Abstracts

Rationale: Childhood epilepsies have a strong genetic contribution, but the longitudinal disease trajectory for many genetic etiologies remains unknown. The adoption of electronic Medical Records (EMR) provides an unprecedented opportunity to leverage clinical data for genomic research. EMR data also contains an inherent temporal dimension that phenotyping algorithms have largely not explored thus far. Maintaining the temporal relationship between clinical features is critical in disorders that follow prominent age-related patterns, such as the childhood epilepsies. Methods: We analyzed 3,251 patient years of EMR data from 658 individuals with known or presumed genetic epilepsies and mapped 62,104 neurology-related Human Phenotype Ontology (HPO) terms to 100 three-month time intervals from birth to 25 years and assessed time-dependent gene-phenotype associations. We then assessed whether pair-wise phenotypic similarities between individuals with a specific gene were higher than pair-wise phenotypic similarities in the remainder of the cohort using the aggregated information content of all shared HPO terms between individuals. Results: We identified 890 nominally significant (p≤0.05) associations between genetic etiologies and phenotypic features. The most significant findings included the association of “Infantile spasms” (HP:0012469; p=2.85e-5) and “Epileptic spasms” (HP:0011097; p=2.85e-5) with STXBP1 at 0.5 years, “Status epilepticus” (HP:0002133; p=1.84e-7) with SCN1A at 1.0 years, and “Severe intellectual disability” (HP:0010864; p=2.96e-6) with PURA at 9.75 years. We also found 19 of the 36 genes present in two or more individuals had a significant phenotypic similarity (p≤0.05) during at least two consecutive time intervals (6 months). Conclusions: Our study demonstrates that EMR data can be used to reconstruct longitudinal disease histories in genetic epilepsies at a scale far beyond the ability of manual phenotyping. We determined age-dependent associations between genetic etiologies and phenotypic terms and significant similarity within groups of individuals with shared genetic etiologies. Many genetic etiologies have unique longitudinal EMR footprints, indicative of previously unanticipated gene-specific disease trajectories. Identifying such trajectories using large-scale phenotypic data will be critically important for decision-support and learning healthcare systems, particularly in rare genetic epilepsies where available clinical information is limited. Funding: The Children's Hospital of Philadelphia