Personalized screening and risk profiles for Mild Cognitive Impairment via a Machine Learning Framework: Implications...

Area of Science:

• Gerontology
• Computational Neuroscience
• Public Health

Background:

• Diagnosing Mild Cognitive Impairment (MCI) often involves lengthy procedures at specialized centers.
• There is a need for accessible diagnostic tools in primary healthcare settings.

Purpose of the Study:

• To evaluate a Machine Learning (ML) framework for identifying MCI and dementia using data available in primary care.
• To assess the performance of ML models compared to existing diagnostic methods.

Main Methods:

• A prospective study utilized demographic, clinical, and cognitive data from 763 individuals (aged 60-93) in Crete, Greece.
• A Balanced Random Forest Classifier was employed for classification tasks (CNI vs MCI, CNI vs Dementia, MCI vs Dementia).
• Model-agnostic analyses were used for feature selection and understanding variable importance.

Main Results:

• ML models achieved improved sensitivity (74%) and comparable specificity (73%) for MCI vs Cognitively Non-Impaired (CNI) classification compared to Mini Mental State Examination (MMSE) alone.
• Higher accuracies were obtained for MCI vs Dementia (87%) and CNI vs Dementia (94%).
• Key predictors included age, education, behavioral changes, multicomorbidity, and polypharmacy, with notable individual variability.

Conclusions:

• Combining readily available demographic and medical information with MMSE scores and behavioral ratings enhances MCI identification in primary care.
• Patient-level explainability of ML models can support personalized risk assessment for clinicians.