Predicting early school-age neurocognitive outcomes from maternal nutrition and sociodemographic factors in a pilot machine learning study

Maternal diet plays an important role in child brain development. However, existing studies mostly focus on dietary nutrients in isolation, neglecting interactions among nutrients. Further, most studies have focused on preterm infants, leaving little knowledge on term-born infants who account for > 90% of all newborns.

Moreover, beyong nutrition, it is important to consider the role of infant factors, family factors, and social determinants of health when assessing outcomes. We hypothesize that advanced data science approaches can fill these gaps. To demonstrate the potential for multivariate machine learning and feature selection to quantify the association between lactating mothers’ diet and childhood neurocognition in early school age with 20 mother-child dyads, we performed a pilot analysis using individual mother-infant dyad features at delivery and during lactation from an existing birth cohort.

The feature selection model identified selected features including phosphatidylcholine, glutathione, glycemic load and other nutrients in the maternal diet during lactation, as well as child demographics and family income, as candidate predictors of early-school-age expressive language. Moreover, the strength of association in our pilot model varied across neurocognitive function domains with the greatest association with expressive language function and least with the letter scores, as measured by the Child Development Inventory. Prediction error of the pilot model was lowest for expressive language (3%) and highest for letter score (72%).

This feasibility study provides an impetus for future larger studies of nutrition-brain interactions which may support hypothesis generation for investigation of individualized nutritional recommendations during pregnancy and lactation. This study was supported, in part, by NIH R03 HD107124, Massachusetts Life Science Center Bits to Byte award, and Abbott Nutrition. Boston Children’s Hospital, Boston, MA, USA Rina Bao, Di Wu, Rutvi Vyas, P.

Morton & Yangming Ou Rina Bao, P. Morton & Yangming Ou Johns Hopkins University, Baltimore, MD, USA Matthew J. Lasekan & Brian Leyshon Department of Bioengineering, University of Illinois at Urbana–Champaign, Urbana, IL, USA Correspondence to Sarah U.

Abbott Inc. funded the parent clinical trial (NCT02058225; Phase 1), which collected demographic, nutrition, and brain MRI data. Abbott-affiliated co-authors were involved in Phase 1 of the study only. The early-school-age neurocognitive outcome data (Phase 2) were collected by investigators from Boston Children’s Hospital (S.U.M., P.E.G., Y.O.) outside the scope of the original trial and without Abbott funding.