Decoding vital variables in predicting different phases of suicide among young adults with childhood sexual abuse: a machine learning approach.

Young adults with childhood sexual abuse (CSA) are an especially vulnerable group to suicide. Suicide encompasses different phases, but for CSA survivors the salient factors precipitating suicide are rarely studied. In this study, from a progressive perspective of suicidal thoughts and behaviors (STB), we aim to identify distinct risk factors for predicting different stages of STB, i.e., suicidal ideation (SI), suicide plan (SP), and suicide attempt (SA), among young adults with CSA experience. Based on mental health profiles of 4,070 young adult CSA survivors from a cross-sectional survey, we constructed five random forest classification models to respectively classify high suicidality, SI, SP, and SA. The common crucial factors for predicting SI, SP, and SA included NSSI and depression. The special important predictors for SI included OCD, anxiety, PTSD, and social rhythm. Co-occurrence of other types of childhood abuse and traumatic events was a special important predictor for SP among participants with SI. Self-compassion was the most crucial factor in classifying SA from those with SI. Social rhythm, co-occurrence of other types of childhood abuse, domestic violence, fear of happiness, and self-compassion made specific contribution to the prediction of SI, SP, and SA. However, the random forest model failed to accurately classify SA from those with SP, which was consistent with existing research. Our findings highlighted the importance of identifying suicidal characteristics for specified interventions at different stages of suicide for young people with CSA experiences.