Leveraging Feature Importance Methods to Characterize Explosive Ordnance Disposal Veterans' Health Profiles

Area of Science:

• Utilizes machine learning and data analytics in military health research.
• Focuses on identifying unique health signatures in veteran populations.

Background:

• Explosive Ordnance Disposal (EOD) Veterans face complex exposures potentially leading to chronic health issues.
• Distinguishing EOD Veteran health features is crucial for understanding their symptom burden.

Purpose of the Study:

• To apply machine learning to identify unique health symptoms differentiating EOD and non-EOD Veterans.
• To characterize EOD Veterans' health profiles for prioritized clinical services and research.

Main Methods:

• Analyzed clinical intake data from 52 EOD and 794 non-EOD Veterans.
• Employed the Boruta algorithm for feature selection from validated questionnaires and clinical measures.
• Trained a balanced random forest classifier to identify discriminating health features.

Main Results:

• A balanced random forest classifier achieved a ROC AUC score of 0.93.
• The Boruta algorithm selected 21 key features from 143 initial clinical intake variables.
• Combat exposure, physical functioning, and Traumatic Brain Injury (TBI) symptom burden were identified as highly significant.

Conclusions:

• Aggregated clinical data and machine learning effectively characterize EOD health profiles.
• Prioritizing attention to combat exposure and physical functioning is vital for EOD Veteran care.
• The methodology can be adapted for other complex exposures and emerging health concerns in veterans.