Scalable Phenotyping to Assay Risk in Childhood after eXposures in Pregnancy: SPARC-XP

About This Grant

PROJECT SUMMARY Infection in pregnancy is a significant risk for neurodevelopmental disorders in offspring, demonstrated in national registries, large epidemiologic studies, and electronic health records cohorts. Mechanistic studies in animals have similarly established adverse effects of maternal infection on offspring neurodevelopment and behavior. Despite this compelling evidence of risk, at present there are no validated methods to predict which exposed children will develop autism, ADHD, or other adverse outcomes. Better risk stratification methods are urgently needed to guide early intervention. Developing reliable, generalizable risk prediction requires large, representative populations and the ability to conduct low-cost phenotyping, valuable in extending the prediction afforded by electronic health records alone. This concept represents the crux of the IMPACT-MH initiative and the organizing principle of this proposal, which seeks to fill a key knowledge gap in predicting neurodevelopmental outcomes. This study will generate risk models to predict neurodevelopmental outcomes by age 8 among offspring of mothers with infections during pregnancy, focusing on two actionable time points: at birth, and between ages 3 and 5. It will first use standard and emerging machine learning methods to analyze electronic health records (EHR) data from at least 150,000 maternal-offspring dyads across 2 health systems, characterizing maternal infections, pregnancy complications, and offspring phenotypes. These EHR data will be augmented by prospective phenotyping of 750 infection- exposed children enrolled across 2 sites: MassGeneral Brigham health system in Massachusetts, and Vanderbilt University Medical Center in Tennessee. The investigators will collect parental report of developmental data, accelerometry data to assess motor function and sleep, and will perform virtual neurocognitive testing at age 3- 5 years. They will compare risk stratification models using electronic health records augmented with natural language processing, parental reports, neurocognitive testing, and accelerometry. Children will again be assessed at age 6-8 years to ascertain diagnoses. Optimal prediction models from the Massachusetts cohort will be externally validated in the Tennessee cohort. Together, these aims will yield tools for risk stratification in children exposed to maternal infection during pregnancy, applying innovative methods to electronic health records integrated with remote behavioral phenotyping. The project will be led by principal investigators with expertise in pregnancy exposures, large-scale electronic health records, machine learning, and neurodevelopmental phenotyping.