Public Health.

Area of Science:

• Neuroscience
• Computational Biology
• Public Health

Background:

• Accurate classification of dementia, Alzheimer's disease (AD), and mild cognitive impairment (MCI) is critical for timely diagnosis and intervention.
• Machine learning (ML) offers potential for improved participant stratification in dementia research.
• Large datasets like the UK Biobank are valuable for identifying predictive factors.

Purpose of the Study:

• To investigate the efficacy of ML methods, particularly XGBoost, in classifying participants with dementia, AD, or MCI against a healthy cohort.
• To identify key comorbidities and lifestyle factors predictive of cognitive impairment.
• To explore the utility of ML for risk prediction in undiagnosed populations.

Main Methods:

• Analysis of over 500,000 UK Biobank participants categorized by cognitive status (healthy, dementia, AD, MCI).
• Classification using XGBoost with features including comorbidities (hypertension, diabetes) and lifestyle factors (smoking, physical activity).
• 5-fold cross-validation with sub-sampling to address class imbalance, with performance measured by mean AUROC.

Main Results:

• XGBoost achieved a mean AUROC of approximately 0.81 in classifying participants.
• Key predictors included hypertension, stroke, and heart disease for comorbidities.
• Vigorous physical activity, alcohol intake, and smoking were significant lifestyle predictors.

Conclusions:

• ML methods, exemplified by XGBoost, demonstrate high accuracy in classifying individuals with diagnosed cognitive impairments.
• Feature importance analysis reveals critical health and lifestyle factors influencing cognitive status.
• This research supports the development of ML-based risk prediction models for early dementia detection and intervention.