(707) Analysis of Extubation Readiness Using Physiologic Time-Series Data in a CTICU

Introduction: The American Association for Thoracic Surgery endorses a class 1 level A recommendation for early extubation after congenital heart surgery. Currently, pediatric extubation readiness guidelines are inconsistent. Premature extubation can lead to extubation failure, increasing morbidity and mortality. This study evaluated the ability of a time-series data streaming platform to predict extubation readiness trial (ERT) outcomes.

Methods: This is a retrospective single center analysis of ERT outcomes at a tertiary hospital from 3/2022-3/2023. Intubated patients undergoing ERTs in a pediatric CTICU were eligible for inclusion. A time-series platform (SickbayTM) was compared to conventional ERT, applying predetermined hemodynamic variations that determined ERT outcome. Conventional ERT assessed variations in respiratory rate, oxygen saturation, near-infrared spectroscopy (NIRS), heart rate, and blood pressure by 20% and ETCO2 change by ≥ 10 torr compared to baseline. Conventional ERT was measured as the percent change between baseline and ERT datapoints. The time-series baseline and ERT was the median of 3500 datapoints/hour. Conventional and time-series ERT outcomes were compared using these hemodynamic variations to evaluate for discrepancies in ERT outcome predictions.

Results: 322 ERTs met inclusion criteria. Of the 211 conventional ERT failures, 69.7% time-series ERTs passed and 30.3% failed. Of the 111 conventional ERT passes, 55.9% time-series ERTs passed and 44.1% failed. In a multivariate non-parametric analysis, there was a significant variation in respiratory rate (p= < 0.001) when time-series ERTs were compared to failed conventional ERTs. Otherwise, there were no statistically significant hemodynamic changes when time-series ERTs were compared to conventional ERTs (p=>0.05). Of the 39 patients extubated within 12 hours of passed conventional ERT, 21 (54%) passed and 18 (46%) failed per time-series ERT. Of the 6 re-intubated patients, 2 failed and 4 passed time-series ERT.

Conclusions: The study concludes that a time-series data streaming platform can be applied to predict ERT outcomes and compares favorably with conventional ERT. This platform may detect subtle hemodynamic variations. Future studies will utilize machine learning to build a predictive algorithm that predicts ERT outcomes.