Computing 3-Step Theory of Suicide Factor Scores From Veterans Health Administration Clinical Progress Notes

Area of Science:

• Psychiatry
• Computational Linguistics
• Health Informatics

Background:

• No established methods exist for scoring 3-step theory of suicide (3ST) factors from clinical notes.
• This study addresses the gap by developing and evaluating scoring options for 3ST factors.

Purpose of the Study:

• To determine the optimal scoring method for 3ST factors (psychological pain, hopelessness, connectedness, capability for suicide) using clinical progress notes.
• To test hypotheses related to the 3ST using natural language processing (NLP) and logistic regression.

Main Methods:

• Utilized terminology-driven NLP to extract data from Veterans Health Administration (VHA) clinical progress notes.
• Evaluated scoring options based on extraction counts for each 3ST factor.
• Employed logistic regression models to test 3ST hypotheses, adjusting for demographic characteristics.

Main Results:

• Optimal discrimination between patient groups was achieved using specific scoring contrasts for psychological pain, hopelessness, and capability for suicide.
• The scoring method for connectedness also showed effectiveness in differentiating patient groups.
• Results indicate that terminology-driven 3ST factor scores can effectively discriminate between patients who attempt or die by suicide and those without suicidal ideation or attempts.

Conclusions:

• Validated 3ST factor scores derived from clinical notes demonstrate the theory's applicability to VHA patients.
• Further research is recommended to confirm findings in a broader VHA population cohort.