Scalable phenotyping to assay risk in childhood after exposures in pregnancy: sparc-XP

The National Institute of Mental Health awarded a $7.01 million Cooperative Agreement under the Mental Health Research Grants program (CFDA 93.242) to Massachusetts General Hospital for the SPARC-XP (Scalable Phenotyping to Assay Risk in Childhood After Exposures in Pregnancy) project. This four-year initiative (September 1, 2025 – August 31, 2029) will develop and validate predictive risk models to identify which offspring of mothers with infections during pregnancy are at highest risk for developing autism spectrum disorder, attention-deficit/hyperactivity disorder, and other neurodevelopmental disorders by age 8. The primary deliverable is a suite of risk prediction models generated through machine learning analysis of electronic health records data from at least 150,000 maternal-offspring dyads across two health systems, with predictions targeted at two critical intervention windows: at birth and between ages 3-5.

To augment electronic health records analysis, the project will conduct prospective phenotyping of 750 infection-exposed children across two clinical sites: Massachusetts General Hospital/Brigham Health System in Massachusetts and Vanderbilt University Medical Center (VUMC) in Tennessee. MGH will lead model development and coordinate overall study execution, while VUMC will establish a matched maternal-offspring cohort with longitudinal outcomes, validate prediction models on their patient population, facilitate recruitment of mothers whose children fall in highest and lowest risk quartiles for detailed neurocognitive assessment, and contribute to collaborative data analysis and manuscript preparation. Deliverables include validated predictive algorithms, phenotypic data from prospective assessments including parental developmental reports, accelerometry-based motor function and sleep metrics, and virtual neurocognitive evaluations, with the ultimate goal of enabling early intervention for at-risk children.

Abstract

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.