Developing a Risk Model to Target High-risk Preventive Interventions for Sexual Assault Victimization among Female...

Area of Science:

• Military health
• Public health
• Epidemiology

Background:

• Sexual violence is a significant issue within the U.S. military, particularly affecting female personnel.
• Effective preventive interventions require accurate identification of high-risk individuals.
• Administrative data on sexual assault is often incomplete, necessitating advanced modeling techniques.

Purpose of the Study:

• To develop a predictive model for identifying female U.S. Army soldiers at high risk of sexual violence victimization.
• To utilize theoretically-guided predictors from administrative data for model development.
• To combine predictions from models of administratively-recorded and self-reported victimization.

Main Methods:

• Developed parallel machine learning models to predict administratively-recorded and self-reported sexual assault victimization.
• Employed capture-recapture methods to integrate predictions from both models.
• Identified key predictors including low status, crime involvement, and treated mental disorders.

Main Results:

• Achieved high predictive accuracy with an Area Under the ROC Curve of 0.83-0.88.
• Identified that 33.7-63.2% of victimizations occurred within the highest-risk 5% of soldiers.
• Demonstrated a significant concentration of risk among a small subgroup of personnel.

Conclusions:

• The developed targeting models show promise for guiding preventive interventions against sexual violence in the U.S. Army.
• Accurate risk stratification can potentially reduce the prevalence of sexual violence.
• Further evaluation is needed to balance intervention costs, effectiveness, and competing risks.