(19) Central Apnea Detection in the Neonatal Intensive Care Unit

Elizabeth C. Smet, University of Iowa Stead Family Children's Hospital, Iowa City, IA, United States; Danielle R Rios, University of Iowa Stead Family Children's Hospital, United States; Regan E. Giesinger, University of Iowa, Iowa City, IA, United States; Jamie L. S. Waugh, Medical Informatics Corp., United States

Background: Apnea spells are characterized by cessation of breathing accompanied by bradycardia ( < 100 beats per minute), cyanosis, or pallor. The incidence of apnea episodes is inversely proportional to gestational age. Currently, heart rate, respiratory rate, oxygen saturation, and chest impedance are monitored and used to assess for apnea; many factors can affect monitoring and documentation of apnea spells. The Sickbay platform allows continuous monitoring of vital signs and waveforms as well as developing and running algorithms both retrospectively and prospectively.

Objectives: To develop an effective algorithm that accurately detects episodes of apnea of prematurity.

Design/Methods: The proposed algorithm identifies episodes of apnea through patterns in cardiorespiratory data. Major (bradycardia or saturation < 75%) and minor apnea events (>20 seconds), not clear events, and non-apnea events ( < 20 seconds) were labeled by a neonatologist. The algorithm was tested on three cohorts: preterm (study group), term (negative control group), and vecuronium (positive control group). Exclusion criteria included congenital heart disease and congenital central sleep apnea.

Results: A total of 199 patients with an average gestational age of 31.9 ± 6.7 weeks and an average birth weight of 2010 ± 1320 grams were included in the study, of which 88 were in the preterm cohort, 83 in the term cohort, and 28 in the vecuronium cohort. The first two detected apnea events per patient were labeled by the research team: 130 in the preterm cohort, 18 in the term cohort, and 8 in the vecuronium cohort. The preterm cohort was more likely to have apnea events (p < 0.001) and be receiving caffeine (p < 0.001) at the time of apnea events than the term or vecuronium cohorts. The algorithm correctly identified 100% of major/minor apnea events while only 41% of major and minor apnea events were documented by nursing staff. Nursing documentation of apnea events were more likely to be present in the preterm cohort. Lastly, 21% of detected apnea events were labeled as not apnea but of these events, 81% were found to have a respiratory rate < 30 breaths per minute and a cessation of breathing < 20 seconds.

Conclusion: The proposed algorithm accurately identifies apnea events while simultaneously ignoring events deemed unrelated. The majority of detected non-apnea events met all criteria except length of apnea event >20 seconds. Future directions of this project include further investigation into cases where the algorithm does not perform as expected as well as a prospective evaluation of the algorithm.