Identifying momentary suicidal ideation using machine learning in patients at high-risk for suicide

Area of Science:

• Psychiatry
• Computational Psychiatry
• Machine Learning in Healthcare

Background:

• Accurate detection of suicidal ideation (SI) is crucial for timely interventions, especially during care transitions.
• Machine learning shows promise in classifying SI using momentary data.
• This study investigates how training data type affects machine learning model accuracy for SI classification.

Purpose of the Study:

• To evaluate the impact of different training data types (baseline vs. momentary) on machine learning model accuracy for classifying suicidal ideation (SI).
• To identify key risk factors contributing to the classification of SI characteristics.
• To explore the potential of machine learning in guiding intervention strategies for SI.

Main Methods:

• 257 psychiatric inpatients participated in a 3-week ecological momentary assessment and suicide risk factor evaluation.
• Machine learning models were trained using baseline and/or momentary suicide risk data.
• Feature importance metrics were analyzed to determine key predictors of SI characteristics.

Main Results:

• Models incorporating both baseline and momentary features demonstrated superior performance in classifying SI presence, duration, and intensity compared to models using single data types.
• The models effectively differentiated individual characteristics of SI.
• Classification accuracy for SI presence, duration, and intensity was comparable across models.

Conclusions:

• Machine learning approaches effectively identify characteristics of suicidal ideation (SI).
• Understanding the factors driving different SI characteristics is vital for developing targeted interventions.
• Further research is needed to generalize findings beyond high-risk inpatient samples and explore temporal relationships.