Predicting 5-year dementia conversion in veterans with mild cognitive impairment

Area of Science:

• Gerontology
• Neurology
• Medical Informatics

Background:

• Identifying patients with mild cognitive impairment (MCI) at high risk for dementia is crucial for timely interventions.
• Electronic health records (EHRs) offer a valuable resource for developing predictive models of disease progression.

Purpose of the Study:

• To develop and validate an EHR-based model for predicting the conversion of MCI to all-cause dementia (ACD) within a 5-year timeframe.
• To identify key demographic and clinical factors associated with MCI to ACD conversion in a veteran population.

Main Methods:

• A Cox proportional hazards model was employed to analyze EHR data from a large cohort of veterans diagnosed with MCI.
• Model performance was rigorously assessed using metrics such as the area under the receiver operating characteristic curve (AUC) and Brier score on a separate validation dataset.
• The study also explored the utility of synthetic data for model development, comparing its performance against models built with real patient data.

Main Results:

• Out of 59,782 individuals with MCI, 15,420 (25.8%) progressed to ACD within 5 years.
• The predictive model demonstrated strong discriminative ability (AUC = 0.73) and good calibration (Brier score = 0.18).
• Significant risk factors for conversion included advanced age, history of stroke, cerebrovascular disease, myocardial infarction, hypertension, and diabetes. Conversely, higher body mass index, alcohol abuse, and sleep apnea were associated with a reduced risk of conversion.

Conclusions:

• The developed EHR-based prediction model exhibits robust performance in identifying individuals with MCI who are likely to convert to ACD within 5 years.
• This model holds significant potential for clinical application in triaging patients and facilitating early interventions for those at higher risk of dementia.
• The comparable performance of models using synthetic data highlights its potential for accelerating research while preserving patient privacy.