Development and validation of an interpretable machine learning model-Predicting mild cognitive impairment in a...

Area of Science:

• Neurology
• Gerontology
• Biostatistics

Background:

• Mild cognitive impairment (MCI) is a preclinical stage of Alzheimer's disease (AD), significantly increasing dementia risk.
• Stroke is a recognized risk factor for MCI, necessitating early identification and intervention in at-risk populations.
• Understanding MCI's risk factors in stroke survivors is crucial for effective prevention strategies.

Purpose of the Study:

• To identify key risk factors for developing MCI in individuals with a history of stroke.
• To develop and validate machine learning models for predicting MCI risk in this high-risk group.
• To create an accessible online tool for assessing MCI risk.

Main Methods:

• The Boruta algorithm was employed for robust feature selection from patient data.
• Eight distinct machine learning models were trained and evaluated for predictive accuracy.
• The best-performing models were utilized to determine variable importance and construct a risk calculator, with Shapley additive explanations for model interpretability.

Main Results:

• Logistic regression achieved the highest AUC (0.8595) in predicting MCI, followed by Elastic Network (0.8312).
• Key predictors identified included transient ischemic attack (TIA), diabetes, education level, and hypertension.
• Variable importance analysis highlighted TIA, diabetes, education, and hypertension as the most significant risk factors for MCI in stroke patients.

Conclusions:

• Transient ischemic attack (TIA), diabetes, education, and hypertension are critical risk factors for MCI in stroke survivors.
• Early and targeted interventions focusing on these factors are recommended to mitigate MCI incidence.
• The study provides a foundation for proactive MCI prevention in high-risk stroke populations.