PheMap: a multi-resource knowledge base for high-throughput phenotyping within electronic health records

Abstract

Objective Developing algorithms to extract phenotypes from electronic health records (EHRs) can be challenging and time-consuming. We developed PheMap, a high-throughput phenotyping approach that leverages multiple independent, online resources to streamline the phenotyping process within EHRs. Materials and Methods PheMap is a knowledge base of medical concepts with quantified relationships to phenotypes that have been extracted by natural language processing from publicly available resources. PheMap searches EHRs for each phenotype’s quantified concepts and uses them to calculate an individual’s probability of having this phenotype. We compared PheMap to clinician-validated phenotyping algorithms from the Electronic Medical Records and Genomics (eMERGE) network for type 2 diabetes mellitus (T2DM), dementia, and hypothyroidism using 84 821 individuals from Vanderbilt Univeresity Medical Center's BioVU DNA Biobank. We implemented PheMap-based phenotypes for genome-wide association studies (GWAS) for T2DM, dementia, and hypothyroidism, and phenome-wide association studies (PheWAS) for variants in FTO, HLA-DRB1, and TCF7L2. Results In this initial iteration, the PheMap knowledge base contains quantified concepts for 841 disease phenotypes. For T2DM, dementia, and hypothyroidism, the accuracy of the PheMap phenotypes were >97% using a 50% threshold and eMERGE case-control status as a reference standard. In the GWAS analyses, PheMap-derived phenotype probabilities replicated 43 of 51 previously reported disease-associated variants for the 3 phenotypes. For 9 of the 11 top associations, PheMap provided an equivalent or more significant P value than eMERGE-based phenotypes. The PheMap-based PheWAS showed comparable or better performance to a traditional phecode-based PheWAS. PheMap is publicly available online. Conclusions PheMap significantly streamlines the process of extracting research-quality phenotype information from EHRs, with comparable or better performance to current phenotyping approaches.

Phe2vec: Automated disease phenotyping based on unsupervised embeddings from electronic health records

“…We also assess the performance of two commonly used phenotyping methods, PheCode 22 and PheMap. 21 PheCode groups ICD-9/10 codes into clinically meaningful phenotypes, thereby collapsing the diagnosis code space. PheMap is a high-throughput phenotyping approach that identified concepts important to phenotypes from publicly available sources, such as MedlinePlus, MedicineNet, and Wikipedia.…”

“… 8 Previous work in this domain used supervised and unsupervised machine learning to derive phenotypes for several diseases, with different strengths and limitations (see literature review in Note S1). 9 , 10 , 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 , 21 , 22 , 23 Supervised models rely on classifiers based on manually labeled gold standards for each specific disease, which is time-consuming and not scalable. Unsupervised approaches discover phenotypes purely from the data, trying to aggregate medical concepts commonly appearing together in the patient records.…”

Automated Coding of Under-Studied Medical Concept Domains: Linking Physical Activity Reports to the International Classification of Functioning, Disability, and Health

“…Automatically coding information in narrative text according to standardized terminologies is a key step in unlocking Electronic Health Record (EHR) documentation for use in health care. Mapping variable descriptions of clinical concepts to welldefined codes-for example, mapping "chronic heart failure" and "chron CHF" to the same ICD-10 code of I50.22-not only improves search and retrieval of medical information from EHRs or published literature (1), but also enables adding evidence from narrative documentation into artificial intelligence-driven predictive analytics and phenotyping (2). Free text is an especially valuable source for information that is not systematically recorded, or difficult to capture in standardized EHR fields, such as social determinants of health (3,4).…”

“…This article presents a general-purpose approach to expanding NLP technologies to assign standardized codes to new types of information in the EHR, and applies this approach to produce new technologies for linking EHR text to the ICF. Existing NLP technologies for coding medical information, as well as for linking text to other kinds of controlled inventories such as realworld named entities, largely rely on curated resources such as standardized vocabularies (32,33), expert knowledge graphs (2,34), and/or large-scale data sets with many thousands of samples (35,36). However, such resources have not yet been developed for the functional status domain (7), and are in fact difficult to procure for most under-studied medical concept domains, necessitating the development of new approaches.…”

Desiderata for the development of next-generation electronic health record phenotype libraries

“…However, like ML, NLP-based phenotype definitions are also often associated with complex processes, especially when used to conduct high-throughput phenotyping. For example, a PheMap phenotype definition consists of a set of linked concepts, the presence of which in a patient’s EHR is used to determine the probability of the patient having the condition represented [ 25 ]. The association of a phenotype with different concepts is defined within the PheMap knowledge base, which is constructed on the basis of a process that uses a specific set of NLP tools to derive these associations on the basis of the content of various text-based resources.…”