Child and adolescent psychiatry and mental health · 2026
Integrating neurobiological markers to prospectively predict adolescent non-suicidal self-injury and suicide attempts: a machine learning approach.
СтатьяНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Reichl Corinna, Fink Erik, von den Driesch Luisa, Lerch Stefan, Koenig Julian, Berger Thomas, Kaess Michael
Аннотация
BACKGROUND: Suicide attempts (SA) and non-suicidal self-injury (NSSI) are closely related phenomena, both common in adolescent clinical samples. This study examined whether neurobiological markers could enhance the prediction of SA and NSSI over a 2-year period, beyond established clinical predictors and age, in a high-risk sample of female adolescents engaging in NSSI.
METHODS: A total of n = 63 (age: M = 15.1 years, SD = 1.45) female adolescents with NSSI were recruited from our outpatient clinic for risk-taking and self-harming behaviour (AtR!Sk) at the Clinic for Child and Adolescent Psychiatry, University Hospital of Heidelberg, Germany. Machine learning models (linear and logistic regression; elastic net regression; support vector machines with repeated cross-validation) were applied. We tested whether the inclusion of biomarkers (thyroid-stimulating hormone [TSH]; free triiodothyronine [fT3]; adrenocorticotropic hormone [ACTH]; dehydroepiandrosterone sulfate [DHEA-S]; resting heart rate variability [rHRV]; pain threshold) improved prediction of SA and NSSI remission - defined as the absence of any NSSI episodes - over a two years period, beyond clinical variables (past SA; symptoms of borderline personality disorder [BPD]) and age.
RESULTS: Models predicting SA that included biomarkers, clinical variables, and age, showed moderate predictive accuracy (AUC: 0.76-0.82), however not higher than when only including clinical variables and age. The strongest predictors were past SA (OR = 7.28), followed by fT3 (OR = 0.16) and TSH (OR = 0.35). The SA prediction models showed high specificity but low sensitivity, indicating strong performance in identifying negative cases but poor detection of positive cases. Models predicting NSSI, including the same variables, did not consistently outperform chance levels and showed low specificity and sensitivity.
CONCLUSIONS: These findings support the use of machine learning models to predict SA in high-risk adolescents but show that prediction is driven mainly by established clinical features (past SA; BPD) and age, with no added benefit from neurobiological markers. Moreover, all models were more effective at identifying negative than positive cases. Given the small sample size and class imbalance, findings should be interpreted cautiously and require replication in larger and independent samples.
METHODS: A total of n = 63 (age: M = 15.1 years, SD = 1.45) female adolescents with NSSI were recruited from our outpatient clinic for risk-taking and self-harming behaviour (AtR!Sk) at the Clinic for Child and Adolescent Psychiatry, University Hospital of Heidelberg, Germany. Machine learning models (linear and logistic regression; elastic net regression; support vector machines with repeated cross-validation) were applied. We tested whether the inclusion of biomarkers (thyroid-stimulating hormone [TSH]; free triiodothyronine [fT3]; adrenocorticotropic hormone [ACTH]; dehydroepiandrosterone sulfate [DHEA-S]; resting heart rate variability [rHRV]; pain threshold) improved prediction of SA and NSSI remission - defined as the absence of any NSSI episodes - over a two years period, beyond clinical variables (past SA; symptoms of borderline personality disorder [BPD]) and age.
RESULTS: Models predicting SA that included biomarkers, clinical variables, and age, showed moderate predictive accuracy (AUC: 0.76-0.82), however not higher than when only including clinical variables and age. The strongest predictors were past SA (OR = 7.28), followed by fT3 (OR = 0.16) and TSH (OR = 0.35). The SA prediction models showed high specificity but low sensitivity, indicating strong performance in identifying negative cases but poor detection of positive cases. Models predicting NSSI, including the same variables, did not consistently outperform chance levels and showed low specificity and sensitivity.
CONCLUSIONS: These findings support the use of machine learning models to predict SA in high-risk adolescents but show that prediction is driven mainly by established clinical features (past SA; BPD) and age, with no added benefit from neurobiological markers. Moreover, all models were more effective at identifying negative than positive cases. Given the small sample size and class imbalance, findings should be interpreted cautiously and require replication in larger and independent samples.