Translating the Clinical Knowledge of Mendelian Diseases to Real-world EHR Data to Improve Identification of Undiagnosed Patients

PROJECT SUMMARY The last two decades have seen extraordinary advances in the cost, accessibility, and interpretability of genetic testing. In the context of this astonishing progress, it is striking that for many rare genetic diseases, diagnostic delay – the time between onset of symptoms and a diagnosis – has not improved. Current health care services are unable to effectively identify patients that would benefit most from genetic testing. As a result, many patients affected by genetic disease are not diagnosed for years after symptoms develop, or are never diagnosed at all, leading to costly diagnostic odysseys, health care disparities in genetic services, and preventable morbidity and mortality for those with conditions that have an effective, targeted treatment. Much of what we know about genetic disease is based on studies of individuals and their families. This has proven to be a powerful method for discerning the clinical characteristics of genetic disease, generating one of the most enduring and useful resources in medical genetics: the online Mendelian inheritance in man (OMIM). However, clinical descriptions in OMIM do not always match the way diseases are described in real- world EHR data. To improve our ability to use genetic testing effectively, we can learn, at scale, from the data clinically captured while testing and diagnosing patients. EHRs provides an opportunity to study genetic disease from a new perspective, enabling scalable methods that augment existing the knowledge base to include phenotypes observed in real-world health care data. This proposal builds on our prior work curating genetic testing data from the EHR and developing tools to identify undiagnosed patients from characteristic genetic disease profiles. Specifically, we have built a database of clinical genetic testing information extracted from the EHR for over 20,000 individuals, with detailed information regarding test results, variant interpretation, and diagnosis. From this resource, we can define EHR-based cases series of individuals with genetically-confirmed clinical diagnoses of genetic disease. We will use a data-driven approach to discern characteristic phenotypes from the EHR-based case series, and merge these results with clinical descriptions from OMIM. This approach seeks to translate the curated, durable knowledge cataloged in OMIM to a portable and scalable product that can layered on any set of EHRs to identify undiagnosed patients with genetic disease. The ultimate goals of this proposal are leverage these data and tools to 1) translate and add to clinical curations of genetic diseases using real world EHR data, 2) assess diagnostic yield of EHR-based tools that identify undiagnosed patients and 3) characterize the contribution of demographic and phenotypic features that lead to earlier or later diagnosis.