Leveraging unstructured electronic medical record notes to derive population-specific suicide risk models

Area of Science:

• Computational psychiatry
• Clinical informatics
• Natural Language Processing (NLP)

Background:

• Traditional suicide risk prediction models use structured Electronic Medical Record (EMR) data.
• Unstructured EMR notes contain nuanced clinical information that could enhance prediction accuracy.

Purpose of the Study:

• To develop and validate a data-driven suicide risk prediction model using NLP on unstructured EMR notes.
• To assess the predictive performance of NLP-derived models compared to existing methods.

Main Methods:

• A matched case-control study of U.S. Department of Veterans Affairs (VA) patients (2015-2016).
• Natural Language Processing (NLP) applied to a large corpus of EMR notes.
• Machine learning classification algorithms used for prediction, with Area Under the Curve (AUC) for accuracy assessment.

Main Results:

• NLP-derived models demonstrated strong predictive accuracy for suicide risk.
• The top 10% of patients identified by the risk model accounted for 29% of suicide decedents.
• The NLP model showed favorable comparison to other leading prediction methods.

Conclusions:

• Leveraging unstructured EMR notes via NLP offers a powerful tool for suicide risk prediction.
• The approach is highly implementable, requiring only text data access and open-source software.
• Future research should explore ensemble models combining NLP-derived and structured data.