Primary-care-focused interpretable machine learning model for depression screening in geriatrics: A comparative study...

Area of Science:

• Geriatric psychiatry
• Computational medicine
• Public health

Background:

• High prevalence of depression in older adults strains healthcare systems.
• Shortage of psychiatrists hinders timely professional evaluations.
• Need for effective screening tools in primary care settings.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for screening geriatric depression.
• To assist primary care practitioners in early detection and management of depression in older adults.
• To address the challenges posed by the shortage of mental health specialists.

Main Methods:

• Utilized National Health and Nutrition Examination Survey (NHANES) data (2011-2018).
• Employed Least Absolute Shrinkage and Selection Operator (LASSO) regression for feature selection.
• Applied Synthetic Minority Over-sampling Technique (SMOTE) for class imbalance and seven ML algorithms for model development.
• Evaluated models using ML metrics, clinical impact measures, and Shapley Additive exPlanations (SHAP) for interpretability.

Main Results:

• Included 3802 participants; 24.54% had depression.
• Extreme Gradient Boosting (XGBoost) model achieved the highest performance (accuracy 0.82, AUC 0.88).
• Key predictors identified: sleep disorders, gender, poverty-income ratio (PIR), serum albumin, and segmented neutrophil count.

Conclusions:

• The developed ML model demonstrates strong predictive performance for geriatric depression screening.
• The model shows clinical applicability, supporting early identification and management by healthcare workers.
• This approach can help mitigate the impact of depression in the aging population.