Abstract
Assessing non-suicidal self-injury (NSSI) in adolescents and young adults is a critical yet challenging aspect of psychiatric evaluations for hospitalized patients. This study aimed to develop a detective model for probability of NSSI in adolescents and young adults utilizing a retrospective cross-sectional analysis. Data from 658 hospitalized adolescents and young adults, including demographic characteristics, hormone levels, and violence risk factors, were collected. Age, history of suicide attempts, gender, and psychiatric diagnosis were identified as key detectors through Boruta and LASSO machine learning algorithms, leading to the construction of a nomogram. Model performance was evaluated based on discrimination, calibration, and decision curve analysis (DCA). The model achieved the area under the receiver operating characteristic curve (AUC) values of 0.803 (training set; 95% CI 0.763-0.843) and 0.745 (validation set; 95% CI 0.676-0.814). Calibration plots demonstrated strong alignment between predicted and actual outcomes. The Hosmer-Lemeshow test indicated good model fit, while DCA revealed clinically relevant threshold ranges for the training and validation sets, highlighting the model's potential to inform clinical decision-making. This detective model can support the rational allocation of medical resources and facilitate the early detection and intervention of NSSI behaviors in hospitalized adolescents and young adults.