Mild Cognitive Impairment: Data-Driven Prediction, Risk Factors, and Workup

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Neurology

Background:

• Dementia is projected to affect over 78 million people by 2030.
• Early identification of mild cognitive impairment (MCI) is crucial for timely intervention.
• Personalized clinical evaluations are needed to diagnose potentially reversible causes of cognitive decline.

Purpose of the Study:

• To develop machine learning models for predicting MCI onset up to one year in advance.
• To identify key clinical factors from electronic health records predictive of MCI.
• To create a data-driven list of clinical procedures for MCI workup.

Main Methods:

• Utilized real-world electronic health records from the OMOP data model.
• Developed and evaluated logistic regression, random forest, and xgboost models.
• Trained and tested models on over 531,000 patient visits.
• Employed association mining techniques to identify predictive clinical factors and procedures.

Main Results:

• The random forest model achieved an ROC-AUC of 68.2±0.7 in predicting MCI onset.
• Identified significant clinical factors from clinician notes that predict MCI.
• Generated a data-driven list of common clinical procedures for MCI evaluation.

Conclusions:

• Machine learning models, particularly random forest, show promise in predicting MCI.
• Identifying predictive clinical factors can enhance early MCI detection.
• A data-driven approach can inform clinical guidelines and order sets for MCI workup.