Machine learning algorithms and traditional statistical models for detection of dementia: a population-based study

Area of Science:

• Gerontology
• Medical Informatics
• Epidemiology

Background:

• Early dementia detection is crucial for timely interventions to slow disease progression.
• This study focuses on diagnosing dementia in a rural Chinese older population.
• Evaluating machine learning and logistic regression for dementia diagnosis.

Purpose of the Study:

• To assess the performance of machine learning algorithms (Random Forest, XGBoost) and logistic regression (LR) for dementia detection.
• To compare the diagnostic accuracy and clinical utility of these models in a rural Chinese older population.
• To identify a practical tool for early dementia screening in resource-limited settings.

Main Methods:

• A population-based study involving 5200 participants.
• Least Absolute Shrinkage and Selection Operator (LASSO) regression for predictor selection in LR.
• Implementation of Random Forest (RF) and eXtreme Gradient Boosting (XGBoost) models.
• Model performance evaluation using Area Under the Receiver Operating Characteristic Curve (AUC), calibration plots, and Decision Curve Analysis (DCA).

Main Results:

• The XGBoost model achieved the highest Area Under the Receiver Operating Characteristic Curve (AUC) of 0.95 (95% CI: 0.94-0.98) in the validation set.
• Logistic Regression (LR) and Random Forest (RF) models showed AUCs of 0.88.
• All models demonstrated good calibration and superior clinical utility compared to no intervention or universal intervention strategies.

Conclusions:

• Machine learning models, particularly XGBoost, show excellent performance in detecting dementia in a rural Chinese older population.
• These models offer high accuracy and clinical utility for dementia detection.
• The models utilize readily available clinical data, facilitating easy implementation in routine practice for rural elderly populations.