Erich Seamon,Jennifer A Mattera,Sarah A Keim,Esther M Leerkes,Jennifer L Rennels,Andrea J Kayl,Kirsty M Kulhanek,Darcia Narvaez,Sarah M Sanborn,Jennifer B Grandits,Christine Dunkel Schetter,Mary Coussons-Read,Amanda R Tarullo,Sarah J Schoppe-Sullivan,Moriah E Thomason,Julie M Braungart-Rieker,Julie C Lumeng,Shannon N Lenze,Lisa M Christian,Darby E Saxbe,Laura R Stroud,Christina M Rodriguez,Stephanie Anzman-Frasca,Maria A Gartstein

Abstract:Background: Preterm birth (birth at <37 completed weeks gestation) is a significant public heatlh concern worldwide. Important health, and developmental consequences of preterm birth include altered temperament development, with greater dysregulation and distress proneness. Aims: The present study leveraged advanced quantitative techniques, namely machine learning approaches, to discern the contribution of narrowly defined and broadband temperament dimensions to birth status classification (full-term vs. preterm). Along with contributing to the literature addressing temperament of infants born preterm, the present study serves as a methodological demonstration of these innovative statistical techniques. Study design: This study represents a metanalysis conducted with multiple samples (N = 19) including preterm (n = 201) children and (n = 402) born at term, with data combined across investigations to perform classification analyses. Subjects: Participants included infants born preterm and term-born comparison children, either matched on chronological age or age adjusted for prematurity. Outcome measures: Infant Behavior Questionnaire-Revised Very Short Form (IBQ-R VSF) was completed by mothers, with factor and item-level data considered herein. Results and conclusions: Accuracy estimates were generally similar regardless of the comparison groups. Results indicated a slightly higher accuracy and efficiency for IBQR-VSF item-based models vs. factor-level models. Divergent patterns of feature importance (i.e., the extent to which a factor/item contributed to classification) were observed for the two comparison groups (chronological age vs. adjusted age) using factor-level scores; however, itemized models indicated that the two most critical items were associated with effortful control and negative emotionality regardless of comparison group.

Leveraging machine learning to study how temperament scores predict pre-term birth status