Canonical text (the exact bytes fingerprinted)

TRUVACE RECORD VERSION
record: TRV-2026-1870
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-10-19T01:08:58.203580Z
status: published
lens: trace
sector: health
headline: Development and Internal Validation of an Interpretable Machine Learning Model for Identifying Past-Year Nonsuicidal Self-Injury Among Adolescents With Depression: Retrospective Study
dek: Background Nonsuicidal self-injury (NSSI) is common among adolescents with depressive disorders and is associated with substantial clinical burden. Machine learning may assist in identifying multidimensional patterns associated with NSSI, although its incremental value beyond direct clinical assessment remains uncertain. Objective This study compares 7 supervised machine learning model variants for classifying past-year NSSI status among adolescents with depressive disorders and conducts detailed analyses of int…
gain_title: In a held-out test set, a random forest classified past-year self-injury among adolescents with depression with AUROC 0.681 and showed greater net benefit than assess-all or assess-none strategies at thresholds of roughly 0.51-0.80.
problem_title: The full random forest did not outperform a suicidal-ideation-only model, and routinely collected biochemical variables showed near-chance discrimination, so the model's added value beyond direct clinical assessment is doubtful.
trace_subject: Machine learning classification of past-year nonsuicidal self-injury in adolescents with depression
gain_reading: In a held-out test set, a random forest classified past-year self-injury among adolescents with depression with AUROC 0.681 and showed greater net benefit than assess-all or assess-none strategies at thresholds of roughly 0.51-0.80.
gain_evidence: the random forest provided greater net benefit than the assess-all and assess-none strategies across threshold probabilities of approximately 0.51-0.80 | The random forest achieved an AUROC of 0.681
problem_reading: The full random forest did not outperform a suicidal-ideation-only model, and routinely collected biochemical variables showed near-chance discrimination, so the model's added value beyond direct clinical assessment is doubtful.
problem_evidence: The full random forest model did not outperform a suicidal-ideation-only model (AUROC 0.701) | A biochemical-only model showed near-chance discrimination (AUROC 0.513)
quick_read: Researchers retrospectively analyzed 410 adolescents aged 10-19 with depressive disorders treated at four hospitals in Zhejiang Province, China, and compared seven supervised machine learning variants for classifying past-year nonsuicidal self-injury. A random forest had the best observed held-out test performance, with moderate discrimination and a calibration slope near 1.

The results matter because the model leaned heavily on suicidality information that clinicians can assess directly, so its incremental value over clinical assessment looks limited. The study cannot show prospective risk prediction, and external and temporal validation would be needed before any clinical use.
limitation: Because this was a retrospective cross-sectional classification study, the findings do not establish prospective risk prediction, and external and prospective temporal validation is required before clinical implementation.
tag: Dual reading
key_points: Of 410 eligible adolescents, 274 reported nonsuicidal self-injury during the preceding year; 67 assessment-time features were used. | The random forest reached an AUROC of 0.681, an AUPRC of 0.808, sensitivity of 0.720 and specificity of 0.561 in the held-out test set. | Pairwise AUROC differences between the random forest and the other models were not significant after Holm correction. | Removing suicidal ideation and previous suicide attempt cut the AUROC to 0.540, and suicidal ideation was the most consistently important feature.
rundown: The data were split with stratified random sampling into a 70% training set of 287 and an untouched 30% test set of 123. Hyperparameters were tuned with 5-fold cross-validation inside the training set only.

The radial basis function support vector machine had the highest mean cross-validated AUROC in training at 0.720, while the random forest had the highest observed test AUROC and AUPRC and the lowest Brier score at 0.212. Interpretability analyses used Gini importance, permutation importance and SHAP.

The authors describe the work as a retrospective cross-sectional classification study, not a demonstration of prospective prediction.
sources:
- peer_reviewed | JMIR Pediatrics and Parenting | https://doi.org/10.2196/95340 | 2026-10-08
prev: 0000000000000000000000000000000000000000000000000000000000000000

sha256

046194ee1b4fd72cd35b73ef68c9df13245bb259809719d357ef0c96b3097c54

previous

0000000000000000000000000000000000000000000000000000000000000000