Inference on summaries of a model-agnostic longitudinal variable importance trajectory with application to suicide...

Area of Science:

• Psychiatry and Mental Health
• Data Science and Machine Learning
• Biostatistics

Background:

• Suicide attempt risk fluctuates over time, necessitating dynamic assessment.
• Identifying key risk factors during mental health visits aids clinical decision-making.
• Longitudinal data analysis is essential for understanding time-varying risk in mental health care.

Purpose of the Study:

• To develop methods for assessing variable importance trajectories in longitudinal mental health data.
• To provide robust estimation and inference procedures for time-varying risk factors.
• To enable the analysis of complex machine learning models for suicide risk prediction.

Main Methods:

• Defined summaries of variable importance trajectories and corresponding estimators.
• Proposed a nonparametric efficient estimation and inference procedure.
• Developed a null hypothesis testing procedure valid for complex prediction algorithms.

Main Results:

• Simulations confirmed the proposed procedures exhibit good operating characteristics.
• Applied the methods to electronic health records data from two large health systems.
• Investigated the longitudinal importance of risk factors for suicide attempts.

Conclusions:

• The developed methods offer a robust approach to analyzing time-varying risk factors in mental health.
• Findings inform future suicide prevention research and clinical workflows by highlighting dynamic risk patterns.
• This work enhances the ability to predict and manage suicide risk over time using electronic health records.