Machine learning-based risk prediction model for cognitive dysfunction in elderly individuals.

Area of Science:

• Gerontology
• Medical Informatics
• Computational Biology

Background:

• Globalisation has increased cognitive dysfunction prevalence in the elderly.
• Early intervention for cognitive impairment reduces disease burden and costs.
• Accurate risk assessment tools are needed for timely intervention.

Purpose of the Study:

• To develop a machine learning (ML)-based risk prediction model for cognitive dysfunction in the elderly.
• To identify key predictors of cognitive impairment using ML algorithms.
• To provide a tool for healthcare professionals and patients for effective risk assessment.

Main Methods:

• 1,325 elderly participants underwent cognitive assessments and blood tests.
• Risk factors were identified using univariate analysis, logistic regression, LASSO, and Boruta algorithms.
• Nine ML models were built and evaluated, with SHAP used for interpretation.

Main Results:

• The Random Forest (RF) model achieved the highest predictive performance (AUC).
• Key predictors identified by SHAP analysis include age, race, education, diabetes, and depression.
• Model calibration and decision curves confirmed strong predictive accuracy and clinical utility.

Conclusions:

• Age, race, education, diabetes, and depression are significant risk factors for cognitive dysfunction.
• The Random Forest model demonstrated superior predictive capability among the evaluated ML algorithms.
• The developed model offers a promising tool for early identification and management of cognitive dysfunction.