A Systematic Review to Summarize and Critically Appraise Existing Phenotype Libraries Using Electronic Health Records.

内容摘要

Pharmacoepidemiology and population health studies using electronic health care records (EHRs) must define study variables through available electronic data. These variables are operationalized through phenotypes, which are a defined set of criteria used to identify specific traits or medical conditions. There is diversity across phenotype libraries (collections of code lists or algorithms) which intend to standardize these sets of criteria. This review aimed to characterize the landscape of phenotype libraries and how phenotypes are constructed, validated, managed, and reused across research settings. We conducted a systematic review of existing phenotype libraries to appraise their attributes. We systematically searched three databases (Scopus, PubMed, and Web of Science) up to November 2025 to identify studies on key characteristics of phenotype libraries. The search combined Medical Subject Headings (MeSH) terms related to "electronic health record," "phenotype algorithm," and "phenotype library". A structured hand search was performed to identify relevant web-based resources without accompanying publications to ensure comprehensive inclusion of libraries available to date. We extracted information on library size, vocabularies, phenotype construction methods, validation practices, management, and portability. Of 336 articles, 37 met eligibility criteria for full-text review, of which 25 were excluded because they were not EHR-based phenotype libraries (representing single algorithms, genomic resources, or study-specific phenotypes rather than reusable libraries), leaving 10 unique libraries described across 12 articles. A structured hand search identified seven more libraries. In total, 17 phenotype libraries met the inclusion criteria, including Education and Child Health Insights from Linked Data (ECHILD) Phenotype Code List Repository, Centralized Interactive Phenomics Resource (CIPHER), Chronic Condition Data Warehouse (CCW), ClinicalCodes Library, Clinical Classifications Software Refined (CCSR), ComPLy, CALIBER (Health Data Research UK (HDR UK) Phenotype Library or CALIBER), Jigsaw Algorithm Repository (JAR), Manitoba Centre for Health Policy (MCHP) Concept Dictionary, Open CodeLists, Observational Health Data Sciences and Informatics (OHDSI) ATLAS, PheCode, Phenotype KnowledgeBase (PheKB), Phenotype Execution and Modeling Architecture (PhEMA) Workbench, PheMap, Sharing and Reusing Computable Phenotype Definitions (SharePhe), Value Set Authority Center (VSAC). Libraries varied substantially in scope, size, and phenotype representation, including rule-based algorithms, probabilistic phenotypes, and standardized code groupings. Validation practices were heterogeneous and reported only for a subset of libraries. All the libraries utilized a web-based platform and met at least the minimum requirements for library management, including phenotype definitions, metadata, and version control. We observed large variations in library construction and validation across diverse libraries built in varied EHR research settings. The transparency of phenotypes and creating computable phenotypes enhance portability and streamline the effective reuse of phenotypes for different systems. Electronic health records (EHRs) contain real‐world information about patients' medical conditions, treatments, and test results. Researchers use this data to study diseases and improve patient outcomes. To this end, researchers must specify how to define specific conditions in EHR data. These definitions are called phenotypes. Phenotype libraries are platforms where such definitions are collected, documented, and shared, allowing researchers to reuse them and ensure consistency across studies. In this study, we reviewed existing phenotype libraries to understand how they are built and how they support health research. We found 17 libraries, each with unique features. Most use rule‐based methods to define conditions, and some use machine learning and natural language processing to construct phenotypes. All are accessible through web platforms and readable by both humans and computers, but not all include validation of their definitions. User interfaces vary across libraries. Our findings show that phenotype libraries play a key role in improving the reliability and reproducibility of research leveraging EHR data. We also suggest improvements to increase their accessibility, quality, and ability to work across different systems.