JMIR Publications

ABSTRACT

Background:

Non-suicidal self-injury (NSSI) is common among adolescents with depression. Early identification of high-risk individuals is essential but challenging.

Objective:

This study aims to develop a predictive model to estimate the risk of NSSI in depressed adolescents using the biopsychosocial approach.

Methods:

This cross-sectional study included 437 adolescents aged 10–19 years with depression from four hospitals in Zhejiang, China (2021–2023). Sixty-eight variables covering sociodemographic, physiological-biochemical, and high-risk behavioral factors were collected. Data were split into validation (70%) and test (30%) sets. Seven machine learning models, including random forest and support vector machine learning models, were applied to identify the best model for predicting the risk of NSSI. We developed a personalized, interpretable risk prediction platform using SHapley Additive exPlanations (SHAP).

Results:

The random forest model achieved the best performance, with an accuracy of 81.2%, recall of 84.3%, F1-score of 81.9%, precision of 79.6%, and AUC of 0.872 on the test set. Key predictors included suicidal ideation, high-sensitivity C-reactive protein (hs-CRP), creatine kinase, sex, and age. The SHAP analysis provided insights into feature contributions.

Conclusions:

This study developed a predictive model to estimate the risk of NSSI in depressed adolescents using the biopsychosocial approach, offering a valuable tool for early risk identification and personalized intervention.