Developing a universal phenotyping algorithm to identify patients with clinically diagnosed and probable Alzheimer's disease using electronic health record data. (20th December 2022)

• Record Type: Journal Article
• Authors: Thakkar, Rut; Liu, Ge; Zhao, Juan; Ong, Henry H.; Li, Bingshan; Wen, Zhexing; Wei, Wei‐Qi
• Is Part Of: Alzheimer's & dementia. Volume 18(2022)Supplement 10
• Page Start: n/a
• Page End: n/a
• Publication Date: 2022-12-20
• Subjects: Alzheimer's disease -- Periodicals; Alzheimer Disease -- Periodicals; Dementia -- Periodicals; Démence; Maladie d'Alzheimer; Périodique électronique (Descripteur de forme); Ressource Internet (Descripteur de forme); 616.83
• DOI: 10.1002/alz.065899
• Languages: English
• ISSNs: 1552-5260

Abstract

Abstract: Background: Despite significant advancements in our understanding of Alzheimer's disease (AD) over the last decade, no disease‐modifying treatments exist for it. Rich clinical information on patients with AD is currently available in electronic health records (EHR) worldwide. Our aim is to develop a portable phenotyping algorithm to mine EHRs and identify patients with clinical and probable preclinical AD. Identified patients can subsequently be used as cohorts in future biomedical research studies. Method: We developed an algorithm using the Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM). This enables other institutes using this standardized EHR data representation to avail the algorithm with minimum effort for local implementation. The algorithm's backbone is based on ubiquitous information readily derived from EHRs such as ICD‐10 billing codes, medication history, cognitive test scores (Mini‐Mental State Exam), imaging reports and lab results. The model classifies identified patients into three categories based on high (category 1), intermediate (category 2) and low (category 3) probability with AD. The algorithm was applied to a deidentified EHR database at the Vanderbilt University Medical Center (VUMC) with clinical information from over 3 million individuals. We performed an manual chart review on 100 randomly selected samples of each category's AD cases to assess its predictive value. Result: Implementation of the phenotyping algorithm yielded 274, 000 patients with probable Alzheimer's disease. 4169 patients were identified based on category 1 criteria, 13270 patients were identified using category 2 and 256, 302 patients were identified in category 3. Independent chart review resulted in a PPV of 94% for category 1, 86% for category 2 and 8% for category 3 patients. Majority of false positive cases occurred because of mild cognitive impairment (MCI) patients in category 3 who were selected through broad billing codes and medication history. Conclusion: A portable phenotyping algorithm can help identify large cohorts of patients with diagnosed and probable Alzheimer's disease for research use by mining EHRs. Based on preliminary results, our algorithm is a promising attempt to achieve this goal. We aim to improve on it till it selects AD patients with an even higher predictive value (>95%).