Background: There are challenges in studying and monitoring Alzheimer’s disease (AD) in large patient populations. Clinicians often have insufficient availability of resources to make the diagnosis (e.g. brain scanning, referral to specialists) and clinical inertia may be tied to the perception that there are no clinical benefits in making the differential diagnosis of this stigmatized disease. Consequently, diagnosis codes specifically for AD are underutilized and prevalence is generally underestimated. This has limited epidemiologic studies of AD using electronic health record data from integrated healthcare systems.
Objectives: Our goal was to study probable AD and estimate its prevalence in a large patient population using a refined search algorithm of computerized clinical notes contained in electronic health records.
Methods: Our methods were developed using records for all Veteran patients in the national Department of Veterans Affairs Healthcare System (VA) in fiscal years 2010-2019. Starting with initial searches for “Alzheimer” and related terms in all clinical notes, the algorithm was optimized through an iterative process. The final algorithm was validated through manual reviews of over 2,400 randomly selected patient charts (predictive value positive = 86.3%; kappa=0.76 among up to 4 reviewers).
Results: When the algorithm was applied to records for the nearly 5 million VA patients over 50 years of age in fiscal year (FY) 2019, we identified 141,816 with probable AD, nearly five times the count based on ICD-10 codes (30,090). Median age of probable AD patients was 75 years. Prevalence, standardized to the 2010 census for age and sex, was 2.70%, with higher prevalence in women (3.26%) than in men (2.06%). AD prevalence also was higher in Black (3.47%) and Hispanic (3.46%) patients, but lower in Asian patients (2.13%), as compared to white patients (2.69%). This method yielded similar prevalence in preceding years (e.g. 2.65% in FY18, 2.71% in FY17).
Conclusions: As disease modifying treatments for AD enter the market, there will be more focus on proper diagnosis of AD, particularly early in the disease process, emphasizing the importance of better identification of the disease in patient populations. This method, based on searches of clinical notes, appears to be promising for the study of AD and its progression in large patient populations. Its application to national VA data resulted in prevalence estimates more consistent with estimates from other sources, such as on-going longitudinal surveys.
