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

BACKGROUND

The prevalence of depression among older adults places a considerable strain on healthcare systems due to a shortage of psychiatrists for professional evaluations. This study reports a machine learning (ML) model to assist in screening for geriatric depression, enabling primary care practitioners to detect and respond to cases effectively and on time.

METHODS

Data from the 2011-2018 National Health and Nutrition Examination Survey (NHANES) were used. To identify relevant variables for depression screening, features were selected using Least Absolute Shrinkage and Selection Operator (LASSO) regression. Synthetic Minority Over-sampling Technique (SMOTE) was applied to address class imbalance in the training set. Seven ML algorithms were employed to develop predictive screening models. Model performance was evaluated using standard ML metrics and clinically relevant impact measures, ensuring a comprehensive assessment. To improve interpretability, Shapley Additive exPlanations (SHAP) were used to visualize the contribution of each feature.

RESULTS

The analysis included 3802 participants, with 933 (24.54 %) identified as having depression. Thirteen key variables were selected for model development. The Extreme Gradient Boosting (XGBoost) model demonstrated the best performance, with an accuracy of 0.82 in test set and a maximum Area Under the Curve (AUC) of 0.88 in Receiver Operating Characteristic (ROC) analysis. Sleep disorders, gender, poverty-income ratio (PIR), serum albumin levels, and segmented neutrophil count were identified as the influential predictors.

CONCLUSION

The ML model developed for screening depression in older adults showed strong predictive performance and clinical applicability, supporting health workers in the early identification and management of depression.