44. A Novel Perioperative Mortality Calculator for Newborn Surgery: An American Association for Thoracic Surgery Quality Gateway Congenital Initiative

Objective:
Risk prediction is notoriously difficult in congenital heart surgery due to a high degree of patient variability and anatomic heterogeneity even within a diagnosis. Prior efforts to develop risk prediction models have had limited ability to discriminate between survivors and non-survivors, and therefore there has been limited patient-level implementation. We sought to develop a predictive model with novel machine learning methodology.

Methods:
Deidentified data were collected from 8 pilot institutions as part of the American Association for Thoracic Surgery Quality Gateway. A subset of procedures typically performed in newborns were selected: Arterial switch operation with or without ventricular septal defect (ASO, ASOVSD); Norwood procedure, Hybrid palliation, repair of hypoplastic arch/coarctation with VSD (CoA-VSD), and repair of truncus arteriosus. Data was split 90/10 for training and testing, respectively. Using relaxed least absolute shrinkage and selection operator and Random Forest, prediction models for perioperative mortality were developed; 95% confidence intervals were developed using a bootstrap technique. Overall model discrimination was assessed using area under the receiver operating characteristic curve (AUC) and discrimination slope. An online calculator was developed based on identified predictors and validated on the test set.

Results:
Among 1320 newborn patients who met inclusion criteria, AUC in the test set for biventricular repair patients was 0.85, and discrimination slope was 0.083. Major contributing factors by Gini importance included weight, age, preoperative mechanical ventilation, anatomic anomaly of the head (not including brain), and need for pulmonary arterioplasty. For the subset of newborns undergoing either the Norwood procedure or hybrid single ventricle palliation, the test AUC was 0.84 with a discrimination slope of 0.161. Major contributing factors for patients undergoing single ventricle palliation by Gini importance included weight, age, obstructed pulmonary venous return, preoperative mechanical ventilation, and aortic atresia.

Conclusions:
A simple-to-use, potentially online interface is able to accurately determine the perioperative risk of mortality associated with newborn congenital heart surgery allowing for individual patient/procedure-level risk-prediction.

Peter Chiu (1), Andrew Goldstone (2), Meena Nathan (3), J. William Gaynor (4), Glen Van Arsdell (5), Mark Galantowicz (6), Michael Ma (7), John Cleveland (8), Pirooz Eghtesady (9), Rachel Vanderlaan (10), Emile Bacha (2), Pedro del Nido (1), (1) Boston Children's Hospital, Boston, MA, (2) NewYork- Presbyterian/Columbia University Medical Center, New York, NY, (3) Boston Children's Hospital, Watertown, MA, (4) The Childrens Hospital of Philadelphia, Philadelphia, PA, (5) UCLA Ronald Reagan Medical Center, LOS ANGELES, CA, (6) Nationwide Children's Hospital, Columbus, OH, (7) Stanford University School of Medicine, Stanford, CA, (8) Children's Hospital of Los Angeles, Los Angeles, CA, (9) Wash U, St. Louis, MO, (10) SickKids, Toronto, ON

Peter Chiu

Abstract Presenter

Peter Chiu, MD, MS is a congenital heart surgeon in the Department of Cardiac Surgery at Boston Children's Hospital. Dr. Chiu completed his residency in cardiac surgery at Stanford Hospital and Clinics and subsequently completed a fellowship in congenital heart surgery at Boston Children's Hospital.

James O'Brien

Commentator

Dr. O'Brien is the Co- Director of the Ward Family Heart Center and Chief of Cardiothoracic Surgery at Children's Mercy Kansas City.  He is also the Joseph Boon Gregg/Missouri Chair in Pediatric Cardiac Surgery and the director of the Congenital Heart Surgery Fellowship.    He is a Professor of Surgery the University of Missouri-Kansas City School of Medicine.

Specialties: Congenital