Team 6: ML Predictive Scoring for Neurological Injury in Pediatric Heart-Lung Bypass (ECMO) Patients

Extracorporeal membrane oxygenation (ECMO) is a form of cardiopulmonary bypass therapy that provides temporary cardiac and pulmonary support to extremely ill patients. Intracranial neurologic injuries occur in approximately 15% of patients treated with ECMO. In this population, expected mortality doubles. In injured survivors, significant neurologic devastation is common.

We know little about how to predict the injuries and how to prevent them. This project aims to identify clinical and lab predictors of neurologic injury, and examine the effect of these injuries on the neuro-developmental outcomes in ECMO survivors. We hypothesize that rapid changes in pre-ECMO to ECMO mean arterial blood pressure and pCO2 are associated with increased risk of intracranial injury. We also hypothesize that hemorrhagic injuries will have a greater neurocognitive morbidity than ischemic injury.

We will have data from all ECMO PICU patients between 2010 and 2017, and have identified patients with post-ECMO imagining, and have subsequently classified the types of neurological injury. Unfortunately as each patient has 1000s of data points between vital signs, laboratory values, near infrared spectroscopy (NIRS), medication/vasoactive changes it makes this project incredibly difficult to approach from a traditional statistics/logistical regression analysis. However our goal is to develop a machine learning prediction algorithm, and cross-validation on the test dataset and potentially use 2018 data as independent validation. As no such data on predictive scoring exists, any strength a machine learning predictive score may have will likely be clinically useful.

Team Lead: Neel Shah, Pediatric Critical Care,