Scholarlink

<p>Suicide attempts (SA) are a common risk in adolescents with non-suicidal self-injury (NSSI). In the present study, we investigated whether a set of biological markers contributed (above clinical features) to the distinction of adolescents with NSSI and SA from those with NSSI alone using machine-based learning approaches. Female adolescents engaging in NSSI (<i>n</i> = 161) were recruited from our outpatient clinic for risk-taking and self-harming behavior (AtR!Sk). Different machine-based learning models (logistic regression, elastic net regression, random forests, gradient boosted trees) with repeated cross-validation were applied. We tested whether a) the full set of neurobiological markers, b) a reduced set including preselected markers based on existing evidence (CRP, interleukin-6, salivary cortisol, DHEA-S, TSH, dopamine, norepinephrine, ACTH), and c) a model with only depressive symptoms and age could distinguish between the two groups (NSSI + SA vs. NSSI alone). Depressive symptoms and age were included as covariates in the reduced set to account for their potential predictive effects. The reduced set of neurobiological markers showed poor to fair predictive performance (AUC between 0.62 and 0.72) for SA depending on the model. Predictors with the highest predictive value were high DHEA-S (OR = 1.47, 95% CI = 1.04–2.09) and low TSH (OR = 0.68, 95% CI = 0.48–0.97). Complex models slightly outperformed simpler ones and feature selection modestly increased predictive performance. The study may suggest a future potential of biomarkers for the assessment of suicide risk among adolescents with NSSI. Further research is needed to replicate these findings longitudinally.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A biological phenotype of suicide attempt in adolescents with nonsuicidal self-injury: a machine-based learning approach

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年7月29日

查看全文
  • Erik Fink,
  • Corinna Reichl,
  • Stefan Lerch,
  • Julian Koenig,
  • Michael Kaess

摘要

Suicide attempts (SA) are a common risk in adolescents with non-suicidal self-injury (NSSI). In the present study, we investigated whether a set of biological markers contributed (above clinical features) to the distinction of adolescents with NSSI and SA from those with NSSI alone using machine-based learning approaches. Female adolescents engaging in NSSI (n = 161) were recruited from our outpatient clinic for risk-taking and self-harming behavior (AtR!Sk). Different machine-based learning models (logistic regression, elastic net regression, random forests, gradient boosted trees) with repeated cross-validation were applied. We tested whether a) the full set of neurobiological markers, b) a reduced set including preselected markers based on existing evidence (CRP, interleukin-6, salivary cortisol, DHEA-S, TSH, dopamine, norepinephrine, ACTH), and c) a model with only depressive symptoms and age could distinguish between the two groups (NSSI + SA vs. NSSI alone). Depressive symptoms and age were included as covariates in the reduced set to account for their potential predictive effects. The reduced set of neurobiological markers showed poor to fair predictive performance (AUC between 0.62 and 0.72) for SA depending on the model. Predictors with the highest predictive value were high DHEA-S (OR = 1.47, 95% CI = 1.04–2.09) and low TSH (OR = 0.68, 95% CI = 0.48–0.97). Complex models slightly outperformed simpler ones and feature selection modestly increased predictive performance. The study may suggest a future potential of biomarkers for the assessment of suicide risk among adolescents with NSSI. Further research is needed to replicate these findings longitudinally.