Interpretable Estimation of Suicide Risk and Severity from Complete Blood Count Parameters with Explainable...

Area of Science:

• Medical research
• Artificial Intelligence in Healthcare
• Clinical Diagnostics

Background:

• Peripheral inflammatory markers play a role in suicidal behavior.
• Practical clinical methods for assessing suicide risk using these markers are underdeveloped.
• Explainable artificial intelligence (xAI) offers a novel approach to analyze complex biological data.

Purpose of the Study:

• To develop predictive models for suicide risk and attempt severity using complete blood count (CBC) values.
• To leverage explainable artificial intelligence (xAI) for understanding the relationship between CBC parameters and suicidal behavior.
• To create a clinically applicable tool for suicide risk assessment.

Main Methods:

• Utilized data from 544 suicide attempters and 458 healthy individuals (2010-2020).
• Employed machine learning algorithms including Random Forest, Logistic Regression, Support Vector Machines, and XGBoost.
• Applied SHAP for explainability to identify key CBC predictors in the optimal XGBoost model.

Main Results:

• The XGBoost model demonstrated superior performance, achieving 0.83 accuracy for suicide risk prediction and 0.943 for suicide attempt severity.
• Identified specific CBC parameters (e.g., NEU, WBC, HCT, PLT) and age as significant contributors to suicide risk prediction.
• Highlighted distinct CBC profiles associated with violent suicide attempts versus general suicide risk.

Conclusions:

• An xAI model utilizing CBC values shows promise for clinical detection of suicide risk.
• The developed models can aid in assessing both the risk and severity of suicidal behavior.
• This research provides a foundation for integrating AI-driven CBC analysis into psychiatric practice.