RESCUE: Rare Disease Detection and Escalation Support via a Learning Health System

PROJECT SUMMARY/ABSTRACT Rare diseases are individually rare yet are collectively common and affect millions of patients and their families in the United States. To date, most efforts for clinical decision support systems (CDSS) for rare diseases are aimed at deriving diagnoses from genomic data. Although those tools can be critical for genetic specialists, it is essential for primary care providers to identify suspected patients and make appropriate referrals at an early stage to get their genomic testing. In this study, we propose a SMART-on-FHIR based Rare Disease Detection and Escalation Support (RESCUE) CDSS. It will use a centralized informatics approach to identify suspected rare disease patients from clinical data warehouse (CDW) and send alerts to physicians with escalation support including phenotype summarization, genetic/genomic test requisition and research opportunity discovery. We will leverage the cutting-edge natural language processing technics to unlock the clinical features in the clinical narratives and convert them to the GA4GH-based standard for optimized genetic disease information exchange. To overcome the challenge of insufficient knowledge and limited sample size for a single rare disease, we propose a hybrid approach to address this challenge. Expert-curated and knowledgebase-derived phenotype- based queries will be issued to identify “silver-standard” cases first, and then a neural network-based analytical model will be trained to further identify potential undiagnosed rare disease patients. We will investigate the efficiency and accuracy of this analytical model by conducting chart-reviews using the clinical data warehouse at the Columbia University Irving Medical Center (CUIMC). To ensure its generalizability, we will further validate the model using Children’s Hospital of Philadelphia as the secondary site. The proposed RESCUE CDSS will apply this analytical model to identify suspected rare disease patients and notify physician by triggering a provider-facing SMARTapp during the patient encounter. By engaging patients (and their families), primary care physicians, genetic specialists, researchers, key policy makers and ELSI experts, our stakeholder-centered participatory approach will ensure that the design of this SMARTapp is interoperable, comprehensible, and actionable. We will collaborate with Epic IT teams to deploy RESCUE and evaluate its usability in ambulatory pediatrics clinics at CUIMC. To optimize the performance and to increase interoperability with various EHR systems, the CUIMC validated software will be further deployed and validated at CHOP. We will further extend RESCUE to enable an enhanced escalation support including genetic/genomic testing support and patient- research matching, and retrospectively validate the extended module. Our transferable software and tools developed in this study will be shared through multiple outlets including CTSAs, the eMERGE network, and the Observational Health Data Sciences and Informatics (OHDSI) community.