Big Data/Machine Learning/AI
Utilizing CIPHER to identify and compare Alzheimer’s Disease Phenotypes: Streamlining EHR-Based Phenotype Discovery and Application Ashley Galloway* Ashley Galloway Vidisha Tanukonda Connor Melley Monika Maripuri Yuk-Lam Ho Jacqueline P. Honerlaw Kelly Cho
Alzheimer’s Disease (AD) is one of the most prevalent neurodegenerative disorders, and leading cause of dementia in the United States (US). Due to its complex pathophysiology and symptomatology, significant challenges exist in accurately identifying individuals with AD. This is further complicated by variations in electronic health record (EHR) systems and documentation. These challenges often lead to inconsistency in definitions and barriers to scientific progress. The Centralized Interactive Phenomics Resource (CIPHER) phenotype library provides solutions to these challenges by allowing for comparison of up to seven different phenotypes, while ensuring each definition contains enough information for replication.
We utilized CIPHER’s phenotype comparison tool to evaluate and select four computable AD definitions: three rules-based definitions using ICD codes and one algorithm utilizing ICD codes and text strings. Details regarding these phenotypes can be found on CIPHER by navigating to phenomics.va.ornl.gov and searching for the full phenotype names shown in Figure1A. All four algorithms were applied to data from the US Department of Veterans Affairs and validated against 100 gold-standard labels derived from clinician chart review. Performance metrics were calculated for each definition and deposited into CIPHER to promote interoperability and reuse (Figure1B).
Due to variation in project goals and resources, some users may prefer a definition that maximizes sensitivity, while a probabilistic definition may be more suitable for others. CIPHER provides an opportunity to enhance scientific knowledge by streamlining algorithm evaluation. By enabling users to report validation in multiple datasets, CIPHER also facilitates interoperability of definitions. This collective approach can also expedite understanding of multifactorial diseases such as AD for which limitations in diagnosis, detection, and treatment exist.
