https://orcid.org/0000-0002-5342-0709
Figures
Abstract
Suicide is now the 2nd leading cause of death. Despite existing prevention initiatives, the incidence of suicide deaths among younger population has risen in recent years. The study aimed to determine the social and behavioral predictors of suicidal ideation (SI) and suicide attempts (SA) among adolescents and young adults. We analyzed data from the 2021 Youth Risk Behavior Surveillance System (YRBSS) involving 17,232 U.S. students in grades 9–12. Among respondents, 20.85% (3,593) reported suicidal ideation, and 10.17% (1,753) attempted suicide. Of the 3,526 participants with depression, 46.23% (1,630) experienced suicidal ideation, and 19.46% (686) attempted suicide. Among the 8,152 female participants, 27.42% (2,335) reported suicidal ideation, and 13.37% (1,090) reported suicide attempts. Also, 29.06% of 3,981 participants who were overweight experienced SI and 14.02% of them attempted suicide. We developed independent SI and SA prediction models using extreme gradient boosting (XGB), logistic regression (LR), and K-Nearest Neighbors (KNN) algorithms. The LR model demonstrated the best performance in predicting both SI and SA, with an accuracy of 0.74 (for both), specificity of 0.71 for SI; 0.69 for SA, and sensitivity (0.77 for SI; 0.81 for SA). It also achieved high positive predictive values (PPV) of 0.90 for SI; 0.95 for SA, and an area under the Receiver Operating Characteristic (AUROC) curve of 0.82 for both outcomes. The model identified cyberbullying (electronically bullied), drug use, forced sexual intercourse, sexual violence, difficulty in concentrating, and early use of alcohol or marijuana, as some of the most important and common predictors of SI and SA. In conclusion, our research advocates for the urgent development of support systems that address the early signs of suicidal behaviors. It also surfaces the fact that predicting with certainty who is likely to attempt suicide remains a challenge.
Citation: Shahsavar Y, Choudhury A (2025) Behavioral and social predictors of suicidal ideation and attempts among adolescents and young adults. PLOS Ment Health 2(1): e0000221. https://doi.org/10.1371/journal.pmen.0000221
Editor: Justus Onu, Nnamdi Azikiwe University, NIGERIA
Received: June 18, 2024; Accepted: December 8, 2024; Published: January 3, 2025
Copyright: © 2025 Shahsavar, Choudhury. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: The datasets used in the study can be retrieved from https://www.cdc.gov/healthyyouth/data/yrbs/data.htm.
Funding: The authors received no specific funding for this work.
Competing interests: The authors have declared that no competing interests exist.
Introduction
The question of ’why’ often lingers after the loss of a loved one to suicide, accompanied by reflections on moments missed and support unoffered. This reality underscores the complex and often unseen nature of mental health struggles. It is often challenging to understand when suicidal thoughts set in, when people plan about it, and when they execute it—the deepest struggles often lie hidden beneath the surface, leaving friends and family wishing for a sign or a clue that they might have missed.
Suicide is now the 2nd leading cause of death, and numbers continue to rise [1]. According to the Center of Disease Control and Prevention (CDC), suicide rates increased 37% between 2000–2018 reaching its peak in 2021 [2]. According to the 2021 Youth Risk Behavior Surveillance System (YRBSS), approximately 10% of high school students attempted suicide and 22% seriously thought about it [3] in the United States. In 2022, about 49,449 people died by suicide in the United States [2]. The increasing prevalence of suicide among youth and its progression from ideation to attempt poses a significant challenge for public health globally [1, 4–9]. Despite substantial investments in research aimed at comprehending the risk factors for suicide and in developing suicide prevention initiatives [10], the incidence of suicide deaths among young Americans has risen in recent years [11, 12]. Studies have acknowledged poor mental health [13], encounters with violence [14], and social pressures [15, 16], as plausible factors influencing suicidal ideation and attempt among youth.
Machine learning (artificial intelligence) offers a promising approach to identifying individuals at risk of suicidal ideation or attempts, which is crucial for preventing suicide [17]. However, a significant limitation of many current models is their dependence on clinical data or psychological factors that are not easily accessible [18–20]. These models require specific data points, like the Patient Health Questionnaire (PHQ) depression score, that are typically only available through medical or psychological evaluations. The reliance on clinical data limits the models’ practical use for the public, who may not have access to such detailed assessments. For example, if someone is worried about a loved one’s mental health, they likely won’t have the necessary clinical mental health assessment scores to use these predictive models effectively. This issue highlights a critical gap in the utility of these models for everyday situations and underscores the challenge of understanding at-risk behaviors in a domestic setting. The limitation points to a critical need for developing models that can operate with more commonly observable data points.
Our study aims to identify and leverage social and behavioral predictors of suicidal ideation and attempts among adolescents and young adults. Note that the objective is not to introduce a new predictive model as a replacement of clinical assessment but to assess if social and behavioral predictors that are observable by friends, family, or individuals themselves, can help predict suicidal ideation or attempt that a layperson can use as a preliminary tool at home.
Method
Ethics statement
The research is a secondary analysis of publicly available data and does not require approval from the institutional review board. The dataset contains no identifiers.
Data description
We used the 2021 YRBSS as the data source [21]. YRBSS is a survey developed by the CDC that tracks behaviors leading to poor health in students grades 9 through 12. The surveys are administered every other year. Some of the health-related behaviors and experiences captured through the survey are student demographics, health behaviors, substance use behaviors, and student experiences. In 2021, YRBSS aimed to gather data from schools and students across the United States. Participation rates varied, with 152 out of the 209 schools agreeing to participate, yielding a school response rate of approximately 72.70%. In total 17,508 students completed the survey. After reviewing the responses for accuracy and completeness, 17,232 responses were deemed usable, leading to a student response rate of around 79.10%. To calculate the overall response rate for the survey, the response rates from both schools and students were combined, resulting in a final participation rate of about 57.50%.
Two variables, each indicative of suicidal ideation (SI) and suicidal attempts (SA), respectively, from the YRBSS were used as an outcome variable (target class) in our study.
Suicidal Ideation (SI)–The variable was measured using the following question: During the past 12 months, did you ever seriously consider attempting suicide? The responses were recorded as ’Yes’ or ’No’. In our model, Yes was coded as 1.
Suicidal Attempt (SA)–The variable was measured using the following question: During the past 12 months, how many times did you actually attempt suicide? The responses were recorded as 0 times, 1 time, 2 or 3 times, 4 or 5 times, 6 or more times. In our study, we recorded all responses indicating at least one attempt as 1 and remaining as 0.
Data preprocessing
Analyses were done in R (version 4.1.0). First, we duplicated and divided the data set into two data files, each having SI and SA as target variable, respectively. In this manuscript, we will address these datasets (D) as D-SA and D-SI. We then conducted correlation analysis on both the datasets (D-SI and D-SA) and removed all variables that were highly correlated either with the target class or among each other. As a result, we eliminated variables pertaining to sadness or hopelessness, suicide planning, and other measures of mental health. Missing values in target class were also deleted from the dataset. We then handled missing values across all input variables by mode replacement. The impute_data function [22] was used from dplyr package in R [23]. Categorical variables were one-hot encoded using dummyVars function [24]. Since the data consisted of variables with varying magnitude, we normalized the dataset [25].
Each of the preprocessed datasets (D-SI and D-SA) were divided into training and testing sets using a sample.split function [26] with an 80–20 ratio. To resolve the class imbalance problem, we used the oversampling of the minority class in the training set (D-SA and D-SI, respectively) by using Ovun.sample function [27]. Class imbalance, where there are many more negative instances (no ideation or attempt) than positive ones (presence of ideation or attempt), is common in sensitive issues like suicidal ideation (D-SI) and attempts (D-SA).
Model selection
We compared three algorithms, extreme gradient boosting (XGB) [28], logistic regression (LR) [29], and K-Nearest neighbors (KNN) [30] on both D-SI and D-DS. We chose these models because of their performance in binary classification and computational efficiency [31–34]. XGB was selected for its performance and feature importance capabilities, offering insightful predictions even with the sparse data characteristic of survey responses. It excels in managing missing values and identifying key variables influencing suicidal ideation and attempts, when compared to models like random forest, which is more prone to overfitting in such a data-rich environment. Logistic regression was selected for its simplicity and interpretability. Lastly, KNN’s flexibility and non-parametric nature allow it to adapt to the patterns within the YRBSS dataset, a feature not as directly accessible in more complex models that might struggle with the interpretability and computational demands presented by the dataset.
Model training and evaluation
For the XGB model from XGBoost package [35], we converted the data (oversampled training data) into a matrix format, excluding the target class (SI for D-SI and SA for D-SA) variable. We used hyperparameter tuning to optimize the model. The best parameters included 0.1 learning rate (eta), 2 maximum depth of trees (max_depth), 1 n_thread and 0.1 gamma for regularization. For the LR model, we used glm function [36] with a binomial setting. Following the model fitting, predicted probabilities were generated, which were then converted to binary classifications using a 0.5 threshold. KNN model was used from Caret package [37].
All the models were tuned using 10-fold cross-validation. To address data dimensionality issues, we used xgb.importance function from the XGBoost package [35] to identify the most important variables for both D-SA and D-SI. All the models were retrained using only the important variables. Model evaluation was achieved by monitoring accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and the area under the receiver operating characteristic curve (AUROC) [38].
Odds ratio
Once the important variables were identified from our ML model, we used JASP [39] for calculating the prevalence odds ratio and 95% confidence interval of the top 10 predictors of ideation and attempts. Likert scale questions were converted to binary format for the analysis. For each question, if a response indicated the occurrence of an event or an action being taken at least once, it was assigned a value of 1; if no event or action was indicated, it was assigned a value of 0.
Results
Descriptive statistics
YRBSS consisted of 17,232 participants, of which 48% were female and 51.9% were male. All participants were students from grades 9 to 12 in the United States. Among respondents, 20.85% (3,593) reported suicidal ideation, and 10.17% (1,753) attempted suicide. Of the 3,526 participants with depression during covid, 46.23% (1,630) experienced suicidal ideation, and 19.46% (686) attempted suicide. Among the 8,152 female participants, 27.42% (2,335) reported suicidal ideation, and 13.37% (1,090) reported suicide attempts. Also, 29.06% of 3,981 participants who were overweight experienced suicide ideation and 14.02% of them attempted suicide. The other important information about the participants in the study is given in Table 1.
Feature importance
Table 2 lists the top 10 important predictors of SI and SA, respectively. For SI, being ‘electronically bullied’ tops the list, underscoring the significant impact of cyberbullying on mental health. Following closely are factors such as the ‘use of electronic vapor products’ and ‘serious difficulty in concentrating, remembering, or making decisions.’ These might initially seem less directly connected to suicide ideation compared to more overtly distressing experiences like ‘physical sexual violence’, yet their prominence highlights the complex nature of mental health risks. Variables like ‘sex’, experience of ‘being bullied on school property’, and early initiation of ‘alcohol use’ were also significant, suggesting that both behavioral patterns and experiences of violence are critical in understanding suicide ideation.
For suicide attempts, the sequence starts similarly with ‘electronic bullying’ but is immediately followed by more severe experiences such as ‘being physically forced to have sexual intercourse’. This indicates a stronger link between direct, physical forms of violence and the act of attempting suicide compared to ideation alone. The ‘use of electronic vapor products’ remains highly relevant, suggesting a possible link between substance use and the escalation from ideation to attempts. The list for attempts also includes more specific variables like ‘early use of alcohol and drugs’, ‘current marijuana use’, and experiences of ‘physical dating violence’, painting a picture of a cumulative risk landscape where substance use, and violent experiences significantly contribute to the likelihood of an attempt. Overall, the results emphasize the point that while some factors are critical in contemplating suicide, others play a more crucial role in the progression to actual attempts, underlining the necessity of a nuanced approach to prevention that addresses both immediate and underlying risk factors.
Model outcomes
Table 3 reports the model performance. For SI, LR shows the highest accuracy (0.74) and specificity (0.71), and it has competitive values for sensitivity (0.77), PPV (0.91), and AUROC (0.82). Its NPV is the lowest among the models for SI, but this is by a marginal difference. The XGB model has slightly lower accuracy and specificity but higher sensitivity, indicating it’s better at identifying true positives but slightly worse overall. The KNN model lags in all aspects except for sensitivity, where it is the lowest, indicating it’s generally less effective for this prediction task.
For SA, the LR model again shows the highest accuracy (0.74) and specificity (0.68), and it has very competitive sensitivity (0.81) and the highest PPV (0.95) and AUROC (0.82). Despite a lower NPV, it demonstrates the best balance across all metrics. The XGB model, while having a slightly higher sensitivity, falls short in accuracy, specificity, and NPV, indicating it might not balance false positives and negatives as well as the LR model. The KNN model, despite its highest sensitivity for suicide attempts, has significantly lower accuracy, specificity, and the lowest NPV, suggesting it might misclassify a greater number of non-attempt cases as attempts. Considering these metrics, LR model emerges as the best fit model for both SI and SA predictions.
Prevalence odds ratio
Table 4 shows the factors associated with increased odds of SI and SA. Electronic bullying emerged as a critical factor, with individuals who experienced electronic bullying showing higher odds of SI (aOR = 1.45) and SA (aOR = 2.52). Similarly, the use of electronic vapor products was associated with a modest increase in SI (aOR = 1.20) and a slightly elevated, though less certain, risk for SA (aOR = 1.14). Cognitive difficulties—specifically issues with concentration, memory, or decision-making—showed a notably strong association with SI (aOR = 4.23), indicating a significant mental health burden linked to SI risk. Individuals exposed to sexual violence demonstrated higher odds for both SI (aOR = 2.10) and SA (aOR = 2.21). Additionally, misuse of prescription pain medication was associated with increased odds of both SI (aOR = 1.95) and SA (aOR = 2.11). Gender differences were observed, with females showing a higher likelihood of SI than males (aOR = 1.50). Social factors, such as being bullied on school property, were also significantly associated with SI (aOR = 1.80). Early substance use, including marijuana and alcohol consumption before age 13, was linked to increased odds of both SI and SA, with early alcohol use (aOR = 1.63; 1.76) and marijuana use (aOR = 1.60; 1.52). Furthermore, experiencing physical dating violence was strongly associated with SA (aOR = 2.00), highlighting the impact of interpersonal violence on suicide-related behaviors.
Discussion
Overall contribution
While we acknowledge that the relationships between the predictors of SI and SA are well established in the literature, our study brings a fresh perspective to predicting thoughts of suicide and suicide attempts by focusing on observable, everyday behaviors and experiences of young people, such as experiences of bullying, drug use (including alcohol, cigarettes, marijuana, and vapor products), experiences of sexual and physical violence, and difficulties with concentration (like remembering things or making decisions). Unlike many existing studies (Table 5), which rely heavily on clinical or latent constructs such as severe depression, anxiety, previous suicide attempts, antisocial score, or perceived warmth, our approach uses data that are more accessible and observable by public, like teachers, parents, or friends. This makes our model more practical for real-world prevention efforts. Besides, the essence of using predictive models in suicide prevention is to uncover hidden patterns that aren’t immediately obvious to humans. If we only used variables that clearly indicate a high risk of reattempt, such as a history of suicide attempts or severe depression, we wouldn’t be fully utilizing preventive approaches. Such information, while crucial, might not help in early prevention because it’s already an established risk factor. Our study reveals new insights that can guide prevention strategies. For example, our model, using the YRBSS, identified predictors like being electronically bullied, using an electronic vapor product, experiencing sexual violence, and misusing prescription pain medicine, achieving an AUROC of 0.82 for suicidal ideation and 0.83 for suicide attempts, with an accuracy of 0.74 for both. In contrast, other studies, such as those using the Health Research Enterprise from French universities [40] or the Minnesota Multiphasic Personality Inventory-2 Restructured Form [41], focus on predictors like suicidal thoughts, anxiety, depression, and emotional dysfunction, achieving similar or slightly higher AUROC scores but often relying on less observable factors. Therefore, our model’s emphasis on accessible data offers a significant advantage in practical applications for early intervention and prevention efforts.
Electronic bullying as a risk factor
We identified electronic bullying (cyberbullying) as a key factor, with individuals who experienced it showing higher odds of both suicidal ideation and attempts. This form of harassment often extends beyond the school environment into the digital lives of young individuals [50], it poses a constant threat to their well-being [51]. Research has shown that victims of cyberbullying, when compared to non-victims, have a higher risk of suicidal ideation, self-harm, and suicide attempts [52]. Additionally, exposure to both cyberbullying and traditional in-school bullying has been associated with depressive symptoms, gradually increasing the risk for suicidal ideation, suicide planning, and suicide attempts [53]. Overall, our research underscores the detrimental impact of bullying, especially cyberbullying, on adolescent mental health, highlighting the need for targeted interventions to address bullying behaviors and support at-risk individuals to prevent suicidal ideation and attempts.
Drug use as a risk factor
Another predictor of suicide ideation and attempts identified in our model is drug use. We noted adolescent indulged with misuse of prescription medicine, marijuana and alcohol use during early and pre-teen age had higher odds of ideation and attempt. Drug use has been consistently linked to an increased risk of suicidal ideation and suicide attempts among various populations [54]. Studies have shown a strong association between illicit drug use and suicidal ideation, self-harm, and suicide attempts [55]. Early initiation of substance use, including smoking cigarettes, alcohol, and drugs like cannabis, has been correlated with suicidal ideation and attempts among adolescents [56]. Additionally, while some research suggests that drug use, particularly marijuana and other drugs, may not directly increase the risk of suicidal ideation, there is a bidirectional relationship where suicidal ideation can lead to an increased risk of illicit drug use [57].
Sexual misconduct as a risk factor
We identified forced sexual encounters and sexual violence as key factors, with adolescents who were physically forced to have sexual intercourse, experienced sexual violence, or physical dating violence, had increased odds of ideation and or attempt. In line with our finding, a recent study in 2024, reported strong association between sexual harassment and suicidal ideation [58]. Additionally, the concurrence of sexual violence has been independently associated with an elevated risk of suicidal behaviors among adolescents [59]. Contrary to some expectations, our findings and 2014 study noted bullying to have a stronger effect on young people’s suicidal ideation compared to sexual abuse or physical violence [60]. Another variable our model acknowledged as a predictor of suicidal ideation is difficulty in concentrating, remembering, or making decisions. The impact of cognitive deficits on decision-making processes has been identified as a significant factor influencing suicidal ideation [61]. Additionally, the inability to suppress unwanted thoughts or lack of executive functions has been recognized as contributing factors to suicidal ideation.
Implications
Our analysis sheds light on early signs of suicide ideation and attempts among youth, offering a plausible pathway for suicide prevention. Identifying electronic bullying as a prominent risk factor underscores its pervasive threat to adolescent mental health. This insight suggests that effective suicide prevention strategies must address the digital landscape where young individuals spend a significant portion of their lives. Interventions aimed at combating cyberbullying could serve as a critical early step in safeguarding against the progression towards suicidal thoughts and behaviors. Furthermore, the association between drug use and increased risk of suicide ideation and attempts highlights the importance of substance abuse prevention as a cornerstone of suicide prevention efforts. Early intervention programs that educate young people about the dangers of substance use and provide support for those struggling could significantly reduce the risk of suicide. Additionally, recognizing signs of forced sexual encounters and sexual violence as predictors of suicide ideation points to the need for accessible support systems and trauma-informed care for victims. Addressing cognitive difficulties, such as problems with concentration, memory, or decision-making, further expands our understanding of suicide prevention, emphasizing the importance of comprehensive mental health support that includes cognitive and emotional aspects. By focusing on these early signs and implementing targeted prevention strategies, we can create a more effective approach to reducing the incidence of suicide ideation and attempts among youth.
Limitations
Our study has few limitations that should be considered. First, the use of a single item to determine suicidal ideation may overestimate its prevalence, potentially affecting the accuracy and reliability of our predictions. Second, while the YRBSS provides a comprehensive dataset, it relies on self-reported data, which may introduce biases such as underreporting or misreporting of sensitive behaviors and experiences. Third, our model was developed and validated using data from students in grades 9 to 12 in the United States, which may limit the generalizability of our findings to other populations or age groups. Additionally, the cross-sectional nature of the YRBSS data prevents us from making causal inferences about the specific directional causal relationships between the predictors and suicidal behaviors. Finally, our study focuses on social and behavioral factors, which, while important, do not encompass all potential risk factors for suicidal ideation and attempts, such as genetic, biological, or environmental influences. Future research should consider integrating these additional dimensions to provide a more holistic understanding of the predictors of suicidal behaviors.
Conclusion
Our study provides critical insights into the predictors of suicidal ideation and attempts among youth, highlighting the significant impact of factors such as cyberbullying, drug use, sexual violence, and misuse of prescription medications. By understanding these influences, families are better equipped to recognize warning signs and intervene effectively, underscoring the need to create supportive home environments that promote mental well-being and resilience. Additionally, this research advocates for the urgent development of support systems that address the early signs of suicidal behaviors. Aligning with the widely accepted conclusion, our study also surfaces the fact that predicting with certainty who is likely to attempt suicide remains a challenge.
References
- 1.
Organization WH. Suicide in the world: global health estimates: World Health Organization; 2019 [cited 2024 October 24]. Available from: https://www.google.com/url?sa=t&source=web&rct=j&opi=89978449&url=https://www.who.int/publications/i/item/suicide-in-the-world&ved=2ahUKEwis39npzqeJAxUlF1kFHcMGKXMQFnoECBAQAQ&usg=AOvVaw0XdFp14vShVVrJZTTDK301.
- 2. CDC. Suicide Data and Statistics 2023 [cited 2024 March 20]. Available from: https://www.cdc.gov/suicide/suicide-data-statistics.html.
- 3. CDC. YOUTH RISK BEHAVIOR SURVEY: DATA SUMMARY & TRENDS REPORT 2021 [cited 2024 March 20]. Available from: https://www.cdc.gov/healthyyouth/data/yrbs/pdf/YRBS_Data-Summary-Trends_Report2023_508.pdf.
- 4.
Organization WH. Preventing suicide: A global imperative: World Health Organization; 2014. Available from: https://iris.who.int/bitstream/handle/10665/341728/9789240026643-eng.pdf.
- 5. Organization WH. Suicide worldwide in 2019: global health estimates 2021 [cited 2024 October 24]. Available from: https://www.google.com/url?sa=t&source=web&rct=j&opi=89978449&url=https://www.who.int/publications/i/item/9789240026643&ved=2ahUKEwi_ro2Wz6eJAxU2K1kFHQEaMqcQFnoECC8QAQ&usg=AOvVaw36Xd8-oStMJSUIyMcjw2_H.
- 6. Beautrais AL, Joyce PR, Mulder RT. Risk factors for serious suicide attempts among youths aged 13 through 24 years. Journal of the American Academy of Child & Adolescent Psychiatry. 1996;35(9):1174–82. pmid:8824061.
- 7. Control CfD, Prevention. Leading causes of death reports, national and regional, 1999–2013 2013. Available from: https://www.google.com/url?sa=t&source=web&rct=j&opi=89978449&url=https://www.cdc.gov/nchs/nvss/leading-causes-of-death.htm&ved=2ahUKEwiX4Za40KeJAxXcEVkFHapUNDsQFnoECBYQAQ&usg=AOvVaw12PdU7JRBGOZWdMqMTS3Wy.
- 8. Ghazali SR, Chen YY, Mohamad M, Yein LP, Razali ZA, Pauzi N, et al. Lifetime trauma, symptoms of psychological disturbance, and suicidal ideation among university students in Malaysia. Current psychology. 2023;42(25):22127–36.
- 9. Miranda-Mendizabal A, Castellví P, Parés-Badell O, Alayo I, Almenara J, Alonso I, et al. Gender differences in suicidal behavior in adolescents and young adults: systematic review and meta-analysis of longitudinal studies. International journal of public health. 2019;64(2):265–83. Epub 20190112. pmid:30635683; PubMed Central PMCID: PMC6439147.
- 10. Glenn CR, Kleiman EM, Kellerman J, Pollak O, Cha CB, Esposito EC, et al. Annual Research Review: A meta‐analytic review of worldwide suicide rates in adolescents. Journal of child psychology and psychiatry. 2020;61(3):294–308. Epub 20190801. pmid:31373003.
- 11. Curtin SC, Heron MP. Death rates due to suicide and homicide among persons aged 10–24: United States, 2000–2017. NCHS Data Brief. 2019;(352):1–8. https://www.ncbi.nlm.nih.gov/pubmed/31751202. pmid:31751202
- 12. Lindsey MA, Sheftall AH, Xiao Y, Joe S. Trends of suicidal behaviors among high school students in the United States: 1991–2017. Pediatrics. 2019;144(5). Epub 20191014. pmid:31611338; PubMed Central PMCID: PMC7299440.
- 13. Aggarwal S, Borschmann R, Patton GC. Tackling stigma in self-harm and suicide in the young. The Lancet Public Health. 2021;6(1):e6–e7. pmid:33417848; PubMed Central PMCID: PMC7611270.
- 14. Castellví P, Miranda‐Mendizábal A, Parés‐Badell O, Almenara J, Alonso I, Blasco M, et al. Exposure to violence, a risk for suicide in youths and young adults. A meta‐analysis of longitudinal studies. Acta psychiatrica scandinavica. 2017;135(3):195–211. Epub 20161220. pmid:27995627.
- 15. Twenge JM. More time on technology, less happiness? Associations between digital-media use and psychological well-being. Current Directions in Psychological Science. 2019;28(4):372–9.
- 16. Luby J, Kertz S. Increasing suicide rates in early adolescent girls in the United States and the equalization of sex disparity in suicide: the need to investigate the role of social media. JAMA network open. 2019;2(5):e193916–e. Epub 20190503. pmid:31099856.
- 17. Jacob N, Scourfield J, Evans R. Suicide prevention via the internet. Crisis. 2014;35(4):261–7. pmid:24984890.
- 18. Weng J-C, Lin T-Y, Tsai Y-H, Cheok MT, Chang Y-PE, Chen VC-H. An autoencoder and machine learning model to predict suicidal ideation with brain structural imaging. Journal of clinical medicine. 2020;9(3):658. Epub 20200229. pmid:32121362; PubMed Central PMCID: PMC7141277.
- 19. Ryu S, Lee H, Lee D-K, Park K. Use of a machine learning algorithm to predict individuals with suicide ideation in the general population. Psychiatry investigation. 2018;15(11):1030. Epub 20181011. pmid:30301301; PubMed Central PMCID: PMC6258996.
- 20. Heckler WF, de Carvalho JV, Barbosa JLV. Machine learning for suicidal ideation identification: A systematic literature review. Computers in Human Behavior. 2022;128:107095.
- 21. Kann L, McManus T, Harris WA, Shanklin SL, Flint KH, Queen B, et al. Youth risk behavior surveillance—United States, 2017. MMWR Surveillance Summaries. 2018;67(8):1. Epub 20180615. pmid:29902162; PubMed Central PMCID: PMC6002027.
- 22. Zhang Z. Missing data imputation: focusing on single imputation. Annals of translational medicine. 2016;4(1):9. pmid:26855945; PubMed Central PMCID: PMC4716933.
- 23. Yarberry W, Yarberry W. Dplyr. CRAN recipes: DPLYR, stringr, lubridate, and regex in R. 2021:1–58.
- 24. Kuhn M, Wing J, Weston S, Williams A, Keefer C, Engelhardt A, et al. Package ‘caret’ 2020 [cited 223 7]. Available from: https://www.google.com/url?sa=t&source=web&rct=j&opi=89978449&url=https://cran.r-project.org/package%3Dcaret&ved=2ahUKEwiix8br0KeJAxUJM1kFHT3CCWsQFnoECAcQAQ&usg=AOvVaw2Xxo7ZXdXq0du5V3jfsVRV.
- 25. Sun J, Nishiyama T, Shimizu K, Kadota K. TCC: an R package for comparing tag count data with robust normalization strategies. BMC bioinformatics. 2013;14:1–14. Epub 20130709. pmid:23837715; PubMed Central PMCID: PMC3716788.
- 26. Hansen BE. Sample splitting and threshold estimation. Econometrica. 2000;68(3):575–603.
- 27. Lunardon N, Menardi G, Torelli N. ROSE: a package for binary imbalanced learning 2014 [cited 6 1]. Available from: https://www.google.com/url?sa=t&source=web&rct=j&opi=89978449&url=https://journal.r-project.org/archive/2014/RJ-2014-008/RJ-2014-008.pdf&ved=2ahUKEwjD8e2O0aeJAxWrMVkFHQnHCTcQFnoECAUQAQ&usg=AOvVaw2mvP4YWtfaCMHstcbccv4P.
- 28. Chen T, He T, Benesty M, Khotilovich V, Tang Y, Cho H, et al. Xgboost: extreme gradient boosting. R package version 04–2. 2015;1(4):1–4.
- 29.
Hosmer DW Jr, Lemeshow S, Sturdivant RX. Applied logistic regression: John Wiley & Sons; 2013. https://doi.org/10.1002/9781118548387
- 30. Peterson LE. K-nearest neighbor. Scholarpedia. 2009;4(2):1883.
- 31. Jung JS, Park SJ, Kim EY, Na K-S, Kim YJ, Kim KG. Prediction models for high risk of suicide in Korean adolescents using machine learning techniques. PLoS one. 2019;14(6):e0217639. Epub 20190606. pmid:31170212; PubMed Central PMCID: PMC6553749.
- 32. Wang Y, Sohn S, Liu S, Shen F, Wang L, Atkinson EJ, et al. A clinical text classification paradigm using weak supervision and deep representation. BMC medical informatics and decision making. 2019;19(1):1–13. Epub 20190107. pmid:30616584; PubMed Central PMCID: PMC6322223.
- 33. Mehedi MAA, Smith V, Hosseiny H, Jiao X. Unraveling the complexities of urban fluvial flood hydraulics through AI. Scientific Reports. 2022;12(1):18738. Epub 20221104. pmid:36333429; PubMed Central PMCID: PMC9636396.
- 34. Lee C, Jo B, Woo H, Im Y, Park RW, Park C. Chronic disease prediction using the common data model: development study. JMIR AI. 2022;1(1):e41030. Epub 20221222. pmid:38875545; PubMed Central PMCID: PMC11041444.
- 35. Chen T, He T, Benesty M, Khotilovich V. Package ‘xgboost’. R version. 2019;90(1–66):40.
- 36. Speelman D. Logistic regression. Corpus methods for semantics: Quantitative studies in polysemy and synonymy. 2014;43:487–533.
- 37. Kuhn M. The caret package. Vienna, Austria: R Found Stat Comput. https://cranr-project_org/package=caret; 2011.
- 38. Kumar R, Indrayan A. Receiver operating characteristic (ROC) curve for medical researchers. Indian pediatrics. 2011;48(4):277–87. pmid:21532099.
- 39. Love J, Selker R, Marsman M, Jamil T, Dropmann D, Verhagen J, et al. JASP: Graphical statistical software for common statistical designs. Journal of Statistical Software. 2019;88:1–17.
- 40. Macalli M, Navarro M, Orri M, Tournier M, Thiébaut R, Côté SM, et al. A machine learning approach for predicting suicidal thoughts and behaviours among college students. Scientific reports. 2021;11(1):11363. Epub 20210615. pmid:34131161; PubMed Central PMCID: PMC8206419.
- 41. Kim S, Lee H-K, Lee K. Detecting suicidal risk using MMPI-2 based on machine learning algorithm. Scientific Reports. 2021;11(1):15310. Epub 20210728. pmid:34321546; PubMed Central PMCID: PMC8319391.
- 42. Shen Y, Zhang W, Chan BSM, Zhang Y, Meng F, Kennon EA, et al. Detecting risk of suicide attempts among Chinese medical college students using a machine learning algorithm. Journal of affective disorders. 2020;273:18–23. Epub 20200511. pmid:32421600.
- 43. Miche M, Studerus E, Meyer AH, Gloster AT, Beesdo-Baum K, Wittchen H-U, et al. Prospective prediction of suicide attempts in community adolescents and young adults, using regression methods and machine learning. Journal of affective disorders. 2020;265:570–8. Epub 20191113. pmid:31786028.
- 44. Navarro MC, Ouellet-Morin I, Geoffroy M-C, Boivin M, Tremblay RE, Côté SM, et al. Machine learning assessment of early life factors predicting suicide attempt in adolescence or young adulthood. JAMA network open. 2021;4(3):e211450–e. Epub 20210301. pmid:33710292; PubMed Central PMCID: PMC7955274.
- 45. Haghish E, Czajkowski NO, von Soest T. Predicting suicide attempts among Norwegian adolescents without using suicide-related items: a machine learning approach. Frontiers in psychiatry. 2023;14:1216791. Epub 20230926. pmid:37822798; PubMed Central PMCID: PMC10562596.
- 46. Wang J, Qiu J, Zhu T, Zeng Y, Yang H, Shang Y, et al. Prediction of suicidal behaviors in the middle-aged population: machine learning analyses of UK Biobank. JMIR public health and surveillance. 2023;9(1):e43419. Epub 20230220. pmid:36805366; PubMed Central PMCID: PMC9989910.
- 47. Lee J, Pak T-Y. Machine learning prediction of suicidal ideation, planning, and attempt among Korean adults: A population-based study. SSM-Population Health. 2022;19:101231. Epub 20220914. pmid:36263295; PubMed Central PMCID: PMC9573904.
- 48. Lim JS, Yang C-M, Baek J-W, Lee S-Y, Kim B-N. Prediction models for suicide attempts among adolescents using machine learning techniques. Clinical psychopharmacology and neuroscience. 2022;20(4):609. pmid:36263637; PubMed Central PMCID: PMC9606439.
- 49. Huang Y, Zhu C, Feng Y, Ji Y, Song J, Wang K, et al. Comparison of three machine learning models to predict suicidal ideation and depression among Chinese adolescents: A cross-sectional study. Journal of Affective Disorders. 2022;319:221–8. Epub 20220916. pmid:36122602.
- 50. Bye A, Carter B, Leightley D, Trevillion K, Liakata M, Branthonne-Foster S, et al. Observational prospective study of social media, smartphone use and self-harm in a clinical sample of young people: Study protocol. BMJ open. 2023;13(2):e069748. Epub 20230201. pmid:36725102; PubMed Central PMCID: PMC9896249.
- 51. Hong JS, Valido A, Espelage DL, Lee J, Dinitto DM. Racial/ethnic differences in the bullying victimization‐suicidality link among LGBQ high school students in the United States. American journal of community psychology. 2024. Epub 20240212. pmid:38344841.
- 52. Peng Z, Klomek AB, Li L, Su X, Sillanmäki L, Chudal R, et al. Associations Between Chinese Adolescents Subjected to Traditional and Cyber Bullying and Suicidal Ideation, Self-Harm and Suicide Attempts. BMC Psychiatry. 2019;19(1):324. Epub 20191028. pmid:31660917; PubMed Central PMCID: PMC6819572.
- 53. Nguyen TH, Shah GH, Muzamil M, Ikhile O, Ayangunna E, Kaur R. Association of in-School and Electronic Bullying With Suicidality and Feelings of Hopelessness Among Adolescents in the United States. Children. 2023;10(4). Epub 20230421. pmid:37190004; PubMed Central PMCID: PMC10136663.
- 54. Kann L. Youth risk behavior surveillance—United States, 2017. MMWR Surveillance Summaries. 2018;67(8):1–114. Epub 20180615. pmid:29902162; PubMed Central PMCID: PMC6002027.
- 55. Robertson HT, Allison DB. Drugs Associated With More Suicidal Ideations Are Also Associated With More Suicide Attempts. Plos One. 2009;4(10):e7312. Epub 20091002. pmid:19798416; PubMed Central PMCID: PMC2749439.
- 56. Peltzer K, Pengpid S. Early Substance Use Initiation and Suicide Ideation and Attempts Among School-Aged Adolescents in Four Pacific Island Countries in Oceania. International Journal of Environmental Research and Public Health. 2015;12(10):12291–303. Epub 20150930. pmid:26437423; PubMed Central PMCID: PMC4626969.
- 57. Zhang X, Wu LT. Suicidal Ideation and Substance Use Among Adolescents and Young Adults: A Bidirectional Relation? Drug and Alcohol Dependence. 2014;142:63–73. Epub 20140612. pmid:24969957; PubMed Central PMCID: PMC4129651.
- 58. Jonasson M, Wiberg M, Dennhag I. Sexual harassment and patterns of symptoms and functional abilities in a psychiatric sample of adolescents. Nordic journal of psychiatry. 2024;78(4):1–11. Epub 20240222. pmid:38385440.
- 59. Li X, Xiang S, Dong J. The Concurrence of Sexual Violence and Physical Fighting Among Adolescent Suicide Ideators and the Risk of Attempted Suicide. Scientific Reports. 2022;12(1):5290. Epub 20220328. pmid:35351957; PubMed Central PMCID: PMC8964683.
- 60. Mossige S, Huang L, Straiton ML, Roen K. Suicidal Ideation and Self‐harm Among Youths in Norway: Associations With Verbal, Physical and Sexual Abuse. Child & Family Social Work. 2014;21(2):166–75.
- 61. Brown VM, Wilson J, Hallquist MN, Szántó K, Dombrovski AY. Ventromedial Prefrontal Value Signals and Functional Connectivity During Decision-Making in Suicidal Behavior and Impulsivity. Neuropsychopharmacology. 2020;45(6):1034–41. Epub 20200208. pmid:32035425; PubMed Central PMCID: PMC7162923.