The role of suicide ideation in assessing near-term suicide risk: A machine learning approach

The majority of suicide attempters do not disclose suicide ideation (SI) prior to making an attempt. When reported, SI is nevertheless associated with increased risk of suicide. This paper implemented machine learning (ML) approaches to assess the degree to which current and lifetime SI affect the predictive validity of the Suicide Crisis Syndrome (SCS), an acute condition indicative of imminent risk, for near-term suicidal behaviors (SB ).

In a sample of 591 high-risk inpatient participants, SCS and SI were respectively assessed using the Suicide Crisis Inventory (SCI) and the Columbia Suicide Severity Rating Scale (C-SSRS). Two ML predictive algorithms, Random Forest and XGBoost, were implemented and framed using optimism adjusted bootstrapping. Metrics collected included AUPRC, AUROC, classification accuracy, balanced accuracy, precision, recall, and brier score. AUROC metrics were compared by computing a z-score.

The combination of current SI and SCI showed slightly higher predictive validity for near-term SB as evidenced by AUROC metrics than the SCI alone, but the difference was not significant (p<0.05). Current SI scored the highest amongst a chi square distribution in regards to predictors of near-term SB.

The addition of SI to the SCS does not materially improve the model's predictive validity for near-term SB, suggesting that patient self-reported SI should not be a requirement for the diagnosis of SCS.