Developing a suicide risk model for use in the Indian Health Service

Area of Science:

• Public Health
• Health Informatics
• Mental Health Research

Background:

• Suicide remains a critical public health issue, particularly within American Indian communities.
• Existing suicide risk screening tools may lack sufficient predictive accuracy for diverse patient populations.
• Electronic Health Records (EHR) contain rich data that could potentially improve suicide risk assessment.

Purpose of the Study:

• To develop and evaluate an EHR-based model for predicting suicide risk in American Indian patients.
• To compare the performance of machine learning models against augmented traditional screening methods.
• To assess the utility of an EHR-derived suicide risk model within the Indian Health Service (IHS) setting.

Main Methods:

• Utilized EHR data from 16,835 American Indian patients (331,588 visits) between 2017 and 2021.
• Developed logistic regression and random forest models incorporating demographics, medications, diagnoses, and screening scores.
• Compared model performance (AUROC) against an enhanced suicide screening protocol including prior attempts/ideation.

Main Results:

• The EHR-based logistic regression and random forest models achieved an AUROC of 0.83 (0.80-0.86).
• Enhanced suicide screening demonstrated a lower predictive performance with an AUROC of 0.64 (0.61-0.67).
• The developed EHR models significantly outperformed the augmented screening method in predicting suicide attempts or death within 90 days.

Conclusions:

• EHR-based suicide risk models show substantial promise for enhancing clinical decision-making in American Indian patient populations.
• These data-driven models offer a more accurate approach to identifying individuals at high risk for suicide compared to current screening practices.
• Implementing such EHR models could improve targeted interventions and reduce suicide rates within IHS clinics.