Risk prediction models for non-suicidal self-injury in adolescents with depression: a systematic review and meta-analysis

Non-suicidal self-injury (NSSI) is prevalent among adolescents with depression and is a significant predictor of suicidal behavior. Although several risk prediction models for NSSI have been developed, their overall performance, methodological quality, and feasibility for clinical translation remain unclear.

This study aimed to comprehensively evaluate the performance of NSSI risk prediction models for adolescents with depression through a systematic review and meta-analysis, summarize their area under the curve (AUC) values, and determine the incidence of NSSI and associated predictors in this population.

PubMed, Web of Science, The Cochrane Library, CINAHL, Embase, Scopus, PsycINFO, CNKI, and Wanfang Database were searched from inception to September 20, 2025. The GRADE approach, PROBAST, and TRIPOD tools were used to assess the overall certainty of evidence, risk of bias, and reporting quality, respectively. Data were extracted using the CHARMS checklist. The PROBAST and TRIPOD tools were used to assess the risk of bias and reporting quality, respectively. Meta-analyses of model AUC values and NSSI incidence were performed using Stata 17.0.

Eighteen studies (comprising 40 models) were included. The meta-analysis revealed a pooled NSSI incidence of 50% (95% CI = 0.29-0.71) among adolescents with depression. The models demonstrated promising discriminative ability: the pooled AUC was 0.82 (95% CI = 0.79-0.86) for development cohorts, 0.90 (95% CI = 0.87-0.93) for internal validation cohorts, and 0.82 (95% CI = 0.77-0.86) for external validation cohorts. The most frequently identified predictors included depression severity, childhood trauma, sex, emotional abuse, and emotional neglect. However, the PROBAST assessment revealed a high risk of bias in most studies, particularly in the domains of participant selection and statistical analysis. The TRIPOD assessment indicated insufficient reporting of key methodological details such as sample size justification and handling of missing data. Only five studies conducted external validation. The certainty of evidence assessed by GRADE was low to very low for the main outcomes, mainly because of high risk of bias, substantial heterogeneity, indirectness, and limited validation evidence.

Existing prediction models show promising discriminative potential for identifying NSSI risk in adolescents with depression. However, high AUC values should be interpreted cautiously because methodological bias, substantial heterogeneity, limited external validation, incomplete reporting, and population differences may have led to overly optimistic performance estimates. Future research should focus on building more robust, trustworthy, and clinically translatable tools through prospective designs, adequate sample sizes, standardized analytical methods, extensive external validation, and complete reporting adhering to the TRIPOD guidelines.