Enhancing Patient Care in Rare Genetic Diseases: An HPO-based Phenotyping Pipeline

Identifying rare genetic diseases poses challenges, often resulting in delayed referrals to genetic specialty care. Consequently, a large number of rare diseases remain undiagnosed for years, with many of these patients often dying without an accurate diagnosis. To help address these diagnostic odysseys, we propose a two-step pipeline using natural language processing (NLP) and machine learning (ML) techniques leveraging raw clinical text narratives of patients from longitudinal clinical data in electronic health records (EHR). Our approach employs two state-of-the-art concept recognition algorithms, ClinPhen and PhenoTagger, to extract human phenotype ontology (HPO) terms from the clinical notes and ML algorithms to identify patients who may benefit from referral. By identifying key phrases indicative of rare diseases, our system can prompt healthcare providers to seek clinical genomics consultations, improving diagnosis and patient outcomes. We evaluated the predictive performance of extracted HPO terms using different ML models. Our strategy and methodological approach are intended to serve as complementary steps in aiding clinicians and researchers to improve the diagnosis efficiency of rare genetic diseases.