Diverse Unity: Exploring Shared Phenotypic Traits Across a Heterogeneous Patient Cohort via Visual Analytics

Presentation Time: 03:30 PM - 03:45 PM

Abstract Keywords: Information Extraction, Population Health, Data Transformation/ETL
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

Clinical data, such as electronic health records (EHRs), hold valuable insights for personalized patient care. However, the complexity of these data often hinders the extraction of actionable clinical evidence. We present an interactive system that leverages patient similarity-based visual analytics to explore clinical evidence embedded in EHRs to promote informed decision-making in clinical practice. Our system constructs patient vectors for EHR data structured in the OMOP Common Data Model (CDM) to represent demographics, conditions, drugs, and laboratory test measurements. We employ dimensionality reduction techniques to project these high-dimensional vectors onto a 2D plane, facilitating intuitive exploration through interactive visualizations. Clinicians can flexibly select and compare patient clusters, gaining insights into similar and dissimilar characteristics. A case study demonstrates the system's capability to identify distinct patient clusters with varying conditions, such as those related to post-acute sequelae of SARS-CoV-2 infection (PASC). The interactive visualizations link algorithmic results back to the original EHR data, enhancing the understanding of patient cohorts and extraction of actionable insights into clinical care.

Speaker(s):
Xingbo Wang, PhD
Weill Cornell Medicine

Author(s):
Xingbo Wang, PhD - Weill Cornell Medicine; Mark Weiner, MD - Weill Cornell Medicine / Population Health Sciences; Qiannan Zhang, PhD - Weill Cornell Medicine; Jyotishman Pathak, PhD - Weill Cornell Medical College, Cornell University; Fei Wang, PhD - Weill Cornell Medicine;

Identifying Subphenotypes that are Predictive of Clinical Outcomes Using Electronic Health Records and Machine Learning

Presentation Time: 03:45 PM - 04:00 PM

Abstract Keywords: Precision Medicine, Machine Learning, Deep Learning
Primary Track: Foundations

Predicting the outcome of the patients receiving specific treatments is an important problem in real-world applications. In practice, there may be several subphenotypes with heterogeneous treatment response or survival distributions. We propose a novel graph neural network based framework to identify subphenotypes for treatment response. Experiments on an EHR-derived advanced non‐small cell lung cancer cohort show that our framework can achieve better prediction performance and subphenotype identification compared to baseline methods.

Speaker(s):
Weishen Pan, PhD
Weill Cornell Medicine

Author(s):
Weishen Pan, PhD - Weill Cornell Medicine; Deep Hathi, PhD - Regeneron Pharmaceuticals, Inc; Zhenxing Xu - Weill Cornell Medical College; Qiannan Zhang, PhD - Weill Cornell Medicine; Ying Li, Ph.D. - Regeneron Pharmaceuticals; Fei Wang, PhD - Weill Cornell Medicine;

1000 Phenotypes: Scalable Automated Phenotyping in i2b2

Presentation Time: 04:00 PM - 04:15 PM

Abstract Keywords: Information Retrieval, Machine Learning, Informatics Implementation, Data Mining
Primary Track: Foundations
Programmatic Theme: Clinical Research Informatics

Diagnosis codes in EHRs often have low precision for predicting the patient’s true condition, or “phenotype”. For example, an ICD-10 code for diabetes could be assigned just for billing purposes, not because the patient has diabetes. Computational phenotypes use machine learning models to predict which patients have the phenotype. In this presentation we describe a new automated highly scalable pipeline that can quickly generate 1000+ phenotypes using EHR data in the i2b2 common data model.

Speaker(s):
Griffin Weber, MD, PhD
Harvard Medical School

Author(s):
Shawn Murphy, MD, Ph.D. - Massachusetts General Hospital;

Promoting the sharing, discovering, and reuse of phenotypes by developing an ontology-driven phenotype library

Presentation Time: 04:15 PM - 04:30 PM

Abstract Keywords: Reproducibility, Knowledge Representation and Information Modeling, Fairness and Elimination of Bias
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

Observational data from Electronic Health Records (EHRs) offer immense potential for clinical research, relying on the precise identification of patients with specific medical conditions, facilitated by phenotype definitions. our study designs an information model for standardized phenotype representation and implements it in a computable format. We establish a centralized repository for enhanced discoverability and develop a user-friendly web portal adhering to FAIR principles, enabling easy searching, downloading, and sharing of standardized phenotype definitions. we developed a unified ontology schema, Phenotype Definition Ontology (PDO) to consistently represent these diverse phenotypes information. The PDO serves as the cornerstone for representing essential phenotype information and supporting various phenotypes collection, curation, sharing, and reuse. We integrated 4 sources of publicly available phenotype definitions. In total, 3,542 phenotype definitions were successfully converted and represented using the unified PDO model and computable formats in our phenotype library. Guided by the PDO framework and adhering to FAIR principles, ComPLy serves as a centralized repository for phenotype sharing, offering the following key services, including Search, Browse, Download, Submit, and API services. Lastly, we self-assessed the FAIRness of our phenotype library resources.

Speaker(s):
Na Hong, PhD
Yale University

Author(s):
Na Hong, PhD - Yale University; Xubing Hao; Yujia Zhou, Ms - Yale University; Ryan Denlinger, PhD - Yale University; Yan Hu - UTHealth Science Center Houston; Xueqing Peng, PhD - Yale University; Fongci Lin, PhD - Yale University; Licong Cui, PhD - The University of Texas Health Science Center at Houston (UTHealth Houston) School of Biomedical Informatics; Yong Chen, PhD - University of Pennsylvania; Hua Xu, Ph.D - Yale University;

Time-Sensitive Modeling of ICU/Critical Care at the Visit-Level: a Study of 48-Hour Mortality and ARDS Phenotyping

Presentation Time: 04:30 PM - 04:45 PM

Abstract Keywords: Natural Language Processing, Machine Learning, Deep Learning, Knowledge Representation and Information Modeling, Reproducibility
Primary Track: Foundations
Programmatic Theme: Clinical Research Informatics

Longitudinal NLP studies often focus on evidence gathered from visit-to-visit without including the time course of events within a visit. We hypothesize that a framework shown to work at a between-visit temporal granularity can be adapted to a within-visit granularity. We have found Pytorch_EHR a reasonable means for enriching NLP models to include within-visit temporality based on two use cases as compared against performance published by other researchers: 48-hour ICU mortality and ARDS phenotyping.

Speaker(s):
Paul Heider, PhD
Medical University of South Carolina

Author(s):
Dmitry Scherbakov, PhD - Medical University of South Carolina; Jihad Obeid, MD - Medical University of South Carolina; Charles Terry, MD - Medical University of South Carolina;

Barriers to Obtaining and Using Interoperable Information among Non-federal Acute Care Hospitals

Presentation Time: 04:45 PM - 05:00 PM

Abstract Keywords: Interoperability and Health Information Exchange, Data Sharing, Data Standards
Primary Track: Policy
Programmatic Theme: Clinical Informatics

Interoperable exchange and use of patient health information is critical to enabling providers to deliver clinically appropriate care by facilitating timely and secure access to information needed to care for patients that have been treated in other hospitals or health systems. This study leverages nationally representative data on US hospitals to describe specific major and minor barriers to obtaining and using information from outside organizations. We also examine how these barriers vary by hospital characteristics and how they relate to use of information by healthcare providers at those hospitals.

Speaker(s):
Jordan Everson
Office of the National Coordinator for Health Information Technology

Author(s):
Chelsea Richwine - Office of the National Coordinator for Health Information Technology; Jordan Everson - Office of the National Coordinator for Health Information Technology;