fund, the United Nations Educational, Scientific and Cultural Organization and the Joint United Nations Programme on HIV/AIDS. 19,23 The survey uses standardized school-based sampling and a set of core questionnaire modules that address Objective To estimate prevalence of suicidal ideation and suicidal ideation with a plan in each surveyed country and to examine crossnational differences in associated risk factors. Methods We analysed data of students aged 13-17 years who participated in the 2003-2012 Global School-based Health Surveys in 32 countries, of which 29 are low-and middle-income. We used random effects meta-analysis to generate regional and overall pooled estimates. Multivariable logistic regression was used to estimate risk ratios for the associated risk factors. Population attributable fractions were estimated based on adjusted risk ratios and the prevalence of the determinants within each exposure level. Findings Across all countries, the pooled 12-month prevalence of suicide ideation were 16.2% (95% confidence interval, CI: 15.6 to 16.7) among females and 12.2% (95% CI: 11.7 to 12.7) among males and ideation with a plan were 8.3% (95% CI: 7.9 to 8.7) among females and 5.8% (95% CI: 5.5 to 6.1) among males. Suicide ideation in the WHO Region of the Americas was higher in females than males, with an estimated prevalence ratio of 1.70 (95% CI: 1.60 to 1.81), while this ratio was 1.04 (95% CI: 0.98 to 1.10) in the WHO African Region. Factors associated with suicidal ideation in most countries included experiences of bullying and physical violence, loneliness, limited parental support and alcohol and tobacco use. Conclusion The prevalence of adolescent suicidal behaviours varies across countries, yet a consistent set of risk factors of suicidal behaviours emerged across all regions and most countries.
leading causes of morbidity and mortality worldwide, including alcohol and drug use, mental health, violence and unintentional injury and sexual behaviours. Some questions include countryspecific examples, options or phrasing to facilitate adaptation of the surveys across diverse global populations. Many questions were adopted from the Youth Risk Behaviour Survey of American Adolescents, for which reliability has been formally evaluated. 24 Reliability studies of the GSHS in low-and middleincome settings are limited, however one study among Fijian girls found high test-retest reliability of the two GSHS items on suicidal behaviours -suicidal ideation and ideation with a plan -with both showing agreement above 90% and kappa coefficients above 0.63. 25 We included countries for which survey data on suicidal behaviours and potential risk factors were publicly available. Our sample included data from 38 surveys in 32 countries (Table 1). The survey dates ranged from 2003, the first year the survey was conducted, to 2012 the most recent year with publicly available data at the time of this analysis. The majority of countries conducted one survey over the time period. Where available, we pooled data from two surveys conducted in the same country. According to 2012 World Bank classification, 29 countries are considered low-and middle-income economies and three are high-income economies (Table 1). 26 Although these three countries by definition are not low-and middle-income countries, we retained them for our analysis given the limited knowledge of adolescent suicidal behaviours in many high-income countries that are not part of the Organisation for Economic Co-operation and Development. Countries were grouped by WHO region. Because of the relatively small number of countries from the South-East Asia and Western Pacific Regions in our sample, data from these two regions were combined. Samples Table 1. Survey year(s) and sample size for countries that participated in the Global School-Based Health Survey, 19

Measures
The questionnaire contained two questions on suicidal ideation and planning where the response option was "yes" or "no" ( Table 2). The questions were: "During the past 12 months, did you ever seriously consider attempting suicide?" and: "During the past 12 months, did you ever make a plan about how you would attempt suicide?" Consistent with a recent study using the 2009 Benin GSHS, we defined variables for suicidal ideation as responding "yes" to the first question and variables for suicidal ideation with planning as responding "yes" to both questions. 14 We selected a priori potential risk factors for suicidal ideation based on previous research on adolescents in low-, middle-and high-income countries and their inclusion in the survey. [27][28][29][30] The determinants included were: psychosocial symptoms (loneliness, having close friends and parental support); substance use (alcohol use and cigarette smoking); being physically attacked; and bullying victimization. Given the survey does not include questions on family socioeconomic conditions, we included information on the frequency of going to bed hungry to capture some socioeconomicrelated variation. The survey in some countries asked about other potentially important determinants (e.g. drug use, sexual assault); these were excluded because they were missing > 50% of the responses across the surveys.
Our analysis included an initial sample size of 164 770 across the 32 countries. The percentages of missing data were 1.8% (2966) for suicide ideation and 2.7% (4449) for planning suicide. Other variables were missing less than 5%, except for bullying (12.0%, 19 772), smoking (15.1%, 24 880), physical attacks (26.8%, 44 158) and alcohol use (28.4%, 46 795). To account for missing data, we performed multiple imputation using the mi impute chained procedure in Stata version 12 (Stata-Corp. LP, College Station, United States of America), which uses an iterative multivariable regression procedure to generate distributions for each variable with missing data that are conditional on all other variables in the imputation models. 31 All variables with missing data were imputed using appropriate distributions to model each variable, During the past 30 days, on how many days did you smoke cigarettes? 1 = 0 days 2 = 1-5 days 3 = 6 or more days a In the survey questionnaire physical attack is defined as "when one or more people hit or strike someone, or when one or more people hurt another person with a weapon (such as a stick, knife, or gun)". The survey questionnaire specifies that "it is not a physical attack when two students of about the same strength or power choose to fight each other". b Bullying is defined as "when a student or group of students say or do bad and unpleasant things to another student" or "when a student is teased a lot in an unpleasant way" or "when a student is left out of things on purpose". The survey questionnaire specifies that "it is not bullying when two students of about the same strength or power argue or fight or when teasing is done in a friendly and fun way". c Drinking alcohol also includes consuming locally produced alcoholic drinks. The survey questionnaire specifies that "drinking alcohol does not include drinking a few sips of wine for religious purposes. A drink is defined as a glass of wine, a bottle of beer, a small glass of liquor, or a mixed drink". for example logistic, ordered logistic regressions. A total of 10 imputed data sets were generated. Results were pooled across imputed data sets using Stata's mi estimate procedures.

Statistical analysis
We estimated the prevalence of suicidal ideation and ideation with a plan among males and females in each country using age-adjusted logistic regression to facilitate comparability of estimates across countries. 32 Ratios and differences comparing the prevalence of suicidal behaviours for females compared to males were estimated for each country. Random effects meta-analysis was used to generate regional and overall pooled estimates, using the DerSimonian and Laird inverse-variance method. 33 We used meta-regression to correlate country-level estimates of adolescent suicidal ideation with estimated national mortality rates from self-harm among young people aged 15-29 years. 2,34 Multivariable logistic regression was used to estimate risk ratios (RR) and 95% confidence intervals (CI) measuring associations between the determinants and suicidal behaviours. RRs were calculated from average marginal probabilities estimated from the logistic coefficients. 32 Multivariable models were estimated separately by WHO region and included fixed effects for country and survey year. Population attributable fractions (PAF) were estimated for the risk factors, based on adjusted risk ratios and the prevalence of the determinants within each exposure level. 35,36 All analyses incorporated sampling weights where available and accounted for clustering at the school level.

Results
The pooled 12-month prevalence of suicide ideation for females was 16.2% (95% CI: 15.6 to 16.7) and for males 12.2% (95% CI: 11.7 to 12.7). For suicide ideation with a plan the pooled 12-month prevalence for females was 8.3% (95% CI: 7.9 to 8.7) and for males 5.8% (95% CI: 5.5 to 6.1). There was considerable heterogeneity between countries in the prevalence of suicide ideation, ranging from 5.1% (95% CI: 2.1 to 8.1) in Indonesia to 28.1% (95% CI: 22.5 to 33.7) in Zambia. For ideation with a plan for both sexes the prevalence ranged from 1.7% (95% CI: −0.1 to 3.5) in the United Republic of Tanzania to 15.3% (95% CI: 11.6 to 19.0) in Benin and Kenya 15.3% (95% CI: 12.6 to 18.1). The African Region showed the highest overall pooled prevalence of suicide ideation (21.6%; 95% CI: 20.4 to 22.9) and no evidence of gender differences. By contrast, suicide ideation in the Region of the Americas was markedly higher in females than males, with an estimated prevalence ratio of 1.7 (95% CI: 1.6 to 1.8). The South-East Asia Region and Western Pacific Region had a relatively low prevalence of suicidal behaviours for both sexes, 10.7% (95% CI: 9.9 to 11.5) for ideation and 5.0% (95% CI: 4.5 to 5.4) for ideation with a plan (Table 3; available at: http://www.who.int/bulwletin/volumes/94/5/15-163295). Fig. 1 and Fig. 2 summarize the country-specific predicted prevalence estimates of suicide ideation and ide-ation with a plan for males and females. Approximately half the countries have statistically significant gender inequality. In all countries with gender differences, suicidal behaviours were more common in females than in males. Correlations between youth suicidal ideation and national mortality rates from self-harm were weak to moderate and stronger among males (r = 0.29) than among females (r = 0.11; Fig. 3). Among the eight explored determinants, the highest correlations were between smoking and drinking (r = 0.31) and having been physically attacked and bullied (r = 0.27) and suicide ideation. Table 4 (available at: http:// w w w . w h o . i n t / b u l l e t i n / v o lumes/94/5/15-163295) presents ad1justed risks and risk ratios for suicide ideation during the past 12 months,  Regional analyses yielded similar results to country-specific analyses, which identified loneliness, bullying, alcohol use and physical attacks as the most consistent risk factors across countries for both suicidal ideation and ideation with a plan. Of the 32 countries, the RRs were significant for bullying and loneliness in 28 countries, for physical attacks in 26 countries and for lack of parental understanding and alcohol use  in 25 countries. Associations between going to bed hungry and smoking and suicide ideation were statistically significant in less than half of the countries. Some determinants showed substantial heterogeneity among countries within the same region in the magnitude of the RRs. For example, there was evidence of substantial heterogeneity within regions for the association between loneliness and suicide ideation (heterogeneity P-values < 0.001). By contrast, the association between bullying and suicide ideation was fairly consistent across countries within regions (all heterogeneity P-values > 0.05; country-specific results are available from corresponding author).
PAFs associated with suicide ideation and ideation with a plan across regions are shown in Fig. 4  In the African Region, PAF estimates were largest for bullying at 16.2% and for physical attacks at 12.0%. Estimates in the African Region for loneliness and parental support were also considerable at 11.3% and 11.9%, respectively. The PAF estimate for alcohol use in the Americas was 15.3%, approximately twice as high as for the other regions. The PAF for going to bed hungry was 6.4% in the African Region compared to just 1.3% in the Americas. In general, PAF estimates for suicide ideation with a plan found similar results as for suicide ideation.

Discussion
Our results confirm that adolescent suicidal behaviours are a common problem in low-income and middle-income countries, with prevalence similar to that seen in Europe and North America. However, we found heterogeneity across countries and regions in the prevalence of adolescent suicidal behaviours and in the magnitude of gender differences, which is consistent with previous research. 5 This variation may in part reflect differences in the meaning of suicidal thoughts and normative attitudes towards suicide across diverse cultural, religious and economic settings. 37,38 The higher prevalence of suicidal behaviours among adolescents in African countries may be partly explained by high human immunodeficiency virus (HIV) and acquired immunodeficiency syndrome (AIDS) prevalence, political instability and food insecurity. 27,39 The fact that country-specific estimates of adolescent suicide ideation did not correlate strongly with national estimates of suicide deaths is perhaps not surprising given the lack of reliable vital registration data in most countries and the underreporting and misclassification of suicide deaths, particularly those occurring among young people. 2 The gender differences seen in the Region of the Americas were similar in magnitude to gender differences seen in high-income countries 9 while other regions showed less or no gender disparity. Over the past few decades, Latin America and the Caribbean have made considerable progress reducing gender disparities in school enrolment and labour force participation. 40 There is evidence that changes in women's traditional family roles and an increase in the share of women working may initially lead to higher female suicide rates. As gender roles evolve in work and education, adolescent girls might face additional stressors. 41 Studies have shown a strong association between adverse childhood experiences -such as physical and sexual  abuse, parental neglect, bullying -and suicidal behaviours during adolescence and adulthood. 27,42 Exposure to these adverse experiences may contribute to suicidal ideation through increasing internalizing behaviours, such as shame, feelings of depression and social isolation, that affect the ability to cope with life stressors. 42 A longitudinal study from South Africa found a strong and graded association between cumulative exposure to adverse childhood experiences and suicidal behaviour among adolescents aged 10-18 years. 27 We find consistency across a large and diverse set of countries in the cross-sectional associations between physical violence, bullying victimization and suicide ideation, suggesting that policy efforts aimed at reducing violence and bullying among school-aged youth may help reduce adolescent suicidal behaviours across low-and middle-income countries. However, as in high-income countries, the environmental and social context underlying acts of bullying and physical violence likely differ across cultures, policy regimes and economic zones. For example, physical violence from school staff is an important contributor to the overall burden of youth violence in some countries, 43 whereas in countries with a high burden of HIV/AIDS, bullying may relate to AIDS-orphanhood and AIDS-related stigma. 44 Policy approaches will need to consider these local contexts. Psychological factors such as depression, low self-esteem, hopelessness and weak social relationships are well established correlates of suicidal behaviours among adolescents in high-income countries 9,45 and in several low-and middle-income countries. 13,46 Although the survey excludes information on common mental disorders, such as depression and generalized anxiety disorder, which are often comorbid with youth suicidal behaviours, 2,13 loneliness and social support are important longitudinal predictors of both adolescent depression and suicide ideation. 47 Across all regions and nearly all countries in our study, psychosocial symptoms, such as loneliness, having few friends and lacking parental support were related to adolescent suicidal behaviours. For countries in the Region of the Americas, Eastern Mediterranean and South-East Asia and Western Pacific Regions, loneliness was the factor most strongly associated with suicide ideation. In the African Region, loneliness was also associated with suicide ideation, but to a lesser extent than in the other regions. Other factors such as bullying, physical attacks and lack of parental control seem to have a stronger influence for suicide ideation in this region.
The strengths of this analysis include the use of standardized measures of suicidal behaviours and risk factors from large number of countries, with most countries having nationally representative samples. In the absence of standardized methods across surveys, cross-national differences in suicidal behaviours and risk factors are more likely to reflect differences in the type of sample, i.e. community-based and school-based, the wording of questions and data collection procedures. 4 However, there are several limitations that should be kept in mind when interpreting our results. First, given the cross-sectional nature of the data, we were unable to assess temporal relationships between the factors associated with suicide ideation. There is also limited evidence of the validity and reliability of the survey's measures across culturally diverse settings. Furthermore, in addition to the sensitive nature of questions about suicide, differences in the willingness of students from different cultural backgrounds to report suicidal behaviours and translation of the questionnaires into different languages may also have affected the results. 48 Finally, restriction of the survey to adolescents currently attending school and present on the day of the survey may have also led to some underreporting of suicidal behaviours. 13 Here we show similar determinants of adolescent suicidal behaviours across a diverse set of countries, including violence, bullying, lacking friends and alcohol use. Given these risk factors appear quite universal, this information could be used to identify adolescents in school settings who are at increased risk for suicidal thoughts. School-based suicide prevention interventions have been shown to effectively reduce suicide thoughts and attempts among adolescents in rich countries, 49,50 however there is a dearth of evidence on effective policies or interventions to reduce youth suicidal behaviours and suicide from low-income and middle-income countries. More research is needed to understand the etiology of suicidal behaviours as well as the types of policies and interventions that can effectively reduce the burden of this critical health challenge. ■