Risk prediction models of non-suicidal self-injury among adolescents: a systematic review and meta-analysis

Non-suicidal self-injury (NSSI) among adolescents has serious adverse effects on individuals, families, and society. There has been a significant increase in research on risk prediction models for adolescent NSSI; however, the quality of these models and their clinical applicability remain unclear. We conducted a systematic review and meta-analysis of predictive performance, common predictors and methodological quality of existing NSSI risk prediction models among adolescents. We searched 10 databases—PubMed, Embase, Web of Science, CINAHL, Cochrane Library, PsycINFO, China National Knowledge Infrastructure (CNKI), Wanfang, VIP, and Chinese Biomedical Literature Database—for all studies about adolescent NSSI risk prediction models from inception to August 15, 2025. Data from included studies were extracted following the CHARMS checklist. We assessed the risk of bias and applicability of the included studies using the PROBAST + AI tool. A total of 27 studies including 58 prediction models were included. Included studies primarily originated from hospital ( n = 17) and community settings ( n = 10). The pooled AUC for 17 training models was 0.850 (95% CI: 0.810–0.890), while the pooled AUC for 5 validation models was 0.840 (95% CI: 0.760–0.930). Common predictors of adolescent NSSI included childhood trauma, depressive mood, anxiety, history of NSSI, sleep disorders, stressful life events, and female gender. The majority of studies were rated as low quality ( n = 24) and high risk of bias ( n = 23) in quality assessment, while two studies were rated as high risk of applicability. The existing risk prediction models for NSSI among adolescents have moderate to good predictive performance, but they perform poorly in methodological quality and clinical applicability. Future research should strengthen the validation and calibration of existing models and improve the methodological quality according to the PROBAST criteria.