Trajectories of self-reported suicide risk with interpretative and predictive values for subsequent suicide attempts: A retrospective cohort study among crisis hotline callers in China

Background Identifying suicide risk and predicting subsequent attempts are critical for crisis hotline interventions. While structured scales are commonly used, self-reported suicide risk scores may serve as a more accessible and caller-centered indicator. Yet, its predictive and interpretive value for future suicide attempts (SAs) remains unclear. This study aims to examine trajectories of self-reported scores and their associations with subsequent SAs. Methods Data were collected from the Guangzhou Psychological Crisis Hotline (Jan 2023–Dec 2024). A total of 1,235 callers reporting suicide attempts, ideation, or plans during the index call included and were followed up on days 1, 7, 15, and 30. Latent class mixed models identified distinct trajectories of self-reported scores. Associations between trajectory-based classes and subsequent SAs were examined using Cox proportional hazards models with inverse probability weighting. An XGBoost model with SHAP values assessed predictor importance. Results Three trajectory groups emerged as consistent high-risk (30.0%), gradual decline (12.0%), and rapid decline (58.0%). Trajectory-based class had the highest SHAP values for subsequent SA prediction. Compared with the consistent high-risk group, the rapid decline group showed an 83% lower risk of SAs ( HR 0.17, 95% CI 0.10–0.29). Lower SA risk was also linked with older age, calls in summer and winter, and suicide ideation (vs. attempts), while prior depression and bipolar disorder increased risk. Conclusions Trajectories of self-reported suicide risk score could reliably interpret and predict for subsequent SAs. Callers with consistently high suicide risk scores, younger age, prior depression, bipolar disorders and SA require prioritized intervention.