Clinical Manifestations.

Area of Science:

• Geriatric Medicine
• Data Science
• Transportation Safety

Background:

• Depression in older adults is often underdiagnosed, leading to serious consequences like impaired driving.
• Older adults' reliance on vehicles for independence necessitates innovative screening solutions.
• Naturalistic driving behaviors are explored as a novel indicator for detecting major depressive disorder (MDD).

Purpose of the Study:

• To investigate the efficacy of machine learning models in identifying MDD in older adults using naturalistic driving data.
• To assess the predictive power of driving metrics, demographic factors, and medication data for depression screening.
• To develop a scalable and functional method for early depression detection in the geriatric population.

Main Methods:

• 157 older adults (81 with MDD) had naturalistic driving data collected over two years.
• Extreme Gradient Boosting (XGBoost) models were trained using driving metrics (braking, trip length, entropy), demographic, and medication data.
• Hyperparameter tuning and 10-fold cross-validation were employed to optimize model performance.

Main Results:

• Driving features alone achieved high predictive accuracy (F1 score 0.82, AUC 0.84).
• Models combining driving data with medication classes showed the highest performance (F1 score 0.81, AUC 0.86).
• Key predictors included hard cornering rate, total medication use, and trip distances.

Conclusions:

• Naturalistic driving data shows significant potential for detecting depression in older adults with high specificity and recall.
• Integrating driving behaviors with medication data enhances predictive accuracy for a scalable, functional screening approach.
• AI-driven tools can aid in geriatric mental health diagnostics, improving interventions and public safety.