Characterizing Performance Gaps of a Code-Based Dementia Algorithm in a Population-Based Cohort of Cognitive Aging

Area of Science:

• Gerontology
• Medical Informatics
• Neurology

Background:

• Electronic health records (EHR) offer valuable data for dementia identification algorithms.
• Existing algorithms using billing codes and medication data show variable performance.
• Understanding misclassification patterns can guide algorithm refinement.

Purpose of the Study:

• To assess a code-based algorithm's accuracy for dementia diagnosis in the Mayo Clinic Study of Aging (MCSA).
• To characterize individuals misclassified by the dementia detection algorithm.
• To compare algorithm performance against a population-based reference standard.

Main Methods:

• Utilized data from 5,316 MCSA participants without baseline dementia.
• Extracted dementia-related ICD-9/10 codes and medications from EHR.
• Employed statistical tests to compare characteristics of correctly and incorrectly classified individuals.

Main Results:

• The algorithm demonstrated moderate-high performance with 0.70 sensitivity and 0.95 specificity.
• False positives were older, had higher comorbidity, and were more likely to have mild cognitive impairment (MCI).
• False negatives were older, more likely to have MCI, or exhibit functional limitations.

Conclusions:

• Code-based dementia diagnosis methods show moderate-high accuracy against a robust reference standard.
• Older age and presence of MCI are key factors associated with misclassification.
• Identifying characteristics of misclassified patients aids in improving dementia diagnostic algorithms.