Risk prediction of mild cognitive impairment using electronic health record data.

Area of Science:

• Gerontology
• Medical Informatics
• Neurology

Background:

• Mild cognitive impairment (MCI) is frequently underdiagnosed by primary care providers, with detection rates between 6% and 15%.
• Predictive models are crucial for identifying at-risk individuals, enabling timely interventions for cognitive health.
• Early detection of MCI is vital for managing cognitive decline and improving patient outcomes.

Purpose of the Study:

• To develop and validate a machine learning model for predicting mild cognitive impairment (MCI) risk using electronic health records (EHRs).
• To identify key demographic factors and comorbidities that serve as predictors for MCI.
• To create a practical triage tool for primary care providers to aid in MCI detection.

Main Methods:

• A retrospective study analyzed 271,054 cognitively unimpaired individuals and 14,501 with confirmed MCI from EHR data.
• A machine learning model was developed using a data-driven variable selection approach, focusing on demographics and comorbidities.
• The model was evaluated for predictive performance using the area under the curve (AUC) metric across different age groups.

Main Results:

• The final model utilized 26 selected variables from an initial 101, achieving an overall AUC of 86%.
• Predictive performance varied by age group, with AUCs of 79.1% (40-49), 77.0% (50-64), 76.9% (65-79), and 74.4% (≥80).
• The model demonstrated higher predictive accuracy in younger age cohorts.

Conclusions:

• Demographic factors and comorbidities are significant predictors of MCI risk.
• The developed machine learning model shows strong predictive capabilities and can function as an effective triage tool for PCPs.
• Implementing this tool can facilitate earlier identification and treatment of individuals with MCI, improving management across aging populations.