评估一种新型的无创连续脉压变异率算法(Evaluation of.pdf

Evaluation of a novel automated non-invasive pulse pressurevariation algorithmMartin Schmida,Helga Prettenthalerb,Christian Wegerc,Karl-Heinz Smolleb,naInstitute for Medical Engineering,University of Technology Graz,AustriabDepartment of Internal Medicine and Intensive Care,University Hospital Graz,AustriacDepartment of Orthopaedic Surgery,University Hospital Graz,Austriaa r t i c l e i n f oArticle history:Received 13 March 2013Accepted 17 July 2013Keywords:Pulse pressure variationFluid managementContinuous non-invasive arterial bloodpressure monitoringMechanical ventilationVascular unloadingCNAPsMonitor 500a b s t r a c tIn mechanically ventilated patients,Pulse Pressure Variation(PPV)has been shown to be a usefulparameter to guide fluid management.We evaluated a real-time automated PPV-algorithm by comparingit to manually calculated PPV-values.In 10 critically ill patients,blood pressure was measured invasively(IBP)and non-invasively(CNAPsMonitor,CNSystems Medizintechnik,Austria).PPV was determinedmanually and compared to automated PPV values:PPVmanIBPvs.PPVautoIBPwas?0.1971.65%(meanbias7standard deviation),PPVmanCNAPvs.PPVautoCNAPwas?1.0272.03%and PPVautoCNAPvs.PPVmanIBPwas?2.1073.14%,suggesting that the automated CNAPsPPV-algorithm works well on both bloodpressure waveforms but needs further clinical evaluation.&2013 Elsevier Ltd.All rights reserved.1.IntroductionSeveral outcome-related studies have demonstrated that goal-directed fluid administration(where the amount of fluid givenduring surgery and on intensive care units is sought to beoptimized based on an objectively quantifiable variable as opposedto supplying fluid on a general basis)can significantly improve theoutcome for the patient 13.Recently,the approach of functionalhemodynamic monitoring in sedated patients receiving mechan-ical ventilation has become the preferred option to predict fluidresponsiveness(i.e.,whether or not the patient reacts withsignificantly increased cardiac output to fluid administration,thusindicating fluid depletion)4.Dynamic indicators have beendemonstrated to be better predictors of fluid responsiveness thanstatic parameters 57.Among these dynamic indices,the varia-tion of pulse pressure(PPV),i.e.the variation of the differencebetween systolic and diastolic blood pressure,has been shown tobe more reliable than other dynamic parameters 8 and is alreadyused for clinical fluid management 9.PPV can be calculated based on blood pressure waveformsdetected with an intra-arterial catheter or based on non-invasiveblood pressure signals(e.g.CNAPsMonitor).Manual off-linecalculation of PPV is considered the“gold standard”in medicalliterature10.Ingeneral,continuousbloodpressureandalso airway pressure signals are required for this calculation.Toeliminate the need for simultaneously acquiring airway pressurefrom the ventilator,elaborate algorithms have been designed toautomatically and continuously estimate PPV from the bloodpressure signal alone 11.Most of them are based on invasiveblood pressure waveforms and show high accuracy 12,13.Reliable PPV-values can also be derived manually from non-invasive blood pressure waveforms 14 which may be usefulespecially in patients without the indication for an arterialcatheter.The CNAPsnon-invasive blood pressure monitor hasrecently been validated in patients undergoing general anesthesiafor abdominal,gynecological,vascular and neurosurgical proce-dures 15,16.The results show good accuracy when comparingbeat-to-beat blood pressure measurements to their invasive coun-terparts.The ability of manually calculated PPV based on bloodpressure waveforms obtained with the CNAPsdevice to predictfluid responsiveness has already been evaluated during vascularsurgery 17 and in critically ill patients 18.In their article,Biaiset al.17 derived PPV manually from the CNAPsblood pressurewaveform and compared it to PPV derived from invasive measure-ments of an ipsilateral radial catheter,while Monnet et al.18compared manually calculated PPV of the CNAPswaveform to PPVderived from invasive measurements of a femoral catheter.Theirresults suggest that the amplitude of the respiratory-inducedvariations in the pulse pressure in the finger can predict fluidContents lists available at ScienceDirectjournal homepage: in Biology and Medicine0010-4825/$-see front matter&2013 Elsevier Ltd.All rights reserved.http:/dx.doi.org/10.1016/pbiomed.2013.07.020nCorresponding author.Tel.:+43 316 385 81414.E-mail addresses:karlheinz.smolleklinikum-graz.at,karl-heinz.smolleklinikum-graz.at(K.-H.Smolle).Computers in Biology and Medicine 43(2013)15831589responsiveness with a similar sensitivity and specificity as PPVderived from invasive readings,both for surgical and intensivecare patients.In this work,we evaluate the automated CNAPsPPV algorithmitself which is used for displaying PPV-values automatically andcontinuously on the CNAPsdevice.The goal of this study was thecomparison of manually calculated PPV values to PPV valuesderived automatically via the PPV algorithm of the CNAPsMonitorusing both invasive as well as non-invasive blood pressure wave-forms as a basis.2.MethodsThe study protocol was approved by the ethics commission forhuman subjects(University Hospital Graz,Austria).All patients ortheir relatives were informed about the study when the patientwas included and could refuse the patients participation at anytime.We studied 10 patients on the medical intensive care unitwho were all sedated,under vasopressor therapy(norepinephrine)and had sinus rhythm.Patients with obvious edema on the upperextremeties,especially the fingers,were not included in the study.Invasive arterial blood pressure(IBP)was monitored via a radialartery catheter(20G,Arterial Cannula,BD Critical Care SystemsLtd.,Singapore).Damping coefficient and natural frequency of thehydrostatic transducer system was tested using the fast flush test19.All patients were mechanically ventilated with an Evita XL(Drger,Germany)in the Biphasic Positive Airway Pressure(BiPAP)mode.Blood pressure signals only of patients without signs ofspontaneous breathing were examined since spontaneous respira-tion has been shown to be ineffective in producing reliablechanges in the arterial waveform to guide fluid management 20.The CNAPssystem(CNAPsMonitor 500,CNSystems Medizin-technik AG,Graz,Austria)consists of a double finger cuff,a pressuretransducer mounted on the forearm and an upper-arm bloodpressure cuff for calibration.The principle of CNAPs,the“volumeclamp method”(or“vascular unloading technique”)was originallydeveloped by Pez 21 in the early 1970s and further improved byFortin et al.22.A finger cuff encompassing two neighboringfingers(see Fig.1)is used for continuous non-invasive bloodpressure monitoring,one finger at a time with automatic switchesbetween fingers every 560 min(set to 30 min for this study asrecommended by the manufacturer).An upper-arm blood pressurecuff derives the measurement of oscillometric blood pressure andserves for calibration of the device every 560 min(set to 15 min forthis study as recommended by the manufacturer).The CNAPsfinger cuff was placed contralaterally to theinvasive catheter.The CNAPsupper-arm cuff was applied to thesame arm as the invasive catheter to eliminate possible pressuredifferences of the arms.While such pressure differences areunimportant when studying only PPV,they might affect compar-ing BP levels between IBP and CNAPsdirectly.IBP and CNAPstransducers were placed approximately at the level of the heart.The CNAPsMonitor was connected to the patient monitor andzero-levelled as recommended by the manufacturer.CNAPsand IBP blood pressure waveforms were synchronouslydisplayed on the bedside patient monitor(Infinity Delta,Drger,Germany)and recorded using data acquisition software(DrgerDataGrabber)which allows the export of CNAPsand IBP systolic,diastolic and mean blood pressure values and the blood pressurewaveforms with a sampling rate of 100 Hz.All off-line dataanalyses were performed using MATLAB-based scripts(MatlabR2008b,The Math Works Inc.,Natick,MA,USA).So that all patients contribute equally to the results,all compar-ison analyses were based on the same number of PPV valuesper patient.For the manually calculated PPV values,the followingprocedure was performed:Using a random number generator,10 time periods of the waveform data were analyzed for each ofthe 10 patients.At randomly generated points of time,the two bloodpressure waveforms were visually checked for artifacts at a length of38 heart beats.This number of heart beats was chosen because itis longer than 3 typical respiratory cycles.If both the invasive as wellas the non-invasive waveforms were visually considered artifact-free,the following procedure was performed:The“gold standard”reference PPV values were calculated by manually selecting threeconsecutive pulse pressure minima and maxima per mouse-click(see Fig.2)and using the standard formula as described in theliterature 10:PPV(PPmax?PPmin)/(PPmax+PPmin)/2?100%.ThePPVmanvalue was defined as the average PPV over these threeconsecutive respiratory cycles.PPVmanwas thus calculated retro-spectively based on the IBP and CNAPsblood pressure waveforms.For every patient,this procedure was repeated until 10 value pairs(PPVmanIBPand PPVmanCNAP)were obtained over the same timeperiods(marked with their end time points T1?T10).These PPVmanvalues were compared to PPVautovalues whichwould have been displayed on the CNAPsMonitor at time pointsT1?T10.These PPVautovalues were retrospectively obtained usinga software extract provided by the manufacturer.Basically,theproprietary PPV algorithm integrated into the CNAPsmonitorautomatically searches for the typical swing patterns in the bloodpressure waveform modulated by mechanical respiration.PPVvalues are computed by employing automated detection of pulsepressure minima and maxima and using the standard PPV formula(see above)before being smoothed using an average over 3consecutiverespiratoryswings.Additionally,anupdatefilterwith an adaptive coefficient corrects for abnormally strong physio-logical changes in PPV and is applied before displaying the finalPPV value on the CNAPsmonitor.The original algorithm part forFig.1.CNAPsmonitor showing the double finger cuff,the pressure transducermounted on the forearm and the upper-arm blood pressure cuff.M.Schmid et al./Computers in Biology and Medicine 43(2013)158315891584PPV calculation was provided by the manufacturer and integratedas a subroutine into Matlab.This subroutine was called first with theCNAPsblood pressure waveform and returned all values PPVau-toCNAP for the complete duration of measurement,tagged by timestamps;then the same procedure was performed with the IBPwaveform to obtain all values of PPVautoIBP.From these PPVautovalues,those at the time points T1?T10were extracted for compar-ison with PPVmanIBPand PPVmanCNAP.Thus,a total of 100 dataquadruplets were included into the analysis.The periods on which the manual and the automated PPV-values were derived were not completely identical:the manuallycalculated PPV values were based on the number of heart beatsrequired to cover three consecutive respiratory cycles(differingover subjects and depending on their current heart rate)whereasthe automated PPV algorithm also includes an update filter whichincorporates information about the past and smoothes PPV valuesdisplayed on the patient monitor.Analysis was divided into two parts:(i)to assess the precisionof the PPV algorithm incorporated into the CNAPsdevice itself,PPV values calculated manually and automatically on the samewaveformwerecompared(i.e.,PPVmanCNAPvs.PPVautoCNAPand PPVmanIBPvs.PPVautoIBP);(ii)to assess the reliability of theautomated CNAPs-PPV for clinical use,PPVautoCNAPwas comparedto the gold standard PPVmanIBP.StatisticalanalysiswasperformedusingMicrosoftOfficeExcel 2007(Microsoft Corporation,Redmond,WA)and SPSS forWindows(SPSS 17.0,Chicago,IL).Data are presented as mean7-standard deviations.Agreement between PPV values was assessedby the calculation of bias,standard deviation of the bias,regres-sion plot and Bland-Altman analysis(accounting for repeatedmeasurements per patient)23.3.ResultsThe characteristics of the 10 patients are reported in Table 1.CNAPsand IBP blood pressure waveforms were recorded for atleast 1 h per patient(minimum duration 1:47,maximum duration3:08,median duration 2:44).Table 2 summarizes the values ofsystolic,diastolic and mean blood pressure,pulse pressure andPPV values for both blood pressure waveforms.3.1.Precision of the PPV algorithm incorporated into the CNAPsdeviceFor both analyses of difference between manual and automatedPPV values(i.e.automated vs.manual PPV,first based on the IBPsignal and then based on the CNAPssignal),the bias of thedifference(automated minus manual PPV)and its standard devia-tion as well as the 95%limits of agreement are listed in Table 3.Both analyses(i.e.,PPVmanIBPvs.PPVautoIBPas well as PPVmanCNAPvs.PPVautoCNAP)showed significant correlation(R0.889 andFig.2.The upper graph shows the non-invasive blood pressure signal from the CNAPsmonitor(left scale)together with the resulting pulse pressure values for each beat(thick blue line with dots,right scale).The lower graph shows the same for the invasive blood pressure signal(IBP).Minima and maxima of the pulse pressure signal selectedmanually by the user are highlighted for both IBP and CNAPssignals.(For interpretation of the references to color in this figure legend,the reader is referred to the webversion of this article.)Table 1Main patients characteristics.Values are median and inter-quartile range(2575%)or numbers.MedianInterquartile Range(2575%)Age(y)4235.051.8Gender(male/female)6/4Weight(kg)7970.587.3Body mass index(kg/m)25.123.429.0Temperature(1C)37.837.338.0Heart rate(beats/min)89.975.7108.1Respiratory rate(cycles/min)151418Tidal volume(ml/kg)7.67.08.0PEEP(cm H2O)109.311.6Artenerol dose(mg/kg/h)4.251.547.70Diagnosis(n)Intoxication4Sepsis1Meningitis1Pneumonia3Myocardial infarction1Table 2Summary of blood pressure(BP),pulse pressure(PP)and PPV values for both bloodpressure waveforms(values are mean7standard deviation).CNAPIBPSystolic BP mmHg121.0719.2127.0718.9Diastolic BP mmHg73.2710.866.2712.7Mean BP mmHg89.4713.785.1715.2PP mmHg47.8714.560.8712.5PPVman%5.873.46.973.6PPVauto%4.872.26.773.2M.Schmid et al./Computers in Biology and Medicine 43(2013)158315891585R0.822,respectively;Po0.001,n100).Fig.3displays(a)regression plots showing regression line and correlationcoefficient and(b)BlandAltman plots showing the bias as wellas 95%limits of agreement(1.96?standard deviation)as horizon-tal dashed and dotted lines,respectively:left for PPVmanCNAPvs.PPVautoCNAPand right for PPVmanIBPvs.PPVautoIB.The underlying data per patient are summarized in Table 4,listing the median as well as the range of difference for all 10patients.3.2.Reliability of the automated CNAPsPPV for clinical usePPVmanIBPandPPVautoCNAPweresignificant