Classifying neonatal spells using real-time temporal analysis of physiological data streams—algorithm development

Abstracts partial pressure of O2 and CO2, respectively. Thus, we were able to decouple the control inputs to the outer patient oriented cascade, where we control physiological target values for arterial O2 saturation and venous CO2. The outmost cascade controls blood flow to operate the oxygenator at an optimal operating point with least flow possible. Results: To validate the designed control structure, we performed a series of 10 animal experiments with 45- to 65-kg domestic pigs. After instrumentation and connection to the extracorporeal circuit, varying degrees of ventilatory insufficiency were simulated by applying a hypoxic gas mixture and using hypoventilation. Under normal operating conditions, the controller was able to reach the desired physiological target concentrations within 20 minutes for CO2 and 3 to 5 minutes for O2. Even with extreme disturbances, hypoxic episodes with SaO2 less than 85 could be prevented. Only in cases where venous wall suction at the inlet cannula prevented further increase of blood flow control targets could not be met. Conclusions: Our experiments support efficacy and performance of the designed control system. Combined with advanced safety functions (not presented here), we show the feasibility of nextgeneration ECLA devices enabling routine application in demanding hospital situations. http://dx.doi.org/10.1016/j.jcrc.2012.10.032

Abstract 17 Classifying neonatal spells using real-time temporal analysis of physiological data streams—algorithm development Edward Pugh a,d,e, Anirudh Thommandram b, Eugene Ng a,e, Carolyn Mcgregor b, Mikael Eklund b, Indra Narang c,e, Jaques Belik d,e, Andrew James d,e a Department of Newborn & Developmental Paediatrics, Sunnybrook Health Sciences Centre, Toronto, Canada b University of Ontario Institute of Technology, Toronto, Canada c Division of Respiratory Medicine, The Hospital for Sick Children, Toronto, Canada d Division of Neonatology, The Hospital for Sick Children, Toronto, Canada e Department of Paediatrics, University of Toronto, Toronto, Canada

Objectives: The term neonatal spell has become a common euphemism in contemporary neonatal intensive care units (NICUs) for cardiorespiratory events that present with variable combinations of cessation of breathing, decrease in blood oxygen saturation and decreased heart rate. There are 5 main types of neonatal spell— central, mixed, and obstructive apnea; isolated bradycardia; and isolated desaturation. Spells may occur in the preterm infant as a consequence of physiological immaturity or as a clinical sign of more serious conditions such as lung collapse, infection, and brain hemorrhage. At present, the need to determine the cause of neonatal spells frequently leads to invasive investigations and interventions, many of which may be unnecessary. The Artemis platform enables real-time, multistream temporal analysis of multipatient, high-fidelity physiological data from bedside monitoring devices to provide meaningful clinical decision support. Our research goal is to develop precise, robust computational algorithms for the Artemis platform to accurately identify and classify spells that occur in the preterm population. Methods: Algorithms representing each type of neonatal spell are being developed based on the classical patterns described in the

e9 medical literature. The Artemis platform enables the capture and analysis in real time of physiological data at the speed generated by the bedside medical device. In our research, we are sampling heart rate, blood oxygen saturation at a rate of 1 Hz together with impedance respiratory wave data at a sampling rate of 62.5 Hz and electrocardiogram data at a sampling rate of 1000 Hz. As part of the algorithm development, we have developed the ability to detect absolute falls in all parameters below set values. We have also developed algorithms to detect fixed percentage fall of parameters from a continuously assessed baseline. We have made a comparison of the output for relative and absolute heart rate and saturation falls for 1 patient over an entire day and compared both to manual data interpretation. Results: The manual comparison shows that there is good concordance between relative falls and clinically significant falls in heart rate and blood oxygen saturation. The absolute method detected 1117 falls in heart rate compared with 484 relative falls in 1 day. There were 92 absolute falls in blood oxygen saturation compared with only 52 relative falls in the same day. Conclusions: The ability to automatically detect clinically significant changes in heart rate and saturation using relative change rather than threshold values will pave the way for a unique system for detecting spells. In addition, an alert system driven by relative falls may help in the future to reduce alarm fatigue in busy NICUs. After this initial development of the detection algorithms, we will perform a comparison study of the Artemis algorithms with the only available current criterion standard of manually interpreted polysomnography in infants breathing without respiratory support. This study will lead to the development of precise, robust algorithms that enable detailed and accurate reports of spells occurring in infants breathing without respiratory support in the NICU. http://dx.doi.org/10.1016/j.jcrc.2012.10.033

Abstract 18 Scale-free metabolic networks in a porcine model of trauma and hemorrhagic shock E.R. Lusczek PhD, D.R. Lexcen PhD, N.E. Witowski PhD, K.E. Mulier MBS, G. Beilman MD Department of Surgery, University of Minnesota, Minneapolis, MN, USA

Objectives: Metabolic rate has been shown to possess scale-free properties under the assumption that oxygen is transported to the organism through the fractal-like vascular system. Cellular metabolic networks have been shown to possess scale-invariant properties via power law behavior of the network's degree distribution with a slope of γ, where − 2 b γ b−3. Changes in oxygen delivery as a result of trauma and hemorrhage induce profound changes in cellular metabolism. We have previously evaluated scale-free metabolic networks of urine and skeletal muscle in an interim data set from our porcine model of trauma and hemorrhagic shock. In muscle metabolic networks, scale invariance is not present at any experimental time point until postresuscitation. We wished to evaluate liver networks in this context and compare them to the muscle networks. We hypothesize that a lack of scale invariance in these networks is related to disruptions in oxygen delivery and use. Methods: Nineteen fasted male Yorkshire pigs were subjected to a standardized protocol consisting of instrumentation and laparotomy,