BackgroundNon‐suicidal self‐injury (NSSI) is common among adolescents receiving inpatient psychiatric treatment and the months post‐discharge is a high‐risk period for self‐injurious behavior. Thus, identifying predictors that shape the course of post‐discharge NSSI may provide insights into ways to improve clinical outcomes. Accordingly, we used machine learning to identify the strongest predictors of NSSI trajectories drawn from a comprehensive clinical assessment.MethodsThe study included adolescents (N = 612; females n = 435; 71.1%) aged 13–19‐years‐old (M = 15.6, SD = 1.4) undergoing inpatient treatment. Youth were administered clinical interviews and symptom questionnaires at intake (baseline) and before termination. NSSI frequency was assessed at 1‐, 3‐, and 6‐month follow‐ups. Latent class growth analyses were used to group adolescents based on their pattern of NSSI across follow‐ups.ResultsThree classes were identified: Low Stable (n = 83), Moderate Fluctuating (n = 260), and High Persistent (n = 269). Important predictors of the High Persistent class in our regularized regression models (LASSO) included baseline psychiatric symptoms and comorbidity, past‐week suicidal ideation (SI) severity, lifetime average and worst‐point SI intensity, and NSSI in the past 30 days (bs = 0.75–2.33). Only worst‐point lifetime suicide ideation intensity was identified as a predictor of the Low Stable class (b = −8.82); no predictors of the Moderate Fluctuating class emerged.ConclusionsThis study found a set of intake clinical variables that indicate which adolescents may experience persistent NSSI post‐discharge. Accordingly, this may help identify youth that may benefit from additional monitoring and support post‐hospitalization.

中文翻译：

---

预测青少年非自杀性自残的轨迹


背景非自杀性自伤（NSSI）在接受住院精神治疗的青少年中很常见，出院后几个月是自伤行为的高危期。因此，确定影响出院后 NSSI 进程的预测因素可能会为改善临床结果的方法提供见解。因此，我们使用机器学习来识别从综合临床评估中得出的 NSSI 轨迹的最强预测因子。方法该研究包括 13-19 岁青少年（N = 612；女性 n = 435；71.1%）（M = 15.6） ，SD = 1.4）正在接受住院治疗。青少年在入学时（基线）和终止前接受了临床访谈和症状问卷调查。在 1、3 和 6 个月的随访中评估 NSSI 频率。潜在类别增长分析用于根据随访期间的 NSSI 模式对青少年进行分组。结果确定了三个类别：低稳定 (n = 83)、中度波动 (n = 260) 和高持续 (n = 269)。我们的正则化回归模型 (LASSO) 中高持续性类别的重要预测因素包括基线精神症状和合并症、过去一周的自杀意念 (SI) 严重程度、终生平均和最坏点 SI 强度以及过去 30 天内的 NSSI（bs = 0.75–2.33）。只有一生最坏点的自杀意念强度被确定为低稳定类别的预测因子（b = -8.82）；没有出现中度波动类别的预测因子。结论这项研究发现了一组摄入临床变量，表明哪些青少年可能会在出院后经历持续的 NSSI。因此，这可能有助于识别可能受益于额外监测和住院后支持的青少年。