Computation of longitudinal phenotypes in 466 individuals with a developmental and epileptic encephalopathy enables clinical trial readiness

Abstract

Electronic health records (EHRs) represent a rich data source to support precision medicine, particularly in disorders with small and heterogeneous populations where longitudinal phenotypes are poorly characterized. However, the impact of EHR data is often limited by incomplete or imperfect source documentation and the inability to leverage unstructured data. Here, we address these shortcomings through a computational analysis of one of the largest cohorts of developmental and epileptic encephalopathies (DEEs), representing 466 individuals across six genetically defined conditions. The DEEs encompass debilitating pediatric-onset disorders with high unmet needs for which treatment development is ongoing. By applying a platform approach to data curation and annotation of 18 clinical data entities from comprehensive medical records, we characterize variation in longitudinal clinical journeys. Assessments of the relative enrichment of phenotypes and semantic similarity analysis highlight commonalities and differences between the six cohorts. Evaluation of medication use reflects unmet needs, particularly in the management of movement disorders. We also present a novel composite measure of seizure severity that is more robust than existing measures of seizure frequency alone. Finally, we show that the attainment of developmental outcomes, including the ability to sit independently and the ability to walk, is correlated with seizure severity scores. Overall, the combined analyses demonstrate that patient-centric real world data generation, including structuring of medical records, holds promise to improve clinical trial success in rare disorders. Applications of this approach support improved understanding of baseline disease progression, selection of relevant endpoints, and definition of inclusion and exclusion criteria.