Predicting Failure of Non-Invasive Ventilation in Children with a Risk Stratification Model
Monday, July 29, 2024
11:00 AM-1:00 PM
BIOMED PhD Thesis Defense
Title:
Predicting Failure of Non-Invasive Ventilation in Children with a Risk Stratification Model
Speaker:
Natalie Napolitano, PhD Candidate
School of Biomedical Engineering, Science and Health Systems
Drexel University
Advisor:
Amy Throckmorton, PhD
Professor
School of Biomedical Engineering, Science and Health Systems
Drexel University
Details:
Respiratory distress and the need for assistance with breathing is one of the most frequent reasons a child is admitted into the pediatric intensive care unit (PICU). Mechanical assistance with breathing is performed non-invasively or invasively and the determination of which level of support is required and when to transition from one level to another is unknown. Although non-invasive ventilation (NIV) has been shown to improve outcomes of care and shorten time in the hospital when successful, the failure of NIV has been reported to cause an increase time on advanced respiratory support, time in the intensive care unit, and time in the hospital. Therefore, it is theorized that if we can predict which patients will not be successfully treated with NIV and at what time-point we should make this decision, we can improve the long-term outcomes of patients and support their faster recovery. The historical definition of NIV failure is flawed and not in line with the traditional framework of therapy failure determination. There is little evidence to support how best to manage NIV, including best approach for success and the appropriate timing of transition to a higher level of care.
Motivated by the need strive for more evidence-based approach to the delivery of respiratory support in the PICU. here we redefined NIV failure in children using the incidence of unfavorable long-term outcomes, provide a landscape of NIV use in a large, busy, tertiary care, referral children’s hospital, and developed a failure prediction model to assist with determination of when to change from NIV to invasive ventilation (IV) to optimize favorable long-term outcome. A new compound outcome definition of NIV failure was designed, and a patient cohort was defined and separated into success and failure groups for comparison. Statistical comparisons were made between groups utilizing Wilcoxon Rank Sum test for non-normally distributed continuous variables and Chi-squared test for categorial variables with our large sample size. A p-value of 0.5 indicates a significant difference between groups.
A retrospective observational study was performed with patient data from the PICU at the Children’s Hospital of Philadelphia from January 1, 2015 – December 31, 2019. The first case of NIV before tracheal intubation (TI) during the first PICU stay was used for data analysis. Complex NIV failure was defined by the incidence of any of the following criteria occurring during the PICU admission: (1) PICU mortality, (2) new tracheostomy, (3) IV for 7 or more days, (4) relevant tracheal intubation associated events during TI (cardiac arrest, hypotension requiring intervention, dysrhythmia, emesis with aspiration, or pneumothorax/ pneumomediastinum), or (5) severe desaturation during TI (SpO2 greater than 90% after preoxygenation and lowest SpO2 during intubation of less than 70%). During the study period, 3,273 patients had 3,844 courses of NIV, during 3,967 PICU admissions that met criteria for inclusion. Of these, 231 (6%) courses met criteria for failure. These cases had a significantly higher rate of TI (72% vs 0.6%, P=<0.001), demonstrated longer time on respiratory support (342.35 days vs 26.68 days, P<0.001), and more time in the PICU (23.31 days vs 2.43 days, P<0.001).
Per these data sets and detailed analysis, nine physiologic metrics or features were identified as important to predicting the worsening of the clinical condition and thus to be important to predicting failure of NIV. These features were isolated for each hour for the first 24-hours of therapy and used in a logistic regression prediction model. This failure prediction model benchmarked and accurately predicted complex NIV failure at 6-hours of therapy (AUPRC = 0.32 and AUROC = 0.803).
Known gaps in the clinical management of NIV failure of definition, physiologic metrics, and timing of NIV failure have been addressed in the development of (1) a new prediction model with metrics of NIV therapy failure and (2) a new clinical tool that will be used to better inform clinical teams who are treating children with respiratory distress and who require the proper data and trends for clinical decision making.
Contact Information
Natalia Broz
njb33@drexel.edu