e8 Abstracts

Conclusions: The proposed response-surface methodology signif-icantly increases computational efficiency over direct numericalintegration, whereas at the same time reducing the number ofvariables to be varied, resulting in more realistic inputs to apopulation-level model of IAV.

http://dx.doi.org/10.1016/j.jcrc.2012.10.030

Abstract 15Development of a low-frequency autoregulation index for calculation ofoptimal CPP in severe traumatic brain injuryFabian Guiza a, Bart Depreitere b, Martin Schuhmann c,

Greet Van Den Berghe a, Geert Meyfroidt a

aDivision of Intensive Care Medicine, University Hospitals Leuven, BelgiumbDivision of Neurosurgery, University Hospitals Leuven, BelgiumcDivision of Neurosurgery, University Hospital Tübingen, Germany

Objectives: Traumatic brain injury (TBI) is an important problemworldwide and a major cause of permanent disability and death inyoung patients. Currently, there is controversy regarding theoptimal cerebral perfusion pressure (CPP) thresholds required inthe management of these patients. There is evidence that thecritical optimal CPP threshold depends on the patient's capacityfor pressure autoregulation, which is an inherent capacity tomaintain cerebral blood flow constant over a broad range of CPPs.In severe TBI, this capacity is frequently and dynamicallydisturbed. A possible tool for monitoring autoregulation is thepressure reactivity index (PRx) defined as a moving correlationcoefficient between the mean arterial blood pressure andintracranial pressure based on signal capture at a frequency of atleast 60 Hz. This need for high-frequency data has constrained theuse of PRx to a few academic centers. A hint of the potential ofautoregulation-driven management of the individual TBI patient isthe recently described association between outcome and thecontinuous calculation of an optimal CPP based on 4 hours ofPRx. Here we present the novel “Leuven autoregulation index”(LAx), developed through machine learning techniques and basedon correlations between routinely available intracranial pressureand mean arterial blood pressure signals at a standard minute-by-minute time resolution. Our main objective is to compare LAx andPRx in their role to determine continuous optimal CPPs forindividual patients.Methods: Nineteen patients with TBI admitted to the UniversityHospital of Tubingen, Germany, with length of stays ranging from7 hours to 19 days, were continuously monitored at the bedsideusing ICM+ software (Cambridge Enterprise, University ofCambridge) allowing for continuous PRx calculation. Autoregula-tion index vs CPP plots for PRx and LAx was computed todetermine an optimal CPP every minute. The 2 indices werecompared with respect to the optimal CPP values and thepercentage of time for which they could be computed.Results: The PRx produced on average an optimal CPP for 42%of the patient's length of stay. Two instantiations of the LAxcomputed on moving windows of either 4 or 8 hours wereconsidered. The 4- and 8-hour LAx produced on average optimalCPPs on 57% and 64% of the stays, respectively. The optimalCPPs of the 4-hour LAx were more similar to those of the PRxwith an average root mean squared error (rmse) of 7.6 mm Hg,whereas the 8-hour LAx had an average rmse of 10 mm Hg.

Conclusions: This study shows, first, a small difference in continuousoptimal CPPs derived from PRx and LAx, noting that differences ofapproximately 5 mm Hg in CPP are not significant from a clinicalperspective. Second, the LAx appears to be computable for longerperiods of the patient's length of stay. Further studies are required toverify the associations between outcome and LAx as well as todetermine which version of the LAx should be used to guide CPPtreatment. If, however, the LAx is thus validated, it will beimmediately applicable in all centers treating TBI without additionalinvestments in software or invasive monitoring.

http://dx.doi.org/10.1016/j.jcrc.2012.10.031

Abstract 16Patient oriented closed loop control of extracorporeal lung assistMarian Walter a, Christian Brendle a, Andre Stollenwerk b, Rüdger Kopp c,

Jutta Arens d, Ralf Bensberg c, Steffen Leonhardt a

aRWTH Aachen University, Aachen, GermanybEmbedded Software Laboratory, RWTH Aachen University, Aachen,

GermanycClinic of Operative Intensive Care, University Hospital Aachen, Aachen,

GermanydDepartment of Cardiovascular Engineering, Institute of Applied Medical

Engineering, Helmholtz Institute, RWTH Aachen University, Aachen,

Germany

Objectives: In cases of severe lung failure, long-term extracorpo-real membrane oxygenation has become a relevant treatmentoption in the intensive care unit (ICU). With the availability ofsmall and long-term stable oxygenation circuits and dedicateddrive units, it is now possible to apply extracorporal lung assist(ECLA) in the ICU on a routine basis. However, compared withthe operating theater, where supervision and optimal setting of anextracorporeal circulation require permanent attention of aspecially trained cardiotechnician/perfusionist, the situation inthe ICU is less satisfactory. Machine operation is limited to lessfrequent care intervals or reaction to machine alarms. To providesimilar levels of control performance, reliability, and safety, ICUdevices should provide advanced autonomous functions. In ourinterdisciplinary project “smart ECLA,” we developed patient-oriented control strategies to address this issue. Other projectpartners focused on dedicated safety mechanisms.Methods: Our proposed ECLA system consists of 2 femoral venouscannulae for venous blood collection, a centrifugal blood pump(Medos DP3), an oxygenator (MedosHilite 7000), and a jugularvein cannula for blood reinfusion. As a central processing andcontrol prototyping unit, we use a DSpaceMicroAutoBox system.Actuators are a self-designed electronic gas blender and a self-designed pump control unit. Sensor components are pressuresensors (at pump inlet, preoxygenator, and return cannula), onlineblood gas measuring system (Terumo CDI500, preoxygenator andpostoxygenator), a blood flow meter in the external circuit(Transsonic), and a fully equipped patient monitor (A/S3 DatexOhmeda including ventilator gas monitoring and continuouscardiac output monitor). Each actuator and sensor component isconnected to the CAN-based data bus via microcontroller-basednetwork nodes, which implement data protocol translation. As acontrol concept, we designed a dedicated cascaded controlstructure. The inner cascade controls oxygenator output blood gas

e9Abstracts

partial pressure of O2 and CO2, respectively. Thus, we were able todecouple 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 bloodflow to operate the oxygenator at an optimal operating point withleast flow possible.Results: To validate the designed control structure, we performed aseries 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 byapplying a hypoxic gas mixture and using hypoventilation. Undernormal operating conditions, the controller was able to reach thedesired physiological target concentrations within 20 minutes forCO2 and 3 to 5 minutes for O2. Even with extreme disturbances,hypoxic episodes with SaO2 less than 85 could be prevented. Onlyin cases where venous wall suction at the inlet cannula preventedfurther increase of blood flow control targets could not be met.Conclusions: Our experiments support efficacy and performance ofthe designed control system. Combined with advanced safetyfunctions (not presented here), we show the feasibility of next-generation ECLA devices enabling routine application in demand-ing hospital situations.

http://dx.doi.org/10.1016/j.jcrc.2012.10.032

Abstract 17Classifying neonatal spells using real-time temporal analysis ofphysiological data streams—algorithm developmentEdward 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

aDepartment of Newborn & Developmental Paediatrics, Sunnybrook Health

Sciences Centre, Toronto, CanadabUniversity of Ontario Institute of Technology, Toronto, CanadacDivision of Respiratory Medicine, The Hospital for Sick Children, Toronto,

CanadadDivision of Neonatology, The Hospital for Sick Children, Toronto, CanadaeDepartment of Paediatrics, University of Toronto, Toronto, Canada

Objectives: The term neonatal spell has become a commoneuphemism in contemporary neonatal intensive care units (NICUs)for cardiorespiratory events that present with variable combinationsof cessation of breathing, decrease in blood oxygen saturation anddecreased heart rate. There are 5 main types of neonatal spell—central, mixed, and obstructive apnea; isolated bradycardia; andisolated desaturation. Spells may occur in the preterm infant as aconsequence of physiological immaturity or as a clinical sign ofmore serious conditions such as lung collapse, infection, and brainhemorrhage. At present, the need to determine the cause of neonatalspells frequently leads to invasive investigations and interventions,many of which may be unnecessary.The Artemis platform enables real-time, multistream temporalanalysis of multipatient, high-fidelity physiological data frombedside monitoring devices to provide meaningful clinical decisionsupport. Our research goal is to develop precise, robust computa-tional algorithms for the Artemis platform to accurately identify andclassify spells that occur in the preterm population.Methods: Algorithms representing each type of neonatal spell arebeing developed based on the classical patterns described in the

medical literature. The Artemis platform enables the capture andanalysis in real time of physiological data at the speed generatedby the bedside medical device. In our research, we are samplingheart rate, blood oxygen saturation at a rate of 1 Hz together withimpedance respiratory wave data at a sampling rate of 62.5 Hzand electrocardiogram data at a sampling rate of 1000 Hz. As partof the algorithm development, we have developed the ability todetect absolute falls in all parameters below set values. We havealso developed algorithms to detect fixed percentage fall ofparameters from a continuously assessed baseline. We have madea comparison of the output for relative and absolute heart rate andsaturation falls for 1 patient over an entire day and compared bothto manual data interpretation.Results: The manual comparison shows that there is goodconcordance between relative falls and clinically significant fallsin heart rate and blood oxygen saturation. The absolute methoddetected 1117 falls in heart rate compared with 484 relative falls in 1day. There were 92 absolute falls in blood oxygen saturationcompared with only 52 relative falls in the same day.Conclusions: The ability to automatically detect clinically signif-icant changes in heart rate and saturation using relative changerather than threshold values will pave the way for a unique systemfor detecting spells. In addition, an alert system driven by relativefalls may help in the future to reduce alarm fatigue in busy NICUs.After this initial development of the detection algorithms, we willperform a comparison study of the Artemis algorithms with the onlyavailable current criterion standard of manually interpreted poly-somnography in infants breathing without respiratory support. Thisstudy will lead to the development of precise, robust algorithms thatenable detailed and accurate reports of spells occurring in infantsbreathing without respiratory support in the NICU.

http://dx.doi.org/10.1016/j.jcrc.2012.10.033

Abstract 18Scale-free metabolic networks in a porcine model of trauma andhemorrhagic shockE.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-freeproperties under the assumption that oxygen is transported to theorganism through the fractal-like vascular system. Cellularmetabolic networks have been shown to possess scale-invariantproperties via power law behavior of the network's degreedistribution with a slope of γ, where −2 b γ b−3. Changes inoxygen delivery as a result of trauma and hemorrhage induceprofound changes in cellular metabolism. We have previouslyevaluated scale-free metabolic networks of urine and skeletalmuscle in an interim data set from our porcine model of trauma andhemorrhagic shock. In muscle metabolic networks, scale invarianceis not present at any experimental time point until postresuscitation.We wished to evaluate liver networks in this context and comparethem to the muscle networks. We hypothesize that a lack of scaleinvariance in these networks is related to disruptions in oxygendelivery and use.Methods: Nineteen fasted male Yorkshire pigs were subjected to astandardized protocol consisting of instrumentation and laparotomy,