This project examines neurodevelopmental outcomes following cardiac surgery in infants. The data set was compiled from several long-term studies which measured pre-existing and in surgery variables, then later followed up on cognitive and motor development in these children using the Bayley scale. Previous studies with linear regression offered insight on meaningful factors that impacted cognitive outcomes.

Our goal was to examine if alternative methods, such as regression and classification trees, could offer additional insights into these neurodevelopmental impacts and outcomes.

First classification models were used to try and predict if a child would score above or below a numerical cut off mark in the separate categories. Then regression trees were used to model several clusters of scores, and what a child was likely to score. The models, complied in R, were validated through statistical computing. In later years in these studies the scoring system switched from Bayley-II to Bayley-III, for which we adjusted, as scores for all children changed slightly under the new test, this provided an additional testing method for accuracy. The models were able to produce accuracy levels around 80% for classification and a similar standard deviation as the Bayley score system. When building only on Bayley-II’s and testing on adjusted Bayley-II’s results were even stronger.

These methods created strong predication models for future outcomes following infant cardiac surgery. These results offer a strong accuracy and are more interpretable to parents and healthcare providers, providing a basis for planning for a child’s future developmental needs immediately following the initial surgery.

Research Institution: Harvard T Chan School of Public Health

Lead Author: David Wypij, PhD

Co-author(s): Carissa Villanueva, Lara Malayeff, Alexa González, Ian Reyes

Co-authors’ Affiliations: Harvard University, University of Puerto Rico at Rio Peidras, University of Texas at El Paso

The group would like to thank their faculty sponsor Brian Clair for their support of this project.