Predicting Next-Day Passive Suicidal Ideation in At-Risk Youth.

Area of Science:

• Adolescent psychiatry
• Computational psychiatry
• Suicidology

Background:

• Passive suicidal ideation (SI) is a significant risk factor for suicidal behavior, yet it is less studied than active SI.
• Machine learning (ML) has shown promise in predicting active SI, but passive SI remains an understudied prediction target.

Purpose of the Study:

• To investigate the feasibility and accuracy of using ML to predict next-day passive SI in psychiatrically hospitalized youth.
• To compare the predictive value of time-varying features versus baseline variables for passive SI.

Main Methods:

• Seventy-eight adolescents (13-17 years) completed daily risk and protective factor ratings for 28 days post-discharge.
• Multiple ML models were trained to predict passive SI, comparing models with and without baseline data.

Main Results:

• ML models achieved high accuracy (AUC = 0.90) in predicting next-day passive SI.
• Within-person 7-day moving averages of passive SI duration and frequency were the strongest predictors.
• Baseline variables had minimal impact on prediction accuracy, even early in the post-discharge period.

Conclusions:

• Forecasting next-day passive SI using ML is feasible and highly accurate, highlighting its potential for suicide prevention.
• Time-varying features are more predictive of passive SI than baseline factors.
• Integrating passive SI prediction into personalized interventions may improve suicide prevention efforts.