Dravet Syndrome Decoded: Deep Learning on Longitudinal Patient-Level Data Predicts Diagnosis Within the First Year of Life
Abstract number :
3.192
Submission category :
4. Clinical Epilepsy / 4B. Clinical Diagnosis
Year :
2023
Submission ID :
1184
Source :
www.aesnet.org
Presentation date :
12/4/2023 12:00:00 AM
Published date :
Authors :
Presenting Author: Shridhar Parthasarathy, BA – Children's Hospital of Philadelphia
Jillian McKee, MD, PhD – Children's Hospital of Philadelphia; Peter Galer, MS – Children's Hospital of Philadelphia; Julie Xian, BS – Children's Hospital of Philadelphia; Casey Sederman, BS – Ambit, Inc.; Chen Chen, MA – Ambit, Inc.; Dan Kim, PhD – Ambit, Inc.; Robert Sederman, MBA – Ambit, Inc.; Ingo Helbig, MD – Children's Hospital of Philadelphia
Rationale:
Dravet syndrome is one of the most common genetic epilepsies with a clinical presentation that is recognizable by epileptologists. Although disease-specific features, including medication-resistant seizures, typically begin within the first year of life, the median age of diagnosis is 4.2 years, representing significant diagnostic lag. This study aims to delineate the disease course in more than 2,000 individuals with Dravet syndrome to facilitate earlier diagnosis of this disorder.
Methods:
Data from 47,419,168 healthcare claims of individuals with epilepsy or neurodevelopmental disorders were analyzed across 874,581 individual-years, including diagnosis, procedure, and medication data. Using the ICD10 code for Dravet syndrome, G40.83, a subcohort of 2,068 individuals were identified. Longitudinal disease trajectories were assessed using clinical features mapped from ICD10 codes to the Human Phenotype Ontology, as well as data on medication prescriptions. Machine learning approaches, namely random forest and neural network models, were trained to classify individuals as having Dravet syndrome or a different neurodevelopmental disorder, in order to assess how accurately and at what age Dravet syndrome is automatically identifiable from health records.
Results:
Administrative claims data recapitulate known clinical features in Dravet syndrome: individuals were more likely than those with other neurodevelopmental disorders to have febrile seizures at 3-6 months of age (p< .0001, OR=11.86, 95%CI=7.01-19.33) and generalized tonic-clonic seizures at 3-6 months (p< .0001, OR=4.68, 95%CI=2.90-7.31), and less likely to have neonatal seizures (p< .0001, OR=0.116, 95%CI=0.0417-0.261) or developmental delays before 3 months of age (p< .0001, OR=0.0856, 95%CI=0.0229-0.226). Medication prescription practices were influenced by Dravet syndrome diagnosis, with frequent initiation of clobazam and cessation of sodium channel blockers after diagnosis, as well as earlier user of valproic acid than in other neurodevelopmental disorders (p< .05, ORs=2.34-27.68 at 0-6.5 years).
Clinical Epilepsy